Merge remote-tracking branch 'origin/main' into overtrue/fix-heal-lock-ci

* perf(ecstore): fuse multipart commit, cache active upload heads, and scale fsync workers (#9077)

- Cache active multipart upload heads, committed part records, and per-part
  transaction state in ACTIVE_MULTIPART_SOURCE_HEADS with invalidation on
  mutation, retirement, abort, and delete paths.
- Fuse destination metadata preparation into MultipartFusedCommitOutcome,
  coalesce prepared metadata writes before sync_data(), and gate direct
  local part transaction settlement on disk.is_local().
- Retain quorum commit receipt publication before physical delete cleanup
  using PREPARED_DELETE_RECEIPTS, batch TARGET and PREPARED syncs, and skip
  redundant pre-reservation directory syncs for inline deletes.
- Scale DiskBlockingLane::Fsync to 8 workers per disk with idle_workers
  accounting and bounded per-disk batch fairness so concurrent fsyncs
  coalesce in the filesystem journal.
- Cache verified small stripe proof records in VERIFIED_STRIPE_RECORDS while
  recomputing payload digests on every read.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: d85aaf47-2a4b-4c2c-97a3-dbe8c7717e2a

* fix(ecstore): align test/prod commit and cache paths and hold source lock across cache publication

* fix(ecstore): preserve fused commit cancellation/panic semantics and retire excess idle fsync workers

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: d85aaf47-2a4b-4c2c-97a3-dbe8c7717e2a

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: d85aaf47-2a4b-4c2c-97a3-dbe8c7717e2a

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=619402ae972a1b97d8684bb3c264091cb77f73d9fbdd59b36b1e002c28765f6b
-sha256-linux=4359905a64855386a5403800c678cbd43dd2a5fda4777b8651aa213fcc66e585
+sha256-darwin=7c705942d3a9e583ebaa85845daa21c4f81fdbf36def24db931aaef38636c53e
+sha256-linux=5c33dc395233671cb77452f8fb12900ecdbeb368632222b96835d310bce1aab8

.config/e2e-repl-nightly-selection.txt

@@ -1 +1 @@
-sha256=4eac269f9a59b2f9be4e7a87ea6aa38454f5ebe10ede35c9b8d4218137d2cd35
+sha256=e4ee3898b840a8374cb02c64342cc15f8d7ecd1becaf4b4a2a4657290f9e7ff1

.config/e2e-smoke-selection.txt

@@ -1 +1 @@
-sha256=879f421f1115d44f930e229510a2e533cfd5ebd4effe38730b82ef61f4fd7a86
+sha256=00f8a56b75990b487a594b97c36610b94476a6944ba00473e37187ea654d42bf

.config/make/lint-fmt.mak

@@ -70,11 +70,6 @@ s3s-footprint-check: ## Check the s3s dependency footprint ratchet stays frozen
 	@echo "📦 Checking s3s footprint ratchet..."
 	./scripts/check_s3s_footprint.sh
 
-.PHONY: test-count-ratchet-check
-test-count-ratchet-check: ## Check no workspace crate lost tests or gained #[ignore] against the committed baseline
-	@echo "🧪 Checking per-crate test-count ratchet..."
-	./scripts/check_test_count_ratchet.sh
-
 .PHONY: fips-wording-check
 fips-wording-check: ## Check docs and crates/kms do not over-claim crypto capabilities
 	@echo "📣 Checking cryptographic capability wording guard..."

.config/make/pre-commit.mak

@@ -20,11 +20,11 @@ planning-docs-check: ## Check that no planning-type documents are committed
 	./scripts/check_no_planning_docs.sh
 
 .PHONY: pre-commit
-pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check test-count-ratchet-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
+pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
 	@echo "✅ All pre-commit checks passed!"
 
 .PHONY: pre-pr
-pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check test-count-ratchet-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check offline-enrollment-e2e-check clippy-check test ## Run full pre-PR checks with clippy and tests
+pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check offline-enrollment-e2e-check clippy-check test ## Run full pre-PR checks with clippy and tests
 	@echo "✅ All pre-PR checks passed!"
 
 .PHONY: dev-check

.config/make/tests.mak

@@ -51,9 +51,7 @@ script-tests: ## Run shell script tests
 	./scripts/test_python_bin.sh
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_check_doc_paths.py
 	./scripts/check_embedded_secrets.sh --self-test
-	./scripts/check_s3s_footprint.sh --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_test_wiring.py --self-test
-	$(RUSTFS_PYTHON_BIN) ./scripts/codemods/s3_error_to_s3_types.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_e2e_binary.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_inspect_btf_prototypes.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_release_build_workflow.py

.config/nextest.toml

@@ -356,12 +356,6 @@ threads-required = "num-test-threads"
 filter = 'package(rustfs-ecstore) & test(=store::init::tests::dispatch_manifest_rollback_bounded_concurrency_reaches_tail_behind_slow_member)'
 threads-required = "num-test-threads"
 
-# These short-deadline real-disk progress probes need the test run's capacity
-# for scanner and object-lock handoffs; keep their assertions and deadlines.
-[[profile.default.overrides]]
-filter = '(package(rustfs-scanner) & (test(=scanner::tests::cycle_persist_failure::later_cycle_post_commit_failure_reloads_durable_state) | test(=scanner::tests::running_main_loop_catches_up_pause_cleared_after_startup_observe))) | (package(rustfs-ecstore) & test(=store::init::tests::restored_transitioned_delete_causally_enqueues_free_version))'
-threads-required = "num-test-threads"
-
 # ---------------------------------------------------------------------------
 # ci profile — the strict CI gate (ci.yml `cargo nextest run --profile ci`)
 # ---------------------------------------------------------------------------
@@ -430,12 +424,6 @@ threads-required = "num-test-threads"
 filter = 'package(rustfs-scanner) & (test(/^scanner::backlog::tests::native_/) | test(=scanner_folder::tests::mrf_ownership::mrf_ownership_manager_completion_preserves_scanner_pending) | test(/^scanner_io::tests::service_cohort::/))'
 threads-required = "num-test-threads"
 
-# Keep short-deadline real-disk progress probes from competing with unrelated
-# tests for scanner and object-lock capacity. Assertions and deadlines stay on.
-[[profile.ci.overrides]]
-filter = '(package(rustfs-scanner) & (test(=scanner::tests::cycle_persist_failure::later_cycle_post_commit_failure_reloads_durable_state) | test(=scanner::tests::running_main_loop_catches_up_pause_cleared_after_startup_observe))) | (package(rustfs-ecstore) & test(=store::init::tests::restored_transitioned_delete_causally_enqueues_free_version))'
-threads-required = "num-test-threads"
-
 # Emit this expensive fixture's stage timings in CI and use tmpfs when safe.
 [[profile.ci.overrides]]
 filter = 'package(rustfs-ecstore) & test(=store::init::tests::decommission_durable_ilm_receipt_pagination_fails_closed_on_second_page)'
@@ -710,14 +698,6 @@ test-group = 'ecstore-serial-flaky'
 # the PR-lane e2e coverage for scanner usage snapshots consumed by quota and
 # admin surfaces. It uses the same single-node, random-port, isolated-temp-dir
 # fixture as the existing smoke modules.
-#
-# Dual-stack differential (rustfs/backlog#2740): `dual_stack_differential::`
-# starts two single-node servers from the same binary, one per
-# RUSTFS_S3_STACK value, on random ports with isolated temp dirs, and runs one
-# SDK script covering every operation of `impl S3 for FS` against both. Its
-# comparator unit tests need no server. It meets the admission criteria
-# unchanged; the module clause admits a new differential test without another
-# filter edit.
 [profile.e2e-smoke]
 default-filter = """
     package(e2e_test) & (
@@ -726,7 +706,6 @@ default-filter = """
         | test(/^reliant::lifecycle::/)
         | test(/^conditional_delete_objects_test::/)
         | test(/^reliant::tiering::/)
-        | test(/^dual_stack_differential::/)
         | test(/^upgrade_compatibility_test::upgrade_write_readiness_tests::/)
         | test(/^on_demand_migration::(get_basic_test::(get_miss_pulls_inline_and_serves_locally_afterwards|head_miss_answers_from_the_source_without_persisting)|interaction_test::test_odm_admin_config_is_redacted_and_status_counts_match_the_source)$/)
     )
@@ -991,10 +970,10 @@ threads-required = "num-test-threads"
 filter = 'package(e2e_test) & (test(/^kms::kms_vault_test::/) | test(/^kms::kms_rekey_sweep_test::/) | test(/^kms::configured_roundtrip_test::test_configured_vault_kms_admin_and_versioned_cleanup$/))'
 test-group = 'e2e-vault'
 
-# Keep the storage-heavy smoke cases isolated in the full suite too. They
-# start multiple server fixtures and issue many sequential S3 operations.
+# Keep storage-heavy cases isolated in the full suite too. The large pagination
+# fixture alone seeds 2,000 objects and can otherwise starve other servers.
 [[profile.e2e-full.overrides]]
-filter = 'package(e2e_test) & (test(/^list_objects_v2_pagination_test::tests::(test_list_objects_v2_delimiter_small_page_traverses_all|test_list_objects_v2_max_keys_above_limit_returns_token|test_list_objects_v2_maxkeys_above_limit_with_delimiter)$/) | test(=reliant::tiering::test_hermetic_transition_restore_failure_expiry_and_retry))'
+filter = 'package(e2e_test) & (test(/^list_objects_v2_pagination_test::tests::(test_list_objects_v2_delimiter_small_page_traverses_all|test_list_objects_v2_max_keys_above_limit_returns_token|test_list_objects_v2_maxkeys_above_limit_with_delimiter|test_list_objects_v2_raw_exceeds_maxkeys_but_visible_below)$/) | test(=reliant::tiering::test_hermetic_transition_restore_failure_expiry_and_retry))'
 threads-required = "num-test-threads"
 
 # Fixed, small contract for every executable server change. These existing

.config/release-validation-plan.json

@@ -0,0 +1,77 @@
+{
+  "schema_version": 1,
+  "plan_id": "rustfs-release-validation",
+  "description": "Versioned certification plan for every public RustFS release exit (rustfs/backlog#2662, R01). preview, prerelease and stable all require this explicit plan. Scripts/release/release_gate.py fail-closes publication on any missing, duplicate, stale, substituted or non-PASS required item.",
+  "release_types": ["preview", "prerelease", "stable"],
+  "targets": [
+    {"os": "linux", "arch": "x86_64", "abi": "musl"},
+    {"os": "linux", "arch": "aarch64", "abi": "musl"},
+    {"os": "linux", "arch": "x86_64", "abi": "gnu"},
+    {"os": "linux", "arch": "aarch64", "abi": "gnu"},
+    {"os": "macos", "arch": "aarch64", "abi": "none"},
+    {"os": "windows", "arch": "x86_64", "abi": "msvc"}
+  ],
+  "targets_basis": "Platform matrix of .github/workflows/build.yml (prepare-platform-matrix). gnu and macOS targets build with the pyroscope feature; musl and Windows targets build with no extra features.",
+  "features": ["pyroscope"],
+  "trusted_workflows": [
+    {"repository": "rustfs/s3chaos", "workflow_id": "release-gate.yml"}
+  ],
+  "trusted_workflows_basis": "The shared evidence envelope producer lives in rustfs/s3chaos (G01, fault/release_gate.rs) and its plan/live split workflow is .github/workflows/release-gate.yml there. Only results from these identities may enter a release certification; PR-uploaded JSON never gains release authority. R02 must extend this list when additional verifier workflows land, never widen it to untrusted sources.",
+  "lanes": [
+    {
+      "lane": "identity",
+      "harness_repository": "rustfs/s3chaos",
+      "harness_sha": "9a48f55c1d6d58ce63a8fb1ea7317e024f32f11d",
+      "harness_basis": "Fixed G01 evidence-envelope producer revision (rustfs/s3chaos PR #102 head recorded in rustfs/backlog#2658). R02 re-pins this when the live verifier ships.",
+      "timeout_seconds": 1800,
+      "environment_capabilities": ["process-exec"],
+      "cases": ["artifact-identity", "startup"],
+      "applies_to": {
+        "release_types": ["preview", "prerelease", "stable"],
+        "platforms": "all"
+      },
+      "applies_basis": "Artifact identity and startup checks are mandatory for every release type, preview included, on every shipped target (rustfs/backlog#2662 frozen design item 5)."
+    },
+    {
+      "lane": "upgrade",
+      "harness_repository": "rustfs/s3chaos",
+      "harness_sha": "9a48f55c1d6d58ce63a8fb1ea7317e024f32f11d",
+      "harness_basis": "Same fixed envelope producer as the identity lane; the workflow SHA and the product build SHA are bound separately and need not be equal.",
+      "timeout_seconds": 5400,
+      "environment_capabilities": ["process-exec", "previous-release-artifact"],
+      "cases": [
+        "direct-upgrade",
+        "mixed-version-upgrade",
+        "bucket-config-upgrade",
+        "bucket-config-rollback",
+        "multipart-layout-upgrade",
+        "multipart-layout-baseline"
+      ],
+      "cases_basis": "Case IDs of the pinned previous-release matrix in .github/workflows/e2e-upgrade.yml (source.version/asset/sha256 and the case list mapped there).",
+      "applies_to": {
+        "release_types": ["prerelease", "stable"],
+        "platforms": [{"os": "linux", "arch": "x86_64", "abi": "gnu"}]
+      },
+      "applies_basis": "preview releases use the pre-declared smaller set (identity lane only). The upgrade lane is only exercised on linux-x86_64-gnu, the platform of the pinned source assets."
+    }
+  ],
+  "upgrade_sources": {
+    "basis": "Exactly the two previous-release sources pinned in .github/workflows/e2e-upgrade.yml are exercised today (source.version / source.asset / source.sha256). No other N-1/N-2 compatibility is claimed, and these digests are not inferred from tags.",
+    "sources": [
+      {
+        "version": "1.0.1",
+        "role": "N-1",
+        "artifact_name": "rustfs-linux-x86_64-gnu-v1.0.1.zip",
+        "artifact_sha256": "9742e183cb0dfb0d48fa909797af572fe47b7f827569036baa4156e0d3958ee5"
+      },
+      {
+        "version": "1.0.0-rc.5",
+        "role": "N-2",
+        "artifact_name": "rustfs-linux-x86_64-gnu-v1.0.0-rc.5.zip",
+        "artifact_sha256": "3ee8df71e8edcfada533be452c4135868f697bc515460ae97b027313eade7a3d"
+      }
+    ],
+    "rollback_supported": false,
+    "rollback_basis": "Only the bucket-config-rollback case (rollback_to_previous_release_reads_current_bucket_metadata) is exercised. Full data-plane rollback is not certified and is never a default claim."
+  }
+}

.config/s3s-edge-allowlist.txt

@@ -1,37 +0,0 @@
-# Files that may still reference s3s (`s3s::` or `s3s_sigv4`) when Phase 1 of
-# the gateway migration ends (rustfs/backlog#2734 task T0.6, rustfs/backlog#2741).
-# Consumed only by `scripts/check_s3s_footprint.sh --mode allowlist`.
-#
-# One entry per line: a repository-relative file path or a git `:(glob)`
-# pathspec. `**` matches across directories, `*` stays within one path
-# component, and a directory entry matches everything under it. Blank lines
-# and lines starting with `#` are ignored; an entry may not contain whitespace,
-# so a note belongs on its own comment line.
-#
-# Every Rust file listed by `git grep -l '\bs3s\(::\|_sigv4\)' -- '*.rs'` that
-# no entry matches fails the allowlist mode. Task T1.9 switches CI to that
-# mode; until then `--dry-run` prints the files still outside this list, which
-# is the Phase 1 tracking list.
-#
-# Never add crates/** other than crates/e2e_test: the engine crates leave s3s
-# entirely, and the script rejects such an entry. Only the s3s-facing edge of
-# the rustfs binary and the e2e oracle keep it.
-
-# rustfs binary: the s3s-facing edge that survives Phase 1
-rustfs/src/storage/ecfs.rs
-rustfs/src/storage/s3_compat/**
-rustfs/src/storage/access.rs
-rustfs/src/auth.rs
-rustfs/src/error.rs
-rustfs/src/storage_api.rs
-rustfs/src/app/storage_api.rs
-rustfs/src/admin/router.rs
-rustfs/src/app/metadata_route.rs
-rustfs/src/server/http.rs
-rustfs/src/server/layer.rs
-rustfs/src/server/hybrid.rs
-rustfs/src/protocols/client.rs
-
-# e2e oracle: the test target that verifies S3 behaviour against s3s
-crates/e2e_test/src/fake_s3_target/**
-crates/e2e_test/src/**

.config/s3s-footprint-baseline.txt

@@ -1,17 +0,0 @@
-# Exact baseline for the direct s3s footprint, consumed only by
-# scripts/check_s3s_footprint.sh (rustfs/backlog#1677; rustfs/backlog#2734 T0.4).
-#
-# Each value is the exact measurement of the tree, not a ceiling. CI fails when a
-# measured counter differs from its line here in either direction:
-#   - a larger measurement means new code widened the s3s surface: route it
-#     through the gateway abstractions instead; never raise a value here;
-#   - a smaller measurement means the PR shrank the surface: lower the matching
-#     value to the measured number in the same PR.
-#
-# files           Rust files referencing s3s paths, repo-wide minus crates/e2e_test
-# s3_error_lines  lines holding s3_error!, same scope, minus those the script
-#                 header's rule clears (rustfs_s3_types's macro); fails closed
-# ecstore_files   Rust files referencing s3s paths under crates/ecstore/src
-files=65
-s3_error_lines=289
-ecstore_files=0

.config/test-count-baseline.txt

@@ -1,64 +0,0 @@
-# Per-crate test-count baseline consumed by scripts/check_test_count_ratchet.sh
-# (rustfs/backlog#2734 section 1 rule 3; T0.3 in rustfs/backlog#2738).
-#
-# One row per workspace member: <package>, tests, ignored, tab-separated.
-# tests counts #[test], #[tokio::test], #[tokio::test(...)] and #[test_case(...)]
-# attribute lines under the package directory; ignored counts #[ignore] and
-# #[ignore = "..."]. Both columns are exact: CI fails when a crate's tests fall
-# below or rise above its row, or when its ignored count changes. Raise a row in
-# the PR that adds tests; lowering one is a reviewed statement that tests were
-# intentionally removed. Regenerate with:
-#   scripts/check_test_count_ratchet.sh --update
-e2e_test	1012	20
-rustfs	6157	64
-rustfs-audit	70	0
-rustfs-checksums	26	0
-rustfs-common	36	0
-rustfs-concurrency	87	0
-rustfs-config	43	0
-rustfs-connect-symbol-catalog	6	0
-rustfs-credentials	24	0
-rustfs-crypto	67	0
-rustfs-data-usage	57	0
-rustfs-ecstore	7561	2
-rustfs-extension-schema	23	0
-rustfs-filemeta	353	0
-rustfs-heal	1097	3
-rustfs-heal-contracts	8	0
-rustfs-iam	401	0
-rustfs-io-core	49	0
-rustfs-io-metrics	154	0
-rustfs-keystone	29	0
-rustfs-kms	715	25
-rustfs-license	2	0
-rustfs-lifecycle	155	0
-rustfs-lock	148	0
-rustfs-log-analyzer	90	0
-rustfs-madmin	210	0
-rustfs-notify	140	0
-rustfs-object-capacity	92	0
-rustfs-object-data-cache	140	0
-rustfs-obs	385	0
-rustfs-policy	734	0
-rustfs-protocols	748	1
-rustfs-protos	73	0
-rustfs-replication	209	0
-rustfs-rio	218	0
-rustfs-rio-v2	34	6
-rustfs-s3-client	69	0
-rustfs-s3-ops	8	0
-rustfs-s3-types	105	0
-rustfs-s3select-api	207	0
-rustfs-s3select-query	204	2
-rustfs-scanner	1340	20
-rustfs-scanner-metrics	90	0
-rustfs-security-governance	20	0
-rustfs-signer	31	0
-rustfs-storage-api	34	0
-rustfs-targets	604	26
-rustfs-test-utils	1	0
-rustfs-tls-runtime	43	0
-rustfs-trusted-proxies	68	0
-rustfs-utils	247	0
-rustfs-zip	30	0
-zstd	11	0

.github/actions/quick-checks/action.yml

@@ -96,10 +96,6 @@ runs:
       shell: bash
       run: ./scripts/check_s3s_footprint.sh
 
-    - name: Check per-crate test-count ratchet
-      shell: bash
-      run: ./scripts/check_test_count_ratchet.sh
-
     - name: Check cryptographic capability wording
       shell: bash
       run: ./scripts/check_fips_wording.sh

.github/workflows/build.yml

@@ -814,13 +814,18 @@ jobs:
           export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
           export AWS_DEFAULT_REGION="auto"
 
-          # Determine upload path based on build type
+          # Determine upload path based on build type. Release and prerelease
+          # builds only stage their bytes here: the official
+          # artifacts/rustfs/release/ path is written solely by the
+          # certification-gated promote-release-artifacts job, from these same
+          # staged objects. Development builds keep the uncertified dev channel
+          # and never touch official or latest channels.
           if [[ "$BUILD_TYPE" == "development" ]]; then
             R2_PATH="s3://${R2_BUCKET}/artifacts/rustfs/dev/"
             echo "📤 Uploading development build to R2 dev directory"
           else
-            R2_PATH="s3://${R2_BUCKET}/artifacts/rustfs/release/"
-            echo "📤 Uploading release build to R2 release directory"
+            R2_PATH="s3://${R2_BUCKET}/artifacts/rustfs/staging/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}/"
+            echo "📤 Staging release build for certification-gated promotion"
           fi
 
           # Upload all rustfs zip files to R2 using S3-compatible API
@@ -869,6 +874,221 @@ jobs:
             echo "This run does not trigger Docker publishing."
           fi
 
+  # Aggregate the parent candidate batch. This job always collects results and
+  # retains its evidence, including on failure, and every publication job
+  # below requires its success. Missing, stale, substituted or non-PASS
+  # evidence fails closed here: only the trusted verification sources
+  # (rustfs/s3chaos release-gate envelopes, rustfs/backlog#2658) can satisfy
+  # it, so a PR-uploaded JSON set never gains release authority.
+  release-certification:
+    name: Release Certification
+    needs: [ build-check, build-rustfs ]
+    if: ${{ !cancelled() && needs.build-check.outputs.should_publish == 'true' && startsWith(github.ref, 'refs/tags/') }}
+    runs-on: ubuntu-latest # gh CLI required; the sm-standard-2 fleet does not ship it
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Download all build artifacts
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          path: ./artifacts
+          pattern: rustfs-*
+          merge-multiple: true
+
+      - name: Write candidate and collect trusted validation results
+        env:
+          # Cross-repo read of the trusted verification run. Falls back to the
+          # workflow token, which cannot read rustfs/s3chaos; a missing token
+          # therefore fail-closes instead of certifying.
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          PRODUCT_COMMIT: ${{ github.sha }}
+          VERSION: ${{ needs.build-check.outputs.version }}
+          BUILD_TYPE: ${{ needs.build-check.outputs.build_type }}
+          FEATURES: ${{ needs.build-check.outputs.build_type == 'development' && '' || 'pyroscope' }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          mkdir -p results release-gate-output
+          PLAN_SHA256=$(python3 -c "import hashlib,json,sys; sys.path.insert(0,'scripts/release'); import release_gate; print(release_gate.canonical_sha256(json.load(open('.config/release-validation-plan.json'))))")
+
+          python3 - "$PLAN_SHA256" > candidate.json <<'PY'
+          import hashlib, json, os, sys, zipfile
+
+          plan_sha256 = sys.argv[1]
+          profile = {"release": "stable", "prerelease": "prerelease", "preview": "preview"}.get(os.environ["BUILD_TYPE"])
+          if profile is None:
+              raise SystemExit("release certification only covers preview, prerelease and stable builds")
+
+          def sha256(path):
+              digest = hashlib.sha256()
+              with open(path, "rb") as handle:
+                  for chunk in iter(lambda: handle.read(1024 * 1024), b""):
+                      digest.update(chunk)
+              return digest.hexdigest()
+
+          deliverables = []
+          for name in sorted(os.listdir("artifacts")):
+              if not name.startswith("rustfs-") or not name.endswith(".zip"):
+                  continue
+              path = os.path.join("artifacts", name)
+              members = zipfile.ZipFile(path).namelist()
+              binary_name = "rustfs.exe" if "rustfs.exe" in members else "rustfs"
+              with zipfile.ZipFile(path) as archive:
+                  binary_bytes = archive.read(binary_name)
+              import tempfile
+              with tempfile.NamedTemporaryFile() as handle:
+                  handle.write(binary_bytes)
+                  handle.flush()
+                  binary_sha = sha256(handle.name)
+              deliverables.append({
+                  "kind": "zip",
+                  "name": name,
+                  "location": "https://github.com/rustfs/rustfs/releases/download/{}/{}".format(os.environ["VERSION"], name),
+                  "sha256": sha256(path),
+                  "size": os.path.getsize(path),
+                  "binary_sha256": binary_sha,
+              })
+          if not deliverables:
+              raise SystemExit("no built deliverables found to certify")
+          candidate = {
+              "schema_version": 1,
+              "product_commit": os.environ["PRODUCT_COMMIT"],
+              "version": os.environ["VERSION"],
+              "release_profile": profile,
+              "features": [f for f in os.environ.get("FEATURES", "").split(",") if f],
+              "plan_sha256": plan_sha256,
+              "validation_batch": {"run_id": os.environ["GITHUB_RUN_ID"], "run_attempt": int(os.environ["GITHUB_RUN_ATTEMPT"])},
+              "deliverables": deliverables,
+          }
+          json.dump(candidate, sys.stdout, indent=2, sort_keys=True)
+          print()
+          PY
+
+          # Trusted verification evidence only: the release-gate envelopes of
+          # rustfs/s3chaos for this exact product commit. Absence blocks below.
+          TRUSTED_REPO="rustfs/s3chaos"
+          TRUSTED_WORKFLOW="release-gate.yml"
+          TRUSTED_RUN=$(gh api "repos/${TRUSTED_REPO}/actions/workflows/${TRUSTED_WORKFLOW}/runs?head_sha=${PRODUCT_COMMIT}&status=success&per_page=1" --jq '.workflow_runs[0].id // empty' || true)
+          if [[ -z "$TRUSTED_RUN" ]]; then
+            echo "❌ No trusted verification run of ${TRUSTED_REPO}/${TRUSTED_WORKFLOW} for ${PRODUCT_COMMIT}; refusing to certify" >&2
+          else
+            gh run download "$TRUSTED_RUN" -R "$TRUSTED_REPO" -n release-validation-evidence -D results || \
+              echo "❌ Trusted verification run ${TRUSTED_RUN} published no release-validation-evidence artifact" >&2
+          fi
+
+          python3 scripts/release/release_gate.py validate \
+            --plan .config/release-validation-plan.json \
+            --candidate candidate.json \
+            --results results \
+            --output release-gate-output
+
+      - name: Upload certification evidence
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: release-certification
+          path: |
+            candidate.json
+            results/
+            release-gate-output/
+          if-no-files-found: warn
+          retention-days: 30
+
+  # The only writer of the official R2 release path. Promotion re-checks the
+  # exact bytes about to be published against the passed certification list
+  # and copies the staged objects byte-for-byte: no rebuild-after-verify.
+  promote-release-artifacts:
+    name: Promote Release Artifacts
+    needs: [ build-check, build-rustfs, release-certification ]
+    # Target-channel lock for the official R2 release path.
+    concurrency:
+      group: release-channel-r2-release
+      cancel-in-progress: false
+    if: needs.build-check.outputs.should_publish == 'true' && startsWith(github.ref, 'refs/tags/') && (needs.build-check.outputs.build_type == 'release' || needs.build-check.outputs.build_type == 'prerelease')
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    env:
+      R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+      R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+      R2_BUCKET: ${{ secrets.R2_BUCKET }}
+      AWS_EC2_METADATA_DISABLED: true
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Download build and certification artifacts
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          path: ./artifacts
+          pattern: rustfs-*
+          merge-multiple: true
+
+      - name: Download certification evidence
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: release-certification
+          path: ./certification
+
+      - name: Verify and promote staged release artifacts
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if [[ -z "$R2_ACCESS_KEY_ID" || -z "$R2_SECRET_ACCESS_KEY" || -z "$R2_ENDPOINT" || -z "$R2_BUCKET" ]]; then
+            echo "❌ R2 credentials missing; refusing to promote to the official channel" >&2
+            exit 1
+          fi
+
+          for zip_file in artifacts/rustfs-*.zip; do
+            [[ -f "$zip_file" ]] || continue
+            name="$(basename "$zip_file")"
+            python3 scripts/release/release_gate.py verify-promotion \
+              --plan .config/release-validation-plan.json \
+              --candidate certification/candidate.json \
+              --summary certification/release-gate-output \
+              --artifact "$zip_file" \
+              --artifact-name "$name"
+          done
+
+          if ! command -v aws >/dev/null 2>&1; then
+            sudo apt-get update && sudo apt-get install -y awscli
+          fi
+          export AWS_ACCESS_KEY_ID="$R2_ACCESS_KEY_ID"
+          export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
+          export AWS_DEFAULT_REGION="auto"
+
+          STAGING_PATH="s3://${R2_BUCKET}/artifacts/rustfs/staging/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}/"
+          RELEASE_PATH="s3://${R2_BUCKET}/artifacts/rustfs/release/"
+          for zip_file in artifacts/rustfs-*.zip; do
+            [[ -f "$zip_file" ]] || continue
+            name="$(basename "$zip_file")"
+            staged="$(mktemp)"
+            aws s3 cp "${STAGING_PATH}${name}" "$staged" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+            staged_sha="$(sha256sum "$staged" | awk '{print $1}')"
+            built_sha="$(sha256sum "$zip_file" | awk '{print $1}')"
+            if [[ "$staged_sha" != "$built_sha" ]]; then
+              echo "❌ Staged object ${name} does not match the verified bytes" >&2
+              exit 1
+            fi
+            aws s3 cp "$staged" "${RELEASE_PATH}${name}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+            rm -f "$staged"
+            echo "✅ Promoted ${name} (${built_sha})"
+          done
+
   # Create GitHub Release for every valid release tag, including previews
   create-release:
     name: Create GitHub Release
@@ -1028,6 +1248,10 @@ jobs:
   update-latest-version:
     name: Update Latest Version
     needs: [ build-check, publish-release ]
+    # Target-channel lock for the latest.json version pointer.
+    concurrency:
+      group: release-channel-oss-latest
+      cancel-in-progress: false
     if: needs.build-check.outputs.should_publish == 'true' && startsWith(github.ref, 'refs/tags/') && (needs.build-check.outputs.build_type == 'release' || needs.build-check.outputs.build_type == 'prerelease')
     runs-on: ubuntu-latest
     timeout-minutes: 30
@@ -1068,6 +1292,45 @@ jobs:
           rm -rf "$OSSUTIL_DIR" "$OSSUTIL_ZIP"
           chmod +x "$OSSUTIL_BIN"
 
+          # Candidate/pointer ordering: an older candidate must never overwrite
+          # a newer pointer, even if its promotion job finishes later.
+          CURRENT_LATEST="${RUNNER_TEMP}/current-latest.json"
+          if "$OSSUTIL_BIN" cp oss://rustfs-version/latest.json "$CURRENT_LATEST" 2>/dev/null; then
+            python3 - "$CURRENT_LATEST" "$VERSION" <<'PY'
+          import json
+          import re
+          import sys
+
+          def key(version):
+              match = re.match(r"^(\d+)\.(\d+)\.(\d+)(?:-([0-9A-Za-z.-]+))?$", version)
+              if not match:
+                  raise SystemExit(3)
+              core = tuple(int(part) for part in match.groups()[:3])
+              prerelease = match.group(4)
+              if prerelease is None:
+                  return core + (1,)
+              return core + (0, tuple(int(p) if p.isdigit() else p for p in re.split(r"[.-]", prerelease)))
+
+          try:
+              current = json.load(open(sys.argv[1], encoding="utf-8")).get("version", "")
+              proposed = sys.argv[2]
+              if current and key(current) > key(proposed):
+                  raise SystemExit(1)
+          except SystemExit:
+              raise
+          except (OSError, ValueError, TypeError):
+              raise SystemExit(3)
+          PY
+            guard_status=$?
+            if [[ "$guard_status" -eq 1 ]]; then
+              echo "Current latest.json points at a newer version; refusing to downgrade" >&2
+              exit 0
+            elif [[ "$guard_status" -ne 0 ]]; then
+              echo "Existing latest.json is unparsable; refusing to overwrite it blindly" >&2
+              exit 1
+            fi
+          fi
+
           # Create latest.json
           cat > latest.json << EOF
           {
@@ -1084,16 +1347,59 @@ jobs:
 
           echo "✅ Updated latest.json for ${RELEASE_TYPE} release $VERSION"
 
-  # Publish release (remove draft status)
+  # Publish release (remove draft status). Requires the release-certification
+  # aggregate: without a complete legal success there is no public write.
   publish-release:
     name: Publish Release
-    needs: [ build-check, create-release, upload-release-assets ]
+    needs: [ build-check, create-release, upload-release-assets, release-certification ]
+    # Target-channel lock: concurrent releases serialize on this channel so an
+    # older candidate cannot interleave its publish with a newer one.
+    concurrency:
+      group: release-channel-github-release
+      cancel-in-progress: false
     if: needs.build-check.outputs.should_publish == 'true' && startsWith(github.ref, 'refs/tags/') && (needs.build-check.outputs.build_type == 'preview' || needs.build-check.outputs.build_type == 'release' || needs.build-check.outputs.build_type == 'prerelease')
     runs-on: ubuntu-latest # gh CLI required; the sm-standard-2 fleet does not ship it
     timeout-minutes: 30
     permissions:
       contents: write
+      actions: read
     steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Download certification evidence
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: release-certification
+          path: ./certification
+
+      - name: Verify release assets against certification
+        env:
+          GH_TOKEN: ${{ github.token }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          TAG="${{ needs.build-check.outputs.version }}"
+          mkdir -p "$RUNNER_TEMP/promotion-assets"
+          gh release download "$TAG" --dir "$RUNNER_TEMP/promotion-assets" --pattern '*.zip'
+          found=0
+          for zip_file in "$RUNNER_TEMP"/promotion-assets/*.zip; do
+            [[ -f "$zip_file" ]] || continue
+            found=1
+            python3 scripts/release/release_gate.py verify-promotion \
+              --plan .config/release-validation-plan.json \
+              --candidate certification/candidate.json \
+              --summary certification/release-gate-output \
+              --artifact "$zip_file" \
+              --artifact-name "$(basename "$zip_file")"
+          done
+          if [[ "$found" -ne 1 ]]; then
+            echo "No release zip assets to verify against the certification" >&2
+            exit 1
+          fi
+
       - name: Publish release
         env:
           GH_TOKEN: ${{ github.token }}

.github/workflows/cache-warm.yml

@@ -96,7 +96,7 @@ env:
   RUST_BACKTRACE: 1
 
 jobs:
-  # Readers: test-and-lint (workspace and ILM), build-rustfs-debug-binary,
+  # Readers: workspace-lint-and-build, workspace-tests (workspace and ILM), build-rustfs-debug-binary,
   # e2e-tests, e2e-full.
   warm-ci-dev:
     name: Warm ci-dev
@@ -146,7 +146,7 @@ jobs:
           if-no-files-found: error
 
       # --all-targets covers the test binaries nextest builds, including
-      # e2e_test, which test-and-lint's own run excludes. The second build adds
+      # e2e_test, which the workspace run excludes. The second build adds
       # the e2e-test-hooks feature resolution that build-rustfs-debug-binary uses
       # and that no lint lane enables.
       - name: Build ci-dev superset

.github/workflows/ci.yml

@@ -16,9 +16,7 @@ name: Continuous Integration
 
 on:
   push:
-    # gateway/integration receives the daily merge of main as a direct push
-    # (rustfs/backlog#2734), so its pushes must prove the result like main's.
-    branches: [ main, gateway/integration ]
+    branches: [ main ]
     paths-ignore:
       - "**.md"
       - "docs/**"
@@ -115,7 +113,7 @@ jobs:
             printf '%s\n' 'mode=full' >> "$GITHUB_OUTPUT"
             echo "Base CI policy unavailable; running the full matrix."
           fi
-          if ! grep -q '^plan=' "$GITHUB_OUTPUT" || ! grep -q '^rehearse_shards=' "$GITHUB_OUTPUT"; then
+          if ! grep -q '^plan=' "$GITHUB_OUTPUT" || ! grep -q '^plan_contract=2$' "$GITHUB_OUTPUT"; then
             printf '%s\n' 'mode=full' 'connect=true' 'full_validation=true' 'rehearse_shards=false' 'packages=' 'plan=' >> "$GITHUB_OUTPUT"
             echo "Base policy does not cover this workflow contract; retaining the full matrix."
           fi
@@ -162,7 +160,7 @@ jobs:
       - name: Run shared quick checks
         uses: ./.github/actions/quick-checks
 
-  test-and-lint:
+  workspace-lint-and-build:
     name: Workspace Lint and Build
     if: contains(fromJSON('["full", "e2e"]'), needs.classify-changes.outputs.mode) && (github.event_name != 'pull_request' || github.event.action != 'closed')
     needs: [ quick-checks, classify-changes ]
@@ -195,13 +193,13 @@ jobs:
 
       - name: Prepare test evidence
         run: |
-          mkdir -p artifacts/test-and-lint
+          mkdir -p artifacts/workspace
           {
             echo "run_id=${GITHUB_RUN_ID}"
             echo "job=${GITHUB_JOB}"
             echo "runner=${RUNNER_NAME}"
             echo "started_at=$(date --utc --iso-8601=seconds)"
-          } > artifacts/test-and-lint/run-metadata.txt
+          } > artifacts/workspace/run-metadata.txt
 
       # Clippy gates every event, pull requests included. When it ran only after
       # merge, a lint regression merged freely and then failed every main run
@@ -235,11 +233,11 @@ jobs:
 
       - name: Run documentation tests
         run: |
-          mkdir -p artifacts/test-and-lint
+          mkdir -p artifacts/workspace
           set +e
           timeout --verbose --signal=TERM --kill-after=30s 15m \
             python3 scripts/ci_test_artifact.py cargo doc --configuration workspace \
-            2>&1 | tee artifacts/test-and-lint/doctest.log
+            2>&1 | tee artifacts/workspace/doctest.log
           status=${PIPESTATUS[0]}
           {
             echo "command=cargo test --all --exclude e2e_test --doc"
@@ -248,7 +246,7 @@ jobs:
             echo
             echo "Remaining test-related processes:"
             pgrep -af 'cargo|rustdoc|target/.*/deps/' || true
-          } > artifacts/test-and-lint/doctest-diagnostics.txt
+          } > artifacts/workspace/doctest-diagnostics.txt
           exit "${status}"
 
       - name: Upload test reports and diagnostics
@@ -257,7 +255,7 @@ jobs:
         with:
           name: junit-test-and-lint-${{ github.run_number }}-${{ github.run_attempt }}
           path: |
-            artifacts/test-and-lint
+            artifacts/workspace
           retention-days: 3
           if-no-files-found: error
 
@@ -270,7 +268,7 @@ jobs:
 
   workspace-tests:
     name: Workspace tests
-    needs: [ test-and-lint, classify-changes ]
+    needs: [ workspace-lint-and-build, classify-changes ]
     # The archive retains compile-time fixture paths and the large test set
     # requires the same runner image until the hosted-runtime pilot passes.
     runs-on: sm-standard-4
@@ -295,14 +293,14 @@ jobs:
 
       - name: Extract and list workspace tests
         run: |
-          mkdir -p artifacts/test-and-lint
+          mkdir -p artifacts/workspace
           cargo nextest list --profile ci --archive-file "$RUNNER_TEMP/workspace-tests.tar.zst" --workspace-remap "$GITHUB_WORKSPACE" --extract-to "$GITHUB_WORKSPACE" --extract-overwrite --message-format json \
-            > artifacts/test-and-lint/core-test-listing.json
-          python3 scripts/check_test_wiring.py --check-core artifacts/test-and-lint/core-test-listing.json
+            > artifacts/workspace/core-test-listing.json
+          python3 scripts/check_test_wiring.py --check-core artifacts/workspace/core-test-listing.json
 
       - name: Run nextest tests
         run: |
-          mkdir -p artifacts/test-and-lint
+          mkdir -p artifacts/workspace
           rm -f target/nextest/ci/junit.xml
           ./scripts/ci/resource_sampler.sh start nextest
           trap './scripts/ci/resource_sampler.sh stop' EXIT
@@ -311,7 +309,7 @@ jobs:
             cargo nextest run --profile ci --cargo-metadata target/nextest/cargo-metadata.json --binaries-metadata target/nextest/binaries-metadata.json \
             --workspace-remap "$GITHUB_WORKSPACE" --target-dir-remap "$GITHUB_WORKSPACE/target" \
             --status-level all --final-status-level all \
-            2>&1 | tee artifacts/test-and-lint/nextest.log
+            2>&1 | tee artifacts/workspace/nextest.log
           status=${PIPESTATUS[0]}
           {
             echo "command=workspace archive execution"
@@ -323,13 +321,13 @@ jobs:
             echo
             echo "Kernel OOM / kill events:"
             dmesg -T 2>/dev/null | grep -iE 'oom|out of memory|killed process' | tail -20 || true
-          } > artifacts/test-and-lint/nextest-diagnostics.txt
+          } > artifacts/workspace/nextest-diagnostics.txt
           exit "${status}"
 
       - name: Verify workspace execution completeness
         run: |
-          python3 scripts/ci_test_artifact.py evidence artifacts/test-and-lint/core-test-listing.json target/nextest/ci/junit.xml
-          ./scripts/check_migration_gate_count.sh evidence artifacts/test-and-lint/core-test-listing.json target/nextest/ci/junit.xml
+          python3 scripts/ci_test_artifact.py evidence artifacts/workspace/core-test-listing.json target/nextest/ci/junit.xml
+          ./scripts/check_migration_gate_count.sh evidence artifacts/workspace/core-test-listing.json target/nextest/ci/junit.xml
 
       - name: Run ignored ILM integration tests serially
         if: needs.classify-changes.outputs.mode == 'full'
@@ -360,7 +358,7 @@ jobs:
           name: workspace-test-evidence-${{ github.run_attempt }}
           path: |
             target/nextest/ci/junit.xml
-            artifacts/test-and-lint
+            artifacts/workspace
             artifacts/ilm-integration
           retention-days: 3
           if-no-files-found: error
@@ -370,7 +368,7 @@ jobs:
         if: needs.classify-changes.outputs.rehearse_shards == 'true'
         run: |
           python3 scripts/ci_test_shards.py prepare \
-            --listing artifacts/test-and-lint/core-test-listing.json \
+            --listing artifacts/workspace/core-test-listing.json \
             --junit target/nextest/ci/junit.xml --output artifacts/workspace-shard-plan
       - name: Upload same-run shard plan
         if: needs.classify-changes.outputs.rehearse_shards == 'true'
@@ -451,7 +449,7 @@ jobs:
 
   connect-feature-boundaries:
     name: Connect Feature Boundaries
-    if: needs.classify-changes.outputs.mode == 'full' && needs.classify-changes.outputs.connect != 'false' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    if: needs.classify-changes.outputs.mode == 'full' && needs.classify-changes.outputs.connect != 'false' && contains(fromJSON('["schedule", "workflow_dispatch"]'), github.event_name)
     needs: [ quick-checks, classify-changes ]
     runs-on: sm-standard-4
     timeout-minutes: 90
@@ -501,11 +499,13 @@ jobs:
 
   test-and-lint-rio-v2:
     name: Test and Lint (rio-v2)
-    # Alternate feature configurations run on the nightly schedule, merge
-    # groups and manual dispatch. The default workspace remains the PR gate and
-    # the per-push main baseline: repeating four feature builds on every merge
-    # bought nothing the nightly run does not report within a day.
-    if: needs.classify-changes.outputs.mode == 'full' && contains(fromJSON('["merge_group", "schedule", "workflow_dispatch"]'), github.event_name)
+    # Specialized boundary, hardware and alternate-feature lanes run only on
+    # the nightly schedule and manual dispatch. They do not gate a merge: the
+    # nightly run reports a regression within a day, and repeating these builds
+    # on every PR revision and merge group bought no earlier signal worth the
+    # runner occupancy. Workspace lint and tests, the core contract, the debug
+    # build, E2E smoke and the S3 lanes remain the pull-request gate.
+    if: needs.classify-changes.outputs.mode == 'full' && contains(fromJSON('["schedule", "workflow_dispatch"]'), github.event_name)
     needs: [ quick-checks, classify-changes ]
     runs-on: sm-standard-4
     timeout-minutes: 90
@@ -542,12 +542,12 @@ jobs:
           # --profile ci so the quarantine list (and its junit flaky markers)
           # covers this leg too; the default profile is the local no-retry
           # profile and silently ignored quarantined flakes here (rustfs#6703).
-          mkdir -p artifacts/test-and-lint
+          mkdir -p artifacts/workspace
           rm -f target/nextest/ci/junit.xml
           ./scripts/ci/resource_sampler.sh start nextest-rio-v2
           trap './scripts/ci/resource_sampler.sh stop' EXIT
           cargo nextest run --profile ci -p rustfs -p rustfs-ecstore --features rio-v2 \
-            2>&1 | tee artifacts/test-and-lint/nextest.log
+            2>&1 | tee artifacts/workspace/nextest.log
           cargo test -p rustfs --doc --features rio-v2
 
       - name: Upload rio-v2 test reports and diagnostics
@@ -557,14 +557,14 @@ jobs:
           name: junit-test-and-lint-rio-v2-${{ github.run_number }}-${{ github.run_attempt }}
           path: |
             target/nextest/ci/junit.xml
-            artifacts/test-and-lint
+            artifacts/workspace
           retention-days: 3
           if-no-files-found: error
 
   test-and-lint-protocols:
     name: "Test and Lint (${{ matrix.features.name }})"
     # Same cadence as test-and-lint-rio-v2 above.
-    if: needs.classify-changes.outputs.mode == 'full' && contains(fromJSON('["merge_group", "schedule", "workflow_dispatch"]'), github.event_name)
+    if: needs.classify-changes.outputs.mode == 'full' && contains(fromJSON('["schedule", "workflow_dispatch"]'), github.event_name)
     needs: [ quick-checks, classify-changes ]
     runs-on: sm-standard-4
     timeout-minutes: 90
@@ -633,7 +633,7 @@ jobs:
           path: |
             target/nextest/ci/junit.xml
             artifacts/protocol-tests
-            artifacts/test-and-lint
+            artifacts/workspace
           retention-days: 3
           if-no-files-found: warn
 
@@ -785,7 +785,7 @@ jobs:
     # job had neither, so each closed/merged PR really ran the whole io_uring
     # suite (measured 4m17s / 7m19s / 7m31s on runs 30678272341 / 30678117601 /
     # 30662728539) and kept the cancellation run in progress for minutes.
-    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    if: needs.classify-changes.outputs.mode == 'full' && contains(fromJSON('["schedule", "workflow_dispatch"]'), github.event_name)
     needs: [ quick-checks, classify-changes ]
     # GitHub-hosted ubuntu-latest runs a recent kernel with io_uring and, unlike
     # a container, applies no seccomp filter that would block io_uring_setup — so
@@ -899,8 +899,6 @@ jobs:
     # Retain the producer image and fixture paths until the runtime pilot passes.
     runs-on: sm-standard-4
     timeout-minutes: 30
-    env:
-      RUSTFS_S3_STACK: legacy
     steps:
       - name: Checkout repository
         uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
@@ -982,7 +980,7 @@ jobs:
           path: |
             ${{ runner.temp }}/rustfs-e2e-smoke-logs/
             ${{ runner.temp }}/rustfs-e2e-smoke-list.json
-            artifacts/test-and-lint
+            artifacts/workspace
           if-no-files-found: warn
 
       - name: Upload e2e smoke JUnit report
@@ -1021,122 +1019,18 @@ jobs:
           path: ${{ runner.temp }}/rustfs-e2e-*/rustfs.log
           retention-days: 3
 
-  # The same e2e-smoke suite against the same debug binary with the gateway
-  # S3 stack selected (rustfs/backlog#2734). Report-only: it is outside the
-  # Test and Lint aggregate and becomes required at T2.10. The test-binary
-  # download, identity, membership, run and completeness steps must stay
-  # identical to e2e-tests; scripts/ci_gate.py enforces that.
-  e2e-tests-gateway:
-    name: e2e-smoke (gateway, report-only until T2.10)
-    needs: [ build-rustfs-debug-binary ]
-    runs-on: sm-standard-2
-    timeout-minutes: 30
-    continue-on-error: true
-    env:
-      RUSTFS_S3_STACK: gateway
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
-        with:
-          persist-credentials: false
-
-      - name: Install verification runtime
-        uses: ./.github/actions/test-runtime
-
-      - name: Download E2E test binaries
-        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
-        with:
-          name: e2e-test-binaries
-          path: ${{ runner.temp }}
-
-      - name: Verify E2E archive identity
-        run: python3 scripts/ci_test_artifact.py verify --configuration e2e --archive "$RUNNER_TEMP/e2e-tests.tar.zst"
-
-      - name: Set up Python
-        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
-        with:
-          python-version: "3.12"
-
-      - name: Install awscurl
-        run: |
-          python3 -m pip install --user --upgrade pip "awscurl==0.44"
-          echo "AWSCURL_PATH=$HOME/.local/bin/awscurl" >> "$GITHUB_ENV"
-
-      - name: Verify awscurl
-        run: test -x "$AWSCURL_PATH"
-
-      # Download after the cache restore so the freshly built binary from the
-      # build job always wins over anything restored into target/debug.
-      - name: Download debug binary
-        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
-        with:
-          name: rustfs-debug-binary
-          path: target/debug
-
-      - name: Make binary executable
-        run: chmod +x ./target/debug/rustfs
-
-      # Build the e2e test graph once. The archive is reused by the smoke
-      # selection guard, security exact-count check, and run below, avoiding a
-      # second compile of the same e2e_test target on cold runners (backlog#1645).
-      - name: Verify e2e smoke test membership
-        env:
-          NEXTEST_ARCHIVE: ${{ runner.temp }}/e2e-tests.tar.zst
-          NEXTEST_LISTING: ${{ runner.temp }}/rustfs-e2e-smoke-list.json
-        run: |
-          cargo nextest list --profile e2e-smoke --archive-file "${NEXTEST_ARCHIVE}" --workspace-remap "$GITHUB_WORKSPACE" --extract-to "$GITHUB_WORKSPACE" --extract-overwrite --message-format json > "${NEXTEST_LISTING}"
-          python3 ./scripts/check_test_wiring.py --check-profile e2e-smoke "${NEXTEST_LISTING}"
-          ./scripts/check_security_smoke_count.sh check "${NEXTEST_LISTING}"
-
-      # PR smoke subset of the in-repo e2e suite (backlog#1149 ci-4). The
-      # profile.e2e-smoke default-filter in .config/nextest.toml is the single
-      # wiring mechanism for e2e tests in CI — extend that filter instead of
-      # adding new e2e jobs here. Each test spawns its own rustfs server on a
-      # random port and reuses the downloaded debug binary above.
-      - name: Run e2e smoke suite
-        env:
-          NEXTEST_ARCHIVE: ${{ runner.temp }}/e2e-tests.tar.zst
-          RUSTFS_E2E_LOG_DIR: ${{ runner.temp }}/rustfs-e2e-smoke-logs
-          SAMPLER_INTERVAL_SECS: "15"
-        run: |
-          ./scripts/ci/resource_sampler.sh start e2e-smoke
-          trap './scripts/ci/resource_sampler.sh stop' EXIT
-          python3 scripts/e2e_binary.py run --features e2e-test-hooks -- cargo nextest run --profile e2e-smoke --cargo-metadata target/nextest/cargo-metadata.json --binaries-metadata target/nextest/binaries-metadata.json --workspace-remap "$GITHUB_WORKSPACE" --target-dir-remap "$GITHUB_WORKSPACE/target" \
-            --status-level all --final-status-level all --failure-output final
-
-      - name: Verify e2e smoke execution completeness
-        run: python3 scripts/ci_test_artifact.py evidence "$RUNNER_TEMP/rustfs-e2e-smoke-list.json" target/nextest/e2e-smoke/junit.xml
-
-      - name: Upload e2e smoke diagnostics
-        if: always()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        with:
-          name: e2e-smoke-gateway-diagnostics-${{ github.run_number }}-${{ github.run_attempt }}
-          path: |
-            ${{ runner.temp }}/rustfs-e2e-smoke-logs/
-            ${{ runner.temp }}/rustfs-e2e-smoke-list.json
-          if-no-files-found: warn
-
-      - name: Upload e2e smoke JUnit report
-        if: always()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        with:
-          name: e2e-smoke-gateway-junit-${{ github.run_number }}-${{ github.run_attempt }}
-          path: target/nextest/e2e-smoke/junit.xml
-          if-no-files-found: warn
-
   e2e-full:
     name: End-to-End Tests (full merge gate)
     # Merge gate only (backlog#1149 ci-5): the never-automated user-visible
     # suites — KMS, object_lock, multipart_auth, quota, checksum, encryption,
     # security-boundary, ... — via the e2e-full nextest profile. Too heavy for
-    # every PR, so it is gated to main and gateway/integration pushes, the merge
-    # queue, and manual dispatch. protocols / the 7 cluster suites / replication
-    # / #[ignore] are owned by other lanes (see .config/nextest.toml profile.e2e-full).
+    # every PR, so it is gated to main pushes, the merge queue, and manual
+    # dispatch. protocols / the 7 cluster suites / replication / #[ignore] are
+    # owned by other lanes (see .config/nextest.toml profile.e2e-full).
     if: >-
       github.event_name == 'workflow_dispatch' ||
       github.event_name == 'merge_group' ||
-      (github.event_name == 'push' && (github.ref == 'refs/heads/main' || github.ref == 'refs/heads/gateway/integration'))
+      (github.event_name == 'push' && github.ref == 'refs/heads/main')
     needs: [ build-rustfs-debug-binary ]
     # Retain the producer image and fixture paths until the runtime pilot passes.
     runs-on: sm-standard-4
@@ -1497,7 +1391,7 @@ jobs:
       - classify-changes
       - typos
       - quick-checks
-      - test-and-lint
+      - workspace-lint-and-build
       - workspace-tests
       - workspace-shards
       - workspace-shard-evidence
@@ -1533,7 +1427,7 @@ jobs:
       - required-checks
       - typos
       - quick-checks
-      - test-and-lint
+      - workspace-lint-and-build
       - workspace-tests
       - core-contract
       - test-and-lint-rio-v2

.github/workflows/docker.yml

@@ -329,13 +329,75 @@ jobs:
           echo "  - Create latest: $create_latest"
           echo "  - Source ref: $source_ref"
 
+  # Consume the trusted release certification for the exact version this
+  # workflow wraps and verify the certified input digests. A manual dispatch
+  # without that certification fails here and pushes no official tag; only
+  # results from the trusted verification sources carry release authority.
+  release-certification:
+    name: Release Certification
+    needs: build-check
+    if: needs.build-check.outputs.should_build == 'true'
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+          ref: ${{ needs.build-check.outputs.source_ref }}
+
+      - name: Download and verify release certification
+        env:
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          TRIGGERING_RUN_ID: ${{ github.event.workflow_run.id }}
+          VERSION: ${{ needs.build-check.outputs.version }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p certification
+          run_id="$TRIGGERING_RUN_ID"
+          if [[ -z "$run_id" ]]; then
+            run_id="$(gh api --method GET \
+              "repos/${GITHUB_REPOSITORY}/actions/workflows/build.yml/runs" \
+              -f branch="$VERSION" -f status=success -F per_page=1 \
+              --jq '.workflow_runs[0].id // empty' || true)"
+          fi
+          if [[ -z "$run_id" ]]; then
+            echo "No certified Build and Release run found for ${VERSION}; failing closed" >&2
+            exit 1
+          fi
+          if ! gh run download "$run_id" -n release-certification -D certification; then
+            echo "Build run ${run_id} published no release-certification artifact; failing closed" >&2
+            exit 1
+          fi
+          python3 scripts/release/release_gate.py verify-promotion \
+            --plan .config/release-validation-plan.json \
+            --candidate certification/candidate.json \
+            --summary certification/release-gate-output \
+            --artifact-sha256 "$(jq -r '.deliverables[0].sha256' certification/candidate.json)" \
+            --artifact-name "$(jq -r '.deliverables[0].name' certification/candidate.json)"
+
+          # Input digest binding: every zip this image wraps must be the
+          # certified byte-for-byte content, never a mutable latest pointer.
+          while IFS=$'\t' read -r name location expected; do
+            echo "Verifying certified input ${name}"
+            actual="$(curl -fsSL "$location" | sha256sum | awk '{print $1}')"
+            if [[ "$actual" != "$expected" ]]; then
+              echo "Certified input ${name} does not match its recorded digest" >&2
+              exit 1
+            fi
+          done < <(jq -r '.deliverables[] | select(.name | contains("linux")) | [.name, .location, .sha256] | @tsv' certification/candidate.json)
+
   # Build multi-arch Docker images
   # Strategy: Build images using pre-built binaries from dl.rustfs.com
   # Supports both release and dev channel binaries based on build context
   # Only runs when should_build is true (which includes workflow success check)
   build-docker:
     name: Build Docker Images
-    needs: build-check
+    needs: [ build-check, release-certification ]
     if: needs.build-check.outputs.should_build == 'true'
     runs-on: ubuntu-latest
     timeout-minutes: 60
@@ -443,6 +505,14 @@ jobs:
           # Output tags
           echo "tags=$TAGS" >> "$GITHUB_OUTPUT"
 
+          # Staging identity for this run's promotion: official tags are only
+          # created after the staged image verifies against the certification,
+          # and they point at this same digest (no rebuild-after-verify).
+          STAGING_TAG="staging-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}${VARIANT_SUFFIX}"
+          STAGING_TAGS="${{ env.REGISTRY_DOCKERHUB }}:${STAGING_TAG},${{ env.REGISTRY_GHCR }}:${STAGING_TAG},${{ env.REGISTRY_QUAY }}:${STAGING_TAG}"
+          echo "staging_tags=$STAGING_TAGS" >> "$GITHUB_OUTPUT"
+          echo "staging_tag=$STAGING_TAG" >> "$GITHUB_OUTPUT"
+
           # Generate labels
           LABELS="org.opencontainers.image.title=RustFS"
           LABELS="$LABELS,org.opencontainers.image.description=RustFS distributed object storage system"
@@ -467,7 +537,9 @@ jobs:
           file: ${{ matrix.file }}
           platforms: ${{ env.DOCKER_PLATFORMS }}
           push: ${{ needs.build-check.outputs.should_push == 'true' }}
-          tags: ${{ steps.meta.outputs.tags }}
+          # Push to staging tags only: official tags are created by the
+          # certification-gated promotion step below from this same digest.
+          tags: ${{ steps.meta.outputs.staging_tags }}
           labels: ${{ steps.meta.outputs.labels }}
           # No layer cache. This build compiles nothing — it downloads a
           # release zip and runs apk/apt — so the cache could only save the
@@ -493,6 +565,114 @@ jobs:
           no-cache: false
           pull: true
 
+      # A later-built wrapper: the staged image must complete this platform's
+      # own binary identity verification before any official tag is written.
+      # It may not inherit an unverified package's conclusion.
+      - name: Verify staged image against certification
+        if: needs.build-check.outputs.should_push == 'true'
+        env:
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          TRIGGERING_RUN_ID: ${{ github.event.workflow_run.id }}
+          VERSION: ${{ needs.build-check.outputs.version }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p certification
+          run_id="$TRIGGERING_RUN_ID"
+          if [[ -z "$run_id" ]]; then
+            run_id="$(gh api --method GET \
+              "repos/${GITHUB_REPOSITORY}/actions/workflows/build.yml/runs" \
+              -f branch="$VERSION" -f status=success -F per_page=1 \
+              --jq '.workflow_runs[0].id // empty' || true)"
+          fi
+          gh run download "$run_id" -n release-certification -D certification
+
+          STAGING_REF="${{ env.REGISTRY_DOCKERHUB }}:staging-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}${{ matrix.suffix }}"
+          case "${{ matrix.variant }}" in
+            musl) libc="musl" ;;
+            glibc) libc="gnu" ;;
+          esac
+          for platform in linux/amd64 linux/arm64; do
+            case "$platform" in
+              linux/amd64) arch_sub="x86_64" ;;
+              linux/arm64) arch_sub="aarch64" ;;
+            esac
+            docker pull --platform "$platform" "$STAGING_REF"
+            container="$(docker create --platform "$platform" "$STAGING_REF")"
+            docker cp "$container:/usr/bin/rustfs" "$RUNNER_TEMP/staged-rustfs"
+            docker rm "$container" >/dev/null
+            actual="$(sha256sum "$RUNNER_TEMP/staged-rustfs" | awk '{print $1}')"
+            expected="$(jq -r --arg needle "linux-${arch_sub}-${libc}" \
+              '[.deliverables[] | select(.name | contains($needle)) | .binary_sha256][0] // empty' \
+              certification/candidate.json)"
+            if [[ -z "$expected" || "$actual" != "$expected" ]]; then
+              echo "Staged image binary for ${platform} does not match the certified binary digest" >&2
+              exit 1
+            fi
+            python3 scripts/release/release_gate.py verify-promotion \
+              --plan .config/release-validation-plan.json \
+              --candidate certification/candidate.json \
+              --summary certification/release-gate-output \
+              --artifact "$RUNNER_TEMP/staged-rustfs" \
+              --artifact-role binary
+          done
+
+      - name: Promote staged image to official tags
+        if: needs.build-check.outputs.should_push == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+          STAGING_REF="${{ env.REGISTRY_DOCKERHUB }}:staging-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}${{ matrix.suffix }}"
+          IFS=',' read -r -a official_tags <<< "${{ steps.meta.outputs.tags }}"
+          for tag in "${official_tags[@]}"; do
+            docker buildx imagetools create -t "$tag" "$STAGING_REF"
+            echo "Promoted $tag -> $STAGING_REF"
+          done
+
+      # Downstream consumers (Helm) verify this record instead of inheriting
+      # an unverified wrapper's conclusion.
+      - name: Record image promotion
+        if: needs.build-check.outputs.should_push == 'true'
+        shell: bash
+        env:
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          TRIGGERING_RUN_ID: ${{ github.event.workflow_run.id }}
+          VERSION: ${{ needs.build-check.outputs.version }}
+        run: |
+          set -euo pipefail
+          mkdir -p certification
+          run_id="$TRIGGERING_RUN_ID"
+          if [[ -z "$run_id" ]]; then
+            run_id="$(gh api --method GET \
+              "repos/${GITHUB_REPOSITORY}/actions/workflows/build.yml/runs" \
+              -f branch="$VERSION" -f status=success -F per_page=1 \
+              --jq '.workflow_runs[0].id // empty' || true)"
+          fi
+          gh run download "$run_id" -n release-certification -D certification
+          STAGING_REF="${{ env.REGISTRY_DOCKERHUB }}:staging-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}${{ matrix.suffix }}"
+          digest="$(docker buildx imagetools inspect "$STAGING_REF" --format '{{json .Manifest}}' | jq -r '.digest')"
+          candidate_sha="$(python3 -c "import json,hashlib,sys; sys.path.insert(0,'scripts/release'); import release_gate; print(release_gate.canonical_sha256(json.load(open('certification/candidate.json'))))")"
+          jq -n \
+            --arg version "$VERSION" \
+            --arg digest "$digest" \
+            --arg staging_ref "$STAGING_REF" \
+            --arg candidate_sha256 "$candidate_sha" \
+            --arg tags "${{ steps.meta.outputs.tags }}" \
+            --slurpfile candidate certification/candidate.json \
+            '{schema_version: 1, version: $version, image_digest: $digest, staging_ref: $staging_ref,
+              promoted_tags: ($tags | split(",")), candidate_sha256: $candidate_sha256,
+              binary_sha256s: ($candidate[0].deliverables | map(.binary_sha256) | unique)}' \
+            > image-promotion-record.json
+
+      - name: Upload image promotion record
+        if: needs.build-check.outputs.should_push == 'true'
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: image-promotion-record${{ matrix.suffix }}
+          path: image-promotion-record.json
+          if-no-files-found: error
+          retention-days: 30
+
   # Note: Manifest creation is no longer needed as we only build one variant
   # Multi-arch manifests are automatically created by docker/build-push-action
 

.github/workflows/helm-package.yml

@@ -30,9 +30,102 @@ permissions:
   contents: read
 
 jobs:
+  # Consume the same trusted release certification as every other publication
+  # exit and verify the image promotion record of the Docker workflow. The
+  # chart wraps no binary of its own, so it may only reference an image whose
+  # platform verification already passed; it must not inherit an unverified
+  # package's conclusion. A run without that evidence fails closed here and
+  # never reaches the chart repository.
+  release-certification:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    outputs:
+      version: ${{ steps.version.outputs.raw_tag }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Normalize release version
+        id: version
+        env:
+          RAW_INPUT: ${{ github.event.inputs.version }}
+          RAW_BRANCH: ${{ github.event.workflow_run.head_branch }}
+        run: |
+          set -euo pipefail
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            RAW="$RAW_INPUT"
+          else
+            RAW="$RAW_BRANCH"
+          fi
+          case "$RAW" in
+            refs/tags/*) RAW_TAG="${RAW#refs/tags/}" ;;
+            *) RAW_TAG="$RAW" ;;
+          esac
+          echo "raw_tag=$RAW_TAG" >> "$GITHUB_OUTPUT"
+
+      - name: Download and verify release certification and image record
+        env:
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          TRIGGERING_RUN_ID: ${{ github.event.workflow_run.id }}
+          RAW_TAG: ${{ steps.version.outputs.raw_tag }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p certification image-record
+
+          build_run="$TRIGGERING_RUN_ID"
+          if [[ -z "$build_run" ]]; then
+            build_run="$(gh api --method GET \
+              "repos/${GITHUB_REPOSITORY}/actions/workflows/build.yml/runs" \
+              -f branch="$RAW_TAG" -f status=success -F per_page=1 \
+              --jq '.workflow_runs[0].id // empty' || true)"
+          fi
+          if [[ -z "$build_run" ]]; then
+            echo "No certified Build and Release run found for ${RAW_TAG}; failing closed" >&2
+            exit 1
+          fi
+          if ! gh run download "$build_run" -n release-certification -D certification; then
+            echo "Build run ${build_run} published no release-certification artifact; failing closed" >&2
+            exit 1
+          fi
+          python3 scripts/release/release_gate.py verify-promotion \
+            --plan .config/release-validation-plan.json \
+            --candidate certification/candidate.json \
+            --summary certification/release-gate-output \
+            --artifact-sha256 "$(jq -r '.deliverables[0].sha256' certification/candidate.json)" \
+            --artifact-name "$(jq -r '.deliverables[0].name' certification/candidate.json)"
+
+          docker_run="$(gh api --method GET \
+            "repos/${GITHUB_REPOSITORY}/actions/workflows/docker.yml/runs" \
+            -f branch="$RAW_TAG" -f status=success -F per_page=1 \
+            --jq '.workflow_runs[0].id // empty' || true)"
+          if [[ -z "$docker_run" ]] || ! gh run download "$docker_run" -n image-promotion-record -D image-record; then
+            echo "No image promotion record for ${RAW_TAG}; a chart may not wrap an unverified image" >&2
+            exit 1
+          fi
+          candidate_sha="$(python3 -c "import json,hashlib,sys; sys.path.insert(0,'scripts/release'); import release_gate; print(release_gate.canonical_sha256(json.load(open('certification/candidate.json'))))")"
+          record_sha="$(jq -r '.candidate_sha256 // empty' image-record/image-promotion-record.json)"
+          if [[ "$record_sha" != "$candidate_sha" ]]; then
+            echo "Image promotion record does not belong to this certification" >&2
+            exit 1
+          fi
+          overlap="$(jq -r --slurpfile candidate certification/candidate.json \
+            '[.binary_sha256s[]] - ($candidate[0].deliverables | map(.binary_sha256)) | length' \
+            image-record/image-promotion-record.json)"
+          if [[ "$overlap" -ne 0 ]]; then
+            echo "Image promotion record carries binary digests outside the certification" >&2
+            exit 1
+          fi
+
   build-helm-package:
     runs-on: ubuntu-latest
     timeout-minutes: 30
+    needs: [ release-certification ]
     if: |
       (github.event_name == 'workflow_dispatch' && !contains(github.event.inputs.version, '-preview')) ||
       (

.github/workflows/package.yml

@@ -225,10 +225,57 @@ jobs:
           echo "  Build run ID: $BUILD_RUN_ID"
           echo "  Build run number: $RUN_NUMBER"
 
+  # Consume the trusted release certification of the resolved build run and
+  # verify the certified input digests before any package is wrapped. A run
+  # without that certification fails closed here; packages built later must
+  # complete their own verification and cannot inherit an unverified
+  # conclusion.
+  release-certification:
+    name: Release Certification
+    needs: resolve
+    if: needs.resolve.result == 'success'
+    runs-on: ubuntu-latest # gh CLI required; the sm-standard-2 fleet does not ship it
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Download and verify release certification
+        env:
+          GH_TOKEN: ${{ secrets.RELEASE_VALIDATION_TOKEN || github.token }}
+          BUILD_RUN_ID: ${{ needs.resolve.outputs.build_run_id }}
+          BUILD_TYPE: ${{ needs.resolve.outputs.build_type }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p certification
+          if [[ "$BUILD_TYPE" == "development" ]]; then
+            # Development artifacts are explicitly uncertified and never touch
+            # official or latest channels; only the staging/dev paths remain.
+            jq -n '{schema_version: 1, certified: false, reason: "development artifacts are explicitly uncertified"}' \
+              > certification/development-uncertified.json
+            exit 0
+          fi
+          if ! gh run download "$BUILD_RUN_ID" -n release-certification -D certification; then
+            echo "Build run ${BUILD_RUN_ID} published no release-certification artifact; failing closed" >&2
+            exit 1
+          fi
+          python3 scripts/release/release_gate.py verify-promotion \
+            --plan .config/release-validation-plan.json \
+            --candidate certification/candidate.json \
+            --summary certification/release-gate-output \
+            --artifact-sha256 "$(jq -r '.deliverables[0].sha256' certification/candidate.json)" \
+            --artifact-name "$(jq -r '.deliverables[0].name' certification/candidate.json)"
+
   # Build DEB and RPM packages for each architecture and libc variant
   package:
     name: Package (${{ matrix.arch }}-${{ matrix.libc }})
-    needs: resolve
+    needs: [ resolve, release-certification ]
     runs-on: ubuntu-latest # gh CLI required; the sm-standard-2 fleet does not ship it
     timeout-minutes: 30
     strategy:
@@ -494,6 +541,55 @@ jobs:
           echo "rpm_file=$RPM_FILE" >> "$GITHUB_OUTPUT"
           echo "✅ RPM built: $RPM_FILE"
 
+      - name: Download certification evidence
+        if: needs.resolve.outputs.build_type != 'development'
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: release-certification
+          path: ./certification
+
+      # Later-built wrapper verification: the binary inside each package must
+      # match the certified binary digest for this platform before anything is
+      # uploaded, and the package itself stays on staging until promotion.
+      - name: Verify wrapped binary against certification
+        if: needs.resolve.outputs.build_type != 'development'
+        shell: bash
+        env:
+          DEB_FILE: ${{ steps.deb.outputs.deb_file }}
+          RPM_FILE: ${{ steps.rpm.outputs.rpm_file }}
+          ARTIFACT_NAME: ${{ matrix.artifact_name }}
+        run: |
+          set -euo pipefail
+          expected="$(jq -r --arg needle "$ARTIFACT_NAME" \
+            '[.deliverables[] | select(.name | contains($needle)) | .binary_sha256][0] // empty' \
+            certification/candidate.json)"
+          if [[ -z "$expected" ]]; then
+            echo "No certified binary digest recorded for ${ARTIFACT_NAME}" >&2
+            exit 1
+          fi
+          if ! command -v rpm2cpio >/dev/null 2>&1; then
+            sudo apt-get update && sudo apt-get install -y rpm2cpio
+          fi
+          if ! command -v cpio >/dev/null 2>&1; then
+            sudo apt-get update && sudo apt-get install -y cpio
+          fi
+          for package_file in "$DEB_FILE" "$RPM_FILE"; do
+            case "$package_file" in
+              *.deb)
+                dpkg-deb --fsys-tarfile "$package_file" | tar -xOf - ./usr/bin/rustfs > "$RUNNER_TEMP/wrapped-rustfs"
+                ;;
+              *.rpm)
+                rpm2cpio "$package_file" | cpio -i --to-stdout ./usr/bin/rustfs > "$RUNNER_TEMP/wrapped-rustfs"
+                ;;
+            esac
+            python3 scripts/release/release_gate.py verify-promotion \
+              --plan .config/release-validation-plan.json \
+              --candidate certification/candidate.json \
+              --summary certification/release-gate-output \
+              --artifact "$RUNNER_TEMP/wrapped-rustfs" \
+              --artifact-role binary
+          done
+
       - name: Upload packages to artifacts
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
@@ -531,10 +627,14 @@ jobs:
           export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
           export AWS_DEFAULT_REGION="auto"
 
+          # Build-phase writes stay on staging for release types: the official
+          # packages/release and packages/latest paths are written only by the
+          # certification-gated promote-packages job, from these same staged
+          # objects. Development packages keep the uncertified dev channel.
           if [[ "$BUILD_TYPE" == "development" ]]; then
             R2_PREFIX="artifacts/rustfs/packages/dev"
           else
-            R2_PREFIX="artifacts/rustfs/packages/release"
+            R2_PREFIX="artifacts/rustfs/packages/staging/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}"
           fi
           R2_PATH="s3://${R2_BUCKET}/${R2_PREFIX}/"
 
@@ -549,18 +649,6 @@ jobs:
 
           echo "✅ Upload complete"
 
-          # Also upload as latest for release/prerelease
-          if [[ "$BUILD_TYPE" == "release" || "$BUILD_TYPE" == "prerelease" ]]; then
-            LATEST_PATH="s3://${R2_BUCKET}/artifacts/rustfs/packages/latest/"
-            for f in "$DEB_FILE" "$RPM_FILE"; do
-              if [[ -n "$f" && -f "$f" ]]; then
-                echo "Uploading latest: $(basename "$f")"
-                aws s3 cp "$f" "$LATEST_PATH" --endpoint-url "$R2_ENDPOINT" --only-show-errors
-              fi
-            done
-            echo "✅ Latest packages updated"
-          fi
-
       - name: Upload packages to GitHub Release
         if: needs.resolve.outputs.tag != ''
         env:
@@ -614,6 +702,114 @@ jobs:
 
           echo "✅ GitHub release assets updated"
 
+  # The only writer of the official packages/release and packages/latest R2
+  # paths. Promotion re-checks each package's wrapped binary against the
+  # passed certification list and copies the staged objects byte-for-byte:
+  # no rebuild-after-verify, no mutable latest resolution downstream.
+  promote-packages:
+    name: Promote Packages
+    needs: [ resolve, package, release-certification ]
+    # Target-channel lock for the official package paths.
+    concurrency:
+      group: release-channel-r2-packages
+      cancel-in-progress: false
+    if: needs.resolve.outputs.tag != '' && (needs.resolve.outputs.build_type == 'release' || needs.resolve.outputs.build_type == 'prerelease')
+    runs-on: ubuntu-latest # gh CLI required; the sm-standard-2 fleet does not ship it
+    timeout-minutes: 30
+    permissions:
+      contents: read
+      actions: read
+    env:
+      R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+      R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+      R2_BUCKET: ${{ secrets.R2_BUCKET }}
+      AWS_EC2_METADATA_DISABLED: true
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Download packages and certification
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          path: ./packages
+          pattern: packages-*
+          merge-multiple: true
+
+      - name: Download certification evidence
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: release-certification
+          path: ./certification
+
+      - name: Verify and promote staged packages
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if [[ -z "$R2_ACCESS_KEY_ID" || -z "$R2_SECRET_ACCESS_KEY" || -z "$R2_ENDPOINT" || -z "$R2_BUCKET" ]]; then
+            echo "❌ R2 credentials missing; refusing to promote to the official channel" >&2
+            exit 1
+          fi
+
+          STAGING_PREFIX="s3://${R2_BUCKET}/artifacts/rustfs/packages/staging/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}/"
+          RELEASE_PATH="s3://${R2_BUCKET}/artifacts/rustfs/packages/release/"
+          LATEST_PATH="s3://${R2_BUCKET}/artifacts/rustfs/packages/latest/"
+
+          if ! command -v aws >/dev/null 2>&1; then
+            sudo apt-get update && sudo apt-get install -y awscli
+          fi
+          export AWS_ACCESS_KEY_ID="$R2_ACCESS_KEY_ID"
+          export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
+          export AWS_DEFAULT_REGION="auto"
+
+          if ! command -v rpm2cpio >/dev/null 2>&1; then
+            sudo apt-get install -y rpm2cpio
+          fi
+          if ! command -v cpio >/dev/null 2>&1; then
+            sudo apt-get install -y cpio
+          fi
+
+          promoted=0
+          for package_file in packages/*.deb packages/*.rpm; do
+            [[ -f "$package_file" ]] || continue
+            name="$(basename "$package_file")"
+            case "$package_file" in
+              *.deb)
+                dpkg-deb --fsys-tarfile "$package_file" | tar -xOf - ./usr/bin/rustfs > "$RUNNER_TEMP/wrapped-rustfs"
+                ;;
+              *.rpm)
+                rpm2cpio "$package_file" | cpio -i --to-stdout ./usr/bin/rustfs > "$RUNNER_TEMP/wrapped-rustfs"
+                ;;
+            esac
+            python3 scripts/release/release_gate.py verify-promotion \
+              --plan .config/release-validation-plan.json \
+              --candidate certification/candidate.json \
+              --summary certification/release-gate-output \
+              --artifact "$RUNNER_TEMP/wrapped-rustfs" \
+              --artifact-role binary
+
+            staged="$(mktemp)"
+            aws s3 cp "${STAGING_PREFIX}${name}" "$staged" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+            staged_sha="$(sha256sum "$staged" | awk '{print $1}')"
+            built_sha="$(sha256sum "$package_file" | awk '{print $1}')"
+            if [[ "$staged_sha" != "$built_sha" ]]; then
+              echo "❌ Staged package ${name} does not match the verified bytes" >&2
+              exit 1
+            fi
+            aws s3 cp "$staged" "${RELEASE_PATH}${name}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+            aws s3 cp "$staged" "${LATEST_PATH}${name}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+            rm -f "$staged"
+            promoted=$((promoted + 1))
+            echo "✅ Promoted ${name} (${built_sha})"
+          done
+          if [[ "$promoted" -eq 0 ]]; then
+            echo "No staged packages found to promote" >&2
+            exit 1
+          fi
+
   # Summary
   summary:
     name: Summary

ARCHITECTURE.md

@@ -155,7 +155,7 @@ default build (lifecycle:
    - ⚠️ PARTIALLY VIOLATED: serving-side `s3s` references remain in ecstore
      (bucket metadata/replication/lifecycle DTOs and error mapping). The
      count is ratcheted shrink-only by `scripts/check_s3s_footprint.sh`
-     (the `ecstore_files` line of `.config/s3s-footprint-baseline.txt`; the `object_lock` module was converted to
+     (`S3S_ECSTORE_FILES_BASELINE`; the `object_lock` module was converted to
      storage-level types as the first ratchet step). Target state: the
      baseline reaches zero and ecstore's `Cargo.toml` drops `s3s`.
 

Cargo.lock

@@ -1321,9 +1321,9 @@ dependencies = [
 
 [[package]]
 name = "aws-smithy-schema"
-version = "1.1.0"
+version = "1.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bec8dae1015a6d959a5faf86fbdf77486cfe19e4ccb6ee009d8a46efc45028a4"
+checksum = "9543fc722ed4865ab29fb66bd7ed0bef7b029d265ba8c587144511e30006d431"
 dependencies = [
  "aws-smithy-runtime-api",
  "aws-smithy-types",
@@ -4198,9 +4198,9 @@ checksum = "34aa73646ffb006b8f5147f3dc182bd4bcb190227ce861fc4a4844bf8e3cb2c0"
 
 [[package]]
 name = "encoding_rs"
-version = "0.8.42"
+version = "0.8.44"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8e985e0451871ad22fb8d2b6b076e2028a502a0d3950998c2c5c0a4f9b5d9679"
+checksum = "89253175e704ab869b04f9db224d38630e0211a53841a79181deae0889e169f5"
 dependencies = [
  "cfg-if",
  "core_detect",
@@ -5277,9 +5277,9 @@ dependencies = [
 
 [[package]]
 name = "hotpath"
-version = "0.30.1"
+version = "0.30.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "502dc37a7c218b9b1d3a0ccfb904da81d10d6dda0e899c722c503092d55576ea"
+checksum = "b39e941010d86063ff233bf8325269bf68097fb24ba67000fbddc0a410e8056e"
 dependencies = [
  "arc-swap",
  "async-channel",
@@ -5312,15 +5312,15 @@ dependencies = [
 
 [[package]]
 name = "hotpath-drain"
-version = "0.30.1"
+version = "0.30.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0ea17d04c231b90b1f68a91f6f80fb0762c5dea07c611dce9ae157b56a50ff4c"
+checksum = "d0faae1a9e1fd8ce5debaaaf452dcbe833fc8a8a4f3779d2ec8828b116c3c16a"
 
 [[package]]
 name = "hotpath-macros"
-version = "0.30.1"
+version = "0.30.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a76e0af911930fc97abc766b393b48cdca77fea9f5f6b388cf2123a6d9c30deb"
+checksum = "a24f5179ff680132003306354efd0ac8cde12e87b1a788719c836e91af35cb58"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -5899,9 +5899,9 @@ dependencies = [
 
 [[package]]
 name = "js-sys"
-version = "0.3.106"
+version = "0.3.107"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7883d941dae510fb2d978fc3fe018c71c9e2892fd38854de3e8b92c2e5ad9cc5"
+checksum = "8dc09d2154098bbf4ef54eab6b117a3c24cbbffe94046b16ee6d6d4510584d3a"
 dependencies = [
  "cfg-if",
  "futures-util",
@@ -6094,9 +6094,9 @@ dependencies = [
 
 [[package]]
 name = "libbz2-rs-sys"
-version = "0.2.5"
+version = "0.2.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "34b357333733e8260735ba5894eb928c02ecc69c78715f01a8019e7fa7f2db4c"
+checksum = "32f454fd6831a29c0e827ed075bca2a15be201bd57f4168f0fd1041124eab17e"
 
 [[package]]
 name = "libc"
@@ -8305,9 +8305,9 @@ dependencies = [
 
 [[package]]
 name = "pulldown-cmark"
-version = "0.13.5"
+version = "0.13.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "278e9268c93b9e498a2da46c34d90ba93a3255eb99009646398f34bc73ab36eb"
+checksum = "3db2be46835f013dd3f4fa7f5f7402c537787b54ebb6fe2b94bbdd7e2669bb7b"
 dependencies = [
  "bitflags 2.13.2",
  "memchr",
@@ -8520,15 +8520,6 @@ version = "1.2.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a1d01941d82fa2ab50be1e79e6714289dd7cde78eba4c074bc5a4374f650dfe0"
 
-[[package]]
-name = "quick-xml"
-version = "0.41.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e660451e55124f798a69a5af3f49ccfbefbd41910eefd25caf2393e1f3473ec1"
-dependencies = [
- "memchr",
-]
-
 [[package]]
 name = "quick-xml"
 version = "0.42.0"
@@ -8832,9 +8823,9 @@ dependencies = [
 
 [[package]]
 name = "redis"
-version = "1.8.0"
+version = "1.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7acd809e80afa754c63f586955699d8990efe1c4b6613183ee5f027a11fbc8b5"
+checksum = "db7f3d75b5a3301746bff39a767bf4c5c42506b08513527308b3875bfb64a031"
 dependencies = [
  "arc-swap",
  "arcstr",
@@ -9393,7 +9384,7 @@ dependencies = [
  "proptest",
  "prost 0.14.4",
  "pyroscope",
- "quick-xml 0.42.0",
+ "quick-xml",
  "quinn",
  "rand 0.10.3",
  "rcgen",
@@ -9412,7 +9403,6 @@ dependencies = [
  "rustfs-ecstore",
  "rustfs-extension-schema",
  "rustfs-filemeta",
- "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-iam",
@@ -9691,7 +9681,6 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-erasure-codec",
  "rustfs-filemeta",
- "rustfs-gateway-types",
  "rustfs-heal-contracts",
  "rustfs-io-metrics",
  "rustfs-lifecycle",
@@ -9797,48 +9786,6 @@ dependencies = [
  "xxhash-rust",
 ]
 
-[[package]]
-name = "rustfs-gateway-stream"
-version = "0.2.0"
-source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
-dependencies = [
- "bitflags 2.13.2",
- "bytes",
- "http 1.5.0",
- "http-body",
-]
-
-[[package]]
-name = "rustfs-gateway-types"
-version = "0.32.0+aws.2026-10-02"
-source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
-dependencies = [
- "bytes",
- "crc-fast",
- "futures-core",
- "hex",
- "http 1.5.0",
- "md-5 0.11.0",
- "percent-encoding",
- "rustfs-gateway-stream",
- "rustfs-gateway-xml",
- "s3s",
- "sha1 0.11.0",
- "sha2 0.11.1",
- "time",
- "twox-hash",
- "unicode-normalization",
- "zeroize",
-]
-
-[[package]]
-name = "rustfs-gateway-xml"
-version = "0.4.0"
-source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
-dependencies = [
- "quick-xml 0.41.0",
-]
-
 [[package]]
 name = "rustfs-heal"
 version = "1.0.2"
@@ -10109,10 +10056,10 @@ dependencies = [
  "metrics-util",
  "proptest",
  "rustfs-config",
- "rustfs-gateway-types",
  "rustfs-replication",
  "rustfs-scanner-metrics",
  "rustfs-storage-api",
+ "s3s",
  "serial_test",
  "temp-env",
  "time",
@@ -10175,6 +10122,7 @@ dependencies = [
  "reqwest",
  "rmp-serde",
  "rustfs-signer",
+ "s3s",
  "serde",
  "serde_json",
  "sysinfo",
@@ -10214,7 +10162,7 @@ dependencies = [
  "jiff",
  "metrics",
  "percent-encoding",
- "quick-xml 0.42.0",
+ "quick-xml",
  "rayon",
  "rustc-hash",
  "rustfs-config",
@@ -10395,13 +10343,10 @@ dependencies = [
  "rustfs-config",
  "rustfs-credentials",
  "rustfs-ecstore",
- "rustfs-gateway-stream",
- "rustfs-gateway-types",
  "rustfs-iam",
  "rustfs-keystone",
  "rustfs-policy",
  "rustfs-rio",
- "rustfs-s3-types",
  "rustfs-storage-api",
  "rustfs-test-utils",
  "rustfs-tls-runtime",
@@ -10409,13 +10354,13 @@ dependencies = [
  "rustfs-trusted-proxies",
  "rustfs-utils",
  "rustls",
+ "s3s",
  "serde",
  "serde_json",
  "sha1 0.11.0",
  "sha2 0.11.1",
  "socket2",
  "subtle",
- "sync_wrapper",
  "tempfile",
  "thiserror 2.0.21",
  "time",
@@ -10465,7 +10410,7 @@ dependencies = [
  "regex",
  "rmp",
  "rmp-serde",
- "rustfs-gateway-types",
+ "s3s",
  "serde",
  "time",
  "tokio",
@@ -10560,13 +10505,11 @@ dependencies = [
  "hyper-util",
  "lazy_static",
  "md-5 0.11.0",
- "quick-xml 0.42.0",
+ "quick-xml",
  "rand 0.10.3",
  "rustfs-checksums",
  "rustfs-config",
- "rustfs-gateway-types",
  "rustfs-rio",
- "rustfs-s3-types",
  "rustfs-signer",
  "rustfs-storage-api",
  "rustfs-tls-runtime",
@@ -10600,19 +10543,9 @@ dependencies = [
 name = "rustfs-s3-types"
 version = "1.0.2"
 dependencies = [
- "bytes",
- "futures-core",
  "hotpath",
- "http 1.5.0",
- "http-body",
- "http-body-util",
- "hyper",
- "hyper-util",
- "pin-project-lite",
- "s3s",
  "serde",
  "serde_json",
- "tokio",
 ]
 
 [[package]]
@@ -10636,10 +10569,9 @@ dependencies = [
  "parking_lot 0.13.0",
  "rustfs-common",
  "rustfs-ecstore",
- "rustfs-gateway-types",
- "rustfs-s3-types",
  "rustfs-storage-api",
  "rustfs-test-utils",
+ "s3s",
  "serde_json",
  "serial_test",
  "thiserror 2.0.21",
@@ -10665,6 +10597,7 @@ dependencies = [
  "parking_lot 0.13.0",
  "rustfs-s3select-api",
  "rustfs-test-utils",
+ "s3s",
  "serde",
  "serde_json",
  "tokio",
@@ -10696,7 +10629,6 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-ecstore",
  "rustfs-filemeta",
- "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-lifecycle",
@@ -10754,10 +10686,9 @@ dependencies = [
  "bytes",
  "hotpath",
  "http 1.5.0",
- "http-body",
  "hyper",
- "rustfs-s3-types",
  "rustfs-utils",
+ "s3s",
  "serde_urlencoded",
  "thiserror 2.0.21",
  "time",
@@ -10815,6 +10746,7 @@ dependencies = [
  "rustfs-utils",
  "rustls",
  "rustls-native-certs",
+ "s3s",
  "serde",
  "serde_json",
  "sha2 0.11.1",
@@ -11205,7 +11137,7 @@ dependencies = [
  "numeric_cast",
  "phf 0.14.0",
  "pin-project-lite",
- "quick-xml 0.42.0",
+ "quick-xml",
  "regex",
  "s3s-multipart",
  "s3s-rfc2047",
@@ -12285,9 +12217,9 @@ dependencies = [
 
 [[package]]
 name = "sysinfo"
-version = "0.40.0"
+version = "0.40.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "110e37d98880fd804cd47509b90a8444f7aaaab655ae625e83cd442a8bc2ebf2"
+checksum = "0e8a3edbe0394ce115f3917a00349dd242be09eb5a091ed509eaf67b55cfb8a8"
 dependencies = [
  "libc",
  "memchr",
@@ -12778,9 +12710,9 @@ checksum = "06bdbd8cfc056b8d2e2e85f29b56a3bdbecb527cef81eb39e3e7b98af4652770"
 
 [[package]]
 name = "tonic"
-version = "0.14.8"
+version = "0.14.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cd1fa9b1a51188f3266a823208078315811269d06d272c3a61be0b9a194194ff"
+checksum = "e2cdfc673aabd30fe8fee92e65837b644c37c5d47f0f97ae6d5e0f1237d15e1b"
 dependencies = [
  "async-trait",
  "axum",
@@ -12811,9 +12743,9 @@ dependencies = [
 
 [[package]]
 name = "tonic-build"
-version = "0.14.8"
+version = "0.14.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2c07bd4e9f99e2a3ed8bebad596ae02a94e53745b68acc6900e22c3475c8bc44"
+checksum = "ac6d916d206ed7bdda1778ddff9bda43303789c258877e7bc22fb44ee5c984b4"
 dependencies = [
  "prettyplease 0.2.37",
  "proc-macro2",
@@ -12823,9 +12755,9 @@ dependencies = [
 
 [[package]]
 name = "tonic-prost"
-version = "0.14.8"
+version = "0.14.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f63c616ffaa648f7c5dd5657e97d9a285fcb44ff3b70da761d436f5a13ff7be6"
+checksum = "eba464054e2226f182a34489b8b079bf4c6e3e7ab5fd7fc30e024f24aa92a484"
 dependencies = [
  "bytes",
  "prost 0.14.4",
@@ -12834,9 +12766,9 @@ dependencies = [
 
 [[package]]
 name = "tonic-prost-build"
-version = "0.14.8"
+version = "0.14.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4a05d8957fac6b3542999796cb4449d45da790751962f974a457e5e00c43d78c"
+checksum = "2ea216ddc448e08e7839ab09e20e60b06b6a1c32745ccb212c0ebd161c199f1b"
 dependencies = [
  "prettyplease 0.2.37",
  "proc-macro2",
@@ -13229,9 +13161,12 @@ checksum = "daf8dba3b7eb870caf1ddeed7bc9d2a049f3cfdfae7cb521b087cc33ae4c49da"
 
 [[package]]
 name = "utf8_iter"
-version = "1.0.4"
+version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b6c140620e7ffbb22c2dee59cafe6084a59b5ffc27a8859a5f0d494b5d52b6be"
+checksum = "af3bd9d3e571c26fcce9d033b87ea5ab061b8d2ffe9b7a5675d810d68f03edf0"
+dependencies = [
+ "utf_types",
+]
 
 [[package]]
 name = "utf8parse"
@@ -13239,6 +13174,12 @@ version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "06abde3611657adf66d383f00b093d7faecc7fa57071cce2578660c9f1010821"
 
+[[package]]
+name = "utf_types"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a3b425286980a367278453f3b6d28cdcb134b574e5a35824a6ff5eebad50bb6"
+
 [[package]]
 name = "uuid"
 version = "1.29.0"
@@ -13353,12 +13294,11 @@ dependencies = [
 
 [[package]]
 name = "wasm-bindgen"
-version = "0.2.129"
+version = "0.2.130"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9bb54f33acc68fd454578d9820b0bde1a1a3d17aa17bb7b6595806d02886d409"
+checksum = "a0181c14ddd6c7b8b0b0132a5f1ea2f32be0d893cd3ea97b6e3b28886b8a3d1c"
 dependencies = [
  "cfg-if",
- "once_cell",
  "rustversion",
  "wasm-bindgen-macro",
  "wasm-bindgen-shared",
@@ -13366,9 +13306,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-bindgen-futures"
-version = "0.4.79"
+version = "0.4.80"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3cbab34de2d982e9b48e18d216d04c4a6f641066ff19ffb699980f591ee3610e"
+checksum = "5e14e0327cb5368ad60ccb22d04a9827a693c95a926d8d125e9fa39e0baec501"
 dependencies = [
  "js-sys",
  "tokio",
@@ -13377,9 +13317,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-bindgen-macro"
-version = "0.2.129"
+version = "0.2.130"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2e29d0c35b16e224a7eeb5cd2d25e3e1968fbd65604117b44d3b789d00ee8535"
+checksum = "6e931a02a39abdbf66966fc35902b5ed58344750a70885bed4ebb3d8eaf28ed9"
 dependencies = [
  "quote",
  "wasm-bindgen-macro-support",
@@ -13387,9 +13327,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-bindgen-macro-support"
-version = "0.2.129"
+version = "0.2.130"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6f501a8bc3719dba86ef8ae4728879c08001bea749eb1333ac5b91e040e2a6b7"
+checksum = "2ca4b54280e455bd831e33a6b8e734baf432fe63fbc35c42ac14ce67a043a53a"
 dependencies = [
  "bumpalo",
  "proc-macro2",
@@ -13400,9 +13340,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-bindgen-shared"
-version = "0.2.129"
+version = "0.2.130"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "23f0c9c52aa7cd7d77769a4cfe2a9adb1b331f489a41d912ce14513d5ab995c6"
+checksum = "3b1f9c3798725016c6ee5f2f57869ec0831e8bbc0ab0310a86e787e97f41389e"
 dependencies = [
  "unicode-ident",
 ]
@@ -13422,9 +13362,9 @@ dependencies = [
 
 [[package]]
 name = "web-sys"
-version = "0.3.106"
+version = "0.3.107"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "88261b9deccee56594c11a3460c462c41f58d148598fe70ad77070126a68aba4"
+checksum = "d7d179596ff5ec0f6de4db64d54011bde6289cc49603d3a6e4ce3f329207ab88"
 dependencies = [
  "js-sys",
  "wasm-bindgen",

Cargo.toml

@@ -48,7 +48,7 @@ members = [
     "crates/replication", # Replication contracts and wire formats
     "crates/concurrency", # Concurrency management for RustFS - timeout, locking, backpressure, and I/O scheduling
     "crates/s3-client", # S3 client for engine-side consumption of remote S3 endpoints (tiering, transition targets)
-    "crates/s3-types", # S3 event and error type definitions
+    "crates/s3-types", # S3 event type definitions
     "crates/s3-ops", # S3 operation definitions and mapping
     "crates/s3select-api", # S3 Select API interface
     "crates/s3select-query", # S3 Select query engine
@@ -143,13 +143,6 @@ rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.2" }
 rustfs-utils = { path = "crates/utils", version = "1.0.2" }
 rustfs-zip = { path = "./crates/zip", version = "1.0.2" }
 
-# The gateway's S3 types: the DTOs and persisted bucket-configuration codecs the
-# engine crates move onto from s3s (rustfs/backlog#2745). The one pin of the
-# gateway in this workspace; no crate may enable a `compat-s3s*` feature except
-# the s3s-facing edge of the rustfs binary.
-rustfs-gateway-types = { git = "https://github.com/rustfs/gateway.git", rev = "21283e874c2381bab611e9347acbc5b92e9e2b3c", version = "0.32.0", default-features = false }
-rustfs-gateway-stream = { git = "https://github.com/rustfs/gateway.git", rev = "21283e874c2381bab611e9347acbc5b92e9e2b3c", version = "0.2.0", default-features = false }
-
 # Async Runtime and Networking
 async-channel = "2.5.0"
 async_zip = { default-features = false, version = "0.0.19" }
@@ -182,9 +175,9 @@ tokio-rustls = { default-features = false, version = "0.26.6" }
 tokio-stream = { version = "0.1.19" }
 tokio-test = "0.4.6"
 tokio-util = { version = "0.7.20" }
-tonic = { version = "0.14.8" }
-tonic-prost = { version = "0.14.8" }
-tonic-prost-build = { version = "0.14.8" }
+tonic = { version = "0.14.9" }
+tonic-prost = { version = "0.14.9" }
+tonic-prost-build = { version = "0.14.9" }
 tower = { version = "0.5.3" }
 tower-http = { version = "0.7.1" }
 h3 = "0.0.8"
@@ -326,17 +319,13 @@ rustfs-erasure-codec = { version = "9.0.0" }
 reed-solomon-simd = "3.1.0"
 regex = { version = "1.13.1" }
 rumqttc = { package = "rumqttc-next", version = "0.34.0" }
-redis = { version = "1.8.0" }
+redis = { version = "1.8.1" }
 rustify = { version = "0.7", default-features = false }
 rustix = { version = "1.1.6" }
 rust-embed = { version = "8.13.0" }
 rustc-hash = { version = "2.1.3" }
 s3s = { git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0", features = ["minio"] }
 s3s-sigv4 = { git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0" }
-# The same s3s pin under the name the legacy edge of the gateway migration uses
-# (crates/s3-types, feature `compat-s3s`). Keep `rev` identical to `s3s` above.
-# DELETE BY T4.2 (rustfs/backlog#2784).
-legacy-s3s = { package = "s3s", git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0", features = ["minio"] }
 serial_test = "4.0.1"
 shadow-rs = { default-features = false, version = "2.0.0" }
 siphasher = "1.0.4"
@@ -345,8 +334,7 @@ compact_str = "0.10.0"
 snap = "1.1.2"
 starshard = { version = "2.4.0" }
 strum = { version = "0.28.0" }
-sync_wrapper = "1.0.2"
-sysinfo = "0.40.0"
+sysinfo = "0.40.1"
 temp-env = "0.3.6"
 tempfile = "3.27.0"
 test-case = "3.4.0"
@@ -398,7 +386,7 @@ dav-server = "0.11.0"
 # Performance Analysis and Memory Profiling
 rustfs-mimalloc = { version = "0.7.2" }
 # Preserve Unicode focus filters until rustfs/backlog#2302 is resolved.
-hotpath = { version = "0.30.1", default-features = false }
+hotpath = { version = "0.30.2", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.49" }
 

crates/e2e_test/src/anonymous_access_test.rs

@@ -17,8 +17,7 @@
 //! when PublicAccessBlock configuration is missing or explicitly set.
 
 use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
-use crate::sts_query_compat_test::{OPA_AUTH_TOKEN, OpaMock};
-use aws_sdk_s3::types::{PublicAccessBlockConfiguration, Tag, Tagging};
+use aws_sdk_s3::types::PublicAccessBlockConfiguration;
 use tracing::info;
 
 async fn setup_public_bucket(
@@ -70,98 +69,6 @@ async fn anonymous_get_object(
     local_http_client().get(&url).send().await
 }
 
-#[tokio::test]
-async fn test_plain_http_matches_secure_transport_false_bucket_policy() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    init_logging();
-
-    let mut env = RustFSTestEnvironment::new().await?;
-    env.start_rustfs_server(vec![]).await?;
-    assert!(env.url.starts_with("http://"), "fixture must exercise plaintext HTTP: {}", env.url);
-
-    let bucket = "secure-transport-policy";
-    let key = "plain-http.txt";
-    let admin = env.create_s3_client();
-    admin.create_bucket().bucket(bucket).send().await?;
-    admin
-        .put_object()
-        .bucket(bucket)
-        .key(key)
-        .body(aws_sdk_s3::primitives::ByteStream::from_static(b"plain http"))
-        .send()
-        .await?;
-    admin
-        .put_bucket_policy()
-        .bucket(bucket)
-        .policy(
-            serde_json::json!({
-                "Version": "2012-10-17",
-                "Statement": [{
-                    "Effect": "Allow",
-                    "Principal": "*",
-                    "Action": "s3:GetObject",
-                    "Resource": format!("arn:aws:s3:::{bucket}/*"),
-                    "Condition": {"Bool": {"aws:SecureTransport": "false"}}
-                }]
-            })
-            .to_string(),
-        )
-        .send()
-        .await?;
-
-    let response = anonymous_get_object(&env, bucket, key).await?;
-    assert_eq!(response.status().as_u16(), 200);
-    assert_eq!(response.bytes().await?.as_ref(), b"plain http");
-    Ok(())
-}
-
-/// Root credentials have no attached IAM policy, so these requests depend on
-/// the owner flag reaching policy evaluation unchanged.
-#[tokio::test]
-async fn test_root_requests_keep_policy_owner_flag() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    init_logging();
-
-    let opa = OpaMock::start().await?;
-    let mut env = RustFSTestEnvironment::new().await?;
-    env.start_rustfs_server_with_env(
-        vec![],
-        &[
-            ("RUSTFS_POLICY_PLUGIN_URL", opa.url.as_str()),
-            ("RUSTFS_POLICY_PLUGIN_AUTH_TOKEN", OPA_AUTH_TOKEN),
-        ],
-    )
-    .await?;
-
-    let bucket = "root-owner-flag";
-    let key = "owner.txt";
-    let client = env.create_s3_client();
-    client.create_bucket().bucket(bucket).send().await?;
-    client
-        .put_object()
-        .bucket(bucket)
-        .key(key)
-        .body(aws_sdk_s3::primitives::ByteStream::from_static(b"owner access"))
-        .send()
-        .await?;
-
-    // Duplicate stored tags are ambiguous. An owner request must not enter the
-    // OPA tag-loading path, which rejects this metadata before evaluation.
-    client
-        .put_bucket_tagging()
-        .bucket(bucket)
-        .tagging(
-            Tagging::builder()
-                .tag_set(Tag::builder().key("department").value("finance").build()?)
-                .tag_set(Tag::builder().key("department").value("engineering").build()?)
-                .build()?,
-        )
-        .send()
-        .await?;
-    let object = client.get_object().bucket(bucket).key(key).send().await?;
-
-    assert_eq!(object.body.collect().await?.into_bytes().as_ref(), b"owner access");
-    Ok(())
-}
-
 /// Issue #2036: Anonymous GetObject should succeed when bucket policy allows it
 /// and no PublicAccessBlock configuration exists (ConfigNotFound).
 #[tokio::test]

crates/e2e_test/src/delete_marker_migration_semantics_test.rs

@@ -16,8 +16,193 @@
 mod tests {
     use crate::common::{RustFSTestEnvironment, init_logging};
     use aws_sdk_s3::Client;
+    use aws_sdk_s3::config::http::HttpResponse;
+    use aws_sdk_s3::error::{ProvideErrorMetadata, SdkError};
     use aws_sdk_s3::primitives::ByteStream;
-    use aws_sdk_s3::types::{BucketVersioningStatus, VersioningConfiguration};
+    use aws_sdk_s3::types::{BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, VersioningConfiguration};
+
+    fn assert_not_found<E: ProvideErrorMetadata + std::fmt::Debug>(error: &SdkError<E, HttpResponse>) {
+        assert_eq!(error.raw_response().map(|r| r.status().as_u16()), Some(404), "{error:?}");
+        assert_eq!(
+            error.as_service_error().and_then(ProvideErrorMetadata::code),
+            Some("NoSuchKey"),
+            "{error:?}"
+        );
+    }
+
+    #[tokio::test]
+    async fn conditional_writes_cannot_replace_delete_marker_with_if_match() {
+        init_logging();
+        let mut env = RustFSTestEnvironment::new().await.expect("create test environment");
+        env.start_rustfs_server(vec![]).await.expect("start RustFS");
+        let client = env.create_s3_client();
+        let bucket = "conditional-delete-marker-proof";
+        create_versioned_bucket(&client, bucket).await;
+        client
+            .put_object()
+            .bucket(bucket)
+            .key("source")
+            .body(ByteStream::from_static(b"source"))
+            .send()
+            .await
+            .expect("write copy source");
+
+        for operation in ["put", "copy", "multipart"] {
+            let old = client
+                .put_object()
+                .bucket(bucket)
+                .key(operation)
+                .body(ByteStream::from_static(b"old"))
+                .send()
+                .await
+                .expect("write historical version");
+            let old_etag = old.e_tag().expect("historical ETag");
+            let marker = client
+                .delete_object()
+                .bucket(bucket)
+                .key(operation)
+                .send()
+                .await
+                .expect("create delete marker");
+            let marker_id = marker.version_id().expect("delete marker version");
+
+            let (upload_id, completed) = if operation == "multipart" {
+                let upload = client
+                    .create_multipart_upload()
+                    .bucket(bucket)
+                    .key(operation)
+                    .send()
+                    .await
+                    .expect("initiate multipart upload");
+                let upload_id = upload.upload_id().expect("upload ID").to_owned();
+                let part = client
+                    .upload_part()
+                    .bucket(bucket)
+                    .key(operation)
+                    .upload_id(&upload_id)
+                    .part_number(1)
+                    .body(ByteStream::from_static(b"new"))
+                    .send()
+                    .await
+                    .expect("upload part");
+                let completed = CompletedMultipartUpload::builder()
+                    .parts(
+                        CompletedPart::builder()
+                            .part_number(1)
+                            .set_e_tag(part.e_tag().map(str::to_owned))
+                            .build(),
+                    )
+                    .build();
+                (Some(upload_id), Some(completed))
+            } else {
+                (None, None)
+            };
+
+            for etag in [old_etag, "\"00000000000000000000000000000000\""] {
+                match operation {
+                    "put" => {
+                        let err = client
+                            .put_object()
+                            .bucket(bucket)
+                            .key(operation)
+                            .if_match(etag)
+                            .body(ByteStream::from_static(b"new"))
+                            .send()
+                            .await
+                            .expect_err("If-Match must reject marker");
+                        assert_not_found(&err);
+                    }
+                    "copy" => {
+                        let err = client
+                            .copy_object()
+                            .bucket(bucket)
+                            .key(operation)
+                            .copy_source(format!("{bucket}/source"))
+                            .if_match(etag)
+                            .send()
+                            .await
+                            .expect_err("If-Match copy must reject marker");
+                        assert_not_found(&err);
+                    }
+                    _ => {
+                        let err = client
+                            .complete_multipart_upload()
+                            .bucket(bucket)
+                            .key(operation)
+                            .upload_id(upload_id.as_deref().expect("upload ID"))
+                            .multipart_upload(completed.clone().expect("completed parts"))
+                            .if_match(etag)
+                            .send()
+                            .await
+                            .expect_err("If-Match completion must reject marker");
+                        assert_not_found(&err);
+                    }
+                }
+                let versions = client
+                    .list_object_versions()
+                    .bucket(bucket)
+                    .prefix(operation)
+                    .send()
+                    .await
+                    .expect("list versions after rejected write");
+                assert_eq!(versions.versions().len(), 1, "{operation}: failed write created a version");
+                assert_eq!(versions.delete_markers().len(), 1);
+                assert_eq!(versions.delete_markers()[0].version_id(), Some(marker_id));
+                assert_eq!(versions.delete_markers()[0].is_latest(), Some(true));
+            }
+
+            match operation {
+                "put" => {
+                    client
+                        .put_object()
+                        .bucket(bucket)
+                        .key(operation)
+                        .if_none_match("*")
+                        .body(ByteStream::from_static(b"new"))
+                        .send()
+                        .await
+                        .expect("If-None-Match PUT over marker");
+                }
+                "copy" => {
+                    client
+                        .copy_object()
+                        .bucket(bucket)
+                        .key(operation)
+                        .copy_source(format!("{bucket}/source"))
+                        .if_none_match("*")
+                        .send()
+                        .await
+                        .expect("If-None-Match copy over marker");
+                }
+                _ => {
+                    client
+                        .complete_multipart_upload()
+                        .bucket(bucket)
+                        .key(operation)
+                        .upload_id(upload_id.as_deref().expect("upload ID"))
+                        .multipart_upload(completed.expect("completed parts"))
+                        .if_none_match("*")
+                        .send()
+                        .await
+                        .expect("If-None-Match completion over marker");
+                }
+            }
+            let current = client
+                .get_object()
+                .bucket(bucket)
+                .key(operation)
+                .send()
+                .await
+                .expect("read successful conditional write")
+                .body
+                .collect()
+                .await
+                .expect("read object body")
+                .into_bytes();
+            let expected: &[u8] = if operation == "copy" { b"source" } else { b"new" };
+            assert_eq!(current.as_ref(), expected, "{operation}: unexpected current object");
+        }
+    }
 
     async fn create_versioned_bucket(client: &Client, bucket: &str) {
         client

crates/e2e_test/src/distributed/scanner_peer_bucket_test.rs

@@ -1636,20 +1636,53 @@ async fn hot_window_with_old_followers(
     let mixed_initial_physical = if mixed { json!(converged_shards(&run)?) } else { Value::Null };
     let scheduler_config = run.assert_default_cadence().await?;
     let baseline = run.quiet_baseline().await?;
-    let leader = run.leader().await?;
-    let leader_pid = run.pid(leader)?;
+    // A leader-lock refresh gap during old-binary replacement can make
+    // `leader()` briefly observe a different node. Poll until the coordinator
+    // is stable before comparing against the measured fleet. A persistent
+    // change to an old-binary node is a genuine election outcome and fails.
+    let (leader, leader_pid) = if mixed {
+        let expected_leader = usize::try_from(
+            mixed_fleet["leader"]
+                .as_u64()
+                .ok_or("mixed fleet omitted its measured leader")?,
+        )?;
+        let expected_pid = mixed_fleet["leader_pid"]
+            .as_u64()
+            .ok_or("mixed fleet omitted its measured leader pid")?;
+        let mut resolved = None;
+        for _ in 0..20 {
+            let candidate = run.leader().await?;
+            let candidate_pid = u64::from(run.pid(candidate)?);
+            if Some(u64::try_from(candidate)?) == mixed_fleet["leader"].as_u64() && Some(candidate_pid) == Some(expected_pid) {
+                resolved = Some((candidate, run.pid(candidate)?));
+                break;
+            }
+            tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+        }
+        match resolved {
+            Some(pair) => pair,
+            None => {
+                let current = match run.leader().await {
+                    Ok(node) => Some((node, run.pid(node).ok())),
+                    Err(_) => None,
+                };
+                return Err(format!(
+                    "the hot window did not retain its measured new coordinator after stabilization \
+                     (expected node {expected_leader} pid {expected_pid}, observed {current:?})"
+                )
+                .into());
+            }
+        }
+    } else {
+        let leader = run.leader().await?;
+        (leader, run.pid(leader)?)
+    };
     let producer = (leader + 1) % 4;
     let producers = if fault.is_some() {
         vec![producer, (leader + 2) % 4]
     } else {
         vec![producer]
     };
-    if mixed
-        && (mixed_fleet["leader"].as_u64() != Some(u64::try_from(leader)?)
-            || mixed_fleet["leader_pid"].as_u64() != Some(u64::from(leader_pid)))
-    {
-        return Err("the hot window did not retain its measured new coordinator".into());
-    }
     let producer_pid = run.pid(producer)?;
     let producer_address = run.peer_address(producer);
     let [cold_a, cold_b, hot] = run.buckets.clone();

crates/e2e_test/src/dual_stack_differential/compare.rs

@@ -1,512 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-//! What one stack answered, and how two answers are compared.
-//!
-//! An [`Observation`] keeps only what a client acts on: the HTTP status, the S3 error code, the
-//! allowlisted response headers and a few typed fields decoded from the response. The error
-//! message is kept for diagnostics and never compared: message wording is judged by the wire
-//! corpus, not here. The comparator is pure so it can be tested without a server.
-
-use serde::Serialize;
-use std::collections::{BTreeMap, BTreeSet};
-use std::path::{Path, PathBuf};
-
-/// Response headers compared exactly (after [`normalized_header_value`]).
-const HEADER_ALLOWLIST: [&str; 7] = [
-    "content-type",
-    "x-amz-version-id",
-    "etag",
-    "x-amz-delete-marker",
-    "content-range",
-    "accept-ranges",
-    "x-amz-request-charged",
-];
-
-/// Every response header with this prefix is compared as well.
-const HEADER_ALLOWLIST_PREFIX: &str = "x-amz-server-side-encryption";
-
-/// Placeholder for a field one side did not send.
-const ABSENT: &str = "<absent>";
-
-/// The lowercase header name if the comparator looks at this header.
-pub(crate) fn compared_header(name: &str) -> Option<String> {
-    let name = name.to_ascii_lowercase();
-    (HEADER_ALLOWLIST.contains(&name.as_str()) || name.starts_with(HEADER_ALLOWLIST_PREFIX)).then_some(name)
-}
-
-/// A version ID is minted per process, so its value never matches across two servers; only
-/// whether it is the literal `null` (an unversioned write) carries meaning.
-fn normalized_header_value(name: &str, value: &str) -> String {
-    if name == "x-amz-version-id" && value != "null" {
-        "<version-id>".to_owned()
-    } else {
-        value.to_owned()
-    }
-}
-
-/// One request's outcome as a client sees it.
-#[derive(Clone, Debug, Serialize)]
-pub(crate) struct Observation {
-    /// The S3 operation, named as in the `s3s::S3` trait (`GetObject`).
-    pub op: &'static str,
-    /// Which request of the script this was (`normal`, `negative`, or a setup label).
-    pub case: &'static str,
-    /// HTTP status; 0 when no response arrived.
-    pub status: u16,
-    /// The S3 error code, or the SDK's error kind when there was no service error.
-    pub error_code: Option<String>,
-    /// Allowlisted headers only, lowercase names, normalized values.
-    pub headers: BTreeMap<String, String>,
-    /// Typed values decoded from the response body, never free text.
-    pub fields: BTreeMap<String, String>,
-    /// Diagnostic only; never compared.
-    #[serde(skip)]
-    pub message: Option<String>,
-}
-
-impl Observation {
-    /// Builds an observation, keeping only the allowlisted headers.
-    pub(crate) fn new<'h>(
-        op: &'static str,
-        case: &'static str,
-        status: u16,
-        error_code: Option<String>,
-        headers: impl IntoIterator<Item = (&'h str, &'h str)>,
-    ) -> Self {
-        let headers = headers
-            .into_iter()
-            .filter_map(|(name, value)| compared_header(name).map(|name| (name, value)))
-            .map(|(name, value)| {
-                let value = normalized_header_value(&name, value);
-                (name, value)
-            })
-            .collect();
-        Self {
-            op,
-            case,
-            status,
-            error_code,
-            headers,
-            fields: BTreeMap::new(),
-            message: None,
-        }
-    }
-
-    pub(crate) fn with_field(mut self, name: &str, value: impl Into<String>) -> Self {
-        self.fields.insert(name.to_owned(), value.into());
-        self
-    }
-
-    pub(crate) fn is_success(&self) -> bool {
-        (200..300).contains(&self.status)
-    }
-}
-
-/// One field on which the two stacks answered differently.
-#[derive(Clone, Debug, PartialEq, Eq, Serialize)]
-pub(crate) struct Difference {
-    pub op: String,
-    pub case: String,
-    pub field: String,
-    pub a: String,
-    pub b: String,
-}
-
-/// Every difference between two transcripts of the same script, matched by `(op, case)`.
-pub(crate) fn compare(a: &[Observation], b: &[Observation]) -> Vec<Difference> {
-    let mut differences = Vec::new();
-    let (index_a, duplicates_a) = index(a);
-    let (index_b, duplicates_b) = index(b);
-    for (op, case) in duplicates_a.union(&duplicates_b) {
-        let count = |side: &[Observation]| {
-            side.iter()
-                .filter(|observation| (observation.op, observation.case) == (*op, *case))
-                .count()
-                .to_string()
-        };
-        differences.push(difference(op, case, "duplicate_case", count(a), count(b)));
-    }
-    let keys: BTreeSet<_> = index_a.keys().chain(index_b.keys()).copied().collect();
-    for key @ (op, case) in keys {
-        match (index_a.get(&key), index_b.get(&key)) {
-            (Some(a), Some(b)) => compare_one(a, b, &mut differences),
-            (a, b) => {
-                let presence = |side: Option<&&Observation>| if side.is_some() { "present" } else { ABSENT }.to_owned();
-                differences.push(difference(op, case, "observation", presence(a), presence(b)));
-            }
-        }
-    }
-    differences
-}
-
-type Key = (&'static str, &'static str);
-
-fn index(side: &[Observation]) -> (BTreeMap<Key, &Observation>, BTreeSet<Key>) {
-    let mut index = BTreeMap::new();
-    let mut duplicates = BTreeSet::new();
-    for observation in side {
-        let key = (observation.op, observation.case);
-        if index.insert(key, observation).is_some() {
-            duplicates.insert(key);
-        }
-    }
-    (index, duplicates)
-}
-
-fn difference(op: &str, case: &str, field: &str, a: String, b: String) -> Difference {
-    Difference {
-        op: op.to_owned(),
-        case: case.to_owned(),
-        field: field.to_owned(),
-        a,
-        b,
-    }
-}
-
-fn compare_one(a: &Observation, b: &Observation, differences: &mut Vec<Difference>) {
-    let mut field = |name: String, value_a: Option<&str>, value_b: Option<&str>| {
-        if value_a != value_b {
-            let show = |value: Option<&str>| value.unwrap_or(ABSENT).to_owned();
-            differences.push(difference(a.op, a.case, &name, show(value_a), show(value_b)));
-        }
-    };
-    field(
-        "status".to_owned(),
-        Some(a.status.to_string().as_str()),
-        Some(b.status.to_string().as_str()),
-    );
-    field("error_code".to_owned(), a.error_code.as_deref(), b.error_code.as_deref());
-    let headers: BTreeSet<_> = a.headers.keys().chain(b.headers.keys()).collect();
-    for name in headers {
-        field(
-            format!("header.{name}"),
-            a.headers.get(name).map(String::as_str),
-            b.headers.get(name).map(String::as_str),
-        );
-    }
-    let fields: BTreeSet<_> = a.fields.keys().chain(b.fields.keys()).collect();
-    for name in fields {
-        field(
-            format!("body.{name}"),
-            a.fields.get(name).map(String::as_str),
-            b.fields.get(name).map(String::as_str),
-        );
-    }
-}
-
-/// Operations observed on both sides.
-pub(crate) fn compared_ops(a: &[Observation], b: &[Observation]) -> BTreeSet<&'static str> {
-    let ops_b: BTreeSet<_> = b.iter().map(|observation| observation.op).collect();
-    a.iter()
-        .map(|observation| observation.op)
-        .filter(|op| ops_b.contains(op))
-        .collect()
-}
-
-/// `(op, case)` pairs observed on both sides.
-pub(crate) fn compared_cases(a: &[Observation], b: &[Observation]) -> usize {
-    let cases_b: BTreeSet<_> = b.iter().map(|observation| (observation.op, observation.case)).collect();
-    a.iter()
-        .map(|observation| (observation.op, observation.case))
-        .collect::<BTreeSet<_>>()
-        .intersection(&cases_b)
-        .count()
-}
-
-/// The JSON report written for every run.
-#[derive(Debug, Serialize)]
-pub(crate) struct Report {
-    pub schema: u32,
-    pub run: String,
-    pub stack_a: String,
-    pub stack_b: String,
-    pub region_a: Option<String>,
-    pub region_b: Option<String>,
-    pub compared_ops: usize,
-    pub compared_cases: usize,
-    pub differences: Vec<Difference>,
-}
-
-impl Report {
-    /// Writes the report as `<directory>/<run>.json` and returns its path.
-    pub(crate) fn write_to(&self, directory: &Path) -> std::io::Result<PathBuf> {
-        std::fs::create_dir_all(directory)?;
-        let path = directory.join(format!("{}.json", self.run));
-        let json = serde_json::to_vec_pretty(self).map_err(std::io::Error::other)?;
-        std::fs::write(&path, json)?;
-        Ok(path)
-    }
-}
-
-/// One line per difference, for a failing assertion.
-pub(crate) fn render(differences: &[Difference]) -> String {
-    differences
-        .iter()
-        .map(|d| format!("({}, {}, {}, a={:?}, b={:?})", d.op, d.case, d.field, d.a, d.b))
-        .collect::<Vec<_>>()
-        .join("\n")
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    fn ok(op: &'static str) -> Observation {
-        Observation::new(op, "normal", 200, None, [("content-type", "application/xml")])
-    }
-
-    fn fields(differences: &[Difference]) -> Vec<&str> {
-        differences.iter().map(|d| d.field.as_str()).collect()
-    }
-
-    #[test]
-    fn identical_transcripts_have_no_difference() {
-        let a = vec![
-            ok("HeadBucket"),
-            ok("GetBucketLocation").with_field("location_constraint", "us-east-1"),
-        ];
-        let b = a.clone();
-        assert_eq!(compare(&a, &b), Vec::new());
-        assert_eq!(compared_ops(&a, &b).len(), 2);
-    }
-
-    #[test]
-    fn a_status_difference_is_reported() {
-        let a = vec![ok("PutBucketAcl")];
-        let b = vec![Observation::new(
-            "PutBucketAcl",
-            "normal",
-            403,
-            None,
-            [("content-type", "application/xml")],
-        )];
-        assert_eq!(
-            compare(&a, &b),
-            vec![Difference {
-                op: "PutBucketAcl".into(),
-                case: "normal".into(),
-                field: "status".into(),
-                a: "200".into(),
-                b: "403".into(),
-            }]
-        );
-    }
-
-    #[test]
-    fn an_error_code_difference_is_reported() {
-        let a = vec![Observation::new("GetObject", "negative", 404, Some("NoSuchKey".into()), [])];
-        let b = vec![Observation::new(
-            "GetObject",
-            "negative",
-            404,
-            Some("NoSuchBucket".into()),
-            [],
-        )];
-        let differences = compare(&a, &b);
-        assert_eq!(fields(&differences), ["error_code"]);
-        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("NoSuchKey", "NoSuchBucket"));
-    }
-
-    #[test]
-    fn an_error_code_on_one_side_only_is_reported() {
-        let a = vec![Observation::new("HeadBucket", "negative", 404, Some("NotFound".into()), [])];
-        let b = vec![Observation::new("HeadBucket", "negative", 404, None, [])];
-        let differences = compare(&a, &b);
-        assert_eq!(fields(&differences), ["error_code"]);
-        assert_eq!(differences[0].b, ABSENT);
-    }
-
-    #[test]
-    fn an_allowlisted_header_difference_is_reported() {
-        let a = vec![Observation::new("HeadObject", "normal", 200, None, [("ETag", "\"aa\"")])];
-        let b = vec![Observation::new("HeadObject", "normal", 200, None, [("etag", "\"bb\"")])];
-        assert_eq!(fields(&compare(&a, &b)), ["header.etag"]);
-    }
-
-    #[test]
-    fn a_header_sent_by_one_side_only_is_reported() {
-        let a = vec![Observation::new(
-            "DeleteObject",
-            "normal",
-            204,
-            None,
-            [("x-amz-delete-marker", "true")],
-        )];
-        let b = vec![Observation::new("DeleteObject", "normal", 204, None, [])];
-        let differences = compare(&a, &b);
-        assert_eq!(fields(&differences), ["header.x-amz-delete-marker"]);
-        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("true", ABSENT));
-    }
-
-    #[test]
-    fn every_server_side_encryption_header_is_compared() {
-        let header = "x-amz-server-side-encryption-customer-key-md5";
-        let a = vec![Observation::new("PutObject", "sse-c", 200, None, [(header, "one")])];
-        let b = vec![Observation::new("PutObject", "sse-c", 200, None, [(header, "two")])];
-        assert_eq!(fields(&compare(&a, &b)), [format!("header.{header}")]);
-    }
-
-    #[test]
-    fn headers_outside_the_allowlist_are_not_compared() {
-        let a = vec![Observation::new(
-            "ListBuckets",
-            "normal",
-            200,
-            None,
-            [
-                ("date", "Mon"),
-                ("x-amz-request-id", "1"),
-                ("server", "a"),
-                ("content-length", "10"),
-            ],
-        )];
-        let b = vec![Observation::new(
-            "ListBuckets",
-            "normal",
-            200,
-            None,
-            [
-                ("date", "Tue"),
-                ("x-amz-request-id", "2"),
-                ("server", "b"),
-                ("content-length", "11"),
-            ],
-        )];
-        assert_eq!(compare(&a, &b), Vec::new());
-        assert!(a[0].headers.is_empty(), "non-allowlisted headers were kept: {:?}", a[0].headers);
-    }
-
-    #[test]
-    fn distinct_version_ids_are_not_a_difference() {
-        let a = vec![Observation::new(
-            "PutObject",
-            "normal",
-            200,
-            None,
-            [("x-amz-version-id", "0b1c")],
-        )];
-        let b = vec![Observation::new(
-            "PutObject",
-            "normal",
-            200,
-            None,
-            [("x-amz-version-id", "9f8e")],
-        )];
-        assert_eq!(compare(&a, &b), Vec::new());
-    }
-
-    #[test]
-    fn a_null_version_id_against_a_real_one_is_reported() {
-        let a = vec![Observation::new(
-            "PutObject",
-            "normal",
-            200,
-            None,
-            [("x-amz-version-id", "null")],
-        )];
-        let b = vec![Observation::new(
-            "PutObject",
-            "normal",
-            200,
-            None,
-            [("x-amz-version-id", "9f8e")],
-        )];
-        assert_eq!(fields(&compare(&a, &b)), ["header.x-amz-version-id"]);
-    }
-
-    #[test]
-    fn a_body_field_difference_is_reported() {
-        let a = vec![ok("GetBucketLocation").with_field("location_constraint", "us-east-1")];
-        let b = vec![ok("GetBucketLocation").with_field("location_constraint", "eu-west-1")];
-        let differences = compare(&a, &b);
-        assert_eq!(fields(&differences), ["body.location_constraint"]);
-        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("us-east-1", "eu-west-1"));
-    }
-
-    #[test]
-    fn a_body_field_on_one_side_only_is_reported() {
-        let a = vec![ok("ListObjectsV2").with_field("keys", "[]")];
-        let b = vec![ok("ListObjectsV2")];
-        assert_eq!(fields(&compare(&a, &b)), ["body.keys"]);
-    }
-
-    #[test]
-    fn error_messages_are_never_compared() {
-        let mut a = Observation::new("GetObject", "negative", 404, Some("NoSuchKey".into()), []);
-        let mut b = a.clone();
-        a.message = Some("The specified key does not exist.".into());
-        b.message = Some("Object not found".into());
-        assert_eq!(compare(&[a], &[b]), Vec::new());
-    }
-
-    #[test]
-    fn a_case_observed_on_one_side_only_is_reported() {
-        let a = vec![ok("HeadBucket"), ok("ListBuckets")];
-        let b = vec![ok("HeadBucket")];
-        let differences = compare(&a, &b);
-        assert_eq!(fields(&differences), ["observation"]);
-        assert_eq!(differences[0].op, "ListBuckets");
-        assert_eq!(compared_ops(&a, &b).into_iter().collect::<Vec<_>>(), ["HeadBucket"]);
-    }
-
-    #[test]
-    fn a_case_recorded_twice_is_reported() {
-        let a = vec![ok("HeadBucket"), ok("HeadBucket")];
-        let b = vec![ok("HeadBucket")];
-        assert_eq!(fields(&compare(&a, &b)), ["duplicate_case"]);
-    }
-
-    #[test]
-    fn compared_ops_counts_operations_not_cases() {
-        let negative = Observation::new("HeadBucket", "negative", 404, Some("NotFound".into()), []);
-        let a = vec![ok("HeadBucket"), negative, ok("ListBuckets")];
-        let b = a.clone();
-        assert_eq!(compared_ops(&a, &b).len(), 2);
-    }
-
-    #[test]
-    fn the_report_lists_each_difference_as_op_field_a_b() {
-        let directory = std::env::temp_dir().join(format!("dual-stack-report-{}", uuid::Uuid::new_v4()));
-        let report = Report {
-            schema: 1,
-            run: "unit".into(),
-            stack_a: "legacy".into(),
-            stack_b: "legacy".into(),
-            region_a: None,
-            region_b: Some("eu-west-1".into()),
-            compared_ops: 1,
-            compared_cases: 1,
-            differences: compare(
-                &[ok("GetBucketLocation").with_field("location_constraint", "us-east-1")],
-                &[ok("GetBucketLocation").with_field("location_constraint", "eu-west-1")],
-            ),
-        };
-        let path = report.write_to(&directory).expect("write the report");
-        let json: serde_json::Value = serde_json::from_slice(&std::fs::read(&path).expect("read the report")).expect("parse");
-        let _ = std::fs::remove_dir_all(&directory);
-        assert_eq!(path.file_name().and_then(|name| name.to_str()), Some("unit.json"));
-        assert_eq!(json["compared_ops"], 1);
-        assert_eq!(
-            json["differences"],
-            serde_json::json!([{
-                "op": "GetBucketLocation",
-                "case": "normal",
-                "field": "body.location_constraint",
-                "a": "us-east-1",
-                "b": "eu-west-1",
-            }])
-        );
-    }
-}

crates/e2e_test/src/dual_stack_differential/mod.rs

@@ -1,199 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-//! Dual-stack differential: one `rustfs` binary, two processes, one SDK script (rustfs/backlog#2740).
-//!
-//! Stack A always runs `RUSTFS_S3_STACK=legacy`. Stack B runs the value of `RUSTFS_E2E_STACK_B`
-//! (default `legacy` until the gateway stack is assembled), and `RUSTFS_E2E_STACK_B_REGION`, when
-//! set, starts B in another region so that a difference can be injected by hand. Both servers
-//! answer the script in `scenarios`, which covers every operation of `impl S3 for FS`; `compare`
-//! reports each `(op, field, a, b)` on which the answers differ, and every run writes its report
-//! to `<target>/dual-stack/<run>.json`. Two processes, not two embedded servers: bucket metadata
-//! is process-global, so two stacks in one process would read each other's buckets.
-
-mod compare;
-mod scenarios;
-
-use crate::common::{RustFSTestEnvironment, init_logging, workspace_root};
-use compare::{Difference, Observation, Report, compare, compared_cases, compared_ops, render};
-use std::path::PathBuf;
-use std::time::{SystemTime, UNIX_EPOCH};
-
-type BoxError = Box<dyn std::error::Error + Send + Sync>;
-
-const STACK_A: &str = "legacy";
-const ENV_STACK_B: &str = "RUSTFS_E2E_STACK_B";
-const ENV_STACK_B_REGION: &str = "RUSTFS_E2E_STACK_B_REGION";
-/// Becomes `gateway` once rustfs/backlog#2734 T2.5 assembles the gateway stack.
-const DEFAULT_STACK_B: &str = "legacy";
-/// The region the negative control gives stack A explicitly, so an inherited `RUSTFS_REGION`
-/// cannot erase the injected difference. Not `us-east-1`: that region answers GetBucketLocation
-/// with an empty LocationConstraint and re-creating an owned bucket with 200, so it would not
-/// report its own name and would add CreateBucket differences to the control.
-const CONTROL_REGION_A: &str = "us-west-2";
-const CONTROL_REGION_B: &str = "eu-west-1";
-/// `normal` cases the legacy stack must answer with a 2xx. The comparison alone cannot tell a
-/// script that exercises real behavior from one in which both stacks fail the same way. Five of
-/// the 73 fail on legacy by design of this script: PutBucketReplication (no registered remote
-/// target), GetBucketReplication (so no configuration), PutBucketNotificationConfiguration (no
-/// notification target configured), RestoreObject (the object is not archived) and
-/// GetObjectTorrent (legacy answers `NoSuchKey`).
-const MIN_NORMAL_SUCCESSES: usize = 68;
-
-struct PairConfig {
-    stack_b: String,
-    region_a: Option<String>,
-    region_b: Option<String>,
-}
-
-impl PairConfig {
-    fn from_env() -> Self {
-        Self {
-            stack_b: std::env::var(ENV_STACK_B).unwrap_or_else(|_| DEFAULT_STACK_B.to_owned()),
-            region_a: None,
-            region_b: std::env::var(ENV_STACK_B_REGION).ok(),
-        }
-    }
-}
-
-/// Starts one server on `stack`, overriding its region only when asked.
-async fn start_stack(stack: &str, region: Option<&str>) -> Result<RustFSTestEnvironment, BoxError> {
-    let mut env = RustFSTestEnvironment::new().await?;
-    let mut vars = vec![("RUSTFS_S3_STACK", stack)];
-    if let Some(region) = region {
-        vars.push(("RUSTFS_REGION", region));
-    }
-    env.start_rustfs_server_without_cleanup_with_env(&vars).await?;
-    Ok(env)
-}
-
-/// Starts stack A and stack B side by side, each with its own port and data directory.
-async fn spawn_pair(config: &PairConfig) -> Result<(RustFSTestEnvironment, RustFSTestEnvironment), BoxError> {
-    tokio::try_join!(
-        start_stack(STACK_A, config.region_a.as_deref()),
-        start_stack(&config.stack_b, config.region_b.as_deref())
-    )
-}
-
-fn report_directory() -> PathBuf {
-    let target = std::env::var_os("CARGO_TARGET_DIR").map_or_else(|| PathBuf::from("target"), PathBuf::from);
-    let target = if target.is_absolute() {
-        target
-    } else {
-        workspace_root().join(target)
-    };
-    target.join("dual-stack")
-}
-
-struct Run {
-    report: Report,
-    path: PathBuf,
-    a: Vec<Observation>,
-}
-
-/// Runs the script against both stacks concurrently and writes the report.
-async fn run_differential(label: &str, config: &PairConfig) -> Run {
-    let (server_a, server_b) = spawn_pair(config).await.expect("start stack A and stack B");
-    let (a, b) = tokio::join!(Box::pin(scenarios::run(&server_a)), Box::pin(scenarios::run(&server_b)));
-    let millis = SystemTime::now()
-        .duration_since(UNIX_EPOCH)
-        .map_or(0, |elapsed| elapsed.as_millis());
-    let report = Report {
-        schema: 1,
-        run: format!("{label}-{millis}-{}", std::process::id()),
-        stack_a: STACK_A.to_owned(),
-        stack_b: config.stack_b.clone(),
-        region_a: config.region_a.clone(),
-        region_b: config.region_b.clone(),
-        compared_ops: compared_ops(&a, &b).len(),
-        compared_cases: compared_cases(&a, &b),
-        differences: compare(&a, &b),
-    };
-    let path = report.write_to(&report_directory()).expect("write the dual-stack report");
-    Run { report, path, a }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[tokio::test]
-    async fn both_stacks_answer_every_operation_identically() {
-        init_logging();
-        let config = PairConfig::from_env();
-        let run = run_differential("self-comparison", &config).await;
-        let report = &run.report;
-
-        let mut observed: Vec<_> = run.a.iter().map(|observation| observation.op).collect();
-        observed.sort_unstable();
-        observed.dedup();
-        assert_eq!(
-            observed,
-            scenarios::OPERATIONS,
-            "the script must drive exactly the operations of impl S3 for FS"
-        );
-        assert_eq!(report.compared_ops, scenarios::OPERATIONS.len(), "report: {}", run.path.display());
-        assert!(
-            report.differences.is_empty(),
-            "stack {} and stack {} differ on {} of {} cases (report: {}):\n{}",
-            report.stack_a,
-            report.stack_b,
-            report.differences.len(),
-            report.compared_cases,
-            run.path.display(),
-            render(&report.differences)
-        );
-
-        let not_4xx: Vec<_> = run
-            .a
-            .iter()
-            .filter(|observation| observation.case == "negative" && !(400..500).contains(&observation.status))
-            .collect();
-        assert!(not_4xx.is_empty(), "negative cases must fail with a 4xx on {STACK_A}: {not_4xx:#?}");
-        let successes = run
-            .a
-            .iter()
-            .filter(|observation| observation.case == "normal" && observation.is_success())
-            .count();
-        assert!(
-            successes >= MIN_NORMAL_SUCCESSES,
-            "only {successes} normal cases succeeded on {STACK_A}: {:#?}",
-            run.a
-        );
-    }
-
-    #[tokio::test]
-    async fn differential_detects_an_injected_difference() {
-        init_logging();
-        let config = PairConfig {
-            stack_b: STACK_A.to_owned(),
-            region_a: Some(CONTROL_REGION_A.to_owned()),
-            region_b: Some(CONTROL_REGION_B.to_owned()),
-        };
-        let run = run_differential("injected-region", &config).await;
-        let expected = Difference {
-            op: "GetBucketLocation".to_owned(),
-            case: "normal".to_owned(),
-            field: "body.location_constraint".to_owned(),
-            a: CONTROL_REGION_A.to_owned(),
-            b: CONTROL_REGION_B.to_owned(),
-        };
-        assert!(
-            run.report.differences.contains(&expected),
-            "B started in {CONTROL_REGION_B} but GetBucketLocation was not reported (report: {}):\n{}",
-            run.path.display(),
-            render(&run.report.differences)
-        );
-    }
-}

crates/e2e_test/src/lib.rs

@@ -31,12 +31,6 @@ pub mod chaos;
 #[cfg(test)]
 pub mod fake_s3_target;
 
-// Dual-stack differential (rustfs/backlog#2740): one binary, two processes on
-// different RUSTFS_S3_STACK values, the same SDK script for every S3 operation,
-// compared on status, error code, allowlisted headers and typed body fields.
-#[cfg(test)]
-mod dual_stack_differential;
-
 // On-demand migration (backlog#2147): shared two-server environment, admin
 // wrappers, and the harness self-test (backlog#2151). Behavior scenarios are
 // added by later ODM tasks.

crates/e2e_test/src/protocols/webdav_core.rs

@@ -32,9 +32,7 @@
 //! Advisory: <https://github.com/rustfs/rustfs/security/advisories/GHSA-3p3x-734c-h5vx>
 
 use crate::common::rustfs_binary_path_with_features;
-use crate::common::{
-    AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via, signed_s3_request,
-};
+use crate::common::{AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via};
 use crate::common::{FAST_DATA_USAGE_SCANNER_ENV, admin_request};
 use crate::protocols::test_env::{DEFAULT_ACCESS_KEY, DEFAULT_SECRET_KEY, ProtocolTestEnvironment};
 use anyhow::Result;
@@ -776,81 +774,6 @@ async fn test_webdav_core_operations_direct() -> Result<()> {
     test_webdav_core_operations().await
 }
 
-#[tokio::test]
-async fn test_webdav_content_type_directory_marker() -> Result<()> {
-    let env = ProtocolTestEnvironment::new().map_err(|e| anyhow::anyhow!("{}", e))?;
-    let admin_base_url = format!("http://{}", S3_TEST_ADDRESS);
-    let binary_path = rustfs_binary_path_with_features(Some("webdav"));
-    let mut server_process = Command::new(&binary_path)
-        .arg("--address")
-        .arg(S3_TEST_ADDRESS)
-        .env("RUSTFS_CONSOLE_ENABLE", "false")
-        .env("RUSTFS_WEBDAV_ENABLE", "true")
-        .env("RUSTFS_WEBDAV_ADDRESS", WEBDAV_ADDRESS)
-        .env("RUSTFS_WEBDAV_TLS_ENABLED", "false")
-        .arg(&env.temp_dir)
-        .spawn()?;
-
-    let result = async {
-        ProtocolTestEnvironment::wait_for_port_ready(WEBDAV_PORT, 30)
-            .await
-            .map_err(|e| anyhow::anyhow!("{}", e))?;
-
-        let client = create_client();
-        let bucket_name = "webdav-content-type-bucket";
-        let directory_name = "content-type-directory";
-        let webdav_base_url = format!("http://{}", WEBDAV_ADDRESS);
-        let mkcol_bucket = client
-            .request(
-                reqwest::Method::from_bytes(b"MKCOL").expect("MKCOL is a valid HTTP method"),
-                format!("{}/{}", webdav_base_url, bucket_name),
-            )
-            .header("Authorization", basic_auth_header())
-            .send()
-            .await?;
-        anyhow::ensure!(
-            mkcol_bucket.status().is_success() || mkcol_bucket.status().as_u16() == 201,
-            "MKCOL bucket should succeed, got: {}",
-            mkcol_bucket.status()
-        );
-
-        let mkcol_directory = client
-            .request(
-                reqwest::Method::from_bytes(b"MKCOL").expect("MKCOL is a valid HTTP method"),
-                format!("{}/{}/{}", webdav_base_url, bucket_name, directory_name),
-            )
-            .header("Authorization", basic_auth_header())
-            .send()
-            .await?;
-        anyhow::ensure!(
-            mkcol_directory.status().is_success() || mkcol_directory.status().as_u16() == 201,
-            "MKCOL directory should succeed, got: {}",
-            mkcol_directory.status()
-        );
-
-        let head_url = format!("{}/{}/{}/", admin_base_url, bucket_name, directory_name);
-        let head = signed_s3_request(http::Method::HEAD, &head_url, None, None, DEFAULT_ACCESS_KEY, DEFAULT_SECRET_KEY)
-            .await
-            .map_err(|error| anyhow::anyhow!("{error}"))?;
-        anyhow::ensure!(head.status().is_success(), "HEAD directory marker should succeed, got: {}", head.status());
-        let content_type = head
-            .headers()
-            .get("content-type")
-            .and_then(|value| value.to_str().ok())
-            .unwrap_or_default();
-        anyhow::ensure!(
-            content_type == "application/x-directory",
-            "WebDAV directory marker must preserve its content type, got: {content_type}"
-        );
-        Ok(())
-    }
-    .await;
-
-    let _ = server_process.kill().await;
-    let _ = server_process.wait().await;
-    result
-}
-
 /// Extract the numeric value of the first occurrence of a DAV property in a
 /// PROPFIND multistatus body. Namespace-prefix agnostic; a self-closing
 /// (unreported) property yields None.

crates/e2e_test/src/reliant/grpc_lock_server.rs

@@ -576,6 +576,34 @@ impl NodeService for MinimalLockNodeService {
         Err(Status::unimplemented("lock-only test server"))
     }
 
+    async fn reserve_object_metadata(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::MetadataReservationRequest>,
+    ) -> Result<Response<SnapshotLeaseResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
+    async fn activate_object_metadata_reservation(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::MetadataReservationActivateRequest>,
+    ) -> Result<Response<SnapshotLeaseMutationResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
+    async fn cancel_object_metadata_reservation(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::MetadataReservationCancelRequest>,
+    ) -> Result<Response<SnapshotLeaseMutationResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
+    async fn rename_reserved_data(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::ReservedRenameDataRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::RenameDataResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
     async fn rename_data_with_pool_meta_cas(
         &self,
         _request: Request<rustfs_protos::proto_gen::node_service::RenameDataRequest>,

crates/e2e_test/src/rpc_fault_proxy.rs

@@ -2146,3 +2146,4 @@ mod tests {
         );
     }
 }
+// event-drill: isolated harness-only change

crates/e2e_test/src/upgrade_compatibility_test.rs

@@ -1112,8 +1112,9 @@ async fn wait_for_upgrade_writes(
             match response {
                 Ok(Ok(_)) => break,
                 Ok(Err(error)) => {
+                    let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
                     if error.raw_response().map(|response| response.status().as_u16()) != Some(503)
-                        || error.as_service_error().and_then(ProvideErrorMetadata::code) != Some("ServiceUnavailable")
+                        || !matches!(code, Some("ServiceUnavailable" | "SlowDown" | "SlowDownRead"))
                     {
                         return Err(format!("{phase}: node {node} {probe} failed: {error:?}").into());
                     }

crates/ecstore/Cargo.toml

@@ -139,7 +139,6 @@ rustfs-lifecycle.workspace = true
 rustfs-s3-client = { workspace = true }
 rustfs-s3-types = { workspace = true }
 rustfs-data-usage.workspace = true
-rustfs-gateway-types.workspace = true
 rustfs-object-capacity.workspace = true
 rustfs-object-data-cache = { workspace = true, features = ["runtime-memory"] }
 arc-swap.workspace = true
@@ -158,6 +157,7 @@ serde = { workspace = true, features = ["derive"] }
 time = { workspace = true, features = ["parsing", "formatting", "macros", "serde"] }
 bytesize.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
+s3s = { workspace = true, features = ["minio"] }
 http.workspace = true
 opentelemetry.workspace = true
 http-body = { workspace = true }
@@ -248,7 +248,6 @@ windows-sys = { workspace = true, features = [
 windows-sys = { workspace = true, features = ["Win32_System_Ioctl"] }
 
 [dev-dependencies]
-s3s = { workspace = true, features = ["minio"] }
 aws-smithy-async.workspace = true
 rustfs-s3-ops.workspace = true
 tokio = { workspace = true, features = ["rt-multi-thread", "macros", "test-util", "fs"] }

crates/ecstore/src/api/mod.rs

@@ -66,8 +66,6 @@ pub mod bucket {
         }
 
         pub mod bucket_lifecycle_ops {
-            pub use rustfs_lifecycle::{RestoreOutputLocation, RestoreRequest, RestoreS3Location, RestoreTag, RestoreTagging};
-
             pub use crate::bucket::lifecycle::bucket_lifecycle_ops::{
                 ExpiryState, LifecycleOps, ManualTransitionCancelCheck, ManualTransitionProgressSink,
                 ManualTransitionQueueSnapshot, ManualTransitionRunExecution, ManualTransitionRunOptions,
@@ -139,7 +137,7 @@ pub mod bucket {
             pub use crate::bucket::lifecycle::lifecycle::{
                 Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate,
                 ObjectOpts, RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, expected_expiry_time,
-                format_lifecycle_date, object_opts_from_object_info, parse_lifecycle_date,
+                object_opts_from_object_info,
             };
         }
 
@@ -237,7 +235,7 @@ pub mod bucket {
     }
 
     pub mod object_lock {
-        pub use crate::bucket::object_lock::ObjectLockApi;
+        pub use crate::bucket::object_lock::{ObjectLockApi, ObjectLockStatusExt};
 
         pub mod types {
             pub use crate::bucket::object_lock::types::{
@@ -332,9 +330,9 @@ pub mod bucket {
 
     pub mod utils {
         pub use crate::bucket::utils::{
-            BucketConfigXml, BucketConfigXmlError, check_bucket_and_object_names, check_list_objs_args,
-            check_object_name_for_length_and_slash, check_valid_bucket_name_strict, deserialize, has_bad_path_component,
-            is_meta_bucketname, is_valid_object_prefix, serialize,
+            check_bucket_and_object_names, check_list_objs_args, check_object_name_for_length_and_slash,
+            check_valid_bucket_name_strict, deserialize, has_bad_path_component, is_meta_bucketname, is_valid_object_prefix,
+            serialize,
         };
     }
 
@@ -361,7 +359,7 @@ pub mod capacity {
 }
 
 pub mod object_api_utils {
-    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader};
+    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader, to_s3s_etag};
 }
 
 pub mod cluster {

crates/ecstore/src/bucket/lifecycle/bucket_lifecycle_ops.rs

@@ -88,14 +88,12 @@ use rustfs_filemeta::{
     FileInfo, FileInfoOpts, NULL_VERSION_ID, RestoreStatus, RestoreStatusOps, TRANSITION_COMPLETE, get_file_info,
     is_restored_object_on_disk,
 };
-use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
-use rustfs_gateway_types::persistence::{PersistedLifecycleConfiguration, PersistedLifecycleFilter};
-use rustfs_lifecycle::{RestoreRequest, RestoreS3Location};
 use rustfs_lock::{NamespaceLockGuard, error::LockError};
 use rustfs_scanner_metrics::metrics::{
     IlmAction, Metrics, ScannerLifecycleExpiryStateUpdate, ScannerLifecycleTransitionStateUpdate, global_metrics,
 };
 use rustfs_utils::{get_env_i64, get_env_usize, path::encode_dir_object, string::parse_bool};
+use s3s::dto::{BucketLifecycleConfiguration, ExpirationStatus, ObjectLockConfiguration, RestoreRequest, RestoreRequestType};
 use sha2::{Digest, Sha256};
 use std::any::Any;
 use std::collections::{BTreeMap, HashMap, HashSet};
@@ -302,7 +300,7 @@ impl LifecycleSys {
         Arc::new(Self)
     }
 
-    pub async fn get(&self, bucket: &str) -> Option<PersistedLifecycleConfiguration> {
+    pub async fn get(&self, bucket: &str) -> Option<BucketLifecycleConfiguration> {
         match metadata_boundary::get_lifecycle_config(bucket).await {
             Ok((lc, _)) => Some(lc),
             Err(Error::ConfigNotFound) => None,
@@ -3559,7 +3557,7 @@ async fn stale_upload_lifecycle_due(
         Err(_) => return None,
     };
 
-    let has_size_filter = |filter: &PersistedLifecycleFilter| {
+    let has_size_filter = |filter: &s3s::dto::LifecycleRuleFilter| {
         filter.object_size_greater_than.is_some()
             || filter.object_size_less_than.is_some()
             || filter
@@ -3568,7 +3566,7 @@ async fn stale_upload_lifecycle_due(
                 .is_some_and(|and| and.object_size_greater_than.is_some() || and.object_size_less_than.is_some())
     };
     let needs_size = lifecycle.rules.iter().any(|rule| {
-        rule.status.as_str() != "Disabled"
+        rule.status.as_str() != ExpirationStatus::DISABLED
             && rule.abort_incomplete_multipart_upload.is_some()
             && rule.filter.as_ref().is_some_and(has_size_filter)
     });
@@ -4002,7 +4000,7 @@ pub fn init_background_stale_multipart_upload_cleanup(api: Arc<ECStore>) {
     });
 }
 
-pub async fn validate_transition_tier(lc: &PersistedLifecycleConfiguration) -> Result<(), std::io::Error> {
+pub async fn validate_transition_tier(lc: &BucketLifecycleConfiguration) -> Result<(), std::io::Error> {
     for rule in &lc.rules {
         if let Some(transitions) = &rule.transitions {
             for transition in transitions {
@@ -4119,9 +4117,8 @@ pub(crate) async fn enqueue_immediate_expiry(api: Arc<ECStore>, oi: &ObjectInfo,
         .map(lifecycle::object_opts_from_object_info)
         .collect::<Vec<ObjectOpts>>();
     let lock_config = configs.object_lock;
-    let evaluator_lock_config = lock_config.clone();
     let events = match Evaluator::new(lifecycle)
-        .with_lock_retention(evaluator_lock_config)
+        .with_lock_retention(lock_config.clone())
         .eval(&object_opts)
         .await
     {
@@ -4601,26 +4598,19 @@ pub async fn enqueue_transition_for_existing_objects_scoped(
     }
 }
 
-fn lifecycle_rule_has_date_expiration(lc: &PersistedLifecycleConfiguration, rule_id: &str) -> bool {
+fn lifecycle_rule_has_date_expiration(lc: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
     lc.rules.iter().any(|rule| {
-        rule.status == "Enabled"
+        rule.status == ExpirationStatus::from_static(ExpirationStatus::ENABLED)
             && rule.id.as_deref() == Some(rule_id)
             && rule.expiration.as_ref().is_some_and(|expiration| expiration.date.is_some())
     })
 }
 
-fn should_defer_date_expiry_for_recent_config_update(lc: &PersistedLifecycleConfiguration, now: OffsetDateTime) -> bool {
-    lc.expiry_updated_at
-        .as_ref()
-        .is_some_and(|updated_at| match rustfs_lifecycle::parse_lifecycle_date(updated_at) {
-            Ok(updated_at) => {
-                now.unix_timestamp().saturating_sub(updated_at.unix_timestamp()) < DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS
-            }
-            Err(error) => {
-                warn!(error = %error, "Deferring date expiry with an unreadable update timestamp");
-                true
-            }
-        })
+fn should_defer_date_expiry_for_recent_config_update(lc: &BucketLifecycleConfiguration, now: OffsetDateTime) -> bool {
+    lc.expiry_updated_at.as_ref().is_some_and(|updated_at| {
+        let updated_at = OffsetDateTime::from(updated_at.clone());
+        now.unix_timestamp().saturating_sub(updated_at.unix_timestamp()) < DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS
+    })
 }
 
 async fn apply_existing_object_expiry(
@@ -4640,7 +4630,7 @@ async fn apply_existing_object_expiry(
 struct ExistingObjectExpiryContext<'a> {
     api: Arc<ECStore>,
     bucket: &'a str,
-    lc: Arc<PersistedLifecycleConfiguration>,
+    lc: Arc<BucketLifecycleConfiguration>,
     lock_config: Option<Arc<ObjectLockConfiguration>>,
     bucket_incarnation_id: Uuid,
     src: &'a LcEventSrc,
@@ -4660,9 +4650,8 @@ async fn enqueue_expiry_for_existing_object_group(
         .iter()
         .map(lifecycle::object_opts_from_object_info)
         .collect::<Vec<ObjectOpts>>();
-    let evaluator_lock_config = context.lock_config.clone();
     let events = match Evaluator::new(context.lc.clone())
-        .with_lock_retention(evaluator_lock_config)
+        .with_lock_retention(context.lock_config.clone())
         .eval(&object_opts)
         .await
     {
@@ -4856,7 +4845,7 @@ fn manual_transition_version_marker(oi: &ObjectInfo) -> String {
 async fn enqueue_transition_with_lifecycle_report(
     api: Option<Arc<ECStore>>,
     oi: &ObjectInfo,
-    lc: &PersistedLifecycleConfiguration,
+    lc: &BucketLifecycleConfiguration,
     src: &LcEventSrc,
     options: &ManualTransitionRunOptions,
     report: &mut ManualTransitionRunReport,
@@ -4922,7 +4911,7 @@ async fn enqueue_transition_with_lifecycle_report(
     false
 }
 
-async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &PersistedLifecycleConfiguration, src: &LcEventSrc) -> bool {
+async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &BucketLifecycleConfiguration, src: &LcEventSrc) -> bool {
     let event = lc.eval(&oi.to_lifecycle_opts()).await;
     match event.action {
         IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
@@ -5436,8 +5425,12 @@ pub async fn post_restore_opts(version_id: &str, bucket: &str, object: &str) ->
     })
 }
 
-fn select_restore_s3_location(rreq: &RestoreRequest) -> Result<Option<&RestoreS3Location>, std::io::Error> {
-    if rreq.type_.as_ref().is_none_or(|type_| type_.as_str() != "SELECT") {
+fn select_restore_s3_location(rreq: &RestoreRequest) -> Result<Option<&s3s::dto::S3Location>, std::io::Error> {
+    if rreq
+        .type_
+        .as_ref()
+        .is_none_or(|type_| type_.as_str() != RestoreRequestType::SELECT)
+    {
         return Ok(None);
     }
     let output_location = rreq
@@ -5479,25 +5472,27 @@ pub async fn put_restore_opts(
     // the OutputLocation metadata instead of a write to `OutputLocation.S3`
     // (backlog#1341). RestoreObject rejects SELECT at the API boundary; this
     // is the fail-closed backstop for any other caller.
-    if rreq.type_.as_ref().is_some_and(|type_| type_.as_str() == "SELECT") {
+    if rreq
+        .type_
+        .as_ref()
+        .is_some_and(|type_| type_.as_str() == RestoreRequestType::SELECT)
+    {
         return Err(std::io::Error::other("SELECT restore requests are not supported"));
     }
     for (k, v) in oi.user_defined.iter() {
         meta.insert(k.to_string(), v.clone());
     }
-    // The fleet capability gate guarantees physical copy-back support. Keep
-    // the completed transition identity for orphan recovery and remote cleanup.
     rustfs_utils::http::metadata_compat::remove_str(&mut meta, rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID);
+    rustfs_utils::http::metadata_compat::remove_str(&mut meta, rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK);
     if !oi.user_tags.is_empty() {
         meta.insert(AMZ_OBJECT_TAGGING.to_string(), (*oi.user_tags).clone());
     }
-    // Finalization publishes completion and its expiry under the object lock.
-    // The local copy-back commit must not make unfinished work eligible for expiry.
+    let restore_expiry = lifecycle::expected_expiry_time(OffsetDateTime::now_utc(), rreq.days.unwrap_or(1));
     meta.insert(
         metadata_keys::RESTORE.to_string(),
         RestoreStatus {
-            is_restore_in_progress: Some(true),
-            restore_expiry_date: None,
+            is_restore_in_progress: Some(false),
+            restore_expiry_date: Some(restore_expiry),
         }
         .to_string(),
     );
@@ -5511,7 +5506,6 @@ pub async fn put_restore_opts(
         // computed MD5 is not the object's public plaintext ETag.
         preserve_etag: oi.etag.clone(),
         shard_integrity_write_mode: Some(oi.shard_integrity_write_mode()),
-        restore_copyback: true,
         //expires:           oi.expires,
         ..Default::default()
     })
@@ -5531,18 +5525,6 @@ impl LifecycleOps for ObjectInfo {
         if self.transitioned_object.status != lifecycle::TRANSITION_COMPLETE {
             return false;
         }
-        if rustfs_utils::http::metadata_compat::get_consistent_str(
-            &self.user_defined,
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
-        ) == Some(rustfs_utils::http::metadata_compat::RESTORE_COPYBACK_PROTOCOL_V1)
-            && self
-                .user_defined
-                .get(metadata_keys::RESTORE)
-                .and_then(|value| rustfs_filemeta::parse_restore_obj_status(value).ok())
-                .is_some_and(|status| status.is_restore_in_progress == Some(true))
-        {
-            return false;
-        }
         !is_restored_object_on_disk(&self.user_defined)
     }
 }
@@ -5554,25 +5536,25 @@ pub trait RestoreRequestOps {
 impl RestoreRequestOps for RestoreRequest {
     fn validate(&self, _api: Arc<ECStore>) -> Result<(), std::io::Error> {
         // SELECT type requires select_parameters, and vice versa
-        if self.type_.as_ref().is_none_or(|t| t.as_str() != "SELECT") && self.select_parameters.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != RestoreRequestType::SELECT) && self.select_parameters.is_some() {
             return Err(std::io::Error::other("Select parameters can only be specified with SELECT request type"));
         }
         if let Some(type_) = &self.type_
-            && type_.as_str() == "SELECT"
+            && type_.as_str() == RestoreRequestType::SELECT
             && self.select_parameters.is_none()
         {
             return Err(std::io::Error::other("SELECT restore request requires select parameters to be specified"));
         }
 
         // OutputLocation is only valid for SELECT requests
-        if self.type_.as_ref().is_none_or(|t| t.as_str() != "SELECT") && self.output_location.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != RestoreRequestType::SELECT) && self.output_location.is_some() {
             return Err(std::io::Error::other("OutputLocation can only be specified with SELECT request type"));
         }
         select_restore_s3_location(self)?;
 
         // Days must not be specified with SELECT requests
         if let Some(type_) = &self.type_
-            && type_.as_str() == "SELECT"
+            && type_.as_str() == RestoreRequestType::SELECT
             && self.days.is_some_and(|d| d > 0)
         {
             return Err(std::io::Error::other("Days cannot be specified with SELECT restore request"));
@@ -5614,7 +5596,7 @@ pub fn lifecycle_version_delete_target(oi: &ObjectInfo) -> Option<ObjectToDelete
 }
 
 pub async fn eval_action_from_lifecycle(
-    lc: &PersistedLifecycleConfiguration,
+    lc: &BucketLifecycleConfiguration,
     lock_config: Option<&ObjectLockConfiguration>,
     oi: &ObjectInfo,
 ) -> lifecycle::Event {
@@ -6319,19 +6301,14 @@ mod tests {
     use rustfs_data_usage::TierStats;
     use rustfs_filemeta::metadata_keys;
     use rustfs_filemeta::{FileInfo, FileMeta};
-    use rustfs_gateway_types::dto::MetadataEntry;
-    use rustfs_gateway_types::persistence::{
-        PersistedDefaultRetention as DefaultRetention, PersistedObjectLockConfiguration as ObjectLockConfiguration,
-        PersistedObjectLockRule as ObjectLockRule,
-    };
-    use rustfs_gateway_types::persistence::{
-        PersistedLifecycleConfiguration, PersistedLifecycleExpiration, PersistedLifecycleRule,
-        PersistedNoncurrentVersionExpiration, PersistedTransition,
-    };
-    use rustfs_lifecycle::{RestoreOutputLocation as OutputLocation, RestoreRequest, RestoreS3Location as S3Location};
     #[cfg(feature = "test-util")]
     use rustfs_s3_client::transition_api::ReaderImpl;
     use rustfs_scanner_metrics::metrics::{IlmAction, global_metrics};
+    use s3s::dto::{
+        BucketLifecycleConfiguration, DefaultRetention, ExpirationStatus, LifecycleExpiration, LifecycleRule, MetadataEntry,
+        NoncurrentVersionExpiration, ObjectLockConfiguration, ObjectLockEnabled, ObjectLockRetentionMode, ObjectLockRule,
+        OutputLocation, RestoreRequest, RestoreRequestType, S3Location, Timestamp, Transition, TransitionStorageClass,
+    };
     use serial_test::serial;
     use sha2::{Digest, Sha256};
     use std::collections::HashMap;
@@ -9338,7 +9315,7 @@ mod tests {
             output_location,
             select_parameters: None,
             tier: None,
-            type_: Some("SELECT".to_owned()),
+            type_: Some(RestoreRequestType::from_static(RestoreRequestType::SELECT)),
         }
     }
 
@@ -9655,12 +9632,12 @@ mod tests {
 
     #[test]
     fn lifecycle_rule_has_date_expiration_detects_enabled_date_rule() {
-        let lc = PersistedLifecycleConfiguration {
+        let lc = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
-                    date: Some(lifecycle::format_lifecycle_date(OffsetDateTime::now_utc()).unwrap()),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
+                    date: Some(Timestamp::from(OffsetDateTime::now_utc())),
                     ..Default::default()
                 }),
                 id: Some("rule-date".to_string()),
@@ -10177,15 +10154,14 @@ mod tests {
     #[test]
     fn should_defer_date_expiry_for_recent_config_update_respects_grace_window() {
         let now = OffsetDateTime::now_utc();
-        let recent = PersistedLifecycleConfiguration {
-            expiry_updated_at: Some(lifecycle::format_lifecycle_date(now - time::Duration::seconds(1)).unwrap()),
+        let recent = BucketLifecycleConfiguration {
+            expiry_updated_at: Some(Timestamp::from(now - time::Duration::seconds(1))),
             rules: Vec::new(),
         };
-        let stale = PersistedLifecycleConfiguration {
-            expiry_updated_at: Some(
-                lifecycle::format_lifecycle_date(now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1))
-                    .unwrap(),
-            ),
+        let stale = BucketLifecycleConfiguration {
+            expiry_updated_at: Some(Timestamp::from(
+                now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1),
+            )),
             rules: Vec::new(),
         };
 
@@ -10313,12 +10289,12 @@ mod tests {
         assert_eq!(deleted.version_id, None);
     }
 
-    fn expired_delete_marker_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn expired_delete_marker_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     expired_object_delete_marker: Some(true),
                     ..Default::default()
                 }),
@@ -10334,12 +10310,12 @@ mod tests {
         }
     }
 
-    fn latest_expiration_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn latest_expiration_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     days: Some(1),
                     ..Default::default()
                 }),
@@ -10355,12 +10331,12 @@ mod tests {
         }
     }
 
-    fn current_and_noncurrent_expiration_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn current_and_noncurrent_expiration_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     days: Some(30),
                     ..Default::default()
                 }),
@@ -10368,7 +10344,7 @@ mod tests {
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("expire-current-and-noncurrent".to_string()),
-                noncurrent_version_expiration: Some(PersistedNoncurrentVersionExpiration {
+                noncurrent_version_expiration: Some(NoncurrentVersionExpiration {
                     noncurrent_days: Some(1),
                     newer_noncurrent_versions: None,
                 }),
@@ -10379,12 +10355,12 @@ mod tests {
         }
     }
 
-    fn all_versions_expiration_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn all_versions_expiration_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     days: Some(1),
                     expired_object_all_versions: Some(true),
                     ..Default::default()
@@ -10403,29 +10379,29 @@ mod tests {
 
     fn lock_enabled_without_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: None,
         }
     }
 
     fn lock_enabled_with_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(30),
-                    mode: Some("COMPLIANCE".to_owned()),
+                    mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
                     years: None,
                 }),
             }),
         }
     }
 
-    fn latest_transition_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn latest_transition_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -10434,10 +10410,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![PersistedTransition {
+                transitions: Some(vec![Transition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some("WARM".to_owned()),
+                    storage_class: Some(TransitionStorageClass::from_static("WARM")),
                 }]),
             }],
         }
@@ -12974,9 +12950,8 @@ mod tests {
 
         for (name, object, lock_config, expected) in cases {
             let object_opts = lifecycle::object_opts_from_object_info(&object);
-            let evaluator_lock_config = lock_config.clone().map(Arc::new);
             let batch_event = Evaluator::new(Arc::new(lifecycle.clone()))
-                .with_lock_retention(evaluator_lock_config)
+                .with_lock_retention(lock_config.clone().map(Arc::new))
                 .eval(&[object_opts])
                 .await
                 .expect("batch lifecycle evaluation should succeed")
@@ -14233,7 +14208,10 @@ mod tests {
         let object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
+                (
+                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
+                    s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+                ),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14245,7 +14223,7 @@ mod tests {
 
     #[tokio::test]
     async fn restored_copy_expiry_is_not_blocked_by_retention() {
-        let lifecycle = PersistedLifecycleConfiguration {
+        let lifecycle = BucketLifecycleConfiguration {
             expiry_updated_at: None,
             rules: Vec::new(),
         };
@@ -14255,7 +14233,10 @@ mod tests {
         let mut object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
+                (
+                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
+                    ObjectLockRetentionMode::COMPLIANCE.to_string(),
+                ),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14472,7 +14453,7 @@ mod tests {
                 "removed" => {
                     metadata_sys::delete(&bucket, BUCKET_LIFECYCLE_CONFIG).await.unwrap();
                 }
-                "disabled" => changed.rules[0].status = "Disabled".to_owned(),
+                "disabled" => changed.rules[0].status = ExpirationStatus::from_static(ExpirationStatus::DISABLED),
                 "longer-days" => {
                     changed.rules[0]
                         .noncurrent_version_expiration
@@ -14588,7 +14569,7 @@ mod tests {
         );
     }
 
-    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<PersistedLifecycleConfiguration> {
+    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<BucketLifecycleConfiguration> {
         let xml = br#"<LifecycleConfiguration><Rule><ID>expire</ID><Status>Enabled</Status><Filter><Prefix></Prefix></Filter><Expiration><Days>0</Days></Expiration><NoncurrentVersionExpiration><NoncurrentDays>0</NoncurrentDays></NoncurrentVersionExpiration></Rule></LifecycleConfiguration>"#;
         metadata_sys::update_in(&ecstore.ctx, bucket, BUCKET_LIFECYCLE_CONFIG, xml.to_vec())
             .await
@@ -15281,11 +15262,7 @@ mod tests {
     #[tokio::test]
     #[serial]
     async fn tier_free_version_recovery_production_entrypoint_enqueues_seeded_item() {
-        let (disk_paths, ecstore) =
-            temp_env::async_with_vars([(super::ENV_TIER_FREE_VERSION_RECOVERY_ENABLED, Some("false"))], async {
-                setup_test_env().await
-            })
-            .await;
+        let (disk_paths, ecstore) = setup_test_env().await;
         let tier = format!("RECOVERY{}", Uuid::new_v4().simple()).to_uppercase();
         crate::services::tier::test_util::register_mock_tier(&ecstore.tier_config_mgr(), &tier).await;
         let runtime_state = install_unconsumed_runtime_expiry_worker(&ecstore, 1).await;

crates/ecstore/src/bucket/lifecycle/core.rs

@@ -17,7 +17,7 @@ use crate::object_api::ObjectInfo;
 pub use rustfs_lifecycle::{
     Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate, ObjectOpts,
     RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, abort_incomplete_multipart_upload_due,
-    expected_expiry_time, expiration_action_has_valid_target, format_lifecycle_date, parse_lifecycle_date,
+    expected_expiry_time, expiration_action_has_valid_target,
 };
 
 pub fn object_opts_from_object_info(oi: &ObjectInfo) -> ObjectOpts {

crates/ecstore/src/bucket/lifecycle/metadata_boundary.rs

@@ -14,8 +14,7 @@
 
 use std::sync::Arc;
 
-use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
-use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
+use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration};
 use time::OffsetDateTime;
 use uuid::Uuid;
 
@@ -25,7 +24,7 @@ use crate::error::{Error, Result};
 
 #[derive(Debug)]
 pub(crate) struct LifecycleExpiryConfigs {
-    pub(crate) lifecycle: Option<Arc<PersistedLifecycleConfiguration>>,
+    pub(crate) lifecycle: Option<Arc<BucketLifecycleConfiguration>>,
     pub(crate) object_lock: Option<Arc<ObjectLockConfiguration>>,
     pub(crate) bucket_incarnation_id: Uuid,
     pub(crate) table_bucket_enabled: bool,
@@ -101,7 +100,7 @@ pub(crate) async fn get_expiry_configs_for_options(
     })
 }
 
-pub(crate) async fn get_lifecycle_config(bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
+pub(crate) async fn get_lifecycle_config(bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
     metadata_sys::get_lifecycle_config(bucket).await
 }
 
@@ -111,15 +110,15 @@ mod tests {
     use crate::bucket::metadata::BucketMetadata;
     use crate::bucket::metadata_sys::{self, test_support::isolated_store_over_temp_disks};
     use crate::storage_api_contracts::bucket::MakeBucketOptions;
-    use rustfs_gateway_types::persistence::{PersistedLifecycleExpiration, PersistedLifecycleRule};
+    use s3s::dto::{ExpirationStatus, LifecycleExpiration, LifecycleRule};
     use serial_test::serial;
 
-    fn lifecycle_config() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn lifecycle_config() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     days: Some(1),
                     ..Default::default()
                 }),

crates/ecstore/src/bucket/lifecycle/object_lock_boundary.rs

@@ -22,7 +22,7 @@ pub(crate) fn is_object_locked_by_metadata(user_defined: &HashMap<String, String
 }
 
 pub(crate) fn check_object_lock_for_deletion_with_config(
-    config: Option<&rustfs_gateway_types::persistence::PersistedObjectLockConfiguration>,
+    config: Option<&s3s::dto::ObjectLockConfiguration>,
     obj_info: &ObjectInfo,
     bypass_governance: bool,
 ) -> crate::error::Result<Option<ObjectLockBlockReason>> {

crates/ecstore/src/bucket/lifecycle/replication_sink.rs

@@ -82,13 +82,12 @@ mod tests {
     use crate::bucket::replication::{DeleteReplicationConfigSnapshot, ReplicationObjectBridge};
     use crate::object_api::{ObjectInfo, ObjectOptions};
     use crate::storage_api_contracts::object::ObjectToDelete;
-    use rustfs_gateway_types::dto::Status;
-    use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
-    use rustfs_gateway_types::persistence::{
-        PersistedOptionalReplicationStatus, PersistedReplicationConfiguration, PersistedReplicationDestination,
-        PersistedReplicationRule, PersistedReplicationStatus,
-    };
     use rustfs_scanner_metrics::metrics::IlmAction;
+    use s3s::dto::{
+        BucketVersioningStatus, DeleteMarkerReplication, DeleteMarkerReplicationStatus, DeleteReplication,
+        DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule, ReplicationRuleStatus,
+        VersioningConfiguration,
+    };
     use uuid::Uuid;
 
     use super::*;
@@ -161,27 +160,27 @@ mod tests {
             for purge_enabled in [false, true] {
                 let snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
                     VersioningConfiguration {
-                        status: Some("Enabled".to_owned()),
+                        status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
                         ..Default::default()
                     },
-                    Some(PersistedReplicationConfiguration {
+                    Some(ReplicationConfiguration {
                         role: String::new(),
-                        rules: vec![PersistedReplicationRule {
-                            delete_marker_replication: Some(PersistedOptionalReplicationStatus {
+                        rules: vec![ReplicationRule {
+                            delete_marker_replication: Some(DeleteMarkerReplication {
                                 status: Some(if marker_enabled {
-                                    Status::ENABLED.to_string()
+                                    DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)
                                 } else {
-                                    Status::DISABLED.to_string()
+                                    DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)
                                 }),
                             }),
-                            delete_replication: Some(PersistedReplicationStatus {
+                            delete_replication: Some(DeleteReplication {
                                 status: if purge_enabled {
-                                    Status::ENABLED.to_string()
+                                    DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED)
                                 } else {
-                                    Status::DISABLED.to_string()
+                                    DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED)
                                 },
                             }),
-                            destination: PersistedReplicationDestination {
+                            destination: Destination {
                                 bucket: "arn:rustfs:replication:target".to_string(),
                                 ..Default::default()
                             },
@@ -191,7 +190,7 @@ mod tests {
                             prefix: Some(String::new()),
                             priority: Some(1),
                             source_selection_criteria: None,
-                            status: Status::ENABLED.to_string(),
+                            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
                         }],
                     }),
                 );

crates/ecstore/src/bucket/lifecycle/runtime_boundary.rs

@@ -14,7 +14,7 @@
 
 use std::sync::Arc;
 
-use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
+use s3s::dto::BucketLifecycleConfiguration;
 use tokio::sync::RwLock;
 use tokio_util::sync::CancellationToken;
 
@@ -52,7 +52,7 @@ pub(crate) fn deployment_id() -> Option<String> {
     sources::deployment_id()
 }
 
-pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<PersistedLifecycleConfiguration> {
+pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<BucketLifecycleConfiguration> {
     sources::bucket_lifecycle_config(bucket).await
 }
 

crates/ecstore/src/bucket/metadata.rs

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 use super::msgp_decode::{read_msgp_ext8_time, skip_msgp_value, write_msgp_time};
-use super::object_lock::ObjectLockApi;
+use super::object_lock::{ObjectLockApi, ObjectLockStatusExt};
 use super::versioning::VersioningApi;
 use super::{quota::BucketQuota, target::BucketTargets};
 use crate::bucket::replication::invalid_replication_config_status_field;
@@ -25,17 +25,12 @@ use crate::object_api::WriteCompletion;
 use crate::runtime::sources as runtime_sources;
 use crate::store::ECStore;
 use byteorder::{BigEndian, ByteOrder, LittleEndian};
-use rustfs_gateway_types::cors_tagging::{PersistedCorsConfiguration as CORSConfiguration, PersistedTagging as Tagging};
-use rustfs_gateway_types::persistence::{
-    PersistedAccelerateConfiguration as AccelerateConfiguration,
-    PersistedBucketEncryptionConfiguration as ServerSideEncryptionConfiguration,
-    PersistedBucketLoggingStatus as BucketLoggingStatus, PersistedLifecycleConfiguration,
-    PersistedNotificationConfiguration as NotificationConfiguration, PersistedObjectLockConfiguration as ObjectLockConfiguration,
-    PersistedPublicAccessBlockConfiguration as PublicAccessBlockConfiguration, PersistedReplicationConfiguration,
-    PersistedRequestPaymentConfiguration as RequestPaymentConfiguration,
-    PersistedVersioningConfiguration as VersioningConfiguration, PersistedWebsiteConfiguration as WebsiteConfiguration,
-};
 use rustfs_policy::policy::BucketPolicy;
+use s3s::dto::{
+    AccelerateConfiguration, BucketLifecycleConfiguration, BucketLoggingStatus, BucketVersioningStatus, CORSConfiguration,
+    NotificationConfiguration, ObjectLockConfiguration, PublicAccessBlockConfiguration, ReplicationConfiguration,
+    RequestPaymentConfiguration, ServerSideEncryptionConfiguration, Tagging, VersioningConfiguration, WebsiteConfiguration,
+};
 use serde::Serializer;
 use sha2::{Digest, Sha256};
 use std::collections::HashMap;
@@ -45,11 +40,23 @@ use time::{Date, OffsetDateTime, PrimitiveDateTime, Time as CivilTime, UtcOffset
 use tracing::error;
 use uuid::Uuid;
 
-// The persisted Object Lock configuration answers the storage-level trait
-// here because this module owns the persisted `ObjectLockConfiguration`.
+// The serving-layer DTO impls for the storage-level Object Lock traits live
+// here because this module owns the persisted `ObjectLockConfiguration`
+// during the s3s ratchet migration (rustfs/backlog#1842).
 impl ObjectLockApi for ObjectLockConfiguration {
     fn enabled(&self) -> bool {
-        self.object_lock_enabled.as_deref() == Some("Enabled")
+        self.object_lock_enabled
+            .as_ref()
+            .is_some_and(|v| v.as_str() == s3s::dto::ObjectLockEnabled::ENABLED)
+    }
+}
+
+impl ObjectLockStatusExt for s3s::dto::ObjectLockLegalHoldStatus {
+    fn valid(&self) -> bool {
+        matches!(
+            self.as_str(),
+            s3s::dto::ObjectLockLegalHoldStatus::ON | s3s::dto::ObjectLockLegalHoldStatus::OFF
+        )
     }
 }
 
@@ -432,8 +439,6 @@ pub struct BucketMetadata {
     pub bucket_incarnation_id: Uuid,
     pub(crate) bucket_incarnation_sidecar: bool,
     pub(crate) bucket_creation_committed: bool,
-    // A legacy default is not proof of an interrupted creation.
-    pub(crate) bucket_creation_commit_record_present: bool,
     pub policy_config_json: Vec<u8>,
     pub notification_config_xml: Vec<u8>,
     pub lifecycle_config_xml: Vec<u8>,
@@ -482,13 +487,13 @@ pub struct BucketMetadata {
 
     pub policy_config: Option<BucketPolicy>,
     pub notification_config: Option<NotificationConfiguration>,
-    pub lifecycle_config: Option<PersistedLifecycleConfiguration>,
+    pub lifecycle_config: Option<BucketLifecycleConfiguration>,
     pub object_lock_config: Option<ObjectLockConfiguration>,
     pub versioning_config: Option<VersioningConfiguration>,
     pub sse_config: Option<ServerSideEncryptionConfiguration>,
     pub tagging_config: Option<Tagging>,
     pub quota_config: Option<BucketQuota>,
-    pub replication_config: Option<PersistedReplicationConfiguration>,
+    pub replication_config: Option<ReplicationConfiguration>,
     pub bucket_target_config: Option<BucketTargets>,
     pub bucket_target_config_meta: Option<HashMap<String, String>>,
     pub cors_config: Option<CORSConfiguration>,
@@ -509,7 +514,6 @@ impl Default for BucketMetadata {
             bucket_incarnation_id: Uuid::nil(),
             bucket_incarnation_sidecar: false,
             bucket_creation_committed: false,
-            bucket_creation_commit_record_present: false,
             policy_config_json: Default::default(),
             notification_config_xml: Default::default(),
             lifecycle_config_xml: Default::default(),
@@ -748,10 +752,7 @@ impl BucketMetadata {
                 "TableBucketConfigUpdatedAt" => self.table_bucket_config_updated_at = read_msgp_time_value(rd)?,
                 "DurabilityConfigUpdatedAt" => self.durability_config_updated_at = read_msgp_time_value(rd)?,
                 "OnDemandMigrationConfigUpdatedAt" => self.on_demand_migration_config_updated_at = read_msgp_time_value(rd)?,
-                "BucketCreationCommitted" => {
-                    self.bucket_creation_committed = read_msgp_bool(rd)?;
-                    self.bucket_creation_commit_record_present = true;
-                }
+                "BucketCreationCommitted" => self.bucket_creation_committed = read_msgp_bool(rd)?,
                 other => {
                     tracing::debug!(field = %other, "BucketMetadata decode_from: skipping unknown field");
                     skip_msgp_value(rd)?;
@@ -764,10 +765,8 @@ impl BucketMetadata {
 
     /// Encode to msgp bytes. Field order follows MinIO BucketMetadata for compatibility.
     pub fn encode_to<W: Write>(&self, wr: &mut W) -> Result<()> {
-        // Preserve the absence of a legacy commit record across config rewrites.
-        let commit_record_present = self.bucket_creation_commit_record_present || self.bucket_creation_committed;
-        // Map size: MinIO fields (25) + RustFS extensions (21) + optional commit record.
-        let map_len: u32 = 46 + u32::from(commit_record_present);
+        // Map size: MinIO fields (25) + RustFS extensions (22)
+        let map_len: u32 = 47;
         rmp::encode::write_map_len(wr, map_len)?;
 
         // MinIO field order (same as Go struct)
@@ -848,10 +847,8 @@ impl BucketMetadata {
         write_msgp_time(wr, self.durability_config_updated_at)?;
         rmp::encode::write_str(wr, "OnDemandMigrationConfigUpdatedAt")?;
         write_msgp_time(wr, self.on_demand_migration_config_updated_at)?;
-        if commit_record_present {
-            rmp::encode::write_str(wr, "BucketCreationCommitted")?;
-            rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
-        }
+        rmp::encode::write_str(wr, "BucketCreationCommitted")?;
+        rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
 
         Ok(())
     }
@@ -1021,11 +1018,9 @@ impl BucketMetadata {
                     None
                 } else {
                     let config = deserialize::<VersioningConfiguration>(&data)?;
-                    if config
-                        .status
-                        .as_ref()
-                        .is_some_and(|status| !matches!(status.as_str(), "Enabled" | "Suspended"))
-                    {
+                    if config.status.as_ref().is_some_and(|status| {
+                        !matches!(status.as_str(), BucketVersioningStatus::ENABLED | BucketVersioningStatus::SUSPENDED)
+                    }) {
                         return Err(Error::other("bucket versioning configuration has an invalid status"));
                     }
                     Some(config)
@@ -1038,7 +1033,7 @@ impl BucketMetadata {
                 let config = if data.is_empty() {
                     None
                 } else {
-                    let config = deserialize::<PersistedReplicationConfiguration>(&data)?;
+                    let config = deserialize::<ReplicationConfiguration>(&data)?;
                     if let Some(field) = invalid_replication_config_status_field(&config) {
                         return Err(Error::other(format!("replication field {field} has an invalid status")));
                     }
@@ -1232,8 +1227,8 @@ impl BucketMetadata {
         }
         self.lifecycle_config = None;
         if !self.lifecycle_config_xml.is_empty()
-            && let Err(e) = deserialize::<PersistedLifecycleConfiguration>(&self.lifecycle_config_xml)
-                .map(|c| self.lifecycle_config = Some(c))
+            && let Err(e) =
+                deserialize::<BucketLifecycleConfiguration>(&self.lifecycle_config_xml).map(|c| self.lifecycle_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1319,8 +1314,8 @@ impl BucketMetadata {
         }
         self.replication_config = None;
         if !self.replication_config_xml.is_empty()
-            && let Err(e) = deserialize::<PersistedReplicationConfiguration>(&self.replication_config_xml)
-                .map(|c| self.replication_config = Some(c))
+            && let Err(e) =
+                deserialize::<ReplicationConfiguration>(&self.replication_config_xml).map(|c| self.replication_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1700,13 +1695,6 @@ mod test {
         let mut bm = BucketMetadata::unmarshal(&body[4..]).expect("unmarshal recovered blob");
         assert_eq!(bm.name, "interop");
         assert!(!bm.bucket_creation_committed, "legacy metadata must default to uncommitted");
-        assert!(!bm.bucket_creation_commit_record_present, "legacy metadata has no creation-commit record");
-        let rewritten = BucketMetadata::unmarshal(&bm.marshal_msg().expect("rewrite legacy metadata"))
-            .expect("read rewritten legacy metadata");
-        assert!(
-            !rewritten.bucket_creation_commit_record_present,
-            "rewriting legacy metadata must not fabricate a creation intent"
-        );
         bm.parse_all_configs().expect("parse recovered configs");
         assert!(bm.lifecycle_config.is_some());
     }
@@ -1724,7 +1712,6 @@ mod test {
 
         assert_eq!(bm.name, new.name);
         assert!(new.bucket_creation_committed);
-        assert!(new.bucket_creation_commit_record_present);
         assert!(!bm.bucket_incarnation_id.is_nil());
         assert_eq!(bm.bucket_incarnation_id, new.bucket_incarnation_id);
     }
@@ -1829,15 +1816,17 @@ mod test {
 
     #[test]
     fn object_locking_requires_lock_metadata_not_plain_versioning() {
+        use s3s::dto::ObjectLockEnabled;
+
         let mut bm = BucketMetadata::new("test-bucket");
         bm.versioning_config = Some(VersioningConfiguration {
-            status: Some("Enabled".to_owned()),
+            status: Some(s3s::dto::BucketVersioningStatus::from_static("Enabled")),
             ..Default::default()
         });
         assert!(!bm.object_locking());
 
         bm.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             ..Default::default()
         });
         assert!(bm.object_locking());

crates/ecstore/src/bucket/metadata_sys.rs

@@ -30,18 +30,14 @@ use crate::storage_api_contracts::heal::HealOperations as _;
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
 use crate::store::{ECStore, await_bucket_namespace_operation};
 use futures::future::join_all;
-use rustfs_gateway_types::cors_tagging::{PersistedCorsConfiguration as CORSConfiguration, PersistedTagging as Tagging};
-use rustfs_gateway_types::persistence::{
-    PersistedAccelerateConfiguration as AccelerateConfiguration,
-    PersistedBucketEncryptionConfiguration as ServerSideEncryptionConfiguration,
-    PersistedBucketLoggingStatus as BucketLoggingStatus, PersistedLifecycleConfiguration,
-    PersistedNotificationConfiguration as NotificationConfiguration, PersistedObjectLockConfiguration as ObjectLockConfiguration,
-    PersistedPublicAccessBlockConfiguration as PublicAccessBlockConfiguration, PersistedReplicationConfiguration,
-    PersistedRequestPaymentConfiguration as RequestPaymentConfiguration,
-    PersistedVersioningConfiguration as VersioningConfiguration, PersistedWebsiteConfiguration as WebsiteConfiguration,
-};
 use rustfs_heal_contracts::heal_channel::HealOpts;
 use rustfs_policy::policy::BucketPolicy;
+use s3s::dto::ReplicationConfiguration;
+use s3s::dto::{
+    AccelerateConfiguration, BucketLifecycleConfiguration, BucketLoggingStatus, CORSConfiguration, NotificationConfiguration,
+    ObjectLockConfiguration, ObjectLockEnabled, ObjectLockRetentionMode, PublicAccessBlockConfiguration,
+    RequestPaymentConfiguration, ServerSideEncryptionConfiguration, Tagging, VersioningConfiguration, WebsiteConfiguration,
+};
 use std::collections::HashSet;
 use std::time::Duration;
 use std::{
@@ -172,7 +168,7 @@ pub(crate) fn object_lock_config_state_from_authoritative_metadata(bm: &BucketMe
     if bm.lock_enabled {
         return Ok(ObjectLockConfigState::Configured {
             config: ObjectLockConfiguration {
-                object_lock_enabled: Some("Enabled".to_owned()),
+                object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
                 rule: None,
             },
             updated_at: bm.object_lock_config_updated_at,
@@ -216,10 +212,17 @@ pub(crate) fn configured_object_lock_state_for_tests(
 ) -> ObjectLockConfigState {
     ObjectLockConfigState::Configured {
         config: ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
-            rule: Some(rustfs_gateway_types::persistence::PersistedObjectLockRule {
-                default_retention: Some(rustfs_gateway_types::persistence::PersistedDefaultRetention {
-                    mode: Some(mode.as_str().to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            rule: Some(s3s::dto::ObjectLockRule {
+                default_retention: Some(s3s::dto::DefaultRetention {
+                    mode: Some(s3s::dto::ObjectLockRetentionMode::from_static(match mode {
+                        crate::bucket::object_lock::types::RetentionMode::Governance => {
+                            s3s::dto::ObjectLockRetentionMode::GOVERNANCE
+                        }
+                        crate::bucket::object_lock::types::RetentionMode::Compliance => {
+                            s3s::dto::ObjectLockRetentionMode::COMPLIANCE
+                        }
+                    })),
                     days: Some(days),
                     years: None,
                 }),
@@ -230,7 +233,7 @@ pub(crate) fn configured_object_lock_state_for_tests(
 }
 
 fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -> Result<()> {
-    if config.object_lock_enabled.as_deref() != Some("Enabled") {
+    if config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str) != Some(ObjectLockEnabled::ENABLED) {
         return Err(Error::other("persisted bucket Object Lock enabled state is invalid"));
     }
 
@@ -243,7 +246,7 @@ fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -
     if !retention
         .mode
         .as_ref()
-        .is_some_and(|mode| matches!(mode.as_str(), "COMPLIANCE" | "GOVERNANCE"))
+        .is_some_and(|mode| matches!(mode.as_str(), ObjectLockRetentionMode::COMPLIANCE | ObjectLockRetentionMode::GOVERNANCE))
     {
         return Err(Error::other("persisted bucket Object Lock retention mode is invalid"));
     }
@@ -1392,7 +1395,7 @@ pub async fn get_public_access_block_config(bucket: &str) -> Result<(PublicAcces
     bucket_meta_sys.get_public_access_block_config(bucket).await
 }
 
-pub async fn get_lifecycle_config(bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
+pub async fn get_lifecycle_config(bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1668,7 +1671,7 @@ pub(crate) async fn get_cached_quota_config_and_incarnation_in(
     bucket_meta_sys.cached_quota_config_and_incarnation(bucket).await
 }
 
-pub async fn get_replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
+pub async fn get_replication_config(bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1678,7 +1681,7 @@ pub async fn get_replication_config(bucket: &str) -> Result<(PersistedReplicatio
 pub(crate) async fn get_replication_config_in(
     ctx: &crate::runtime::instance::InstanceContext,
     bucket: &str,
-) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
+) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = bucket_metadata_sys_of(ctx)?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1816,27 +1819,9 @@ pub(crate) fn current_local_meta_mutation_gen() -> u64 {
 
 type LocalAuthorityCache = Arc<StdMutex<HashMap<String, (u64, Arc<BucketMetadata>)>>>;
 
-#[derive(Debug, Clone)]
-pub(crate) struct PendingBucketCreationProof {
-    pub(crate) generation: Uuid,
-    pub(crate) intent: Vec<u8>,
-    pub(crate) versioning_requested: bool,
-    pub(crate) witnesses: Vec<crate::disk::os::NativeBucketCreationWitness>,
-}
-
-impl PendingBucketCreationProof {
-    pub(crate) fn matches_metadata(&self, metadata: &BucketMetadata) -> Result<bool> {
-        Ok(!metadata.bucket_creation_committed
-            && metadata.bucket_incarnation_sidecar
-            && metadata.bucket_incarnation_id == self.generation
-            && metadata.marshal_msg()? == self.intent)
-    }
-}
-
 #[derive(Debug, Clone)]
 pub struct BucketMetadataSys {
     metadata_map: Arc<RwLock<HashMap<String, Arc<BucketMetadata>>>>,
-    pending_bucket_creations: Arc<RwLock<HashMap<String, PendingBucketCreationProof>>>,
     local_authority_cache: LocalAuthorityCache,
     /// Serializes metadata-map commits and their derived cache updates for one
     /// bucket. Namespace locks, when present, are acquired before this lock.
@@ -1881,7 +1866,6 @@ impl BucketMetadataSys {
     pub fn new(api: Arc<ECStore>) -> Self {
         Self {
             metadata_map: Arc::new(RwLock::new(HashMap::new())),
-            pending_bucket_creations: Arc::new(RwLock::new(HashMap::new())),
             local_authority_cache: Arc::new(StdMutex::new(HashMap::new())),
             metadata_publish_locks: Arc::new(MetadataPublishLockRegistry {
                 locks: StdMutex::new(HashMap::new()),
@@ -1913,85 +1897,19 @@ impl BucketMetadataSys {
             .expect("bucket metadata object store should still be live")
     }
 
-    pub(crate) async fn pending_bucket_creation(&self, bucket: &str) -> Option<PendingBucketCreationProof> {
-        self.pending_bucket_creations.read().await.get(bucket).cloned()
-    }
-
-    pub(crate) async fn grant_pending_bucket_creation(&self, bucket: &str, proof: PendingBucketCreationProof) {
-        self.pending_bucket_creations.write().await.insert(bucket.to_owned(), proof);
-    }
-
-    pub(crate) async fn revoke_pending_bucket_creation(&self, bucket: &str) {
-        self.pending_bucket_creations.write().await.remove(bucket);
-    }
-
-    pub(crate) async fn consume_pending_bucket_creation(
-        &self,
-        bucket: &str,
-        metadata: &BucketMetadata,
-        witnesses: &[crate::disk::os::NativeBucketCreationWitness],
-    ) -> Result<bool> {
-        let mut pending = self.pending_bucket_creations.write().await;
-        let Some(proof) = pending.get(bucket) else {
-            return Ok(false);
-        };
-        if !proof.matches_metadata(metadata)?
-            || proof.witnesses.len() != witnesses.len()
-            || !proof
-                .witnesses
-                .iter()
-                .zip(witnesses)
-                .all(|(old, current)| old.matches(current))
-            || !crate::disk::os::consume_native_creation_witnesses(witnesses)
-        {
-            return Ok(false);
-        }
-        pending.remove(bucket);
-        Ok(true)
-    }
-
-    #[cfg(test)]
-    pub(crate) async fn refresh_bucket_for_test(&self, bucket: &str) -> Result<()> {
-        let expected = self.metadata_map.read().await.get(bucket).cloned();
-        let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
-        let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
-        self.load_bucket_under_namespace(bucket, MetadataLoadMode::Refresh, expected.as_ref(), &guard)
-            .await
-    }
-
-    #[cfg(test)]
-    pub(crate) async fn publish_persisted_metadata_for_test(&self, bucket: &str) -> Result<()> {
-        let expected = self.metadata_map.read().await.get(bucket).cloned();
-        let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
-        let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
-        let (metadata, persisted) = await_bucket_namespace_operation(
-            Some(&guard),
-            bucket,
-            "test cached metadata load",
-            load_bucket_metadata_parse_with_presence(self.object_store(), bucket, true),
-        )
-        .await?;
-        self.publish_if_unchanged(bucket, expected.as_ref(), metadata, persisted, &guard)
-            .await
-    }
-
-    #[cfg(test)]
-    pub(crate) async fn migrate_bucket_creation_commit_for_test(&self, bucket: &str) -> Result<()> {
-        self.migrate_bucket_creation_commit(bucket).await.map(|_| ())
-    }
-
-    #[cfg(test)]
-    pub(crate) async fn legacy_migration_reports_missing_bucket_for_test(&self, bucket: &str) -> Result<bool> {
-        Ok(matches!(
-            self.migrate_legacy_metadata(bucket).await?,
-            BucketMetadataAuthority::MissingBucket
-        ))
-    }
-
     fn object_store_if_live(&self) -> Option<Arc<ECStore>> {
         self.api.upgrade()
     }
 
+    /// Whether the backing `ECStore` has been dropped.
+    ///
+    /// Used by `InstanceContext::clear_bucket_metadata_sys_if_stale` during
+    /// `ECStore::drop` so a shared bootstrap context does not retain a
+    /// `BucketMetadataSys` whose weak handle is dangling.
+    pub(crate) fn is_store_stale(&self) -> bool {
+        self.api.upgrade().is_none()
+    }
+
     fn metadata_publish_lock(&self, bucket: &str) -> Arc<Mutex<MetadataPublishLockState>> {
         Self::bucket_lock_in(&self.metadata_publish_locks, bucket)
     }
@@ -2178,7 +2096,6 @@ impl BucketMetadataSys {
         expected: Option<&Arc<BucketMetadata>>,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
-        self.revoke_pending_bucket_creation(bucket).await;
         if !await_bucket_namespace_operation(
             Some(namespace_guard),
             bucket,
@@ -2262,7 +2179,6 @@ impl BucketMetadataSys {
         persisted: bool,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
-        self.revoke_pending_bucket_creation(bucket).await;
         if !persisted {
             let _publish_guard = self
                 .lock_metadata_publish(bucket, namespace_guard, "refreshed bucket metadata absence publish")
@@ -2324,7 +2240,6 @@ impl BucketMetadataSys {
     }
 
     pub async fn set(&self, bucket: String, bm: Arc<BucketMetadata>) {
-        self.revoke_pending_bucket_creation(&bucket).await;
         bump_local_meta_mutation_gen();
         if !is_meta_bucketname(&bucket) {
             let publish_lock = self.metadata_publish_lock(&bucket);
@@ -2348,7 +2263,6 @@ impl BucketMetadataSys {
         if is_meta_bucketname(bucket) {
             return false;
         }
-        self.revoke_pending_bucket_creation(bucket).await;
         let publish_lock = self.metadata_publish_lock(bucket);
         let _publish_guard = publish_lock.lock().await;
         let mut map = self.metadata_map.write().await;
@@ -2366,7 +2280,6 @@ impl BucketMetadataSys {
     }
 
     async fn _reset(&mut self) {
-        self.pending_bucket_creations.write().await.clear();
         bump_local_meta_mutation_gen();
         let mut map = self.metadata_map.write().await;
         map.clear();
@@ -2506,7 +2419,6 @@ impl BucketMetadataSys {
     /// server's metadata never leaks into the ambient (first) instance.
     pub(crate) async fn persist_and_set(&self, bm: BucketMetadata) -> Result<()> {
         let mut bm = bm;
-        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.save_with_store(self.object_store()).await?;
 
         self.set(bm.name.clone(), Arc::new(bm)).await;
@@ -2515,7 +2427,6 @@ impl BucketMetadataSys {
     }
 
     async fn persist_new_and_set(&self, mut bm: BucketMetadata) -> Result<()> {
-        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.bucket_creation_committed = true;
         bm.save_with_store_committed(self.object_store()).await?;
         save_bucket_incarnation(self.object_store(), &bm.name, bm.bucket_incarnation_id).await?;
@@ -2525,9 +2436,7 @@ impl BucketMetadataSys {
     }
 
     async fn persist_new_bucket_metadata_intent(&self, mut bm: BucketMetadata) -> Result<()> {
-        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.bucket_creation_committed = false;
-        bm.bucket_creation_commit_record_present = true;
         bm.save_with_store(self.object_store()).await?;
         save_bucket_incarnation(self.object_store(), &bm.name, bm.bucket_incarnation_id).await?;
         bm.bucket_incarnation_sidecar = true;
@@ -2537,7 +2446,6 @@ impl BucketMetadataSys {
     }
 
     async fn commit_bucket_metadata(&self, mut bm: BucketMetadata) -> Result<()> {
-        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.parse_all_configs()?;
         bm.bucket_incarnation_sidecar = true;
         // The caller holds the full bucket creation fence (lifecycle, metadata
@@ -2586,7 +2494,6 @@ impl BucketMetadataSys {
         bucket: &str,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
-        self.revoke_pending_bucket_creation(bucket).await;
         let expected = self.metadata_map.read().await.get(bucket).cloned();
         if !self
             .bucket_exists(bucket, namespace_guard, "peer bucket metadata existence check")
@@ -2643,7 +2550,6 @@ impl BucketMetadataSys {
             #[cfg(test)]
             self.lazy_disk_loads.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
 
-            self.revoke_pending_bucket_creation(bucket).await;
             let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
             let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
             #[cfg(test)]
@@ -2969,7 +2875,6 @@ impl BucketMetadataSys {
     /// revalidated at write quorum, so a stale snapshot can neither revert an
     /// acknowledged configuration update nor outlive a delete/recreate.
     async fn migrate_bucket_creation_commit(&self, bucket: &str) -> Result<BucketMetadataAuthority> {
-        self.revoke_pending_bucket_creation(bucket).await;
         let transaction_lock = self
             .object_store()
             .new_ns_lock(RUSTFS_META_BUCKET, &bucket_metadata_transaction_lock_key(bucket))
@@ -3047,7 +2952,6 @@ impl BucketMetadataSys {
         &self,
         bucket: &str,
     ) -> Result<BucketMetadataAuthority> {
-        self.revoke_pending_bucket_creation(bucket).await;
         #[cfg(test)]
         if self.object_lock_disk_read_errors.write().await.remove(bucket) {
             return Err(Error::other(format!("injected Object Lock metadata disk read failure: {bucket}")));
@@ -3270,7 +3174,7 @@ impl BucketMetadataSys {
         }
     }
 
-    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
+    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         match ConfigState::of(&bm.lifecycle_config_xml, &bm.lifecycle_config).require(bucket, BUCKET_LIFECYCLE_CONFIG)? {
@@ -3393,7 +3297,7 @@ impl BucketMetadataSys {
         }))
     }
 
-    pub async fn get_replication_config(&self, bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
+    pub async fn get_replication_config(&self, bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         if !bm.replication_config_xml.is_empty() && bm.replication_config.is_none() {
@@ -3495,7 +3399,6 @@ mod tests {
     };
     use crate::bucket::target::{BucketTarget, BucketTargetType, Credentials};
     use crate::config::com::read_config;
-    use crate::disk::{DiskAPI as _, RUSTFS_META_TMP_BUCKET};
     use crate::storage_api_contracts::bucket::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions};
     use byteorder::{ByteOrder as _, LittleEndian};
     use serial_test::serial;
@@ -3550,180 +3453,6 @@ mod tests {
         ),
     ];
 
-    #[tokio::test]
-    #[serial(storage_class_env)]
-    async fn persistence_entry_failures_revoke_pending_creation_before_cache_publish() {
-        let (dirs, store) = isolated_store_over_temp_disks().await;
-        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
-        let sys = bucket_metadata_sys_of(&store.ctx).expect("instance metadata system");
-        let sys = sys.read().await.clone();
-        let options = MakeBucketOptions {
-            lock_enabled: true,
-            ..Default::default()
-        };
-        for entry in ["save", "persist", "persist-new", "commit"] {
-            let bucket = format!("revoke-{entry}-{}", Uuid::new_v4());
-            store.fail_next_bucket_creation_after_intent_for_test(&bucket);
-            store
-                .make_bucket(&bucket, &options)
-                .await
-                .expect_err("stop the real producer before physical publication");
-            sys.pending_bucket_creation(&bucket)
-                .await
-                .expect("the producer must leave a real live creation permission");
-            let (intent, persisted) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-                .await
-                .expect("read the original producer intent");
-            assert!(persisted, "{entry}: the complete original intent must be persisted");
-            let original_intent = intent.marshal_msg().expect("encode the original intent");
-
-            // Keep the creation fence order used by the real producer.
-            let lifecycle_guard = store
-                .acquire_bucket_lifecycle_write_lock(&bucket)
-                .await
-                .expect("hold the bucket lifecycle fence");
-            let transaction_guard = acquire_bucket_metadata_transaction_lock_in(&store.ctx, &bucket)
-                .await
-                .expect("hold the metadata transaction fence");
-            let namespace_lock = store
-                .new_ns_lock(&bucket, &bucket)
-                .await
-                .expect("create the exact bucket namespace lock");
-            let namespace_guard = namespace_lock
-                .get_write_lock(crate::set_disk::get_lock_acquire_timeout())
-                .await
-                .expect("hold the exact bucket namespace fence");
-
-            // A file at each native staging root causes actual IO to fail before
-            // metadata can reach the cache-publishing set call. Preserve the
-            // original staging directories so no fixture data is discarded.
-            for dir in &dirs {
-                let staging = dir.path().join(RUSTFS_META_TMP_BUCKET);
-                let saved = dir.path().join(".rustfs.sys/revocation-saved-tmp");
-                std::fs::rename(&staging, &saved).expect("preserve the native staging directory");
-                std::fs::write(&staging, b"block metadata staging with a non-directory")
-                    .expect("install an actual non-directory IO obstruction");
-            }
-            let result = match entry {
-                "save" => sys.save(intent).await,
-                "persist" => sys.persist_and_set(intent).await,
-                "persist-new" => sys.persist_new_and_set(intent).await,
-                "commit" => sys.commit_bucket_metadata(intent).await,
-                _ => unreachable!(),
-            };
-            for dir in &dirs {
-                let staging = dir.path().join(RUSTFS_META_TMP_BUCKET);
-                let saved = dir.path().join(".rustfs.sys/revocation-saved-tmp");
-                std::fs::remove_file(&staging).expect("remove only the fixture's staging obstruction");
-                std::fs::rename(&saved, &staging).expect("restore the complete original staging directory");
-            }
-            drop((namespace_guard, transaction_guard, lifecycle_guard));
-            result.expect_err("native persistence must fail before its downstream cache set");
-            assert!(
-                sys.pending_bucket_creation(&bucket).await.is_none(),
-                "{entry}: the entry revoker must retire live permission even when persistence fails"
-            );
-            let (after_failure, present) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-                .await
-                .expect("read the intent after the failed persistence entry");
-            assert!(present, "{entry}: a failed persistence entry must retain the original metadata");
-            assert_eq!(
-                after_failure
-                    .marshal_msg()
-                    .expect("encode the original intent after persistence failure"),
-                original_intent,
-                "{entry}: native IO failure must not change the original generation or complete intent"
-            );
-            let error = store
-                .make_bucket(&bucket, &options)
-                .await
-                .expect_err("a failed ordinary persistence entry cannot preserve creation permission");
-            assert!(error.to_string().contains("administrator recovery"), "{entry}: {error}");
-            let (after_retry, _) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-                .await
-                .expect("read the original intent after the rejected retry");
-            assert_eq!(
-                after_retry.marshal_msg().expect("encode the intent after the rejected retry"),
-                original_intent,
-                "{entry}: the rejected retry must preserve the complete original intent"
-            );
-            for disk in store.pools[0].disk_set[0].disks.read().await.iter() {
-                assert!(
-                    matches!(
-                        disk.as_ref().expect("native disk").stat_volume(&bucket).await,
-                        Err(crate::disk::error::DiskError::VolumeNotFound)
-                    ),
-                    "{entry}: failed persistence and the rejected retry must not publish a physical volume"
-                );
-            }
-        }
-    }
-
-    #[tokio::test]
-    #[serial(storage_class_env)]
-    async fn metadata_reset_revokes_real_pending_creation_without_changing_persisted_intent() {
-        let (_dirs, store) = isolated_store_over_temp_disks().await;
-        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
-        let bucket = format!("revoke-reset-{}", Uuid::new_v4());
-        let options = MakeBucketOptions {
-            lock_enabled: true,
-            ..Default::default()
-        };
-        store.fail_next_bucket_creation_after_intent_for_test(&bucket);
-        store
-            .make_bucket(&bucket, &options)
-            .await
-            .expect_err("stop the real producer before physical publication");
-        let sys = bucket_metadata_sys_of(&store.ctx).expect("instance metadata system");
-        let mut sys = sys.read().await.clone();
-        sys.pending_bucket_creation(&bucket)
-            .await
-            .expect("the real producer must leave live creation permission before reset");
-        let (intent, persisted) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-            .await
-            .expect("read the original intent before cache reset");
-        assert!(persisted, "cache reset starts from a complete real persisted intent");
-        let original_intent = intent.marshal_msg().expect("encode the original reset intent");
-        sys._reset().await;
-        assert!(
-            sys.pending_bucket_creation(&bucket).await.is_none(),
-            "reset must revoke the producer-issued permission shared with the instance metadata system"
-        );
-        let (after_reset, present) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-            .await
-            .expect("read the original intent after cache reset");
-        assert!(present, "cache reset must retain the persisted intent");
-        assert_eq!(
-            after_reset.marshal_msg().expect("encode the intent after cache reset"),
-            original_intent,
-            "cache reset must leave the original generation and complete persisted intent unchanged"
-        );
-        let error = store
-            .make_bucket(&bucket, &options)
-            .await
-            .expect_err("reset cannot regrant the producer's old creation permission");
-        assert!(error.to_string().contains("administrator recovery"), "{error}");
-        let (after_retry, _) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
-            .await
-            .expect("read the intent after the reset-induced retry rejection");
-        assert_eq!(
-            after_retry
-                .marshal_msg()
-                .expect("encode the intent after the rejected reset retry"),
-            original_intent,
-            "the reset-induced rejection must preserve the original complete intent"
-        );
-        for disk in store.pools[0].disk_set[0].disks.read().await.iter() {
-            assert!(
-                matches!(
-                    disk.as_ref().expect("native disk").stat_volume(&bucket).await,
-                    Err(crate::disk::error::DiskError::VolumeNotFound)
-                ),
-                "cache reset and the rejected retry must not publish a physical volume"
-            );
-        }
-    }
-
     #[tokio::test]
     async fn g_d3_003_new_writer_replication_loads_without_fail_closed_state() {
         let (dirs, store) = isolated_store_over_temp_disks().await;
@@ -3912,7 +3641,7 @@ mod tests {
         let mut versioned = BucketMetadata::new("authoritative-versioned");
         versioned.versioning_config_xml = b"<VersioningConfiguration><Status>Enabled</Status></VersioningConfiguration>".to_vec();
         versioned.versioning_config = Some(VersioningConfiguration {
-            status: Some("Enabled".to_owned()),
+            status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
             ..Default::default()
         });
         sys.set("authoritative-versioned".to_string(), Arc::new(versioned)).await;
@@ -4035,24 +3764,25 @@ mod tests {
         else {
             panic!("legacy lock-enabled metadata must remain configured");
         };
-        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
+        assert_eq!(
+            config.object_lock_enabled.as_ref().map(|value| value.as_str()),
+            Some(ObjectLockEnabled::ENABLED)
+        );
         assert!(config.rule.is_none());
     }
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_retention_periods() {
-        use rustfs_gateway_types::persistence::{
-            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as ObjectLockRule,
-        };
+        use s3s::dto::{DefaultRetention, ObjectLockRule};
 
         for (days, years) in [(Some(0), None), (Some(-1), None), (Some(1), Some(1)), (None, None)] {
             let mut metadata = BucketMetadata::new("invalid-object-lock");
             metadata.object_lock_config = Some(ObjectLockConfiguration {
-                object_lock_enabled: Some("Enabled".to_owned()),
+                object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
                 rule: Some(ObjectLockRule {
                     default_retention: Some(DefaultRetention {
                         days,
-                        mode: Some("COMPLIANCE".to_owned()),
+                        mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
                         years,
                     }),
                 }),
@@ -4067,11 +3797,9 @@ mod tests {
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_enabled_mode_and_rule() {
-        use rustfs_gateway_types::persistence::{
-            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as ObjectLockRule,
-        };
+        use s3s::dto::{DefaultRetention, ObjectLockRule};
 
-        for enabled in [None, Some("Disabled".to_string())] {
+        for enabled in [None, Some(ObjectLockEnabled::from("Disabled".to_string()))] {
             let mut metadata = BucketMetadata::new("invalid-object-lock-enabled");
             metadata.lock_enabled = true;
             metadata.object_lock_config = Some(ObjectLockConfiguration {
@@ -4083,11 +3811,11 @@ mod tests {
 
         let mut invalid_mode = BucketMetadata::new("invalid-object-lock-mode");
         invalid_mode.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(1),
-                    mode: Some("invalid".to_string()),
+                    mode: Some(ObjectLockRetentionMode::from("invalid".to_string())),
                     years: None,
                 }),
             }),
@@ -4096,7 +3824,7 @@ mod tests {
 
         let mut missing_retention = BucketMetadata::new("missing-object-lock-retention");
         missing_retention.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: Some(ObjectLockRule { default_retention: None }),
         });
         assert!(object_lock_config_state_from_authoritative_metadata(&missing_retention).is_err());
@@ -4106,7 +3834,7 @@ mod tests {
     fn object_lock_ever_enabled_marker_survives_config_removal() {
         let mut metadata = BucketMetadata::new("lock-config-removed");
         let config = ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: None,
         };
         metadata
@@ -4124,7 +3852,10 @@ mod tests {
         else {
             panic!("an ever-enabled bucket must remain Object Lock configured");
         };
-        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
+        assert_eq!(
+            config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str),
+            Some(ObjectLockEnabled::ENABLED)
+        );
     }
 
     #[tokio::test]
@@ -4140,7 +3871,7 @@ mod tests {
 
         let node = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore.clone())));
         let config = ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: None,
         };
         update_with_sys(
@@ -5121,7 +4852,7 @@ mod tests {
     #[tokio::test]
     async fn update_config_with_persists_tagging_rewrite_across_disk_reload() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
+        use s3s::dto::Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
 
@@ -5285,7 +5016,7 @@ mod tests {
     #[tokio::test]
     async fn concurrent_update_config_with_calls_do_not_lose_writes() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
+        use s3s::dto::Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
         let sys = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore)));
@@ -5385,7 +5116,7 @@ mod tests {
         let mut persisted = BucketMetadata::new("reload-bucket");
         persisted.policy_config_json = b"persisted-marker".to_vec();
         let lock_config = ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             rule: None,
         };
         persisted
@@ -5664,7 +5395,7 @@ mod tests {
     #[serial]
     async fn bucket_delete_waits_for_config_mutation_fence() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
+        use s3s::dto::Tag;
 
         let (_dirs, store) = isolated_store_over_temp_disks().await;
         init_bucket_metadata_sys(store.clone(), Vec::new()).await;

crates/ecstore/src/bucket/object_lock/mod.rs

@@ -17,8 +17,15 @@ pub mod objectlock_sys;
 pub mod types;
 
 /// Whether a bucket Object Lock configuration has locking enabled. The
-/// persisted `ObjectLockConfiguration` implements this in the bucket-metadata
-/// module, which owns the persisted configuration type.
+/// serving-layer `ObjectLockConfiguration` DTO implements this in
+/// the bucket-metadata module, which owns the persisted configuration type
+/// during the s3s ratchet migration (rustfs/backlog#1842).
 pub trait ObjectLockApi {
     fn enabled(&self) -> bool;
 }
+
+/// Whether a legal-hold status value is one of the two valid wire values.
+/// Implemented for the serving-layer DTO in the bucket-metadata module.
+pub trait ObjectLockStatusExt {
+    fn valid(&self) -> bool;
+}

crates/ecstore/src/bucket/replication/replication_config_boundary.rs

@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-pub use rustfs_replication::config_cache_serde;
 pub use rustfs_replication::{
     ObjectOpts, OperatorRuleContract, REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION, REMOTE_TARGET_READ_ONLY_HISTORICAL_FIELDS,
     REMOTE_TARGET_UNSUPPORTED_FIELDS, REMOTE_TARGET_WRITABLE_FIELDS, REPLICATION_CAPABILITY_CONTRACT_VERSION,

crates/ecstore/src/bucket/replication/replication_lifecycle_bridge.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
+use s3s::dto::ReplicationConfiguration;
 
 use super::replication_config_boundary::ReplicationConfigurationExt as _;
 use super::replication_filemeta_boundary::{
@@ -42,7 +42,7 @@ impl ReplicationLifecycleBridge {
         dead_code,
         reason = "declared boundary surface for the ECStore replication split plan; no caller in this port (backlog#1823)"
     )]
-    pub(crate) fn new_config(config: PersistedReplicationConfiguration) -> ReplicationLifecycleConfig {
+    pub(crate) fn new_config(config: ReplicationConfiguration) -> ReplicationLifecycleConfig {
         ReplicationConfig::new(Some(config), None)
     }
 
@@ -118,17 +118,16 @@ impl ReplicationLifecycleBridge {
 
 #[cfg(test)]
 mod tests {
-    use rustfs_gateway_types::dto::Status;
-    use rustfs_gateway_types::persistence::{PersistedReplicationDestination, PersistedReplicationRule};
+    use s3s::dto::{Destination, ReplicationRule, ReplicationRuleStatus};
 
     use super::super::replication_filemeta_boundary::{ReplicateTargetDecision, VersionPurgeStatusType};
     use super::*;
 
-    fn replication_rule() -> PersistedReplicationRule {
-        PersistedReplicationRule {
+    fn replication_rule() -> ReplicationRule {
+        ReplicationRule {
             delete_marker_replication: None,
             delete_replication: None,
-            destination: PersistedReplicationDestination {
+            destination: Destination {
                 bucket: "arn:aws:s3:::target-bucket".to_string(),
                 ..Default::default()
             },
@@ -138,13 +137,13 @@ mod tests {
             prefix: Some(String::new()),
             priority: Some(1),
             source_selection_criteria: None,
-            status: Status::ENABLED.to_string(),
+            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
         }
     }
 
     #[test]
     fn has_pending_version_purge_preserves_replication_active_rule_behavior() {
-        let config = ReplicationLifecycleBridge::new_config(PersistedReplicationConfiguration {
+        let config = ReplicationLifecycleBridge::new_config(ReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });

crates/ecstore/src/bucket/replication/replication_metadata_boundary.rs

@@ -17,8 +17,8 @@ use std::sync::Arc;
 use crate::bucket::{metadata::BucketMetadata, metadata_sys};
 use crate::disk::{BUCKET_META_PREFIX, RUSTFS_META_BUCKET};
 use crate::runtime::instance::InstanceContext;
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 use rustfs_utils::path::path_join_buf;
+use s3s::dto::ReplicationConfiguration;
 use time::OffsetDateTime;
 
 use super::replication_error_boundary::{Error, Result};
@@ -36,11 +36,11 @@ impl ReplicationMetadataStore {
     pub(crate) const FORCE_DELETE_REPLICATION_FILE: &'static str = "config/replication/force-delete.bin";
     pub(crate) const FORCE_DELETE_REPLICATION_TRANSACTION_LOCK: &'static str = "config/replication/force-delete.bin.transaction";
 
-    pub(crate) async fn replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
+    pub(crate) async fn replication_config(bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
         metadata_sys::get_replication_config(bucket).await
     }
 
-    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<PersistedReplicationConfiguration>> {
+    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<ReplicationConfiguration>> {
         let config = match Self::replication_config(bucket).await {
             Ok((config, _)) => Some(config),
             Err(err) => {

crates/ecstore/src/bucket/replication/replication_object_config.rs

@@ -16,9 +16,7 @@ use std::{collections::HashMap, fmt, sync::Arc};
 
 use super::replication_filemeta_boundary::metadata_keys;
 use crate::bucket::metadata::BucketMetadata;
-use rustfs_gateway_types::dto::Status;
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
-use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+use s3s::dto::{BucketVersioningStatus, ReplicationConfiguration, ReplicationRuleStatus, VersioningConfiguration};
 use serde::{Deserialize, Serialize};
 use time::OffsetDateTime;
 use tracing::error;
@@ -41,7 +39,7 @@ use super::replication_target_boundary::{BucketTargets, ReplicationTargetStore};
 use super::replication_versioning_boundary::ReplicationVersioningStore;
 use super::runtime_boundary as runtime_sources;
 
-pub(crate) async fn get_replication_config(bucket: &str) -> Result<Option<PersistedReplicationConfiguration>> {
+pub(crate) async fn get_replication_config(bucket: &str) -> Result<Option<ReplicationConfiguration>> {
     let config = ReplicationMetadataStore::optional_replication_config(bucket).await?;
     validate_delete_replication_config(&VersioningConfiguration::default(), config.as_ref())?;
     Ok(config)
@@ -66,7 +64,7 @@ impl DeleteReplicationConfigSnapshot {
     #[cfg(test)]
     pub(crate) fn from_configs_for_test(
         versioning: VersioningConfiguration,
-        replication: Option<PersistedReplicationConfiguration>,
+        replication: Option<ReplicationConfiguration>,
     ) -> Self {
         let metadata = replication.map(|config| {
             let mut metadata = BucketMetadata::new("test-bucket");
@@ -80,7 +78,7 @@ impl DeleteReplicationConfigSnapshot {
         &self.versioning
     }
 
-    pub fn replication_config(&self) -> Option<&PersistedReplicationConfiguration> {
+    pub fn replication_config(&self) -> Option<&ReplicationConfiguration> {
         self.metadata
             .as_ref()
             .and_then(|metadata| metadata.replication_config.as_ref())
@@ -103,7 +101,7 @@ impl DeleteReplicationConfigSnapshot {
     pub(crate) fn active_delete_marker_rules_require_tags(&self, object: &str) -> bool {
         self.replication_config().is_some_and(|config| {
             config.rules.iter().any(|rule| {
-                if rule.status == Status::DISABLED.as_str() {
+                if rule.status == ReplicationRuleStatus::from_static(ReplicationRuleStatus::DISABLED) {
                     return false;
                 }
                 if !object.starts_with(rule.prefix()) {
@@ -124,12 +122,12 @@ impl DeleteReplicationConfigSnapshot {
 
 fn validate_delete_replication_config(
     versioning: &VersioningConfiguration,
-    config: Option<&PersistedReplicationConfiguration>,
+    config: Option<&ReplicationConfiguration>,
 ) -> Result<()> {
     if versioning
         .status
         .as_ref()
-        .is_some_and(|status| !matches!(status.as_str(), "Enabled" | "Suspended"))
+        .is_some_and(|status| !matches!(status.as_str(), BucketVersioningStatus::ENABLED | BucketVersioningStatus::SUSPENDED))
     {
         return Err(super::replication_error_boundary::Error::other(
             "bucket versioning configuration has an invalid status",
@@ -146,7 +144,7 @@ fn validate_delete_replication_config(
         let role = config.role.trim();
         let mut role_destination = None;
         for rule in &config.rules {
-            if rule.status.as_str() == Status::ENABLED.as_str() {
+            if rule.status.as_str() == ReplicationRuleStatus::ENABLED {
                 let destination = rule.destination.bucket.trim();
                 if role.is_empty() && destination.is_empty() {
                     return Err(super::replication_error_boundary::Error::other(
@@ -171,7 +169,7 @@ fn validate_delete_replication_config(
     Ok(())
 }
 
-fn replication_config_from_metadata(metadata: &BucketMetadata) -> Result<Option<&PersistedReplicationConfiguration>> {
+fn replication_config_from_metadata(metadata: &BucketMetadata) -> Result<Option<&ReplicationConfiguration>> {
     if !metadata.replication_config_xml.is_empty() && metadata.replication_config.is_none() {
         return Err(super::replication_error_boundary::Error::other(
             "persisted bucket replication configuration is invalid",
@@ -252,15 +250,12 @@ pub(crate) async fn load_delete_replication_config_in(
 
 #[derive(Debug, Clone, Default, Serialize, Deserialize)]
 pub struct ReplicationConfig {
-    /// Carried in the scanner's persisted data usage cache, so its serde form
-    /// is the one the s3s DTO wrote there (see `config_cache_serde` in the replication crate).
-    #[serde(with = "super::replication_config_boundary::config_cache_serde")]
-    pub config: Option<PersistedReplicationConfiguration>,
+    pub config: Option<ReplicationConfiguration>,
     pub remotes: Option<BucketTargets>,
 }
 
 impl ReplicationConfig {
-    pub fn new(config: Option<PersistedReplicationConfiguration>, remotes: Option<BucketTargets>) -> Self {
+    pub fn new(config: Option<ReplicationConfiguration>, remotes: Option<BucketTargets>) -> Self {
         Self { config, remotes }
     }
 
@@ -473,7 +468,7 @@ fn check_replicate_delete_with_config(
     oi: &ObjectInfo,
     del_opts: &ObjectOptions,
     source_error: bool,
-    config: Option<&PersistedReplicationConfiguration>,
+    config: Option<&ReplicationConfiguration>,
     existing_delete_marker: bool,
     trust_persisted_replica_status: bool,
 ) -> ReplicateDecision {
@@ -592,7 +587,7 @@ pub(crate) async fn must_replicate(bucket: &str, object: &str, mopts: MustReplic
 }
 
 fn metadata_target_should_replicate(
-    cfg: &PersistedReplicationConfiguration,
+    cfg: &ReplicationConfiguration,
     opts: &ObjectOpts,
     mopts: &MustReplicateOptions,
     arn: &str,
@@ -602,21 +597,20 @@ fn metadata_target_should_replicate(
 
 #[cfg(test)]
 mod tests {
-    use rustfs_gateway_types::dto::Status;
-    use rustfs_gateway_types::persistence::{
-        PersistedOptionalReplicationStatus, PersistedReplicationDestination, PersistedReplicationFilter,
-        PersistedReplicationRule, PersistedReplicationStatus, PersistedReplicationTag, PersistedSourceSelectionCriteria,
+    use s3s::dto::{
+        DeleteMarkerReplication, DeleteMarkerReplicationStatus, DeleteReplication, DeleteReplicationStatus, Destination,
+        ReplicaModifications, ReplicationRule, ReplicationRuleFilter, ReplicationRuleStatus, SourceSelectionCriteria, Tag,
     };
 
     use super::super::replication_filemeta_boundary::VersionPurgeStatusType;
     use super::super::replication_target_boundary::BucketTarget;
     use super::*;
 
-    fn replication_rule() -> PersistedReplicationRule {
-        PersistedReplicationRule {
+    fn replication_rule() -> ReplicationRule {
+        ReplicationRule {
             delete_marker_replication: None,
             delete_replication: None,
-            destination: PersistedReplicationDestination {
+            destination: Destination {
                 bucket: "arn:aws:s3:::target-bucket".to_string(),
                 ..Default::default()
             },
@@ -626,7 +620,7 @@ mod tests {
             prefix: Some(String::new()),
             priority: Some(1),
             source_selection_criteria: None,
-            status: Status::ENABLED.to_string(),
+            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
         }
     }
 
@@ -635,11 +629,11 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.filter = Some(PersistedReplicationFilter {
+        rule.filter = Some(ReplicationRuleFilter {
             prefix: Some("admitted/".to_string()),
             ..Default::default()
         });
-        let cfg = PersistedReplicationConfiguration {
+        let cfg = ReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
@@ -677,7 +671,7 @@ mod tests {
         assert!(!empty.replicate(&ObjectOpts::default()));
 
         let config = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
                 rules: vec![replication_rule()],
             }),
@@ -731,7 +725,7 @@ mod tests {
     fn delete_snapshot_rejects_enabled_rules_without_a_destination() {
         let mut rule = replication_rule();
         rule.destination.bucket.clear();
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
@@ -747,7 +741,7 @@ mod tests {
         let first = replication_rule();
         let mut second = replication_rule();
         second.destination.bucket = "arn:aws:s3:::other-target".to_string();
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: "arn:aws:s3:::role-target".to_string(),
             rules: vec![first, second],
         };
@@ -758,47 +752,47 @@ mod tests {
     #[test]
     fn delete_snapshot_rejects_unknown_status_values() {
         let invalid_versioning = VersioningConfiguration {
-            status: Some("Enabld".to_string()),
+            status: Some("Enabld".to_string().into()),
             ..Default::default()
         };
         assert!(validate_delete_replication_config(&invalid_versioning, None).is_err());
 
         let mut invalid_rule = replication_rule();
-        invalid_rule.status = "Enabld".to_string();
-        let config = PersistedReplicationConfiguration {
+        invalid_rule.status = "Enabld".to_string().into();
+        let config = ReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_rule],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_delete = replication_rule();
-        invalid_delete.delete_replication = Some(PersistedReplicationStatus {
-            status: "Enabld".to_string(),
+        invalid_delete.delete_replication = Some(DeleteReplication {
+            status: "Enabld".to_string().into(),
         });
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_delete],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_delete_marker = replication_rule();
-        invalid_delete_marker.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some("Enabld".to_string()),
+        invalid_delete_marker.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some("Enabld".to_string().into()),
         });
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_delete_marker],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_replica_modifications = replication_rule();
-        invalid_replica_modifications.source_selection_criteria = Some(PersistedSourceSelectionCriteria {
-            replica_modifications: Some(PersistedReplicationStatus {
-                status: "Enabld".to_string(),
+        invalid_replica_modifications.source_selection_criteria = Some(SourceSelectionCriteria {
+            replica_modifications: Some(ReplicaModifications {
+                status: "Enabld".to_string().into(),
             }),
             sse_kms_encrypted_objects: None,
         });
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_replica_modifications],
         };
@@ -810,11 +804,11 @@ mod tests {
         let mut metadata = BucketMetadata::new("bucket");
         metadata.versioning_config_xml = b"configured".to_vec();
         metadata.versioning_config = Some(VersioningConfiguration {
-            status: Some("Enabled".to_owned()),
+            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
             ..Default::default()
         });
         metadata.replication_config_xml = b"configured".to_vec();
-        metadata.replication_config = Some(PersistedReplicationConfiguration {
+        metadata.replication_config = Some(ReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });
@@ -854,7 +848,7 @@ mod tests {
         );
 
         let mut inconsistent = BucketMetadata::new("bucket");
-        inconsistent.replication_config = Some(PersistedReplicationConfiguration {
+        inconsistent.replication_config = Some(ReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });
@@ -869,11 +863,11 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some(Status::ENABLED.to_string()),
+        rule.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
         });
         let mut metadata = BucketMetadata::new("bucket");
-        metadata.replication_config = Some(PersistedReplicationConfiguration {
+        metadata.replication_config = Some(ReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         });
@@ -904,27 +898,27 @@ mod tests {
     fn delete_marker_source_read_is_required_only_for_tag_filtered_rules() {
         let mut prefix_rule = replication_rule();
         prefix_rule.prefix = Some("logs/".to_string());
-        prefix_rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some(Status::ENABLED.to_string()),
+        prefix_rule.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
         });
 
         let prefix_snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some("Enabled".to_owned()),
+                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
                 ..Default::default()
             },
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
                 rules: vec![prefix_rule],
             }),
         );
 
         let mut tag_rule = replication_rule();
-        tag_rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some(Status::ENABLED.to_string()),
+        tag_rule.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
         });
-        tag_rule.filter = Some(PersistedReplicationFilter {
-            tag: Some(PersistedReplicationTag {
+        tag_rule.filter = Some(ReplicationRuleFilter {
+            tag: Some(Tag {
                 key: Some("class".to_string()),
                 value: Some("audit".to_string()),
             }),
@@ -932,10 +926,10 @@ mod tests {
         });
         let tag_snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some("Enabled".to_owned()),
+                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
                 ..Default::default()
             },
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
                 rules: vec![tag_rule],
             }),
@@ -950,14 +944,14 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_replication = Some(PersistedReplicationStatus {
-            status: Status::ENABLED.to_string(),
+        rule.delete_replication = Some(DeleteReplication {
+            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
         });
-        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some(Status::DISABLED.to_string()),
+        rule.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)),
         });
         let config = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
                 rules: vec![rule],
             }),
@@ -991,22 +985,22 @@ mod tests {
     #[test]
     fn live_delete_does_not_trust_persisted_replica_status() {
         let mut rule = replication_rule();
-        rule.delete_replication = Some(PersistedReplicationStatus {
-            status: Status::ENABLED.to_string(),
+        rule.delete_replication = Some(DeleteReplication {
+            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
         });
-        rule.source_selection_criteria = Some(PersistedSourceSelectionCriteria {
-            replica_modifications: Some(PersistedReplicationStatus {
-                status: Status::DISABLED.to_string(),
+        rule.source_selection_criteria = Some(SourceSelectionCriteria {
+            replica_modifications: Some(ReplicaModifications {
+                status: s3s::dto::ReplicaModificationsStatus::from_static(s3s::dto::ReplicaModificationsStatus::DISABLED),
             }),
             sse_kms_encrypted_objects: None,
         });
-        let replication = PersistedReplicationConfiguration {
+        let replication = ReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
         let snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some("Enabled".to_owned()),
+                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
                 ..Default::default()
             },
             Some(replication.clone()),
@@ -1042,14 +1036,14 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_replication = Some(PersistedReplicationStatus {
-            status: Status::ENABLED.to_string(),
+        rule.delete_replication = Some(DeleteReplication {
+            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
         });
-        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
-            status: Some(Status::DISABLED.to_string()),
+        rule.delete_marker_replication = Some(DeleteMarkerReplication {
+            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)),
         });
         let config = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
                 rules: vec![rule],
             }),

crates/ecstore/src/bucket/replication/replication_pool.rs

@@ -4794,10 +4794,9 @@ mod tests {
     #[tokio::test]
     async fn heal_queue_marks_missing_versioning_state_as_missed() {
         use super::super::replication_target_boundary::BucketTargets;
-        use rustfs_gateway_types::dto::Status;
-        use rustfs_gateway_types::persistence::{
-            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationRule,
-            PersistedReplicationStatus,
+        use s3s::dto::{
+            DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule,
+            ReplicationRuleStatus,
         };
 
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
@@ -4812,14 +4811,14 @@ mod tests {
                 ..Default::default()
             },
             ReplicationConfig::new(
-                Some(PersistedReplicationConfiguration {
+                Some(ReplicationConfiguration {
                     role: String::new(),
-                    rules: vec![PersistedReplicationRule {
+                    rules: vec![ReplicationRule {
                         delete_marker_replication: None,
-                        delete_replication: Some(PersistedReplicationStatus {
-                            status: Status::ENABLED.to_string(),
+                        delete_replication: Some(DeleteReplication {
+                            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
                         }),
-                        destination: PersistedReplicationDestination {
+                        destination: Destination {
                             bucket: arn.to_string(),
                             ..Default::default()
                         },
@@ -4829,7 +4828,7 @@ mod tests {
                         prefix: Some(String::new()),
                         priority: Some(1),
                         source_selection_criteria: None,
-                        status: Status::ENABLED.to_string(),
+                        status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
                     }],
                 }),
                 Some(BucketTargets::default()),

crates/ecstore/src/bucket/replication/replication_resyncer.rs

@@ -85,8 +85,6 @@ use http_body_util::StreamBody;
 use metrics::counter;
 #[cfg(test)]
 use rmp_serde;
-#[cfg(test)]
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 use rustfs_s3_types::EventName;
 use rustfs_utils::http::{
     AMZ_OBJECT_LOCK_LEGAL_HOLD, AMZ_OBJECT_LOCK_MODE, AMZ_OBJECT_LOCK_RETAIN_UNTIL_DATE, AMZ_TAGGING_DIRECTIVE,
@@ -94,6 +92,8 @@ use rustfs_utils::http::{
     insert_str, replication_target_versions,
 };
 use rustfs_utils::{DEFAULT_SIP_HASH_KEY, get_env_usize, sip_hash};
+#[cfg(test)]
+use s3s::dto::ReplicationConfiguration;
 use std::collections::{HashMap, HashSet};
 use std::fmt::Display;
 use std::pin::Pin;
@@ -6574,10 +6574,9 @@ mod tests {
 
     use super::super::replication_target_boundary::{BucketTarget, BucketTargets};
     use super::*;
-    use rustfs_gateway_types::dto::Status;
-    use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
-    use rustfs_gateway_types::persistence::{
-        PersistedReplicationDestination, PersistedReplicationRule, PersistedReplicationStatus,
+    use s3s::dto::{
+        BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ExcludedPrefix, ReplicationRule,
+        ReplicationRuleStatus, VersioningConfiguration,
     };
     use std::collections::HashMap;
     use time::OffsetDateTime;
@@ -8688,7 +8687,7 @@ mod tests {
             ..Default::default()
         };
         let rcfg = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: role.to_string(),
                 rules: vec![],
             }),
@@ -8714,7 +8713,7 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some("Enabled".to_owned()),
+                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
                 ..Default::default()
             },
         );
@@ -8747,20 +8746,22 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some("Enabled".to_owned()),
-                excluded_prefixes: Some(vec![Some("archive/".to_string())]),
+                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                excluded_prefixes: Some(vec![ExcludedPrefix {
+                    prefix: Some("archive/".to_string()),
+                }]),
                 ..Default::default()
             },
         );
         let rcfg = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
-                rules: vec![PersistedReplicationRule {
+                rules: vec![ReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(PersistedReplicationStatus {
-                        status: Status::ENABLED.to_string(),
+                    delete_replication: Some(DeleteReplication {
+                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
                     }),
-                    destination: PersistedReplicationDestination {
+                    destination: Destination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
@@ -8770,7 +8771,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: Status::ENABLED.to_string(),
+                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
                 }],
             }),
             Some(BucketTargets {
@@ -8806,14 +8807,14 @@ mod tests {
         let admitted_arn = "arn:rustfs:replication:us-east-1:target:admitted";
         let current_role = "arn:rustfs:replication:us-east-1:target:current";
         let rcfg = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: current_role.to_string(),
-                rules: vec![PersistedReplicationRule {
+                rules: vec![ReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(PersistedReplicationStatus {
-                        status: Status::DISABLED.to_string(),
+                    delete_replication: Some(DeleteReplication {
+                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED),
                     }),
-                    destination: PersistedReplicationDestination {
+                    destination: Destination {
                         bucket: current_role.to_string(),
                         ..Default::default()
                     },
@@ -8823,7 +8824,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: Status::ENABLED.to_string(),
+                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
                 }],
             }),
             Some(BucketTargets {
@@ -8871,12 +8872,12 @@ mod tests {
     #[tokio::test]
     async fn heal_rejects_semantically_invalid_replication_config() {
         let rcfg = ReplicationConfig::new(
-            Some(PersistedReplicationConfiguration {
+            Some(ReplicationConfiguration {
                 role: String::new(),
-                rules: vec![PersistedReplicationRule {
+                rules: vec![ReplicationRule {
                     delete_marker_replication: None,
                     delete_replication: None,
-                    destination: PersistedReplicationDestination {
+                    destination: Destination {
                         bucket: "arn:rustfs:replication:us-east-1:target:bucket".to_string(),
                         ..Default::default()
                     },
@@ -8886,7 +8887,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: "Enabld".to_owned(),
+                    status: ReplicationRuleStatus::from_static("Enabld"),
                 }],
             }),
             Some(BucketTargets::default()),

crates/ecstore/src/bucket/replication/replication_target_config_bridge.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
+use s3s::dto::ReplicationConfiguration;
 
 use super::replication_config_boundary::{ObjectOpts, ReplicationConfigurationExt};
 use super::replication_filemeta_boundary::ReplicationType;
@@ -20,7 +20,7 @@ use super::replication_filemeta_boundary::ReplicationType;
 pub(crate) struct ReplicationTargetConfigBridge;
 
 impl ReplicationTargetConfigBridge {
-    pub(crate) fn target_is_used_by_rules(config: &PersistedReplicationConfiguration, arn: &str) -> bool {
+    pub(crate) fn target_is_used_by_rules(config: &ReplicationConfiguration, arn: &str) -> bool {
         config
             .filter_target_arns(&ObjectOpts {
                 op_type: ReplicationType::All,
@@ -37,7 +37,7 @@ mod tests {
 
     #[test]
     fn target_config_bridge_matches_role_target() {
-        let config = PersistedReplicationConfiguration {
+        let config = ReplicationConfiguration {
             role: " arn:target ".to_string(),
             rules: Vec::new(),
         };

crates/ecstore/src/bucket/replication/replication_versioning_boundary.rs

@@ -15,7 +15,7 @@
 use super::replication_error_boundary::Result;
 use crate::bucket::{versioning::VersioningApi as _, versioning_sys::BucketVersioningSys};
 #[cfg(test)]
-use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+use s3s::dto::VersioningConfiguration;
 #[cfg(test)]
 use std::{collections::HashMap, sync::LazyLock, sync::Mutex};
 

crates/ecstore/src/bucket/tagging/mod.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
+use s3s::dto::Tag;
 use std::cmp::Ordering;
 use std::collections::HashMap;
 use url::form_urlencoded;

crates/ecstore/src/bucket/utils.rs

@@ -15,23 +15,9 @@
 use crate::disk::{MIGRATING_META_BUCKET, RUSTFS_META_BUCKET};
 use crate::error::{Error, Result, StorageError};
 use regex::Regex;
-use rustfs_gateway_types::cors_tagging::{
-    CorsTaggingCodecError, PersistedCorsConfiguration, PersistedTagging, parse_cors, parse_tagging, serialize_cors,
-    serialize_tagging,
-};
-use rustfs_gateway_types::persistence::{
-    PersistedAccelerateConfiguration, PersistedBucketEncryptionConfiguration, PersistedBucketLoggingStatus,
-    PersistedLifecycleConfiguration, PersistedNotificationConfiguration, PersistedObjectLockConfiguration,
-    PersistedPublicAccessBlockConfiguration, PersistedReplicationConfiguration, PersistedRequestPaymentConfiguration,
-    PersistedVersioningConfiguration, PersistedWebsiteConfiguration, PersistenceCodecError, parse_accelerate,
-    parse_bucket_encryption, parse_bucket_logging, parse_lifecycle_legacy_dates, parse_notification, parse_object_lock,
-    parse_public_access_block, parse_replication, parse_request_payment, parse_versioning, parse_website, serialize_accelerate,
-    serialize_bucket_encryption, serialize_bucket_logging, serialize_lifecycle_legacy_dates, serialize_notification,
-    serialize_object_lock, serialize_public_access_block, serialize_replication, serialize_request_payment, serialize_versioning,
-    serialize_website,
-};
 use rustfs_utils::WindowsFileNameError;
 use rustfs_utils::path::SLASH_SEPARATOR;
+use s3s::xml;
 use tracing::instrument;
 
 pub fn is_meta_bucketname(name: &str) -> bool {
@@ -88,173 +74,37 @@ pub fn check_valid_bucket_name_strict(bucket_name: &str) -> Result<()> {
     check_bucket_name_common(bucket_name, true)
 }
 
-/// A bucket configuration persisted as XML in the bucket metadata, together
-/// with the codec that reads and writes its stored bytes.
-///
-/// Every family reads and writes its stored bytes through the gateway
-/// persistence codecs (`rustfs_gateway_types::persistence` and
-/// `rustfs_gateway_types::cors_tagging`, rustfs/backlog#2745). The codecs keep
-/// the persisted bytes the `bucket_config_goldens` pin, so a document the old
-/// s3s codec wrote reads back and re-encodes unchanged.
-pub trait BucketConfigXml: Sized {
-    /// Decodes the stored bytes of one configuration document.
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError>;
-
-    /// Encodes the value as RustFS stores it.
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError>;
+// RUSTFS_COMPAT_TODO(s3gate-metadata-xml): the s3s codec reads persisted XML during migration. Remove after every supported writer uses the gateway codec and every retained metadata object and backup archive is verified or rewritten.
+pub fn deserialize<T>(input: &[u8]) -> xml::DeResult<T>
+where
+    T: for<'xml> xml::Deserialize<'xml>,
+{
+    let mut d = xml::Deserializer::new(input);
+    let ans = T::deserialize(&mut d)?;
+    d.expect_eof()?;
+    Ok(ans)
 }
 
-/// Why a persisted bucket configuration could not be decoded or encoded.
-#[derive(Debug, thiserror::Error)]
-pub enum BucketConfigXmlError {
-    /// A gateway persistence codec refused the bytes or the value.
-    #[error(transparent)]
-    Persisted(#[from] PersistenceCodecError),
-    /// The gateway CORS or Tagging codec refused the bytes or the value.
-    #[error(transparent)]
-    CorsTagging(#[from] CorsTaggingCodecError),
+#[allow(
+    dead_code,
+    reason = "xml serialize helper with no caller in this port; the live sibling is deserialize (backlog#1823)"
+)]
+pub fn serialize_content<T: xml::SerializeContent>(val: &T) -> xml::SerResult<String> {
+    let mut buf = Vec::with_capacity(256);
+    {
+        let mut ser = xml::Serializer::new(&mut buf);
+        val.serialize_content(&mut ser)?;
+    }
+    Ok(String::from_utf8(buf).unwrap())
 }
 
-/// Decodes one persisted bucket configuration document.
-pub fn deserialize<T: BucketConfigXml>(input: &[u8]) -> std::result::Result<T, BucketConfigXmlError> {
-    T::decode_config_xml(input)
-}
-
-/// Encodes one bucket configuration as RustFS persists it.
-pub fn serialize<T: BucketConfigXml>(val: &T) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-    val.encode_config_xml()
-}
-
-impl BucketConfigXml for PersistedReplicationConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_replication(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_replication(self)?)
-    }
-}
-
-impl BucketConfigXml for PersistedLifecycleConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_lifecycle_legacy_dates(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_lifecycle_legacy_dates(self)?)
-    }
-}
-
-// The gateway writers of these families are infallible: a value that exists is
-// always encodable, so only the readers can refuse bytes.
-impl BucketConfigXml for PersistedAccelerateConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_accelerate(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_accelerate(self))
-    }
-}
-
-impl BucketConfigXml for PersistedBucketLoggingStatus {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_bucket_logging(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_bucket_logging(self))
-    }
-}
-
-impl BucketConfigXml for PersistedCorsConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_cors(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_cors(self))
-    }
-}
-
-impl BucketConfigXml for PersistedNotificationConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_notification(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_notification(self))
-    }
-}
-
-impl BucketConfigXml for PersistedObjectLockConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_object_lock(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_object_lock(self))
-    }
-}
-
-impl BucketConfigXml for PersistedPublicAccessBlockConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_public_access_block(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_public_access_block(self))
-    }
-}
-
-impl BucketConfigXml for PersistedRequestPaymentConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_request_payment(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_request_payment(self))
-    }
-}
-
-impl BucketConfigXml for PersistedBucketEncryptionConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_bucket_encryption(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_bucket_encryption(self))
-    }
-}
-
-impl BucketConfigXml for PersistedTagging {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_tagging(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_tagging(self))
-    }
-}
-
-impl BucketConfigXml for PersistedVersioningConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_versioning(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_versioning(self))
-    }
-}
-
-impl BucketConfigXml for PersistedWebsiteConfiguration {
-    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
-        Ok(parse_website(input)?)
-    }
-
-    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
-        Ok(serialize_website(self))
+pub fn serialize<T: xml::Serialize>(val: &T) -> xml::SerResult<Vec<u8>> {
+    let mut buf = Vec::with_capacity(256);
+    {
+        let mut ser = xml::Serializer::new(&mut buf);
+        val.serialize(&mut ser)?;
     }
+    Ok(buf)
 }
 
 pub fn has_bad_path_component(path: &str) -> bool {
@@ -584,40 +434,6 @@ mod tests {
     use super::*;
     use proptest::prelude::*;
 
-    /// Every bucket-configuration family reads and writes through its own
-    /// gateway codec (rustfs/backlog#2745): the family's golden document
-    /// decodes and re-encodes to the same bytes, and a document of another
-    /// family is refused instead of decoding into an empty configuration.
-    #[test]
-    fn every_bucket_config_family_uses_its_own_gateway_codec() {
-        fn assert_family<T: BucketConfigXml>(family: &str, own: &[u8], foreign: &[u8]) {
-            let value = deserialize::<T>(own).unwrap_or_else(|err| panic!("{family}: own document must decode: {err}"));
-            let encoded = serialize(&value).unwrap_or_else(|err| panic!("{family}: value must encode: {err}"));
-            assert_eq!(encoded, own, "{family}: re-encoding must keep the persisted bytes");
-            assert!(deserialize::<T>(foreign).is_err(), "{family}: a foreign root element must be refused");
-        }
-
-        macro_rules! golden {
-            ($family:literal) => {
-                include_bytes!(concat!("../../tests/fixtures/bucket-config-goldens/", $family, "/minio.out.xml"))
-            };
-        }
-        let tagging = golden!("tagging");
-        let versioning = golden!("versioning");
-
-        assert_family::<PersistedAccelerateConfiguration>("accelerate", golden!("accelerate"), tagging);
-        assert_family::<PersistedBucketLoggingStatus>("logging", golden!("logging"), tagging);
-        assert_family::<PersistedCorsConfiguration>("cors", golden!("cors"), tagging);
-        assert_family::<PersistedNotificationConfiguration>("notification", golden!("notification"), tagging);
-        assert_family::<PersistedObjectLockConfiguration>("object_lock", golden!("object_lock"), tagging);
-        assert_family::<PersistedPublicAccessBlockConfiguration>("public_access_block", golden!("public_access_block"), tagging);
-        assert_family::<PersistedRequestPaymentConfiguration>("request_payment", golden!("request_payment"), tagging);
-        assert_family::<PersistedBucketEncryptionConfiguration>("encryption", golden!("encryption"), tagging);
-        assert_family::<PersistedTagging>("tagging", tagging, versioning);
-        assert_family::<PersistedVersioningConfiguration>("versioning", versioning, tagging);
-        assert_family::<PersistedWebsiteConfiguration>("website", golden!("website"), tagging);
-    }
-
     /// rustfs#5785: keys whose path segments exceed the on-disk NAME_MAX
     /// budget must be rejected up front as ObjectNameInvalid (4xx), not leak
     /// ENAMETOOLONG as InternalError 500 from the disk layer.

crates/ecstore/src/bucket/versioning/mod.rs

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 use rustfs_utils::string::match_simple;
+use s3s::dto::{BucketVersioningStatus, VersioningConfiguration};
 
 pub trait VersioningApi {
     fn enabled(&self) -> bool;
@@ -28,10 +28,10 @@ pub trait VersioningApi {
 
 impl VersioningApi for VersioningConfiguration {
     fn enabled(&self) -> bool {
-        self.status.as_deref() == Some("Enabled")
+        self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED))
     }
     fn prefix_enabled(&self, prefix: &str) -> bool {
-        if self.status.as_deref() != Some("Enabled") {
+        if self.status != Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
             return false;
         }
 
@@ -44,10 +44,12 @@ impl VersioningApi for VersioningConfiguration {
         }
 
         if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-            for sprefix in excluded_prefixes.iter().flatten() {
-                let pattern = format!("{sprefix}*");
-                if match_simple(&pattern, prefix) {
-                    return false;
+            for p in excluded_prefixes.iter() {
+                if let Some(ref sprefix) = p.prefix {
+                    let pattern = format!("{sprefix}*");
+                    if match_simple(&pattern, prefix) {
+                        return false;
+                    }
                 }
             }
         }
@@ -56,11 +58,11 @@ impl VersioningApi for VersioningConfiguration {
     }
 
     fn prefix_suspended(&self, prefix: &str) -> bool {
-        if self.status.as_deref() == Some("Suspended") {
+        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)) {
             return true;
         }
 
-        if self.status.as_deref() == Some("Enabled") {
+        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
             if prefix.is_empty() {
                 return false;
             }
@@ -73,10 +75,12 @@ impl VersioningApi for VersioningConfiguration {
             }
 
             if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-                for sprefix in excluded_prefixes.iter().flatten() {
-                    let pattern = format!("{sprefix}*");
-                    if match_simple(&pattern, prefix) {
-                        return true;
+                for p in excluded_prefixes.iter() {
+                    if let Some(ref sprefix) = p.prefix {
+                        let pattern = format!("{sprefix}*");
+                        if match_simple(&pattern, prefix) {
+                            return true;
+                        }
                     }
                 }
             }
@@ -88,12 +92,14 @@ impl VersioningApi for VersioningConfiguration {
         self.prefix_enabled(prefix) || self.prefix_suspended(prefix)
     }
     fn suspended(&self) -> bool {
-        self.status.as_deref() == Some("Suspended")
+        self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED))
     }
 }
 
 #[cfg(test)]
 mod tests {
+    use s3s::dto::{BucketVersioningStatus, ExcludedPrefix};
+
     use super::*;
 
     struct LegacyVersioning;
@@ -128,8 +134,10 @@ mod tests {
     #[test]
     fn delete_state_treats_excluded_prefixes_as_unversioned() {
         let config = VersioningConfiguration {
-            status: Some("Enabled".to_owned()),
-            excluded_prefixes: Some(vec![Some("archive/".to_string())]),
+            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+            excluded_prefixes: Some(vec![ExcludedPrefix {
+                prefix: Some("archive/".to_string()),
+            }]),
             ..Default::default()
         };
 
@@ -138,7 +146,7 @@ mod tests {
         assert_eq!(config.delete_state("live/object"), (true, false));
 
         let suspended = VersioningConfiguration {
-            status: Some("Suspended".to_owned()),
+            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)),
             ..Default::default()
         };
         assert_eq!(suspended.delete_state("archive/object"), (false, true));

crates/ecstore/src/bucket/versioning_sys.rs

@@ -17,7 +17,7 @@ use super::metadata::unreadable_config_refusal;
 use super::{metadata_sys::get_bucket_metadata_sys, versioning::VersioningApi};
 use crate::disk::RUSTFS_META_BUCKET;
 use crate::error::Result;
-use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+use s3s::dto::VersioningConfiguration;
 use tracing::{error, warn};
 
 pub struct BucketVersioningSys {}
@@ -163,7 +163,7 @@ mod tests {
 
     fn enabled() -> VersioningConfiguration {
         VersioningConfiguration {
-            status: Some("Enabled".to_owned()),
+            status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
             ..Default::default()
         }
     }

crates/ecstore/src/cache_value/metacache_set.rs

@@ -747,6 +747,16 @@ async fn list_path_raw_inner(
             // Start every missing head read in the same round so one stalled
             // disk cannot multiply the wait budget by the erasure-set width.
             // Outcomes are still consumed below in stable disk-index order.
+            if !revjob_rx.is_cancelled() {
+                for (i, reader) in readers.iter_mut().enumerate() {
+                    if errs[i].is_none()
+                        && pending_entries[i].is_none()
+                        && let Some(Ok(Some(entry))) = reader.try_peek_buffered()
+                    {
+                        pending_entries[i] = Some(entry);
+                    }
+                }
+            }
             let concurrent_peeks = readers.iter_mut().enumerate().filter_map(|(i, reader)| {
                 if errs[i].is_some() || pending_entries[i].is_some() {
                     return None;
@@ -878,7 +888,7 @@ async fn list_path_raw_inner(
                 }
                 if entry.name == current.name {
                     // If exact match, we agree.
-                    if current.matches_strict(&entry) {
+                    if current.matches_strict_borrowed(&entry) {
                         top_entries[i] = Some(entry);
                         agree += 1;
                     } else {
@@ -986,7 +996,9 @@ async fn list_path_raw_inner(
 
             if agree == readers.len() && !opts.preserve_replica_metadata {
                 for r in readers.iter_mut() {
-                    let _ = r.skip(1).await;
+                    if !r.skip_current() {
+                        let _ = r.skip(1).await;
+                    }
                 }
 
                 if let Some(agreed_fn) = opts.agreed.as_ref() {
@@ -1009,7 +1021,7 @@ async fn list_path_raw_inner(
             }
 
             for (i, r) in readers.iter_mut().enumerate() {
-                if top_entries[i].is_some() {
+                if top_entries[i].is_some() && !r.skip_current() {
                     let _ = r.skip(1).await;
                 }
             }

crates/ecstore/src/cluster/rpc/peer_rest_client.rs

@@ -1965,15 +1965,6 @@ impl PeerRestClient {
         Ok((self.topology_member.clone(), epoch))
     }
 
-    pub async fn probe_restore_copyback_capability(&self, topology_fingerprint: String) -> Result<(String, Uuid)> {
-        let probe = rustfs_protos::restore_copyback_capability_probe(Uuid::new_v4().as_bytes());
-        let result = self
-            .heal_control(rustfs_protos::HEAL_CONTROL_PROTOCOL_VERSION, topology_fingerprint, probe)
-            .await?;
-        let epoch = decode_remote_version_state_capability(&self.topology_member, &result)?;
-        Ok((self.topology_member.clone(), epoch))
-    }
-
     pub async fn probe_cross_pool_fence(&self, topology_fingerprint: String) -> Result<(String, u32, Uuid)> {
         let mut probe = rustfs_protos::CROSS_POOL_FENCE_CAPABILITY_PROBE_PREFIX.to_vec();
         probe.extend_from_slice(Uuid::new_v4().as_bytes());

crates/ecstore/src/cluster/rpc/remote_disk.rs

@@ -67,14 +67,15 @@ use rustfs_protos::proto_gen::node_service::{
     BatchReadVersionRequest, BatchReadVersionResponse, CheckPartsRequest, CompareAndUpdateFileOutcome,
     CompareAndUpdateFileRequest, DeletePathsRequest, DeleteRequest, DeleteVersionRequest, DeleteVersionsRequest,
     DeleteVersionsResponse, DeleteVolumeRequest, DiskInfoRequest, GuardedDeleteVersionsRequest, ListDirRequest,
-    ListVolumesRequest, MakeVolumeRequest, MakeVolumesRequest, PoolMetaNativeRecoveryRequest as PoolMetaNativeRecoveryRpcRequest,
-    PreparePartTransactionRequest, ReadAllRequest, ReadMetadataRequest, ReadMultipleRequest, ReadMultipleResponse,
-    ReadPartsRequest, ReadVersionRequest, ReadXlRequest, RecoverDeleteTransactionRequest,
-    RecoverMultipartSourceRequest as RecoverMultipartSourceRpcRequest,
+    ListVolumesRequest, MakeVolumeRequest, MakeVolumesRequest, MetadataReservationActivateRequest,
+    MetadataReservationCancelRequest, MetadataReservationRequest,
+    PoolMetaNativeRecoveryRequest as PoolMetaNativeRecoveryRpcRequest, PreparePartTransactionRequest, ReadAllRequest,
+    ReadMetadataRequest, ReadMultipleRequest, ReadMultipleResponse, ReadPartsRequest, ReadVersionRequest, ReadXlRequest,
+    RecoverDeleteTransactionRequest, RecoverMultipartSourceRequest as RecoverMultipartSourceRpcRequest,
     RecoverMultipartSourceResponse as RecoverMultipartSourceRpcResponse, RenameDataRequest, RenameFileRequest,
-    SettlePartTransactionRequest, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest, SnapshotLeaseRequest,
-    SnapshotLeaseResponse, StatVolumeRequest, UpdateMetadataRequest, VerifyFileRequest, WriteAllRequest, WriteMetadataRequest,
-    node_service_client::NodeServiceClient,
+    ReservedRenameDataRequest, SettlePartTransactionRequest, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest,
+    SnapshotLeaseRequest, SnapshotLeaseResponse, StatVolumeRequest, UpdateMetadataRequest, VerifyFileRequest, WriteAllRequest,
+    WriteMetadataRequest, node_service_client::NodeServiceClient,
 };
 use serde::{Serialize, de::DeserializeOwned};
 use std::{
@@ -2452,6 +2453,212 @@ fn validate_decoded_file_info(file_info: &FileInfo) -> Result<()> {
 }
 
 impl RemoteDisk {
+    /// Prepare a remote token without retrying the RPC. It owns the bucket
+    /// publication fence but does not enter the object metadata queue yet.
+    pub async fn reserve_object_metadata(
+        &self,
+        bucket: &str,
+        object: &str,
+        incarnation: Uuid,
+        timeout: Duration,
+    ) -> Result<Uuid> {
+        if timeout.is_zero() {
+            return Err(DiskError::Timeout);
+        }
+        self.execute_with_timeout_for_op_and_health_action(
+            "reserve_object_metadata",
+            || async {
+                let mut client = self.get_client().await?;
+                let mut request = Request::new(MetadataReservationRequest {
+                    disk: self.endpoint.to_string(),
+                    bucket: bucket.to_owned(),
+                    object: object.to_owned(),
+                    timeout_ms: u64::try_from(timeout.as_millis()).map_err(|_| Error::other("reservation timeout overflow"))?,
+                    bucket_incarnation_id: incarnation.as_bytes().to_vec().into(),
+                });
+                let body = rustfs_protos::canonical_metadata_reservation_request_body(request.get_ref())
+                    .map_err(|_| Error::other("reservation request length overflow"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .reserve_object_metadata(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                if !response.success {
+                    return Err(response.error.map(Into::into).unwrap_or(DiskError::Unexpected));
+                }
+                if response.protocol_version != 2 {
+                    return Err(DiskError::UnsupportedDisk);
+                }
+                Uuid::from_slice(&response.token)
+                    .ok()
+                    .filter(|token| !token.is_nil())
+                    .ok_or(DiskError::Unexpected)
+            },
+            timeout.min(Duration::from_secs(5)),
+            FailureHealthAction::IgnoreFailure,
+        )
+        .await
+    }
+
+    /// Enqueue a prepared token. The caller must cancel it on an indeterminate
+    /// result before allowing any subsequent unprotected mutation.
+    pub async fn activate_object_metadata(&self, bucket: &str, object: &str, token: Uuid, timeout: Duration) -> Result<()> {
+        self.update_object_metadata_reservation(bucket, object, token, timeout, false)
+            .await
+    }
+
+    pub async fn renew_object_metadata(&self, bucket: &str, object: &str, token: Uuid, timeout: Duration) -> Result<()> {
+        self.update_object_metadata_reservation(bucket, object, token, timeout, true)
+            .await
+    }
+
+    async fn update_object_metadata_reservation(
+        &self,
+        bucket: &str,
+        object: &str,
+        token: Uuid,
+        timeout: Duration,
+        renew_only: bool,
+    ) -> Result<()> {
+        if timeout.is_zero() {
+            return Err(DiskError::Timeout);
+        }
+        self.execute_with_timeout_for_op_and_health_action(
+            "activate_object_metadata_reservation",
+            || async {
+                let mut client = self.get_client().await?;
+                let mut request = Request::new(MetadataReservationActivateRequest {
+                    disk: self.endpoint.to_string(),
+                    bucket: bucket.to_owned(),
+                    object: object.to_owned(),
+                    token: token.as_bytes().to_vec().into(),
+                    renew_only,
+                });
+                let body = rustfs_protos::canonical_metadata_reservation_activate_request_body(request.get_ref())
+                    .map_err(|_| Error::other("reservation activation request length overflow"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .activate_object_metadata_reservation(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                if !response.success {
+                    return Err(response.error.map(Into::into).unwrap_or(DiskError::Unexpected));
+                }
+                Ok(())
+            },
+            timeout.min(Duration::from_secs(5)),
+            FailureHealthAction::IgnoreFailure,
+        )
+        .await
+    }
+
+    /// Cancel an unclaimed remote reservation. False cannot prove that an
+    /// earlier claim did not start a physical mutation.
+    pub async fn cancel_object_metadata_reservation(&self, token: Uuid) -> Result<bool> {
+        self.execute_with_timeout_for_op(
+            "cancel_object_metadata_reservation",
+            || async {
+                let mut client = self.get_client().await?;
+                let mut request = Request::new(MetadataReservationCancelRequest {
+                    disk: self.endpoint.to_string(),
+                    token: token.as_bytes().to_vec().into(),
+                });
+                let body = rustfs_protos::canonical_metadata_reservation_cancel_request_body(request.get_ref())
+                    .map_err(|_| Error::other("reservation cancellation request length overflow"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .cancel_object_metadata_reservation(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                if let Some(error) = response.error {
+                    return Err(error.into());
+                }
+                Ok(response.success)
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+
+    /// Publish a staged user-object write using a remote one-shot reservation.
+    /// The caller must retain the bucket lifecycle fence until the tail drains.
+    pub async fn rename_data_with_metadata_reservation(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+        token: Uuid,
+    ) -> Result<RenameDataResp> {
+        self.execute_with_timeout_for_op(
+            "rename_reserved_data",
+            || async {
+                let mut client = self.get_client().await?;
+                let rename = RenameDataRequest {
+                    disk: self.endpoint.to_string(),
+                    src_volume: src_volume.to_owned(),
+                    src_path: src_path.to_owned(),
+                    file_info: compat_json(fi)?,
+                    dst_volume: dst_volume.to_owned(),
+                    dst_path: dst_path.to_owned(),
+                    file_info_bin: encode_file_info_msgpack(fi)?.into(),
+                    scanner_publication_lease_token: Bytes::new(),
+                    bucket_incarnation_id: Bytes::new(),
+                };
+                let mut request = Request::new(ReservedRenameDataRequest {
+                    rename: Some(rename),
+                    token: token.as_bytes().to_vec().into(),
+                });
+                let body = rustfs_protos::canonical_reserved_rename_data_request_body(request.get_ref())
+                    .map_err(|_| Error::other("reserved rename request length overflow"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .rename_reserved_data(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                if !response.success {
+                    return Err(response.error.map(Into::into).unwrap_or(DiskError::Unexpected));
+                }
+                decode_msgpack_or_json::<RenameDataResp>(
+                    &response.rename_data_resp_bin,
+                    &response.rename_data_resp,
+                    "RenameDataResp",
+                )
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+
     #[tracing::instrument(level = "trace", skip_all)]
     pub(crate) async fn rename_data_borrowed(
         &self,
@@ -5109,6 +5316,63 @@ mod tests {
         peer.stop().await;
     }
 
+    #[tokio::test]
+    async fn metadata_reservation_old_peer_rejects_without_legacy_fallback() {
+        runtime_sources::ensure_test_rpc_secret();
+        let peer = TestGrpcPeer::spawn(Bytes::new(), Bytes::new())
+            .await
+            .expect("old peer must bind");
+        let remote = remote_disk_for_addr(&peer.addr).await;
+        let token = Uuid::new_v4();
+        assert!(matches!(
+            remote
+                .reserve_object_metadata("bucket", "object", Uuid::new_v4(), Duration::from_secs(5))
+                .await,
+            Err(DiskError::UnsupportedDisk)
+        ));
+        assert!(matches!(
+            remote
+                .activate_object_metadata("bucket", "object", token, Duration::from_secs(5))
+                .await,
+            Err(DiskError::UnsupportedDisk)
+        ));
+        assert!(matches!(
+            remote
+                .renew_object_metadata("bucket", "object", token, Duration::from_secs(5))
+                .await,
+            Err(DiskError::UnsupportedDisk)
+        ));
+        assert!(matches!(
+            remote.cancel_object_metadata_reservation(token).await,
+            Err(DiskError::UnsupportedDisk)
+        ));
+        assert!(matches!(
+            remote
+                .rename_data_with_metadata_reservation(
+                    crate::disk::RUSTFS_META_TMP_BUCKET,
+                    "staged",
+                    &sample_rename_data_file_info(),
+                    "bucket",
+                    "object",
+                    token,
+                )
+                .await,
+            Err(DiskError::UnsupportedDisk)
+        ));
+        assert_eq!(
+            *peer.peer.unsupported_methods.lock().expect("observed methods lock"),
+            vec![
+                "/node_service.NodeService/ReserveObjectMetadata",
+                "/node_service.NodeService/ActivateObjectMetadataReservation",
+                "/node_service.NodeService/ActivateObjectMetadataReservation",
+                "/node_service.NodeService/CancelObjectMetadataReservation",
+                "/node_service.NodeService/RenameReservedData",
+            ],
+            "reservation operations must never downgrade to legacy methods",
+        );
+        peer.stop().await;
+    }
+
     #[tokio::test]
     async fn quota_fence_old_peer_rejects_without_legacy_fallback() {
         runtime_sources::ensure_test_rpc_secret();

crates/ecstore/src/cluster/rpc/remote_locker.rs

@@ -1315,6 +1315,122 @@ impl LockClient for RemoteClient {
         Ok(resp.success)
     }
 
+    async fn downgrade_committed_tail(&self, lock_id: &LockId) -> Result<bool> {
+        let marker = rustfs_lock::client::COMMITTED_TAIL_CONVERSION_MARKER;
+        let mut request = Self::create_unlock_request(lock_id);
+        request.metadata.tags.insert(marker.to_owned(), marker.to_owned());
+        let mut client = self.get_client().await?;
+        let resource_summary = request.resource.to_string();
+        let mut req = Request::new(GenerallyLockRequest {
+            args: serde_json::to_string(&request)
+                .map_err(|error| LockError::internal(format!("Failed to serialize request: {error}")))?,
+        });
+        attach_lock_mutation_body_digest(&mut req)?;
+        let response = self
+            .execute_rpc(
+                "refresh",
+                &resource_summary,
+                Self::rpc_timeout(),
+                async move { client.refresh(req).await },
+                RpcCleanup::new(None, None, None),
+            )
+            .await?
+            .into_inner();
+        if let Some(error) = response.error_info {
+            return Err(LockError::internal(error));
+        }
+        // Older nodes refresh normally but return no conversion marker.
+        Ok(response.success && response.lock_info.as_deref() == Some(marker))
+    }
+
+    async fn try_upgrade_committed_tail(&self, lock_id: &LockId) -> Result<bool> {
+        let marker = rustfs_lock::client::COMMITTED_TAIL_UPGRADE_MARKER;
+        let mut request = Self::create_unlock_request(lock_id);
+        request.metadata.tags.insert(marker.to_owned(), marker.to_owned());
+        let mut client = self.get_client().await?;
+        let resource_summary = request.resource.to_string();
+        let mut req = Request::new(GenerallyLockRequest {
+            args: serde_json::to_string(&request)
+                .map_err(|error| LockError::internal(format!("Failed to serialize request: {error}")))?,
+        });
+        attach_lock_mutation_body_digest(&mut req)?;
+        let response = self
+            .execute_rpc(
+                "refresh",
+                &resource_summary,
+                Self::rpc_timeout(),
+                async move { client.refresh(req).await },
+                RpcCleanup::new(None, None, None),
+            )
+            .await?
+            .into_inner();
+        if let Some(error) = response.error_info {
+            return Err(LockError::internal(error));
+        }
+        Ok(response.success && response.lock_info.as_deref() == Some(marker))
+    }
+
+    async fn supports_heal_lock_transition(&self, lock_id: &LockId) -> Result<bool> {
+        let marker = rustfs_lock::client::HEAL_LOCK_TRANSITION_PROBE_MARKER;
+        let mut request = Self::create_unlock_request(lock_id);
+        request.metadata.tags.insert(marker.to_owned(), marker.to_owned());
+        let mut client = self.get_client().await?;
+        let resource_summary = request.resource.to_string();
+        let mut req = Request::new(GenerallyLockRequest {
+            args: serde_json::to_string(&request)
+                .map_err(|error| LockError::internal(format!("Failed to serialize request: {error}")))?,
+        });
+        attach_lock_mutation_body_digest(&mut req)?;
+        let response = self
+            .execute_rpc(
+                "refresh",
+                &resource_summary,
+                Self::rpc_timeout(),
+                async move { client.refresh(req).await },
+                RpcCleanup::new(None, None, None),
+            )
+            .await?
+            .into_inner();
+        if let Some(error) = response.error_info {
+            return Err(LockError::internal(error));
+        }
+        Ok(response.success && response.lock_info.as_deref() == Some(marker))
+    }
+
+    async fn waiting_writer(&self, lock_id: &LockId) -> Result<Option<bool>> {
+        let marker = rustfs_lock::client::HEAL_WAITING_WRITER_MARKER;
+        let mut request = Self::create_unlock_request(lock_id);
+        request.metadata.tags.insert(marker.to_owned(), marker.to_owned());
+        let mut client = self.get_client().await?;
+        let resource_summary = request.resource.to_string();
+        let mut req = Request::new(GenerallyLockRequest {
+            args: serde_json::to_string(&request)
+                .map_err(|error| LockError::internal(format!("Failed to serialize request: {error}")))?,
+        });
+        attach_lock_mutation_body_digest(&mut req)?;
+        let response = self
+            .execute_rpc(
+                "refresh",
+                &resource_summary,
+                Self::rpc_timeout(),
+                async move { client.refresh(req).await },
+                RpcCleanup::new(None, None, None),
+            )
+            .await?
+            .into_inner();
+        if let Some(error) = response.error_info {
+            return Err(LockError::internal(error));
+        }
+        if !response.success {
+            return Ok(None);
+        }
+        Ok(match response.lock_info.as_deref().and_then(|info| info.strip_prefix(marker)) {
+            Some(":0") => Some(false),
+            Some(":1") => Some(true),
+            _ => None,
+        })
+    }
+
     async fn force_release(&self, lock_id: &LockId) -> Result<bool> {
         info!("remote force_release for {}", lock_id);
         let force_request = Self::create_unlock_request(lock_id);

crates/ecstore/src/config/com.rs

@@ -3506,6 +3506,7 @@ mod tests {
                 replication_decision: String::new(),
                 checksum: None,
                 multipart_completion_replayed: false,
+                backfilled_old_current_size: None,
             }
         }
     }

crates/ecstore/src/core/pools.rs

@@ -84,13 +84,12 @@ use rand::RngExt as _;
 use rmp_serde::Deserializer;
 use rmp_serde::Serializer;
 use rustfs_filemeta::{FileInfo, FileInfoVersions, MetaCacheEntries, MetaCacheEntry, MetadataResolutionParams};
-use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
-use rustfs_gateway_types::persistence::{PersistedLifecycleConfiguration, PersistedReplicationConfiguration};
 use rustfs_heal_contracts::heal_channel::HealOpts;
 use rustfs_utils::crypto::{hex_sha256, is_sha256_checksum};
 use rustfs_utils::path::{
     decode_dir_object, encode_dir_object, path_join, path_to_bucket_object, path_to_bucket_object_with_base_path,
 };
+use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration, ReplicationConfiguration};
 use serde::{Deserialize, Serialize};
 use sha2::{Digest, Sha256};
 use std::collections::{HashMap, HashSet};
@@ -12884,7 +12883,7 @@ pub(crate) async fn should_skip_lifecycle_for_data_movement(
     store: Arc<ECStore>,
     bucket: &str,
     version: &rustfs_filemeta::FileInfo,
-    lifecycle_config: Option<&PersistedLifecycleConfiguration>,
+    lifecycle_config: Option<&BucketLifecycleConfiguration>,
     object_lock_config: Option<&ObjectLockConfiguration>,
     apply_actions: bool,
     event_source: &LcEventSrc,
@@ -16813,9 +16812,9 @@ impl ECStore {
         generation: OffsetDateTime,
         bucket: String,
         set: Arc<SetDisks>,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
         object_lock_config: Option<ObjectLockConfiguration>,
-        replication_config: Option<(PersistedReplicationConfiguration, OffsetDateTime)>,
+        replication_config: Option<(ReplicationConfiguration, OffsetDateTime)>,
         expected_bucket_incarnation_id: Option<uuid::Uuid>,
         source_changed_exhaustions: Arc<AtomicUsize>,
         entry_budget: Arc<Semaphore>,
@@ -16987,9 +16986,9 @@ impl ECStore {
         mut deferred_entry: DeferredDecommissionEntry,
         bucket: String,
         set: Arc<SetDisks>,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
         object_lock_config: Option<ObjectLockConfiguration>,
-        replication_config: Option<(PersistedReplicationConfiguration, OffsetDateTime)>,
+        replication_config: Option<(ReplicationConfiguration, OffsetDateTime)>,
         expected_bucket_incarnation_id: Option<uuid::Uuid>,
         source_changed_exhaustions: Arc<AtomicUsize>,
         entry_budget: Arc<Semaphore>,
@@ -17096,9 +17095,9 @@ impl ECStore {
         generation: OffsetDateTime,
         set: Arc<SetDisks>,
         bi: DecomBucketInfo,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
         object_lock_config: Option<ObjectLockConfiguration>,
-        replication_config: Option<(PersistedReplicationConfiguration, OffsetDateTime)>,
+        replication_config: Option<(ReplicationConfiguration, OffsetDateTime)>,
         expected_bucket_incarnation_id: Option<uuid::Uuid>,
         source_changed_exhaustions: Arc<AtomicUsize>,
         entry_budget: Arc<Semaphore>,
@@ -17773,9 +17772,9 @@ impl ECStore {
         entry: MetaCacheEntry,
         bucket: String,
         set: Arc<SetDisks>,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
         object_lock_config: Option<ObjectLockConfiguration>,
-        replication_config: Option<(PersistedReplicationConfiguration, OffsetDateTime)>,
+        replication_config: Option<(ReplicationConfiguration, OffsetDateTime)>,
         expected_bucket_incarnation_id: Option<uuid::Uuid>,
         source_changed_exhaustions: Arc<AtomicUsize>,
         attempts: &mut usize,
@@ -17959,9 +17958,9 @@ impl ECStore {
         entry: MetaCacheEntry,
         bucket: String,
         set: Arc<SetDisks>,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
         object_lock_config: Option<ObjectLockConfiguration>,
-        replication_config: Option<(PersistedReplicationConfiguration, OffsetDateTime)>,
+        replication_config: Option<(ReplicationConfiguration, OffsetDateTime)>,
         expected_bucket_incarnation_id: Option<uuid::Uuid>,
         entry_attempt: usize,
         scanner_backlog_handoff_attempt: usize,
@@ -19247,7 +19246,7 @@ impl ECStore {
         entry: MetaCacheEntry,
         bucket: String,
         set: Arc<SetDisks>,
-        lifecycle_config: Option<PersistedLifecycleConfiguration>,
+        lifecycle_config: Option<BucketLifecycleConfiguration>,
     ) -> Result<()> {
         let expected_bucket_incarnation_id = if is_meta_bucketname(&bucket) {
             None

crates/ecstore/src/core/sets.rs

@@ -65,38 +65,6 @@ use uuid::Uuid;
 
 const LIST_MULTIPART_SETS_CONCURRENCY: usize = 4;
 
-#[cfg(test)]
-type TestSetLockers = HashMap<usize, Vec<Arc<dyn rustfs_lock::LockClient>>>;
-
-#[cfg(test)]
-static TEST_SET_LOCKERS: std::sync::LazyLock<std::sync::Mutex<TestSetLockers>> =
-    std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
-
-/// Keep a fixture's real locker domain scoped to its owning instance.
-#[cfg(test)]
-pub(crate) struct ScopedSetLockers(Arc<InstanceContext>);
-
-#[cfg(test)]
-impl ScopedSetLockers {
-    pub(crate) fn install(ctx: Arc<InstanceContext>, lockers: Vec<Arc<dyn rustfs_lock::LockClient>>) -> Self {
-        let key = Arc::as_ptr(&ctx) as usize;
-        let mut domains = TEST_SET_LOCKERS.lock().expect("test set lockers mutex");
-        assert!(!domains.contains_key(&key), "the instance already owns a test locker domain");
-        domains.insert(key, lockers);
-        Self(ctx)
-    }
-}
-
-#[cfg(test)]
-impl Drop for ScopedSetLockers {
-    fn drop(&mut self) {
-        TEST_SET_LOCKERS
-            .lock()
-            .expect("test set lockers mutex")
-            .remove(&(Arc::as_ptr(&self.0) as usize));
-    }
-}
-
 fn is_idempotent_delete_prefix_error(err: &Error) -> bool {
     is_err_object_not_found(err) || is_err_strict_volume_not_found(err)
 }
@@ -298,14 +266,6 @@ impl Sets {
                 }
             }
 
-            #[cfg(test)]
-            let pool_lockers = TEST_SET_LOCKERS
-                .lock()
-                .expect("test set lockers mutex")
-                .get(&(Arc::as_ptr(&instance_ctx) as usize))
-                .cloned()
-                .unwrap_or_else(|| pool_lockers.clone());
-
             let set_disks = SetDisks::new_with_instance_ctx(
                 runtime_sources::local_node_name().await,
                 Arc::new(RwLock::new(set_drive)),
@@ -695,6 +655,19 @@ impl Sets {
             .await
     }
 
+    pub(crate) async fn put_object_with_prepared_reservations(
+        &self,
+        bucket: &str,
+        object: &str,
+        data: &mut PutObjReader,
+        opts: &ObjectOptions,
+        reservations: Option<Vec<Option<crate::set_disk::RenameMetadataReservation>>>,
+    ) -> Result<(ObjectInfo, Option<crate::disk::OldCurrentSize>)> {
+        self.get_disks_by_key(object)
+            .put_object_with_prepared_reservations(bucket, object, data, opts, reservations)
+            .await
+    }
+
     pub(crate) async fn put_object_with_old_current_size_for_data_movement(
         &self,
         bucket: &str,

crates/ecstore/src/disk/disk_store.rs

@@ -604,8 +604,69 @@ impl LocalDiskWrapper {
         let super::RenameDataGuards {
             external_guard,
             namespace_owner,
+            metadata_reservation,
+            metadata_reservation_token,
             ..
         } = guards;
+        let metadata_reservation = match (metadata_reservation, metadata_reservation_token) {
+            (Some(_), Some(_)) => return super::RenameDataObservation::unknown(Err(DiskError::InvalidPath)),
+            (Some(reservation), None) => Some(reservation),
+            (None, Some(token)) => match self.disk.claim_object_metadata_reservation(token, dst_volume, dst_path) {
+                Ok(reservation) => Some(reservation),
+                // A missing token may already have been claimed by an earlier
+                // request that is still publishing; do not claim it was safe.
+                Err(DiskError::DiskOngoingReq) => {
+                    return super::RenameDataObservation::unknown(Err(DiskError::DiskOngoingReq));
+                }
+                Err(error) => return super::RenameDataObservation::not_dispatched(error),
+            },
+            (None, None) => None,
+        };
+        // A queued reservation must acquire its metadata lease before disk
+        // admission. Otherwise a waiting background mutation could hold the
+        // disk permit needed by the earlier lease owner.
+        let metadata_reservation = match metadata_reservation {
+            Some(reservation) => Some(reservation.wait().await),
+            None => None,
+        };
+        // The claimed reservation stays in this cancellable future while disk
+        // admission can wait. Only physical work moves into an owned task.
+        let prechecked_permit = if metadata_reservation.is_some() {
+            // Background admission is exclusive. Waiting for it while holding
+            // this object's lease can deadlock an admitted mutation waiting for
+            // the same lease. An already owned permit does not need to wait.
+            if current_disk_io_context().is_some_and(|context| {
+                context.class == DiskIoClass::Background
+                    && !context
+                        .permit
+                        .as_ref()
+                        .is_some_and(|permit| permit.is_for(&self.io_admission))
+            }) {
+                return super::RenameDataObservation::not_dispatched(DiskError::DiskOngoingReq);
+            }
+            let preflight = self
+                .with_native_resources(async {
+                    if self.health.is_faulty() {
+                        return Err(DiskError::FaultyDisk);
+                    }
+                    self.check_disk_stale().await?;
+                    self.acquire_disk_io_permit().await
+                })
+                .await;
+            match preflight {
+                Ok(Ok(Some(permit))) => Some(permit),
+                Ok(Ok(None)) => return super::RenameDataObservation::not_dispatched(DiskError::DiskOngoingReq),
+                Ok(Err(error)) | Err(error) => return super::RenameDataObservation::not_dispatched(error),
+            }
+        } else {
+            None
+        };
+        // The reserved metadata lease must outlive a cancelled RPC waiter.
+        let external_guard: Option<Arc<dyn Send + Sync>> = if metadata_reservation.is_some() {
+            Some(Arc::new(external_guard))
+        } else {
+            external_guard
+        };
         let operation = self.clone();
         let src_volume = src_volume.to_owned();
         let src_path = src_path.to_owned();
@@ -624,27 +685,54 @@ impl LocalDiskWrapper {
         };
         let observed = run_owned_mutation(external_guard, move || async move {
             let mut preflight_rejection = None;
-            let result = operation
-                .track_disk_health_mutation(
-                    "rename_data",
-                    DiskMetricMutation::Write,
-                    || async {
-                        // Preserve the former DiskAPI future's single boxing boundary.
-                        let observed = Box::pin(operation.disk.rename_data_observed(
-                            &src_volume,
-                            &src_path,
-                            &fi,
-                            &dst_volume,
-                            &dst_path,
-                            namespace_owner,
+            let rename = || async {
+                // Preserve the former DiskAPI future's single boxing boundary.
+                let observed = if let Some(reservation) = metadata_reservation {
+                    Box::pin(operation.disk.rename_data_observed_with_reservation(
+                        &src_volume,
+                        &src_path,
+                        &fi,
+                        &dst_volume,
+                        &dst_path,
+                        (namespace_owner, Some(reservation)),
+                    ))
+                    .await
+                } else {
+                    Box::pin(operation.disk.rename_data_observed(
+                        &src_volume,
+                        &src_path,
+                        &fi,
+                        &dst_volume,
+                        &dst_path,
+                        namespace_owner,
+                    ))
+                    .await
+                };
+                preflight_rejection = observed.preflight_rejection;
+                observed.result
+            };
+            let result = if let Some(permit) = prechecked_permit {
+                with_disk_io_permit(Some(permit), async {
+                    operation
+                        .with_native_resources(operation.track_admitted_disk_operation(
+                            "rename_data",
+                            move |io_permit| async move {
+                                let _io_permit = io_permit;
+                                rename().await
+                            },
+                            Duration::ZERO,
+                            TimeoutHealthAction::MarkFailure,
+                            DiskMetricMutation::Write,
+                            true,
                         ))
-                        .await;
-                        preflight_rejection = observed.preflight_rejection;
-                        observed.result
-                    },
-                    timeout_duration,
-                )
-                .await;
+                        .await?
+                })
+                .await
+            } else {
+                operation
+                    .track_disk_health_mutation("rename_data", DiskMetricMutation::Write, rename, timeout_duration)
+                    .await
+            };
             // Health tracking must observe the real disk error, not an Ok tuple.
             Ok(super::RenameDataObservation {
                 result,
@@ -1598,6 +1686,67 @@ pub struct LocalDiskWrapper {
 }
 
 impl LocalDiskWrapper {
+    /// Reserve the metadata mutation lock for one object on this disk instance.
+    /// Unclaimed reservations expire after 60 seconds.
+    pub async fn reserve_object_metadata(&self, bucket: &str, object: &str, timeout: Duration) -> Result<Uuid> {
+        if self.health.is_faulty() {
+            return Err(DiskError::FaultyDisk);
+        }
+        self.check_disk_stale().await?;
+        self.disk.reserve_object_metadata(bucket, object, timeout).await
+    }
+
+    /// Register an object metadata reservation in FIFO order on this disk.
+    /// The returned token may be claimed before its physical lease is available.
+    pub async fn enqueue_object_metadata(&self, bucket: &str, object: &str) -> Result<Uuid> {
+        if self.health.is_faulty() {
+            return Err(DiskError::FaultyDisk);
+        }
+        self.check_disk_stale().await?;
+        self.disk.enqueue_object_metadata(bucket, object).await
+    }
+
+    /// Prepare an inert reservation token bound to this disk and object.
+    pub async fn prepare_object_metadata(&self, bucket: &str, object: &str) -> Result<Uuid> {
+        self.prepare_object_metadata_with_owner(bucket, object, None).await
+    }
+
+    pub(crate) async fn prepare_object_metadata_with_owner(
+        &self,
+        bucket: &str,
+        object: &str,
+        owner: Option<Arc<dyn super::os::MetadataReservationFence>>,
+    ) -> Result<Uuid> {
+        if self.health.is_faulty() {
+            return Err(DiskError::FaultyDisk);
+        }
+        self.check_disk_stale().await?;
+        self.disk.prepare_object_metadata_with_owner(bucket, object, owner)
+    }
+
+    /// Refresh an inert token while the caller is still staging the write.
+    pub async fn renew_object_metadata(&self, bucket: &str, object: &str, token: Uuid) -> Result<()> {
+        if self.health.is_faulty() {
+            return Err(DiskError::FaultyDisk);
+        }
+        self.check_disk_stale().await?;
+        self.disk.renew_object_metadata(bucket, object, token)
+    }
+
+    /// Enqueue a prepared token; a cancelled token cannot be reactivated.
+    pub async fn activate_object_metadata(&self, bucket: &str, object: &str, token: Uuid) -> Result<()> {
+        if self.health.is_faulty() {
+            return Err(DiskError::FaultyDisk);
+        }
+        self.check_disk_stale().await?;
+        self.disk.activate_object_metadata(bucket, object, token).await
+    }
+
+    /// Cancel an unclaimed reservation. A claimed physical operation retains its own lock.
+    pub fn cancel_object_metadata_reservation(&self, token: Uuid) -> bool {
+        self.disk.cancel_object_metadata_reservation(token)
+    }
+
     /// Create a new LocalDiskWrapper
     pub fn new(disk: Arc<LocalDisk>, health_check: bool) -> Self {
         let admission = DiskIoAdmission::new(disk.endpoint().to_string());
@@ -2398,6 +2547,10 @@ impl LocalDiskWrapper {
         .await
     }
 
+    pub(crate) fn has_clean_active_multipart_source(&self, upload_or_part_path: &str, part_path: bool) -> bool {
+        self.disk.has_clean_active_multipart_source(upload_or_part_path, part_path)
+    }
+
     pub(crate) async fn try_publish_prepared_delete_local(
         &self,
         volume: &str,
@@ -2577,6 +2730,7 @@ impl LocalDiskWrapper {
             timeout_duration,
             timeout_health_action,
             mutation,
+            false,
         ))
         .await?
     }
@@ -2588,6 +2742,7 @@ impl LocalDiskWrapper {
         timeout_duration: Duration,
         timeout_health_action: TimeoutHealthAction,
         mutation: DiskMetricMutation,
+        identity_prechecked: bool,
     ) -> Result<T>
     where
         F: FnOnce(Option<crate::disk::io_admission::DiskIoPermit>) -> Fut,
@@ -2609,7 +2764,7 @@ impl LocalDiskWrapper {
         }
 
         // Check if disk is stale
-        if let Err(err) = self.check_disk_stale().await {
+        if !identity_prechecked && let Err(err) = self.check_disk_stale().await {
             self.metrics.record_availability_error();
             return Err(err);
         }
@@ -2916,15 +3071,6 @@ impl DiskAPI for LocalDiskWrapper {
         })
     }
 
-    async fn bucket_creation_witness(&self, volume: &str) -> Result<Option<super::os::NativeBucketCreationWitness>> {
-        self.track_disk_health_with_op(
-            "bucket_creation_witness",
-            || async { self.disk.bucket_creation_witness(volume).await },
-            Duration::ZERO,
-        )
-        .await
-    }
-
     async fn make_volume(&self, volume: &str) -> Result<()> {
         // Scoped heal must drain directory creation before releasing its lifecycle owner.
         let timeout = if crate::store::bucket_heal_scope(volume).is_some() {
@@ -5586,6 +5732,115 @@ mod tests {
         drop(metadata_lease);
     }
 
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn cancelled_reserved_rename_releases_lease_before_disk_admission() {
+        use crate::disk::{RUSTFS_META_TMP_BUCKET, RenameDataGuards};
+
+        let dir = tempfile::tempdir().expect("disk directory");
+        let mut endpoint = Endpoint::try_from(dir.path().to_str().expect("disk path")).expect("endpoint");
+        endpoint.set_pool_index(0);
+        endpoint.set_set_index(0);
+        endpoint.set_disk_index(0);
+        let meta = dir.path().join(RUSTFS_META_BUCKET);
+        tokio::fs::create_dir_all(&meta).await.expect("metadata directory");
+        let mut format = FormatV3::new(1, 1);
+        format.erasure.this = format.erasure.sets[0][0];
+        tokio::fs::write(meta.join(FORMAT_CONFIG_FILE), format.to_json().expect("format bytes"))
+            .await
+            .expect("write format");
+        let disk = Arc::new(LocalDisk::new(&endpoint, false).await.expect("formatted disk"));
+        disk.make_volume("bucket").await.expect("bucket volume");
+        let wrapper = Arc::new(LocalDiskWrapper::new(Arc::clone(&disk), false));
+        let token = wrapper
+            .reserve_object_metadata("bucket", "object", Duration::from_secs(5))
+            .await
+            .expect("reserve key");
+        wrapper.set_disk_id_state(Some(format.erasure.this)).await;
+        disk.format_info.write().await.last_check = None;
+        let (entered_tx, entered_rx) = tokio::sync::oneshot::channel();
+        let (release_tx, release_rx) = std::sync::mpsc::channel();
+        disk.install_format_probe_hook(false, move || {
+            let _ = entered_tx.send(());
+            let _ = release_rx.recv_timeout(Duration::from_secs(30));
+        });
+        let publisher = Arc::clone(&wrapper);
+        let rename = tokio::spawn(async move {
+            publisher
+                .rename_data_observed_with_guards(
+                    RUSTFS_META_TMP_BUCKET,
+                    "staging",
+                    &FileInfo::default(),
+                    "bucket",
+                    "object",
+                    RenameDataGuards {
+                        metadata_reservation_token: Some(token),
+                        ..Default::default()
+                    },
+                )
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(10), entered_rx)
+            .await
+            .expect("disk identity preflight starts")
+            .expect("preflight hook");
+        rename.abort();
+        assert!(matches!(rename.await, Err(error) if error.is_cancelled()));
+        let key = disk.get_object_path("bucket", "object").expect("object path");
+        let mut successor = Box::pin(crate::disk::os::acquire_metadata_mutation_lease(&key, None));
+        let released_before_io = tokio::time::timeout(Duration::from_secs(1), &mut successor).await.is_ok();
+        release_tx.send(()).expect("release disk identity preflight");
+        if !released_before_io {
+            drop(
+                tokio::time::timeout(Duration::from_secs(10), successor)
+                    .await
+                    .expect("preflight finishes"),
+            );
+        }
+        assert!(
+            released_before_io,
+            "a cancelled request must not retain the reservation before physical mutation starts"
+        );
+    }
+
+    #[tokio::test]
+    async fn reserved_background_rename_does_not_wait_for_disk_admission() {
+        use crate::disk::{RUSTFS_META_TMP_BUCKET, RenameDataGuards};
+
+        let dir = tempfile::tempdir().expect("disk directory");
+        let endpoint = Endpoint::try_from(dir.path().to_str().expect("disk path")).expect("endpoint");
+        let disk = Arc::new(LocalDisk::new(&endpoint, false).await.expect("local disk"));
+        let wrapper = LocalDiskWrapper::new(Arc::clone(&disk), false);
+        let token = wrapper
+            .reserve_object_metadata("bucket", "object", Duration::from_secs(5))
+            .await
+            .expect("reserve key");
+        let held = wrapper
+            .io_admission
+            .background(&CancellationToken::new())
+            .await
+            .expect("background permit");
+        let fi = FileInfo::default();
+        let rename = crate::disk::with_background_disk_io(
+            CancellationToken::new(),
+            wrapper.rename_data_observed_with_guards(
+                RUSTFS_META_TMP_BUCKET,
+                "staging",
+                &fi,
+                "bucket",
+                "object",
+                RenameDataGuards {
+                    metadata_reservation_token: Some(token),
+                    ..Default::default()
+                },
+            ),
+        );
+        let result = tokio::time::timeout(Duration::from_secs(1), rename).await;
+        drop(held);
+        let observation = result.expect("reserved mutation must not wait while holding the object lease");
+        assert!(matches!(&observation.result, Err(DiskError::DiskOngoingReq)));
+        assert!(observation.rejected_before_publication());
+    }
+
     #[tokio::test]
     async fn delete_versions_counts_faulty_drive_availability_rejection() {
         let dir = tempfile::tempdir().expect("temp dir should be created");

crates/ecstore/src/disk/fs.rs

@@ -92,9 +92,6 @@ pub const O_APPEND: FileMode = 0x00400;
 //     create_new: bool,
 
 pub async fn open_file(path: impl AsRef<Path>, mode: FileMode) -> io::Result<File> {
-    if mode & O_CREATE != 0 {
-        super::os::record_native_publication(path.as_ref());
-    }
     let path = path.as_ref().to_path_buf();
     native_io(move || {
         let mut options = std::fs::OpenOptions::new();
@@ -147,7 +144,6 @@ pub fn create_dir_all_std(path: &Path) -> io::Result<()> {
 }
 
 pub async fn make_dir_all(path: impl AsRef<Path>) -> io::Result<()> {
-    super::os::record_native_publication(path.as_ref());
     let path = path.as_ref().to_path_buf();
     native_io(move || create_dir_all_std(&path)).await
 }
@@ -209,13 +205,11 @@ pub fn remove_all_std(path: impl AsRef<Path>) -> io::Result<()> {
 
 #[cfg(test)]
 pub async fn mkdir(path: impl AsRef<Path>) -> io::Result<()> {
-    super::os::record_native_publication(path.as_ref());
     let path = path.as_ref().to_path_buf();
     native_io(move || std::fs::create_dir(path)).await
 }
 
 pub async fn rename(from: impl AsRef<Path>, to: impl AsRef<Path>) -> io::Result<()> {
-    super::os::record_native_publication(to.as_ref());
     let from = from.as_ref().to_path_buf();
     let to = to.as_ref().to_path_buf();
     native_io(move || std::fs::rename(from, to)).await
@@ -223,7 +217,6 @@ pub async fn rename(from: impl AsRef<Path>, to: impl AsRef<Path>) -> io::Result<
 
 #[cfg(any(not(windows), test))]
 pub fn rename_std(from: impl AsRef<Path>, to: impl AsRef<Path>) -> io::Result<()> {
-    super::os::record_native_publication(to.as_ref());
     std::fs::rename(from, to)
 }
 

crates/ecstore/src/disk/local/commit.rs

@@ -175,17 +175,30 @@ pub(super) fn create_local_inline_rollback_backup(
     dst_file_path: &Path,
     staging_file_path: &Path,
     old_metadata: &[u8],
+    sync: bool,
 ) -> std::io::Result<PathBuf> {
     let Some(staging_parent) = staging_file_path.parent() else {
         return Err(std::io::Error::new(ErrorKind::InvalidInput, "missing staging metadata parent"));
     };
     let backup_path = staging_parent.join(STORAGE_FORMAT_FILE_BACKUP);
     remove_file_if_exists(&backup_path)?;
-    if (should_fail_local_inline_rollback_hardlink(dst_file_path) || std::fs::hard_link(dst_file_path, &backup_path).is_err())
-        && let Err(err) = std::fs::write(&backup_path, old_metadata)
-    {
-        let _ = remove_file_if_exists(&backup_path);
-        return Err(err);
+    if should_fail_local_inline_rollback_hardlink(dst_file_path) || std::fs::hard_link(dst_file_path, &backup_path).is_err() {
+        let write_res = (|| -> std::io::Result<()> {
+            #[cfg(test)]
+            run_owned_file_before_open(&backup_path);
+            let mut file = std::fs::File::create(&backup_path)?;
+            std::io::Write::write_all(&mut file, old_metadata)?;
+            if sync && !old_metadata.is_empty() {
+                file.sync_data()?;
+            }
+            Ok(())
+        })();
+        if let Err(err) = write_res {
+            let _ = remove_file_if_exists(&backup_path);
+            return Err(err);
+        }
+    } else if sync && !old_metadata.is_empty() {
+        std::fs::File::open(&backup_path)?.sync_data()?;
     }
     Ok(backup_path)
 }
@@ -356,9 +369,12 @@ fn prepare_rename_commit_and_destination_metadata_std(
 
 #[cfg(target_os = "linux")]
 enum FusedCommitAttempt {
-    Committed(RenameDataResp),
+    Committed(Box<RenameDataResp>),
     CommitErr(DiskError),
-    Fallback(os::RenameCommitGuard, Option<Bytes>, FileInfo),
+    RestoreDataSource(os::RenameCommitGuard, DiskError),
+    Fallback(os::RenameCommitGuard, Option<Bytes>, Box<FileInfo>),
+    #[cfg(test)]
+    ResumePanic(Box<dyn std::any::Any + Send + 'static>),
 }
 
 async fn prepare_rename_commit_and_destination_metadata(
@@ -428,12 +444,29 @@ async fn prepare_rename_commit_and_destination_metadata(
 #[derive(Debug)]
 pub(in crate::disk) struct LocalRenamePreflightRejection(());
 
+impl LocalRenamePreflightRejection {
+    pub(in crate::disk) fn not_dispatched() -> Self {
+        Self(())
+    }
+}
+
 #[derive(Default)]
 pub(super) struct RenameDataState {
     namespace_owner: Option<Arc<dyn Send + Sync>>,
+    metadata_reservation: Option<os::MetadataMutationReservation>,
     preflight_rejection: Option<LocalRenamePreflightRejection>,
 }
 
+#[cfg(test)]
+impl RenameDataState {
+    pub(super) fn with_metadata_reservation(reservation: os::MetadataMutationReservation) -> Self {
+        Self {
+            metadata_reservation: Some(reservation),
+            ..Default::default()
+        }
+    }
+}
+
 // Keep optional native MPU state machines out of ordinary rename frames.
 // Construct the future here as well as pinning it, so its temporary stays off
 // the caller's poll stack. Call only from the existing native MPU branches.
@@ -475,6 +508,15 @@ impl LocalDisk {
             wait_for_heal_commit_test_barrier(&self.root, dst_volume, dst_path).await?;
         }
         let mut fi = fi;
+        // Pre-acquired metadata must never reverse quota or multipart-source
+        // lock order. These paths need their own reservation protocol.
+        if state.metadata_reservation.is_some()
+            && (src_volume == super::super::RUSTFS_META_MULTIPART_BUCKET
+                || rustfs_utils::http::metadata_compat::contains_key_str(&fi.metadata, QUOTA_MUTATION_FENCE_METADATA_SUFFIX))
+        {
+            state.preflight_rejection = Some(LocalRenamePreflightRejection(()));
+            return Err(DiskError::DiskOngoingReq);
+        }
         let pool_meta_condition = if super::pool_meta_native::is_object(&self.root, dst_volume, dst_path) {
             Some(super::pool_meta_native::PoolMetaNativeCasCondition::take(&mut fi).inspect_err(|_| {
                 state.preflight_rejection = Some(LocalRenamePreflightRejection(()));
@@ -540,7 +582,9 @@ impl LocalDisk {
             None
         };
         let multipart_publication = if let Some(staging) = multipart_staging.as_ref() {
-            Some(pin_multipart_future(|| self.prepare_native_multipart_publication(src_path, &mut fi, staging)).await?)
+            Some(Arc::new(
+                pin_multipart_future(|| self.prepare_native_multipart_publication_deferred(src_path, &mut fi, staging)).await?,
+            ))
         } else {
             None
         };
@@ -568,7 +612,27 @@ impl LocalDisk {
             Some(staging) => Some(Arc::new((state.namespace_owner.take(), staging.lease.clone()))),
             None => state.namespace_owner.take(),
         };
-        let metadata_lease = os::acquire_metadata_mutation_lease(&root_destination_object_path, metadata_owner).await;
+        // A reserved lease is single-use and bound to the exact disk path.
+        // Keep the independent namespace/source owner alongside it; a pre-held
+        // mutex cannot be acquired again without deadlocking.
+        let reservation_fence = state
+            .metadata_reservation
+            .as_ref()
+            .and_then(|reservation| reservation.fence());
+        let metadata_owner: Arc<dyn Send + Sync> = if let Some(reservation) = state.metadata_reservation.take() {
+            if multipart_staging.is_some() || quota_fence_claim.is_some() {
+                state.preflight_rejection = Some(LocalRenamePreflightRejection(()));
+                return Err(DiskError::DiskOngoingReq);
+            }
+            let lease = reservation
+                .into_lease_for(&root_destination_object_path, self.reservation_instance)
+                .inspect_err(|_| {
+                    state.preflight_rejection = Some(LocalRenamePreflightRejection(()));
+                })?;
+            Arc::new((lease, metadata_owner))
+        } else {
+            os::acquire_metadata_mutation_lease(&root_destination_object_path, metadata_owner).await
+        };
         drop(metadata_wait);
         #[cfg(unix)]
         let parent_receipt = if native_timing {
@@ -582,17 +646,23 @@ impl LocalDisk {
         };
         #[cfg(unix)]
         let mut receipt_mutation = native_timing
-            .then(|| super::pool_meta_native::ReceiptMutationGuard::new(&self.pool_meta_durable_head, metadata_lease.clone()));
+            .then(|| super::pool_meta_native::ReceiptMutationGuard::new(&self.pool_meta_durable_head, metadata_owner.clone()));
         let namespace_wait = os::StageTimer::new(native_timing, "pool_meta_native_namespace_lease_wait");
         let mutation_lease = os::acquire_rename_data_mutation_lease_with_owner(
             &self.root,
             dst_volume,
             &root_destination_object_path,
             staging_prefix.as_deref(),
-            Some(metadata_lease),
+            Some(metadata_owner),
         )
         .await;
         drop(namespace_wait);
+        // The volume read lease prevents deletion of this disk's bucket until
+        // publication finishes. Recheck after its wait, before any mutation.
+        if reservation_fence.as_ref().is_some_and(|fence| !fence.is_valid()) {
+            state.preflight_rejection = Some(LocalRenamePreflightRejection(()));
+            return Err(DiskError::DiskOngoingReq);
+        }
         if let Some(claim) = quota_fence_claim {
             mutation_lease.attach_external_guard(claim);
         }
@@ -677,20 +747,8 @@ impl LocalDisk {
         };
         let dst_file_path = self.io_get_object_open_path(dst_volume, format!("{}/{}", dst_path, STORAGE_FORMAT_FILE).as_str())?;
 
-        // Multipart restore completion has validated staged local parts,
-        // while its public status stays pending until finalization.
-        let pending_multipart_restore = src_volume == crate::disk::RUSTFS_META_MULTIPART_BUCKET
-            && fi
-                .metadata
-                .get(rustfs_filemeta::metadata_keys::RESTORE)
-                .and_then(|value| rustfs_filemeta::parse_restore_obj_status(value).ok())
-                .is_some_and(|status| status.is_restore_in_progress == Some(true));
-        if pending_multipart_restore && fi.data_dir.is_none_or(|dir| dir.is_nil()) {
-            return Err(DiskError::FileCorrupt);
-        }
-
         // data_dir path (only used by the non-inline commit branch)
-        let has_data_dir_path = if no_inline && (!fi.is_remote() || pending_multipart_restore) {
+        let has_data_dir_path = if no_inline && !fi.is_remote() {
             if let Some(data_dir) = fi
                 .data_dir
                 .map(|dir| rustfs_utils::path::retain_slash(dir.to_string().as_str()))
@@ -753,11 +811,6 @@ impl LocalDisk {
             && !fi_healing
             && !os::dst_dir_fsync_group_commit_enabled()
             && !os::file_fdatasync_group_commit_enabled()
-            && !super::has_rename_data_test_failpoints(dst_path)
-            && !crash_inject::is_armed(CrashPoint::RenameAfterDataRename, dst_path)
-            && !crash_inject::is_armed(CrashPoint::RenameAfterBackupBeforeMetaCommit, dst_path)
-            && !crash_inject::is_armed(CrashPoint::RenameAfterMetaCommit, dst_path)
-            && !super::has_delete_version_test_failpoints()
         {
             #[cfg(test)]
             if !no_inline {
@@ -797,10 +850,11 @@ impl LocalDisk {
                     &dst_volume_dir_owned,
                     &publication_root_owned,
                 )?;
-                // An existing xl.meta can share versions or require rollback
-                // restoration when a post-rename directory sync fails.
-                if is_cancelled() || (sync_commit_metadata && has_dst_buf.is_some()) {
-                    return Ok(FusedCommitAttempt::Fallback(guard, has_dst_buf, fi_moved));
+                // An existing xl.meta that replaces an existing version slot
+                // requires a rollback directory; appending a new version keeps
+                // an inline rollback backup for post-rename directory sync errors.
+                if is_cancelled() {
+                    return Ok(FusedCommitAttempt::Fallback(guard, has_dst_buf, Box::new(fi_moved)));
                 }
                 let mut xlmeta = FileMeta::new();
                 let mut dst_meta_unparsable = false;
@@ -823,7 +877,19 @@ impl LocalDisk {
                     }
                 });
                 if rollback_data_dir.is_some() {
-                    return Ok(FusedCommitAttempt::Fallback(guard, has_dst_buf, fi_moved));
+                    return Ok(FusedCommitAttempt::Fallback(guard, has_dst_buf, Box::new(fi_moved)));
+                }
+                #[cfg(test)]
+                let probe_active = os::file_sync_probe::is_active();
+                #[cfg(not(test))]
+                let probe_active = false;
+                if no_inline
+                    && sync_data_shards
+                    && let Some((src_data_path, _)) = has_data_dir_path_owned.as_ref()
+                    && (probe_active || !super::is_presynced_tmp_data_dir(src_data_path))
+                    && os::regular_files(src_data_path).is_ok_and(|files| files.len() >= os::PARALLEL_FILE_SYNC_THRESHOLD)
+                {
+                    return Ok(FusedCommitAttempt::Fallback(guard, has_dst_buf, Box::new(fi_moved)));
                 }
                 let (old_current_size, old_current_source_checked, old_current_source) = observe_old_current(
                     has_dst_buf.is_some(),
@@ -842,13 +908,31 @@ impl LocalDisk {
                     skip_parent = dst_parent_owned.clone();
                 }
 
+                #[cfg(test)]
+                run_owned_file_before_open(&src_file_path_owned);
                 let mut prepared_metadata_source =
                     os::create_prepared_rename_source_with_commit_guard(&src_file_path_owned, &dst_file_path_owned, &guard)?;
-                prepared_metadata_source.write_all(&new_dst_buf, sync_commit_metadata)?;
+                prepared_metadata_source.write_all(&new_dst_buf, false)?;
+                if !no_inline {
+                    run_inline_preparation_before_backup(&dst_path_str);
+                }
+
+                let mut staged_rollback_path = None;
+                if let Some(ref old_metadata) = has_dst_buf
+                    && (sync_commit_metadata || cfg!(test))
+                {
+                    let backup_path = create_local_inline_rollback_backup(
+                        &dst_file_path_owned,
+                        &src_file_path_owned,
+                        old_metadata,
+                        sync_commit_metadata,
+                    )?;
+                    staged_rollback_path = Some(StagedRollbackBackup(backup_path));
+                }
 
                 if no_inline {
                     if sync_data_shards && let Some((src_data_path, _)) = has_data_dir_path_owned.as_ref() {
-                        let sync_result = if super::take_presynced_tmp_data_dir(src_data_path) {
+                        let sync_result = if !probe_active && super::take_presynced_tmp_data_dir(src_data_path) {
                             // The payload was synced, but its part.1 directory entry
                             // must also be persisted before moving the data directory.
                             os::fsync_dir_std(src_data_path)
@@ -858,49 +942,157 @@ impl LocalDisk {
                         if let Err(err) = sync_result
                             && err.kind() != ErrorKind::NotFound
                         {
+                            if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                                let _ = remove_file_if_exists(backup_path);
+                            }
                             return Err(err);
                         }
                     }
+                    if sync_commit_metadata && let Err(err) = prepared_metadata_source.sync_data() {
+                        if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                            let _ = remove_file_if_exists(backup_path);
+                        }
+                        return Err(err);
+                    }
+                } else if sync_commit_metadata && let Err(err) = prepared_metadata_source.sync_data() {
+                    if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                        let _ = remove_file_if_exists(backup_path);
+                    }
+                    return Err(err);
+                }
+
+                if is_cancelled() {
+                    drop(prepared_metadata_source);
+                    if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                        let _ = remove_file_if_exists(backup_path);
+                    }
+                    return Err(std::io::Error::new(ErrorKind::Interrupted, "commit cancelled"));
+                }
+
+                let guard = match super::remove_dst_base_before_commit_std(
+                    &dst_path_str,
+                    guard,
+                    &src_parent_owned,
+                    &dst_parent_owned,
+                    &dst_volume_dir_owned,
+                    &publication_root_owned,
+                ) {
+                    Ok(guard) => guard,
+                    Err(err) => {
+                        if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                            let _ = remove_file_if_exists(backup_path);
+                        }
+                        return Ok(FusedCommitAttempt::CommitErr(err));
+                    }
+                };
+                if should_remove_staged_meta_before_commit(&dst_path_str) {
+                    drop(prepared_metadata_source);
+                    if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                        let _ = remove_file_if_exists(backup_path);
+                    }
+                    std::fs::remove_file(&src_file_path_owned)?;
+                    return Ok(FusedCommitAttempt::CommitErr(DiskError::FileNotFound));
+                }
+
+                if no_inline {
                     if let Some((src_data_path, dst_data_path)) = has_data_dir_path_owned.as_ref()
                         && let Err(err) = os::rename_all_with_commit_guard_on_thread(src_data_path, dst_data_path, &guard)
                     {
-                        return Ok(FusedCommitAttempt::CommitErr(err));
+                        if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                            let _ = remove_file_if_exists(backup_path);
+                        }
+                        return Ok(FusedCommitAttempt::RestoreDataSource(guard, err));
                     }
                     #[cfg(test)]
-                    if has_data_dir_path_owned.is_some() {
-                        run_rename_data_after_first_publication(&root_owned, &dst_volume_str, &dst_path_str);
+                    if has_data_dir_path_owned.is_some()
+                        && let Err(panic_payload) = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
+                            run_rename_data_after_first_publication(&root_owned, &dst_volume_str, &dst_path_str);
+                        }))
+                    {
+                        return Ok(FusedCommitAttempt::ResumePanic(panic_payload));
+                    }
+                    if crash_inject::should_crash_at(CrashPoint::RenameAfterDataRename, &dst_path_str) {
+                        return Ok(FusedCommitAttempt::CommitErr(DiskError::Unexpected));
+                    }
+                    if should_fail_before_old_metadata_backup(&dst_path_str) {
+                        if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                            let _ = remove_file_if_exists(backup_path);
+                        }
+                        return Ok(FusedCommitAttempt::RestoreDataSource(guard, DiskError::Unexpected));
+                    }
+                    if crash_inject::should_crash_at(CrashPoint::RenameAfterBackupBeforeMetaCommit, &dst_path_str) {
+                        return Ok(FusedCommitAttempt::CommitErr(DiskError::Unexpected));
                     }
                 }
 
-                if let Err(err) = os::rename_prepared_source_on_thread(
-                    prepared_metadata_source,
-                    &src_file_path_owned,
-                    &dst_file_path_owned,
-                    &skip_parent,
-                    &publication_root_owned,
-                    &guard,
-                ) {
-                    if no_inline && let Some((src_data_path, dst_data_path)) = has_data_dir_path_owned.as_ref() {
-                        let _ = std::fs::rename(dst_data_path, src_data_path);
+                let rename_res = if should_fail_commit_rename(&dst_path_str) {
+                    drop(prepared_metadata_source);
+                    Err(DiskError::other("test fail during metadata commit rename"))
+                } else {
+                    os::rename_prepared_source_on_thread(
+                        prepared_metadata_source,
+                        &src_file_path_owned,
+                        &dst_file_path_owned,
+                        &skip_parent,
+                        &publication_root_owned,
+                        &guard,
+                    )
+                };
+                if let Err(err) = rename_res {
+                    if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                        let _ = remove_file_if_exists(backup_path);
+                    }
+                    if no_inline {
+                        return Ok(FusedCommitAttempt::RestoreDataSource(guard, err));
                     }
                     return Ok(FusedCommitAttempt::CommitErr(err));
                 }
 
-                if sync_commit_metadata
-                    && let Err(err) =
-                        fsync_inline_commit_std(&dst_file_path_owned, &dst_volume_dir_owned, destination_was_absent, None, &guard)
-                {
-                    if no_inline && let Some((src_data_path, dst_data_path)) = has_data_dir_path_owned.as_ref() {
-                        let _ = std::fs::rename(dst_data_path, src_data_path);
+                if should_fail_after_metadata_commit(&dst_path_str) {
+                    if no_inline {
+                        rollback_inline_metadata_std(&dst_file_path_owned, None, staged_rollback_path.as_ref())?;
+                        if let Some((_, new_data_path)) = has_data_dir_path_owned.as_ref() {
+                            remove_dir_all_if_exists(new_data_path)?;
+                        }
+                        return Ok(FusedCommitAttempt::CommitErr(DiskError::Unexpected));
                     }
-                    return Err(err);
+                    let err = rollback_inline_metadata_std(&dst_file_path_owned, None, staged_rollback_path.as_ref())
+                        .err()
+                        .unwrap_or_else(|| std::io::Error::other("test fail after metadata commit"));
+                    return Ok(FusedCommitAttempt::CommitErr(DiskError::from(err)));
+                }
+
+                if crash_inject::should_crash_at(CrashPoint::RenameAfterMetaCommit, &dst_path_str) {
+                    return Ok(FusedCommitAttempt::CommitErr(DiskError::Unexpected));
+                }
+
+                if sync_commit_metadata
+                    && let Err(err) = fsync_inline_commit_std(
+                        &dst_file_path_owned,
+                        &dst_volume_dir_owned,
+                        destination_was_absent,
+                        staged_rollback_path.as_ref(),
+                        &guard,
+                    )
+                {
+                    if no_inline {
+                        if let Some((_, new_data_path)) = has_data_dir_path_owned.as_ref() {
+                            remove_dir_all_if_exists(new_data_path)?;
+                        }
+                        return Ok(FusedCommitAttempt::CommitErr(to_file_error(err).into()));
+                    }
+                    return Ok(FusedCommitAttempt::CommitErr(DiskError::from(err)));
+                }
+
+                if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
+                    let _ = remove_file_if_exists(backup_path);
                 }
 
                 if let Some(parent) = src_file_path_owned.parent() {
                     let _ = std::fs::remove_dir(parent);
                 }
 
-                Ok(FusedCommitAttempt::Committed(RenameDataResp {
+                Ok(FusedCommitAttempt::Committed(Box::new(RenameDataResp {
                     multipart_source: None,
                     old_data_dir: None,
                     rollback_data_dir: None,
@@ -909,7 +1101,7 @@ impl LocalDisk {
                     old_current_size,
                     old_current_source_checked,
                     old_current_source,
-                }))
+                })))
             };
 
             let attempt_res = if let Some(admission) = file_sync_admission.as_ref() {
@@ -929,7 +1121,23 @@ impl LocalDisk {
                     if let Some(receipt_mutation) = receipt_mutation.as_mut() {
                         receipt_mutation.complete();
                     }
-                    return Ok(resp);
+                    return Ok(*resp);
+                }
+                Ok(FusedCommitAttempt::RestoreDataSource(directory_guard, err)) => {
+                    let commit = RenameCommitContext {
+                        directory_guard,
+                        publication_root: &self.publication_root,
+                        mutation_lease,
+                        source_metadata: &src_file_path,
+                        destination_metadata: &dst_file_path,
+                        source_volume_dir: &src_volume_dir,
+                        data_paths: has_data_dir_path.as_ref(),
+                    };
+                    commit.restore_published_data_source().await?;
+                    for part_path in &invalidate_part_paths {
+                        self.io_backend.invalidate_cached_fd(dst_volume, part_path).await;
+                    }
+                    return Err(err);
                 }
                 Ok(FusedCommitAttempt::CommitErr(err)) | Err(err) => {
                     for part_path in &invalidate_part_paths {
@@ -937,11 +1145,269 @@ impl LocalDisk {
                     }
                     return Err(err);
                 }
+                #[cfg(test)]
+                Ok(FusedCommitAttempt::ResumePanic(panic_payload)) => {
+                    std::panic::resume_unwind(panic_payload);
+                }
                 Ok(FusedCommitAttempt::Fallback(guard, dst_buf, restored_fi)) => {
-                    fi = restored_fi;
+                    fi = *restored_fi;
                     (guard, dst_buf)
                 }
             }
+        } else if let Some(publication) = multipart_publication.as_ref()
+            && no_inline
+            && !preserve_upload_metadata
+            && !publication.legacy
+            && !fi_healing
+            && !os::dst_dir_fsync_group_commit_enabled()
+        {
+            enum MultipartFusedCommitOutcome {
+                Committed(Box<RenameDataResp>),
+                AbortBeforeDataRename(DiskError),
+                RestoreDataSource(os::RenameCommitGuard, DiskError),
+                RestoreTargetAfterError(DiskError, bool),
+            }
+            let publication_owned = Arc::clone(publication);
+            let src_parent_owned = src_file_parent.to_path_buf();
+            let dst_parent_owned = dst_file_parent.to_path_buf();
+            let dst_file_path_owned = dst_file_path.clone();
+            let src_file_path_owned = src_file_path.clone();
+            let dst_volume_dir_owned = dst_volume_dir.clone();
+            let publication_root_owned = self.publication_root.clone();
+            let has_data_dir_path_owned = has_data_dir_path.clone();
+            let sync_data_shards = durability.syncs_data_shards();
+            let sync_commit_metadata = durability.syncs_commit_metadata();
+            let staging_cleanup = multipart_staging.as_ref().map(|s| s.cleanup_handle());
+            let dst_volume_str = dst_volume.to_string();
+            let dst_path_str = dst_path.to_string();
+            let fi_moved = fi;
+            #[cfg(test)]
+            let root_owned = self.root.clone();
+
+            let outcome = os::run_blocking_namespace_operation_with_sync(
+                mutation_lease.clone(),
+                true,
+                move || -> std::io::Result<MultipartFusedCommitOutcome> {
+                    let (directory_guard_owned, has_dst_buf_owned) = match prepare_rename_commit_and_destination_metadata_std(
+                        false,
+                        &src_parent_owned,
+                        &dst_parent_owned,
+                        &dst_file_path_owned,
+                        &dst_volume_dir_owned,
+                        &publication_root_owned,
+                    ) {
+                        Ok(res) => res,
+                        Err(err) => return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(to_file_error(err).into())),
+                    };
+                    let mut xlmeta = FileMeta::new();
+                    let mut dst_meta_unparsable = false;
+                    if let Some(dst_buf) = has_dst_buf_owned.as_ref() {
+                        if FileMeta::is_xl2_v1_format(dst_buf)
+                            && let Ok(nmeta) = FileMeta::load(dst_buf)
+                        {
+                            xlmeta = nmeta;
+                        } else {
+                            dst_meta_unparsable = true;
+                        }
+                    }
+                    let skip_parent_owned = if has_dst_buf_owned.is_some() {
+                        dst_parent_owned.clone()
+                    } else {
+                        dst_volume_dir_owned.clone()
+                    };
+                    let version_id = fi_moved.version_id.unwrap_or_default();
+                    let (old_current_size, old_current_source_checked, old_current_source) = observe_old_current(
+                        has_dst_buf_owned.is_some(),
+                        dst_meta_unparsable,
+                        &xlmeta,
+                        &dst_volume_str,
+                        &dst_path_str,
+                        version_id,
+                    );
+                    let has_old_data_dir = xlmeta.find_unshared_data_dir_for_version(Some(version_id));
+                    if let Some(old_data_dir) = has_old_data_dir.as_ref() {
+                        let _ = xlmeta.data.remove_two(version_id, *old_data_dir);
+                    }
+                    if let Err(err) = xlmeta.add_version(fi_moved) {
+                        return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(err.into()));
+                    }
+                    let version_signature = rename_data_versions_signature(&xlmeta);
+                    let new_dst_buf_owned = match xlmeta.marshal_msg() {
+                        Ok(buf) => buf,
+                        Err(err) => return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(err.into())),
+                    };
+
+                    let prepared_metadata_source = match publication_owned.prepare_and_install_head_on_thread(
+                        &new_dst_buf_owned,
+                        has_dst_buf_owned.as_deref(),
+                        &dst_file_path_owned,
+                        &directory_guard_owned,
+                    ) {
+                        Ok(source) => source,
+                        Err(err) => return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(err)),
+                    };
+                    #[cfg(test)]
+                    let probe_active = os::file_sync_probe::is_active();
+                    #[cfg(not(test))]
+                    let probe_active = false;
+                    // When presynced, shard payloads were synced during UploadPart and
+                    // prepare_and_install_head_on_thread just fsynced src_data_path after
+                    // writing RETIREMENT_ANCHOR.
+                    if sync_data_shards
+                        && let Some((src_data_path, _)) = has_data_dir_path_owned.as_ref()
+                        && (probe_active || !super::take_presynced_tmp_data_dir(src_data_path))
+                        && let Err(err) = os::sync_dir_files_std(src_data_path)
+                        && err.kind() != ErrorKind::NotFound
+                    {
+                        return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(to_file_error(err).into()));
+                    }
+                    let directory_guard_owned = match super::remove_dst_base_before_commit_std(
+                        &dst_path_str,
+                        directory_guard_owned,
+                        &src_parent_owned,
+                        &dst_parent_owned,
+                        &dst_volume_dir_owned,
+                        &publication_root_owned,
+                    ) {
+                        Ok(guard) => guard,
+                        Err(err) => return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(err)),
+                    };
+                    if should_remove_staged_meta_before_commit(&dst_path_str) {
+                        drop(prepared_metadata_source);
+                        if let Err(err) = std::fs::remove_file(&src_file_path_owned) {
+                            return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(to_file_error(err).into()));
+                        }
+                        return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(DiskError::FileNotFound));
+                    }
+                    if let Some((src_data_path, dst_data_path)) = has_data_dir_path_owned.as_ref() {
+                        let Some(src_parent) = src_data_path.parent() else {
+                            return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(DiskError::InvalidPath));
+                        };
+                        let payload_guard = match os::multipart_payload_commit_guard_std(&directory_guard_owned, src_parent) {
+                            Ok(guard) => guard,
+                            Err(err) => return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(to_file_error(err).into())),
+                        };
+                        if let Err(err) = os::rename_all_with_commit_guard_on_thread(src_data_path, dst_data_path, &payload_guard)
+                        {
+                            return Ok(MultipartFusedCommitOutcome::RestoreDataSource(directory_guard_owned, err));
+                        }
+                        #[cfg(test)]
+                        run_rename_data_after_first_publication(&root_owned, &dst_volume_str, &dst_path_str);
+                        if crash_inject::should_crash_at(CrashPoint::RenameAfterDataRename, &dst_path_str) {
+                            return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(DiskError::Unexpected));
+                        }
+                        if should_fail_before_old_metadata_backup(&dst_path_str) {
+                            return Ok(MultipartFusedCommitOutcome::RestoreDataSource(
+                                directory_guard_owned,
+                                DiskError::Unexpected,
+                            ));
+                        }
+                        if crash_inject::should_crash_at(CrashPoint::RenameAfterBackupBeforeMetaCommit, &dst_path_str) {
+                            return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(DiskError::Unexpected));
+                        }
+                    }
+                    let rename_res = if should_fail_commit_rename(&dst_path_str) {
+                        drop(prepared_metadata_source);
+                        Err(DiskError::other("test fail during metadata commit rename"))
+                    } else {
+                        os::rename_prepared_source_on_thread(
+                            prepared_metadata_source,
+                            &src_file_path_owned,
+                            &dst_file_path_owned,
+                            &skip_parent_owned,
+                            &publication_root_owned,
+                            &directory_guard_owned,
+                        )
+                    };
+                    if let Err(err) = rename_res {
+                        return Ok(MultipartFusedCommitOutcome::RestoreDataSource(directory_guard_owned, err));
+                    }
+                    if should_fail_after_metadata_commit(&dst_path_str) {
+                        return Ok(MultipartFusedCommitOutcome::RestoreTargetAfterError(DiskError::Unexpected, false));
+                    }
+                    if crash_inject::should_crash_at(CrashPoint::RenameAfterMetaCommit, &dst_path_str) {
+                        return Ok(MultipartFusedCommitOutcome::AbortBeforeDataRename(DiskError::Unexpected));
+                    }
+                    if sync_commit_metadata {
+                        if let Some(parent) = dst_file_path_owned.parent()
+                            && let Err(err) = os::fsync_commit_directory_limited_std(parent, &directory_guard_owned)
+                        {
+                            return Ok(MultipartFusedCommitOutcome::RestoreTargetAfterError(to_file_error(err).into(), true));
+                        }
+                        if has_dst_buf_owned.is_none() {
+                            let mut ancestor = dst_file_path_owned.parent().and_then(|object_dir| object_dir.parent());
+                            while let Some(dir) = ancestor {
+                                if !dir.starts_with(&dst_volume_dir_owned) {
+                                    break;
+                                }
+                                if let Err(err) = os::fsync_dir_limited_std(dir) {
+                                    return Ok(MultipartFusedCommitOutcome::RestoreTargetAfterError(
+                                        to_file_error(err).into(),
+                                        true,
+                                    ));
+                                }
+                                if dir == dst_volume_dir_owned.as_path() {
+                                    break;
+                                }
+                                ancestor = dir.parent();
+                            }
+                        }
+                    }
+                    match publication_owned.finish_on_thread(sync_commit_metadata, staging_cleanup) {
+                        Ok(result) => Ok(MultipartFusedCommitOutcome::Committed(Box::new(RenameDataResp {
+                            multipart_source: Some(Box::new(result)),
+                            old_data_dir: has_old_data_dir,
+                            rollback_data_dir: None,
+                            cleanup_data_dir: has_old_data_dir,
+                            sign: version_signature,
+                            old_current_size,
+                            old_current_source_checked,
+                            old_current_source,
+                        }))),
+                        Err(err) => Ok(MultipartFusedCommitOutcome::RestoreTargetAfterError(err, false)),
+                    }
+                },
+            )
+            .await
+            .map_err(to_file_error)
+            .map_err(DiskError::from)?;
+
+            match outcome {
+                MultipartFusedCommitOutcome::Committed(resp) => {
+                    for part_path in &invalidate_part_paths {
+                        self.io_backend.invalidate_cached_fd(dst_volume, part_path).await;
+                    }
+                    return Ok(*resp);
+                }
+                MultipartFusedCommitOutcome::AbortBeforeDataRename(err) => return Err(err),
+                MultipartFusedCommitOutcome::RestoreDataSource(directory_guard, err) => {
+                    let commit = RenameCommitContext {
+                        directory_guard,
+                        publication_root: &self.publication_root,
+                        mutation_lease,
+                        source_metadata: &src_file_path,
+                        destination_metadata: &dst_file_path,
+                        source_volume_dir: &src_volume_dir,
+                        data_paths: has_data_dir_path.as_ref(),
+                    };
+                    commit.restore_published_data_source().await?;
+                    return Err(err);
+                }
+                MultipartFusedCommitOutcome::RestoreTargetAfterError(err, invalidate_before_return) => {
+                    if !invalidate_before_return {
+                        for part_path in &invalidate_part_paths {
+                            self.io_backend.invalidate_cached_fd(dst_volume, part_path).await;
+                        }
+                    }
+                    pin_multipart_future(|| publication.restore_target_after_error(mutation_lease.clone())).await?;
+                    if invalidate_before_return {
+                        for part_path in &invalidate_part_paths {
+                            self.io_backend.invalidate_cached_fd(dst_volume, part_path).await;
+                        }
+                    }
+                    return Err(err);
+                }
+            }
         } else {
             prepare_rename_commit_and_destination_metadata(
                 !no_inline && !preserve_upload_metadata,
@@ -1065,71 +1531,81 @@ impl LocalDisk {
             // after the commit rename, and a crash before the rename means the
             // PUT was never acknowledged. A metadata commit: relaxed tiers
             // leave it to the page cache.
-            if let Some(publication) = multipart_publication.as_ref() {
-                pin_multipart_future(|| publication.prepare_metadata(&new_dst_buf, has_dst_buf.as_deref())).await?;
-            }
-            let tmp_meta_sync = if durability.syncs_commit_metadata() || multipart_publication.is_some() {
-                SyncMode::FileOnly
-            } else {
-                SyncMode::None
-            };
-            // The tmp xl.meta write and the shard-file fdatasync are independent
-            // (disjoint paths) and both only need to be durable before the commit
-            // renames below, so run them concurrently to drop a blocking
-            // round-trip from the PUT commit critical path (rustfs/backlog#922
-            // step 2). The "contents durable -> rename -> dst dir fsync" ordering
-            // is unchanged — both futures complete before any rename — which the
-            // rename_data crash-consistency harness (backlog#935) exercises.
-            //
-            // Shard durability: once rename_data succeeds the write is
-            // acknowledged, so data must not live only in the page cache.
-            // Multipart parts were already synced during rename_part, so their
-            // fdatasync here is a cheap no-op. A missing source dir is left for the
-            // rename below to report through the existing rollback path. Payload
-            // durability is kept by both strict and relaxed.
-            let tmp_meta_write = {
-                let src_file_path = multipart_publication
-                    .as_ref()
-                    .map_or_else(|| commit.source_metadata.to_path_buf(), |publication| publication.prepared_path.clone());
-                let dst_file_path = dst_file_path.clone();
-                let owned_directory_guard = commit.directory_guard.clone();
-                let owned_mutation_lease = commit.mutation_lease.clone();
-                async move {
-                    os::run_blocking_namespace_operation(owned_mutation_lease, move || {
-                        #[cfg(test)]
-                        run_owned_file_before_open(&src_file_path);
-                        let mut prepared_metadata_source = os::create_prepared_rename_source_with_commit_guard(
-                            &src_file_path,
-                            &dst_file_path,
-                            &owned_directory_guard,
-                        )?;
-                        prepared_metadata_source.write_all(&new_dst_buf, tmp_meta_sync != SyncMode::None)?;
-                        Ok(prepared_metadata_source)
-                    })
-                    .await
-                    .map_err(to_file_error)
-                    .map_err(DiskError::from)
-                }
-            };
+            #[cfg(all(test, target_os = "linux"))]
+            let probe_active = os::file_sync_probe::is_active();
+            #[cfg(all(not(test), target_os = "linux"))]
+            let probe_active = false;
             let shard_sync = async {
                 if durability.syncs_data_shards()
                     && let Some((src_data_path, _)) = has_data_dir_path.as_ref()
-                    && let Err(err) = os::sync_dir_files_with_limiter(src_data_path, self.file_sync_permits.clone()).await
-                    && err.kind() != ErrorKind::NotFound
                 {
-                    return Err::<(), DiskError>(to_file_error(err).into());
+                    #[cfg(target_os = "linux")]
+                    if !probe_active && super::take_presynced_tmp_data_dir(src_data_path) {
+                        if multipart_publication.is_none()
+                            && let Err(err) = os::fsync_dir_with_owner(src_data_path, Some(commit.mutation_lease.clone())).await
+                            && err.kind() != ErrorKind::NotFound
+                        {
+                            return Err::<(), DiskError>(to_file_error(err).into());
+                        }
+                        return Ok(());
+                    }
+                    if let Err(err) = os::sync_dir_files_with_limiter(src_data_path, self.file_sync_permits.clone()).await
+                        && err.kind() != ErrorKind::NotFound
+                    {
+                        return Err::<(), DiskError>(to_file_error(err).into());
+                    }
                 }
                 Ok(())
             };
-            let (tmp_meta_res, shard_sync_res) = tokio::join!(tmp_meta_write, shard_sync);
-            // Surface a tmp-meta failure first (its prior serial position), then a
-            // shard-sync failure; either aborts before any rename, exactly as the
-            // sequential version did.
-            let prepared_metadata_source = tmp_meta_res?;
-            shard_sync_res?;
-            let prepared_metadata_source = match multipart_publication.as_ref() {
-                Some(publication) => pin_multipart_future(|| publication.install_head(prepared_metadata_source)).await?,
-                None => prepared_metadata_source,
+            let prepared_metadata_source = if let Some(publication) = multipart_publication.as_ref() {
+                let dst_file_path_ref = &dst_file_path;
+                let dir_guard_ref = &commit.directory_guard;
+                let lease_clone = commit.mutation_lease.clone();
+                let prepare_fut = pin_multipart_future(|| {
+                    publication.prepare_and_install_head(
+                        &new_dst_buf,
+                        has_dst_buf.as_deref(),
+                        dst_file_path_ref,
+                        dir_guard_ref,
+                        lease_clone,
+                    )
+                });
+                let (tmp_meta_res, shard_sync_res) = tokio::join!(prepare_fut, shard_sync);
+                let prepared_metadata_source = tmp_meta_res?;
+                shard_sync_res?;
+                prepared_metadata_source
+            } else {
+                let tmp_meta_sync = if durability.syncs_commit_metadata() {
+                    SyncMode::FileOnly
+                } else {
+                    SyncMode::None
+                };
+                let tmp_meta_write = {
+                    let src_file_path = commit.source_metadata.to_path_buf();
+                    let dst_file_path = dst_file_path.clone();
+                    let owned_directory_guard = commit.directory_guard.clone();
+                    let owned_mutation_lease = commit.mutation_lease.clone();
+                    async move {
+                        os::run_blocking_namespace_operation(owned_mutation_lease, move || {
+                            #[cfg(test)]
+                            run_owned_file_before_open(&src_file_path);
+                            let mut prepared_metadata_source = os::create_prepared_rename_source_with_commit_guard(
+                                &src_file_path,
+                                &dst_file_path,
+                                &owned_directory_guard,
+                            )?;
+                            prepared_metadata_source.write_all(&new_dst_buf, tmp_meta_sync != SyncMode::None)?;
+                            Ok(prepared_metadata_source)
+                        })
+                        .await
+                        .map_err(to_file_error)
+                        .map_err(DiskError::from)
+                    }
+                };
+                let (tmp_meta_res, shard_sync_res) = tokio::join!(tmp_meta_write, shard_sync);
+                let prepared_metadata_source = tmp_meta_res?;
+                shard_sync_res?;
+                prepared_metadata_source
             };
             commit.directory_guard = remove_dst_base_before_commit(
                 dst_path,
@@ -1469,13 +1945,22 @@ impl LocalDisk {
             }
 
             let multipart_source = match multipart_publication.as_ref() {
-                Some(publication) => match pin_multipart_future(|| publication.finish(commit.mutation_lease.clone())).await {
-                    Ok(result) => Some(Box::new(result)),
-                    Err(error) => {
-                        pin_multipart_future(|| publication.restore_target_after_error(commit.mutation_lease.clone())).await?;
-                        return Err(error);
+                Some(publication) => {
+                    let target_already_synced = durability.syncs_commit_metadata();
+                    let staging_cleanup = multipart_staging.as_ref().map(|s| s.cleanup_handle());
+                    match pin_multipart_future(|| {
+                        publication.finish(commit.mutation_lease.clone(), target_already_synced, staging_cleanup)
+                    })
+                    .await
+                    {
+                        Ok(result) => Some(Box::new(result)),
+                        Err(error) => {
+                            pin_multipart_future(|| publication.restore_target_after_error(commit.mutation_lease.clone()))
+                                .await?;
+                            return Err(error);
+                        }
                     }
-                },
+                }
                 None => None,
             };
             Ok(RenameDataResp {
@@ -1675,13 +2160,8 @@ impl LocalDisk {
                     #[cfg(not(windows))]
                     let backup_path = {
                         let _backup = os::StageTimer::new(native_timing, "pool_meta_native_backup_create");
-                        create_local_inline_rollback_backup(&dst, &src, old_metadata)?
+                        create_local_inline_rollback_backup(&dst, &src, old_metadata, sync)?
                     };
-                    #[cfg(not(windows))]
-                    if sync {
-                        let _backup_sync = os::StageTimer::new(native_timing, "pool_meta_native_backup_file_sync");
-                        std::fs::File::open(&backup_path)?.sync_data()?;
-                    }
                     staged_rollback_path = Some(StagedRollbackBackup(backup_path));
                 }
 
@@ -2121,8 +2601,23 @@ impl LocalDisk {
         dst_path: &str,
         namespace_owner: Option<Arc<dyn Send + Sync>>,
     ) -> super::super::RenameDataObservation {
+        self.rename_data_observed_with_reservation(src_volume, src_path, fi, dst_volume, dst_path, (namespace_owner, None))
+            .await
+    }
+
+    pub(in crate::disk) async fn rename_data_observed_with_reservation(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+        owners: (Option<Arc<dyn Send + Sync>>, Option<os::MetadataMutationReservation>),
+    ) -> super::super::RenameDataObservation {
+        let (namespace_owner, metadata_reservation) = owners;
         let mut state = RenameDataState {
             namespace_owner,
+            metadata_reservation,
             ..Default::default()
         };
         let result = self

crates/ecstore/src/disk/local/delete_recovery.rs

@@ -1000,6 +1000,25 @@ pub(super) fn inspection_with_markers(
     Ok((inspection, has_physical_data_dirs_or_markers))
 }
 
+type PreparedDeleteReceiptEntry = (DeleteRecoveryJournal, bool);
+
+static PREPARED_DELETE_RECEIPTS: std::sync::LazyLock<parking_lot::Mutex<HashMap<PathBuf, PreparedDeleteReceiptEntry>>> =
+    std::sync::LazyLock::new(|| parking_lot::Mutex::new(HashMap::new()));
+
+fn put_prepared_delete_receipt(rollback: PathBuf, entry: PreparedDeleteReceiptEntry) {
+    let mut map = PREPARED_DELETE_RECEIPTS.lock();
+    if map.len() >= 4096
+        && let Some(evict_key) = map.keys().next().cloned()
+    {
+        map.remove(&evict_key);
+    }
+    map.insert(rollback, entry);
+}
+
+fn take_prepared_delete_receipt(rollback: &Path) -> Option<PreparedDeleteReceiptEntry> {
+    PREPARED_DELETE_RECEIPTS.lock().remove(rollback)
+}
+
 impl LocalDisk {
     // Native MPU source closure has its own conditional protocol. Generic
     // object recovery must never restore its head or acquire SourceRW under
@@ -1523,6 +1542,9 @@ impl LocalDisk {
         owner: Option<Arc<dyn Send + Sync>>,
     ) -> Result<DeleteRecoveryInspection> {
         self.reject_multipart_delete_recovery_mutation(volume, object)?;
+        if let Ok(root) = self.io_get_object_path(volume, object) {
+            let _ = take_prepared_delete_receipt(&root.join(transaction.to_string()));
+        }
 
         let (mut observed, mut has_data_dir_work) = self
             .delete_recovery_inspect_owned_with_markers(volume, object, transaction, owner.clone())
@@ -1911,8 +1933,14 @@ impl LocalDisk {
                     return Err(DiskError::DiskOngoingReq);
                 }
                 let bucket_path = self.io_get_bucket_path(volume)?;
-                let mut file_leases = Vec::with_capacity(3);
-                for name in [TARGET, PREPARED, FINISHED] {
+                let mut file_leases = Vec::with_capacity(5);
+                for name in [
+                    STORAGE_FORMAT_FILE_BACKUP,
+                    DELETE_MARKER_ROLLBACK_FILE,
+                    TARGET,
+                    PREPARED,
+                    FINISHED,
+                ] {
                     let path = rollback.join(name);
                     let lease = os::acquire_namespace_mutation_lease_with_owner(&path, owner.clone()).await;
                     file_leases.push((path, lease));
@@ -2132,119 +2160,151 @@ impl LocalDisk {
                     Err(err) if err.kind() == ErrorKind::NotFound => return Ok(Ok(None)),
                     Err(err) => return Err(err),
                 };
-                let (old_dir, journal_raw) = match (|| -> Result<Option<(Option<Uuid>, Vec<u8>)>> {
-                    let mut meta = FileMeta::load(&buf)?;
-                    let deleted_versions: Vec<_> = meta
-                        .versions
-                        .iter()
-                        .map(|v| v.header.version_id.unwrap_or_default())
-                        .collect();
-                    let original_versions: Vec<_> = meta
-                        .versions
-                        .iter()
-                        .map(|v| generation(v.into_fileinfo(&volume_str, &object_str, true)?))
-                        .collect::<Result<_>>()?;
-                    let old_dir = meta.delete_version(&fi_clone)?;
-                    if old_dir.is_some_and(|dir| dir == transaction || dir.is_nil()) {
-                        return Err(DiskError::FileCorrupt);
-                    }
-                    if let Some(uuid) = old_dir {
-                        meta.data.remove(vec![fi_clone.version_id.unwrap_or_default(), uuid])?;
-                        let old_path = path_join(&[file_path.as_path(), Path::new(uuid.to_string().as_str())]);
-                        check_path_length(old_path.to_string_lossy().as_ref())?;
-                        if old_path.join(STORAGE_FORMAT_FILE).exists() {
-                            return Ok(None);
+                type PreparedDeleteTuple = (Option<Uuid>, Option<Vec<u8>>, Vec<u8>, DeleteRecoveryJournal, bool);
+                let (old_dir, target, journal_raw, journal, sole_transaction) =
+                    match (|| -> Result<Option<PreparedDeleteTuple>> {
+                        let mut meta = FileMeta::load(&buf)?;
+                        let mut deleted_headers: Vec<_> = meta
+                            .versions
+                            .iter()
+                            .map(|v| (v.header.version_id, std::mem::discriminant(&v.header.version_type)))
+                            .collect();
+                        let original_versions: Vec<_> = meta
+                            .versions
+                            .iter()
+                            .map(|v| generation(v.into_fileinfo(&volume_str, &object_str, true)?))
+                            .collect::<Result<_>>()?;
+                        let old_dir = meta.delete_version(&fi_clone)?;
+                        if old_dir.is_some_and(|dir| dir == transaction || dir.is_nil()) {
+                            return Err(DiskError::FileCorrupt);
                         }
-                    }
-                    if !meta.versions.is_empty() {
-                        return Ok(None);
-                    }
-                    let mut predecessors = BTreeSet::new();
-                    if let Ok(entries) = std::fs::read_dir(&file_path) {
-                        for (visited, entry) in entries.enumerate() {
-                            if visited >= 65536 {
-                                return Err(DiskError::DiskOngoingReq);
+                        if let Some(uuid) = old_dir {
+                            meta.data.remove(vec![fi_clone.version_id.unwrap_or_default(), uuid])?;
+                            let old_path = path_join(&[file_path.as_path(), Path::new(uuid.to_string().as_str())]);
+                            check_path_length(old_path.to_string_lossy().as_ref())?;
+                            if old_path.join(STORAGE_FORMAT_FILE).exists() {
+                                return Ok(None);
                             }
-                            let entry = entry.map_err(to_file_error)?;
-                            let Some(id) = entry
-                                .file_name()
-                                .to_str()
-                                .and_then(|name| Uuid::parse_str(name).ok())
-                                .filter(|id| !id.is_nil() && *id != transaction)
-                            else {
-                                continue;
-                            };
-                            if !entry.file_type().map_err(to_file_error)?.is_dir() {
-                                return Err(DiskError::InvalidPath);
-                            }
-                            let prepared = regular_bytes(&entry.path().join(PREPARED))?;
-                            let receipt = regular_bytes(&entry.path().join(DELETE_QUORUM_COMMITTED_FILE))?;
-                            let mut records = Vec::new();
-                            if let Some(bytes) = prepared {
-                                records.push(
-                                    serde_json::from_slice::<DeleteRecoveryJournal>(&bytes)
-                                        .map_err(|_| DiskError::FileCorrupt)?,
-                                );
-                            }
-                            if let Some(bytes) = receipt.filter(|bytes| !bytes.is_empty()) {
-                                records.push(
-                                    serde_json::from_slice::<CommitReceipt>(&bytes)
-                                        .map_err(|_| DiskError::FileCorrupt)?
-                                        .journal,
-                                );
-                            }
-                            for record in records {
-                                if record.protocol_version != DELETE_RECOVERY_PROTOCOL_VERSION
-                                    || record.volume != volume_str
-                                    || record.object != object_str
-                                    || record.transaction != id
-                                    || incarnation.zip(record.bucket_incarnation).is_some_and(|(a, b)| a != b)
-                                    || record.predecessors.iter().any(|before| before.is_nil() || *before == id)
-                                {
-                                    return Err(DiskError::FileCorrupt);
+                        }
+                        let surviving: HashSet<_> = meta
+                            .versions
+                            .iter()
+                            .map(|v| (v.header.version_id, std::mem::discriminant(&v.header.version_type)))
+                            .collect();
+                        deleted_headers.retain(|header| !surviving.contains(header));
+                        let deleted_versions = deleted_headers.into_iter().map(|(id, _)| id.unwrap_or_default()).collect();
+                        let target = (!meta.versions.is_empty()).then(|| meta.marshal_msg()).transpose()?;
+                        let mut predecessors = BTreeSet::new();
+                        let mut sole_transaction = true;
+                        if let Ok(entries) = std::fs::read_dir(&file_path) {
+                            for (visited, entry) in entries.enumerate() {
+                                if visited >= 65536 {
+                                    return Err(DiskError::DiskOngoingReq);
+                                }
+                                let entry = entry.map_err(to_file_error)?;
+                                let file_name = entry.file_name();
+                                let Some(name_str) = file_name.to_str() else {
+                                    sole_transaction = false;
+                                    continue;
+                                };
+                                let Some(id) = Uuid::parse_str(name_str).ok().filter(|id| !id.is_nil() && *id != transaction)
+                                else {
+                                    continue;
+                                };
+                                if !entry.file_type().map_err(to_file_error)?.is_dir() {
+                                    return Err(DiskError::InvalidPath);
+                                }
+                                let dir_path = entry.path();
+                                let prepared = regular_bytes(&dir_path.join(PREPARED))?;
+                                let receipt = regular_bytes(&dir_path.join(DELETE_QUORUM_COMMITTED_FILE))?;
+                                if sole_transaction && name_str == id.hyphenated().to_string() && old_dir != Some(id) {
+                                    let is_surviving_data_dir = original_versions.iter().any(|v| v.data_dir == Some(id))
+                                        && prepared.is_none()
+                                        && receipt.is_none()
+                                        && !dir_path.join(STORAGE_FORMAT_FILE_BACKUP).exists();
+                                    if !is_surviving_data_dir {
+                                        sole_transaction = false;
+                                    }
+                                }
+                                let mut records = Vec::new();
+                                if let Some(bytes) = prepared {
+                                    records.push(
+                                        serde_json::from_slice::<DeleteRecoveryJournal>(&bytes)
+                                            .map_err(|_| DiskError::FileCorrupt)?,
+                                    );
+                                }
+                                if let Some(bytes) = receipt.filter(|bytes| !bytes.is_empty()) {
+                                    records.push(
+                                        serde_json::from_slice::<CommitReceipt>(&bytes)
+                                            .map_err(|_| DiskError::FileCorrupt)?
+                                            .journal,
+                                    );
+                                }
+                                for record in records {
+                                    if record.protocol_version != DELETE_RECOVERY_PROTOCOL_VERSION
+                                        || record.volume != volume_str
+                                        || record.object != object_str
+                                        || record.transaction != id
+                                        || incarnation.zip(record.bucket_incarnation).is_some_and(|(a, b)| a != b)
+                                        || record.predecessors.iter().any(|before| before.is_nil() || *before == id)
+                                    {
+                                        return Err(DiskError::FileCorrupt);
+                                    }
+                                    predecessors.insert(id);
                                 }
-                                predecessors.insert(id);
                             }
                         }
-                    }
-                    let mut sparse = fi_clone;
-                    sparse.data = None;
-                    let journal = DeleteRecoveryJournal {
-                        protocol_version: DELETE_RECOVERY_PROTOCOL_VERSION,
-                        volume: volume_str,
-                        object: object_str,
-                        transaction,
-                        bucket_incarnation: incarnation,
-                        original_digest: Some(digest(&buf)),
-                        target_digest: None,
-                        data_dirs: old_dir.into_iter().collect(),
-                        deleted_versions,
-                        requests: vec![sparse],
-                        original_versions,
-                        predecessors: predecessors.into_iter().collect(),
+                        sole_transaction &= predecessors.is_empty();
+                        let mut sparse = fi_clone;
+                        sparse.data = None;
+                        let journal = DeleteRecoveryJournal {
+                            protocol_version: DELETE_RECOVERY_PROTOCOL_VERSION,
+                            volume: volume_str,
+                            object: object_str,
+                            transaction,
+                            bucket_incarnation: incarnation,
+                            original_digest: Some(digest(&buf)),
+                            target_digest: target.as_deref().map(digest),
+                            data_dirs: old_dir.into_iter().collect(),
+                            deleted_versions,
+                            requests: vec![sparse],
+                            original_versions,
+                            predecessors: predecessors.into_iter().collect(),
+                        };
+                        let raw = journal_bytes(&journal)?;
+                        Ok(Some((old_dir, target, raw, journal, sole_transaction)))
+                    })() {
+                        Ok(Some(tuple)) => tuple,
+                        Ok(None) => return Ok(Ok(None)),
+                        Err(err) => return Ok(Err(err)),
                     };
-                    Ok(Some((old_dir, journal_bytes(&journal)?)))
-                })() {
-                    Ok(Some(pair)) => pair,
-                    Ok(None) => return Ok(Ok(None)),
-                    Err(err) => return Ok(Err(err)),
-                };
                 std::fs::create_dir_all(&staging_root)?;
-                let staged = tempfile::Builder::new()
-                    .prefix("delete-preparation-")
-                    .tempdir_in(&staging_root)?;
-                for (name, bytes) in [
-                    (STORAGE_FORMAT_FILE_BACKUP, buf.as_slice()),
-                    (PREPARED, journal_raw.as_slice()),
-                ] {
-                    let mut file = std::fs::File::create(staged.path().join(name))?;
-                    std::io::Write::write_all(&mut file, bytes)?;
-                    if sync && !bytes.is_empty() {
+                let staged = tempfile::Builder::new().prefix("delete-prep-").tempdir_in(&staging_root)?;
+                let staged_dir = staged.path().to_path_buf();
+                let backup_path = staged_dir.join(STORAGE_FORMAT_FILE_BACKUP);
+                if std::fs::hard_link(&xl_path, &backup_path).is_err() {
+                    let mut file = std::fs::File::create(&backup_path)?;
+                    std::io::Write::write_all(&mut file, &buf)?;
+                    if sync && !buf.is_empty() {
                         file.sync_data()?;
                     }
                 }
+                let mut written_files = Vec::with_capacity(2);
+                for (name, bytes) in [(TARGET, target.as_deref()), (PREPARED, Some(journal_raw.as_slice()))] {
+                    let Some(bytes) = bytes else {
+                        continue;
+                    };
+                    let mut file = std::fs::File::create(staged_dir.join(name))?;
+                    std::io::Write::write_all(&mut file, bytes)?;
+                    if sync && !bytes.is_empty() {
+                        written_files.push(file);
+                    }
+                }
+                for file in written_files {
+                    file.sync_data()?;
+                }
                 if sync {
-                    os::fsync_dir_std(staged.path())?;
+                    os::fsync_dir_std(&staged_dir)?;
                 }
                 match std::fs::symlink_metadata(&destination) {
                     Ok(metadata) if metadata.is_dir() => match std::fs::remove_dir(&destination) {
@@ -2257,28 +2317,62 @@ impl LocalDisk {
                     Err(err) if err.kind() == ErrorKind::NotFound => {}
                     Err(err) => return Err(err),
                 }
-                if let Some(parent) = destination.parent() {
-                    std::fs::create_dir_all(parent)?;
+                let staged_dir = staged.keep();
+                if let Err(err) = std::fs::rename(&staged_dir, &destination) {
+                    let _ = std::fs::remove_dir_all(&staged_dir);
+                    return Err(err);
                 }
-                os::record_native_publication(&destination);
-                std::fs::rename(staged.path(), &destination)?;
-                if sync {
-                    os::fsync_dir_std(&file_path)?;
+                let old_data_path = old_dir.map(|uuid| file_path.join(uuid.to_string())).filter(|p| p.is_dir());
+                if sync && let Err(err) = os::fsync_dir_std(&file_path) {
+                    let _ = std::fs::remove_dir_all(&destination);
+                    return Err(err);
                 }
                 let mut reserved_data_path = None;
-                if let Some(uuid) = old_dir {
-                    let old_path = file_path.join(uuid.to_string());
-                    if old_path.is_dir() {
-                        let reserved_marker = old_path.join(format!("{RESERVED_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}"));
+                if let Some(old_path) = old_data_path {
+                    let reserved_marker = old_path.join(format!("{RESERVED_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}"));
+                    let reserve_res = (|| -> std::io::Result<()> {
                         let marker = std::fs::File::create(&reserved_marker)?;
                         if sync {
                             marker.sync_all()?;
                             os::fsync_dir_std(&old_path)?;
                         }
-                        reserved_data_path = Some((old_path, reserved_marker));
+                        Ok(())
+                    })();
+                    if let Err(err) = reserve_res {
+                        let _ = std::fs::remove_file(&reserved_marker);
+                        let _ = std::fs::remove_dir_all(&destination);
+                        return Err(err);
                     }
+                    reserved_data_path = Some((old_path, reserved_marker));
                 }
-                if let Err(err) = std::fs::remove_file(&xl_path)
+                if let Some(ref target_bytes) = target {
+                    let target_in_rollback = destination.join(TARGET);
+                    let linked_xl = staging_root.join(format!("delete-xl-{transaction}"));
+                    let _ = std::fs::remove_file(&linked_xl);
+                    let write_res = if std::fs::hard_link(&target_in_rollback, &linked_xl).is_ok() {
+                        let rename_res = std::fs::rename(&linked_xl, &xl_path);
+                        if rename_res.is_err() {
+                            let _ = std::fs::remove_file(&linked_xl);
+                        }
+                        rename_res
+                    } else {
+                        (|| -> std::io::Result<()> {
+                            let mut staged_xl = tempfile::NamedTempFile::new_in(&staging_root)?;
+                            std::io::Write::write_all(staged_xl.as_file_mut(), target_bytes)?;
+                            if sync && !target_bytes.is_empty() {
+                                staged_xl.as_file().sync_data()?;
+                            }
+                            std::fs::rename(staged_xl.path(), &xl_path)
+                        })()
+                    };
+                    if let Err(err) = write_res {
+                        if let Some((_, reserved_marker)) = &reserved_data_path {
+                            let _ = std::fs::remove_file(reserved_marker);
+                        }
+                        let _ = std::fs::remove_dir_all(&destination);
+                        return Err(err);
+                    }
+                } else if let Err(err) = std::fs::remove_file(&xl_path)
                     && err.kind() != ErrorKind::NotFound
                 {
                     if let Some((_, reserved_marker)) = &reserved_data_path {
@@ -2299,6 +2393,7 @@ impl LocalDisk {
                         Err(err) => return Err(err),
                     }
                 }
+                put_prepared_delete_receipt(destination, (journal, sole_transaction));
                 Ok(Ok(Some(old_dir)))
             },
         )
@@ -2344,6 +2439,26 @@ impl LocalDisk {
             lease,
             sync,
             move || -> std::io::Result<Result<Option<(DeleteRecoveryJournal, bool, bool)>>> {
+                if let Some((journal, sole_transaction)) = take_prepared_delete_receipt(&rollback)
+                    && rollback.join(PREPARED).is_file()
+                {
+                    let receipt = CommitReceipt {
+                        journal: journal.clone(),
+                        finished: false,
+                    };
+                    let receipt_bytes = serde_json::to_vec(&receipt).map_err(std::io::Error::other)?;
+                    std::fs::create_dir_all(&staging_root)?;
+                    let mut staged = tempfile::NamedTempFile::new_in(&staging_root)?;
+                    std::io::Write::write_all(staged.as_file_mut(), &receipt_bytes)?;
+                    if sync && !receipt_bytes.is_empty() {
+                        staged.as_file().sync_data()?;
+                    }
+                    std::fs::rename(staged.path(), &receipt_path)?;
+                    if sync {
+                        os::fsync_dir_std(&rollback)?;
+                    }
+                    return Ok(Ok(Some((journal, sole_transaction, true))));
+                }
                 let prepared = match regular_bytes(&rollback.join(PREPARED)) {
                     Ok(Some(bytes)) => bytes,
                     Ok(None) => return Ok(Ok(None)),
@@ -2361,12 +2476,18 @@ impl LocalDisk {
                 {
                     return Ok(Err(DiskError::FileCorrupt));
                 }
-                let current_exists = match regular_file_exists(&root.join(STORAGE_FORMAT_FILE)) {
-                    Ok(exists) => exists,
-                    Err(e) => return Ok(Err(e)),
+                let target_satisfied = match journal.target_digest {
+                    None => match regular_file_exists(&root.join(STORAGE_FORMAT_FILE)) {
+                        Ok(exists) => !exists,
+                        Err(e) => return Ok(Err(e)),
+                    },
+                    Some(expected_target) => match regular_bytes(&root.join(STORAGE_FORMAT_FILE)) {
+                        Ok(Some(bytes)) => digest(&bytes) == expected_target,
+                        Ok(None) => false,
+                        Err(e) => return Ok(Err(e)),
+                    },
                 };
-                let unversioned_target_absent = journal.target_digest.is_none() && !current_exists;
-                if !unversioned_target_absent {
+                if !target_satisfied {
                     return Ok(Ok(None));
                 }
                 let mut sole_transaction = journal.predecessors.is_empty();
@@ -2383,8 +2504,15 @@ impl LocalDisk {
                                     && id != transaction
                                     && !journal.data_dirs.contains(&id)
                                 {
-                                    sole_transaction = false;
-                                    break;
+                                    let dir_path = entry.path();
+                                    let is_surviving_data_dir = journal.original_versions.iter().any(|v| v.data_dir == Some(id))
+                                        && !dir_path.join(PREPARED).exists()
+                                        && !dir_path.join(DELETE_QUORUM_COMMITTED_FILE).exists()
+                                        && !dir_path.join(STORAGE_FORMAT_FILE_BACKUP).exists();
+                                    if !is_surviving_data_dir {
+                                        sole_transaction = false;
+                                        break;
+                                    }
                                 }
                             }
                         }
@@ -2403,12 +2531,11 @@ impl LocalDisk {
                 if sync && !receipt_bytes.is_empty() {
                     staged.as_file().sync_data()?;
                 }
-                os::record_native_publication(&receipt_path);
                 std::fs::rename(staged.path(), &receipt_path)?;
                 if sync {
                     os::fsync_dir_std(&rollback)?;
                 }
-                Ok(Ok(Some((journal, sole_transaction, unversioned_target_absent))))
+                Ok(Ok(Some((journal, sole_transaction, target_satisfied))))
             },
         )
         .await
@@ -2460,12 +2587,15 @@ impl LocalDisk {
         }
 
         let bucket_path = self.io_get_bucket_path(volume)?;
+        let staging_root = self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?;
         let root = self.io_get_object_path(volume, object)?;
         let rollback = root.join(transaction.to_string());
         let receipt_path = rollback.join(DELETE_QUORUM_COMMITTED_FILE);
         let sync = effective_durability(volume).syncs_commit_metadata();
-        let lease = os::acquire_namespace_mutation_lease_with_owner(&receipt_path, owner).await;
         let data_dirs = journal.data_dirs.clone();
+        let has_target = journal.target_digest.is_some();
+        let journal_for_receipt = journal.clone();
+        let lease = os::acquire_namespace_mutation_lease_with_owner(&receipt_path, owner).await;
         let prune_paths: Vec<_> = rollback
             .ancestors()
             .take_while(|dir| *dir != bucket_path.as_path() && dir.starts_with(&bucket_path))
@@ -2476,11 +2606,13 @@ impl LocalDisk {
             let committed = format!("{DELETE_DATA_DIR_MARKER_PREFIX}{transaction}");
             let recovery_committed = format!("{RECOVERY_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}");
             let reserved = format!("{RESERVED_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}");
+            let mut had_physical_data_dirs = false;
             for data_dir in &data_dirs {
                 for base in [&root, &rollback] {
                     let dir_path = base.join(data_dir.to_string());
                     match std::fs::read_dir(&dir_path) {
                         Ok(entries) => {
+                            had_physical_data_dirs = true;
                             let mut removable = Vec::new();
                             let mut preserve_dir = false;
                             for entry in entries {
@@ -2521,23 +2653,45 @@ impl LocalDisk {
                     }
                 }
             }
-            if !data_dirs.is_empty() && sync && root.is_dir() {
+            if had_physical_data_dirs && sync && root.is_dir() {
                 os::fsync_dir_std(&root)?;
             }
-            for name in [
-                STORAGE_FORMAT_FILE_BACKUP,
-                DELETE_MARKER_ROLLBACK_FILE,
-                TARGET,
-                PREPARED,
-                FINISHED,
-            ] {
+            for name in [STORAGE_FORMAT_FILE_BACKUP, DELETE_MARKER_ROLLBACK_FILE] {
                 ignore_not_found(std::fs::remove_file(rollback.join(name)))?;
             }
+            if has_target || had_physical_data_dirs {
+                let finished_receipt = CommitReceipt {
+                    journal: journal_for_receipt,
+                    finished: true,
+                };
+                let receipt_bytes = serde_json::to_vec(&finished_receipt).map_err(std::io::Error::other)?;
+                std::fs::create_dir_all(&staging_root)?;
+                let mut staged_receipt = tempfile::NamedTempFile::new_in(&staging_root)?;
+                std::io::Write::write_all(staged_receipt.as_file_mut(), &receipt_bytes)?;
+                if sync && !receipt_bytes.is_empty() {
+                    staged_receipt.as_file().sync_data()?;
+                }
+                std::fs::rename(staged_receipt.path(), &receipt_path)?;
+                let _ = std::fs::File::create(rollback.join(FINISHED))?;
+                if sync {
+                    os::fsync_dir_std(&rollback)?;
+                }
+            }
+            for name in [TARGET, PREPARED, FINISHED] {
+                let path = rollback.join(name);
+                #[cfg(all(any(test, feature = "test-util"), not(windows)))]
+                os::prepared_publication_test_hooks::run(os::prepared_publication_test_hooks::Stage::Remove, &path);
+                ignore_not_found(std::fs::remove_file(path))?;
+            }
             if sync {
                 os::fsync_dir_std(&rollback)?;
             }
+            #[cfg(all(any(test, feature = "test-util"), not(windows)))]
+            os::prepared_publication_test_hooks::run(os::prepared_publication_test_hooks::Stage::Remove, &receipt_path);
             ignore_not_found(std::fs::remove_file(&receipt_path))?;
             for dir_path in &prune_paths {
+                #[cfg(all(any(test, feature = "test-util"), not(windows)))]
+                os::prepared_publication_test_hooks::run(os::prepared_publication_test_hooks::Stage::Remove, dir_path);
                 if let Err(err) = std::fs::remove_dir(dir_path) {
                     if is_dir_not_empty_error(&err) {
                         break;
@@ -6284,6 +6438,214 @@ mod tests {
             b"outside bytes"
         );
     }
+
+    #[tokio::test]
+    async fn delete_recovery_fused_versioned_delete_marker_and_publish_fast_path() {
+        let (_temp, disk, info, _) = fixture().await;
+        let root = disk.io_get_object_path("recovery", "object").expect("object root");
+        let transaction = Uuid::new_v4();
+        let mut marker = FileInfo {
+            name: "object".to_owned(),
+            version_id: Some(Uuid::new_v4()),
+            deleted: true,
+            mark_deleted: true,
+            mod_time: info.mod_time.map(|t| t + time::Duration::seconds(1)),
+            ..Default::default()
+        };
+        let opts = DeleteOptions {
+            old_data_dir: Some(transaction),
+            recoverable_delete: true,
+            ..Default::default()
+        };
+        let fused = disk
+            .try_fused_recoverable_delete_version("recovery", "object", &marker, &opts, transaction, None)
+            .await
+            .expect("fused delete marker preparation must succeed");
+        assert_eq!(fused, Some(()), "versioned delete marker must use fused path");
+        assert!(root.join(transaction.to_string()).join(TARGET).is_file());
+        assert!(root.join(STORAGE_FORMAT_FILE).is_file());
+
+        let published = disk
+            .try_publish_prepared_delete_local("recovery", "object", transaction, None)
+            .await
+            .expect("local publish must succeed")
+            .expect("versioned target must satisfy local publish");
+        assert!(published.1, "sole transaction must be true");
+        assert!(published.2, "target_satisfied must be true");
+
+        let finalized = disk
+            .try_finish_and_finalize_prepared_delete_local("recovery", "object", transaction, &published.0, None)
+            .await
+            .expect("local finalize must succeed");
+        assert!(finalized);
+        assert!(!root.join(transaction.to_string()).exists());
+        let final_meta = FileMeta::load(&fs::read(root.join(STORAGE_FORMAT_FILE)).await.expect("read final xl.meta"))
+            .expect("decode final xl.meta");
+        assert_eq!(final_meta.versions.len(), 2);
+        assert_eq!(final_meta.versions[0].header.version_id, marker.version_id);
+        marker.deleted = false;
+    }
+
+    #[tokio::test]
+    async fn delete_recovery_fused_versioned_delete_with_surviving_version_and_interrupted_target_cleanup() {
+        let (_temp, disk, mut v1, _) = fixture().await;
+        let root = disk.io_get_object_path("recovery", "object").expect("object path");
+
+        // Configure v1 and v2 with distinct physical data directories so deleting v2
+        // exercises both `old_dir` reservation and `is_surviving_data_dir`.
+        v1.data = None;
+        rustfs_utils::http::metadata_compat::remove_str(
+            &mut v1.metadata,
+            rustfs_utils::http::metadata_compat::SUFFIX_INLINE_DATA,
+        );
+        let v1_dir = Uuid::new_v4();
+        v1.data_dir = Some(v1_dir);
+        fs::create_dir_all(root.join(v1_dir.to_string()))
+            .await
+            .expect("create v1 dir");
+        fs::write(root.join(v1_dir.to_string()).join("part.1"), b"v1-data")
+            .await
+            .expect("write v1 part");
+
+        let mut v2 = v1.clone();
+        v2.version_id = Some(Uuid::new_v4());
+        let v2_dir = Uuid::new_v4();
+        v2.data_dir = Some(v2_dir);
+        fs::create_dir_all(root.join(v2_dir.to_string()))
+            .await
+            .expect("create v2 dir");
+        fs::write(root.join(v2_dir.to_string()).join("part.1"), b"v2-data")
+            .await
+            .expect("write v2 part");
+
+        let mut meta = FileMeta::new();
+        meta.add_version(v1.clone()).expect("add v1");
+        meta.add_version(v2.clone()).expect("add v2");
+        fs::write(root.join(STORAGE_FORMAT_FILE), meta.marshal_msg().expect("marshal meta"))
+            .await
+            .expect("write xl.meta");
+
+        let transaction = Uuid::new_v4();
+        let opts = DeleteOptions {
+            recoverable_delete: true,
+            ..Default::default()
+        };
+        let fused = disk
+            .try_fused_recoverable_delete_version("recovery", "object", &v2, &opts, transaction, None)
+            .await
+            .expect("fused versioned delete preparation must succeed");
+        assert_eq!(fused, Some(()));
+        assert!(
+            root.join(v2_dir.to_string())
+                .join(format!("{RECOVERY_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}"))
+                .is_file(),
+            "deleted version data_dir must have recovery marker"
+        );
+
+        let (journal, sole_transaction, target_satisfied) = disk
+            .try_publish_prepared_delete_local("recovery", "object", transaction, None)
+            .await
+            .expect("local publish must succeed")
+            .expect("versioned target must satisfy local publish");
+        assert!(sole_transaction, "surviving v1 data_dir must not clear sole_transaction");
+        assert!(target_satisfied);
+        assert!(journal.target_digest.is_some());
+
+        // Simulate interruption during finish-before-finalize right after TARGET is removed
+        // (after finished receipt + FINISHED marker were persisted, before receipt_path is removed).
+        let rollback = root.join(transaction.to_string());
+        let finished_receipt = CommitReceipt {
+            journal: journal.clone(),
+            finished: true,
+        };
+        fs::write(
+            rollback.join(DELETE_QUORUM_COMMITTED_FILE),
+            serde_json::to_vec(&finished_receipt).expect("serialize finished receipt"),
+        )
+        .await
+        .expect("write finished receipt");
+        fs::write(rollback.join(FINISHED), []).await.expect("write FINISHED");
+        fs::remove_file(rollback.join(STORAGE_FORMAT_FILE_BACKUP))
+            .await
+            .expect("remove backup");
+        fs::remove_file(rollback.join(TARGET)).await.expect("remove TARGET");
+
+        let inspected = disk
+            .inspect_delete_transaction("recovery", "object", transaction)
+            .await
+            .expect("interrupted post-TARGET state must remain inspectable without FileCorrupt");
+        assert_eq!(inspected.state, DeleteRecoveryState::CommitFinished);
+
+        // Complete cleanup via try_finish_and_finalize_prepared_delete_local.
+        let finalized = disk
+            .try_finish_and_finalize_prepared_delete_local("recovery", "object", transaction, &journal, None)
+            .await
+            .expect("local finalize must succeed after interruption");
+        assert!(finalized);
+        assert!(!rollback.exists(), "rollback directory must be pruned");
+        assert!(!root.join(v2_dir.to_string()).exists(), "deleted v2 data_dir must be removed");
+        assert!(
+            root.join(v1_dir.to_string()).join("part.1").is_file(),
+            "surviving v1 data_dir must remain intact"
+        );
+    }
+
+    #[tokio::test]
+    async fn delete_recovery_fused_unversioned_physical_delete_retains_receipt_until_finalize() {
+        let (_temp, disk, mut info, _) = fixture().await;
+        let root = disk.io_get_object_path("recovery", "object").expect("object path");
+
+        info.data = None;
+        rustfs_utils::http::metadata_compat::remove_str(
+            &mut info.metadata,
+            rustfs_utils::http::metadata_compat::SUFFIX_INLINE_DATA,
+        );
+        let data_dir = Uuid::new_v4();
+        info.data_dir = Some(data_dir);
+        fs::create_dir_all(root.join(data_dir.to_string()))
+            .await
+            .expect("create data_dir");
+        fs::write(root.join(data_dir.to_string()).join("part.1"), b"physical-payload")
+            .await
+            .expect("write part.1");
+        let mut meta = FileMeta::new();
+        meta.add_version(info.clone()).expect("add version");
+        fs::write(root.join(STORAGE_FORMAT_FILE), meta.marshal_msg().expect("marshal meta"))
+            .await
+            .expect("write xl.meta");
+
+        let transaction = Uuid::new_v4();
+        let opts = DeleteOptions {
+            recoverable_delete: true,
+            ..Default::default()
+        };
+        let fused = disk
+            .try_fused_recoverable_delete_version("recovery", "object", &info, &opts, transaction, None)
+            .await
+            .expect("fused unversioned physical delete must succeed");
+        assert_eq!(fused, Some(()));
+
+        let (journal, sole_transaction, target_satisfied) = disk
+            .try_publish_prepared_delete_local("recovery", "object", transaction, None)
+            .await
+            .expect("local publish must succeed")
+            .expect("absent target must satisfy local publish");
+        assert!(sole_transaction);
+        assert!(target_satisfied);
+        assert!(journal.target_digest.is_none());
+
+        let rollback = root.join(transaction.to_string());
+        assert!(rollback.join(DELETE_QUORUM_COMMITTED_FILE).is_file());
+
+        let finalized = disk
+            .try_finish_and_finalize_prepared_delete_local("recovery", "object", transaction, &journal, None)
+            .await
+            .expect("local finalize must succeed");
+        assert!(finalized);
+        assert!(!rollback.exists(), "rollback directory must be pruned");
+        assert!(!root.join(data_dir.to_string()).exists(), "physical data_dir must be removed");
+        assert!(!root.exists(), "empty object directory must be pruned");
+    }
 }
 
 #[cfg(test)]

crates/ecstore/src/disk/local/multipart_source_abort.rs

@@ -421,6 +421,7 @@ impl LocalDisk {
                     return Err(DiskError::ConditionalDeleteConflict);
                 }
                 let permit = AbortDeletePermit::new(&store, &expected, certificate.clone())?;
+                super::multipart_source_state::invalidate_active_multipart_source_head(&store.source);
                 let destination = self.io_get_object_path(RUSTFS_META_TMP_DELETED_BUCKET, &Uuid::new_v4().to_string())?;
                 let source = store.source.clone();
                 let base = store.base.clone();
@@ -618,6 +619,9 @@ pub(super) struct OrdinaryMultipartDeleteCheck {
     root: os::PublicationRoot,
 }
 impl OrdinaryMultipartDeleteCheck {
+    pub(super) fn source(&self) -> &Path {
+        &self.source
+    }
     pub(super) fn validate(&self) -> Result<()> {
         let pinned = match os::pin_multipart_source_directory(&self.source, &self.base, &self.root) {
             Ok(pinned) => pinned,

crates/ecstore/src/disk/local/multipart_source_state.rs

@@ -238,9 +238,192 @@ pub(super) fn read_optional_in_pinned(
         Err(error) => Err(error.into()),
     }
 }
+#[derive(Clone, Copy, PartialEq, Eq)]
+struct FileStamp {
+    len: u64,
+    modified: Option<std::time::SystemTime>,
+    #[cfg(unix)]
+    ino: u64,
+}
+
+impl FileStamp {
+    fn from_path(path: &Path) -> Option<Self> {
+        let meta = std::fs::symlink_metadata(path).ok()?;
+        Some(Self {
+            len: meta.len(),
+            modified: meta.modified().ok(),
+            #[cfg(unix)]
+            ino: {
+                use std::os::unix::fs::MetadataExt;
+                meta.ino()
+            },
+        })
+    }
+}
+
+#[derive(Clone)]
+struct ActiveMultipartSourceEntry {
+    raw: Bytes,
+    fi: FileInfo,
+    head_stamp: FileStamp,
+    unsettled_parts: std::collections::HashSet<usize>,
+    parts: std::collections::HashMap<usize, (FileStamp, rustfs_filemeta::ObjectPartInfo)>,
+}
+
+static ACTIVE_MULTIPART_SOURCE_HEADS: std::sync::LazyLock<
+    parking_lot::Mutex<std::collections::HashMap<PathBuf, ActiveMultipartSourceEntry>>,
+> = std::sync::LazyLock::new(|| parking_lot::Mutex::new(std::collections::HashMap::new()));
+
+pub(super) fn put_active_multipart_source_head(source: PathBuf, raw: Bytes, fi: FileInfo) {
+    let Some(head_stamp) = FileStamp::from_path(&source.join(STORAGE_FORMAT_FILE)) else {
+        return;
+    };
+    let mut map = ACTIVE_MULTIPART_SOURCE_HEADS.lock();
+    if map.len() >= 4096
+        && let Some(evict_key) = map.keys().next().cloned()
+    {
+        map.remove(&evict_key);
+    }
+    map.insert(
+        source,
+        ActiveMultipartSourceEntry {
+            raw,
+            fi,
+            head_stamp,
+            unsettled_parts: std::collections::HashSet::new(),
+            parts: std::collections::HashMap::new(),
+        },
+    );
+}
+
+fn stamp_matches_active_disk(source: &Path, expected_head_stamp: FileStamp) -> bool {
+    FileStamp::from_path(&source.join(STORAGE_FORMAT_FILE)) == Some(expected_head_stamp)
+        && !source.join(STATE).exists()
+        && !source.join(crate::disk::MULTIPART_RECOVERY_DIR).exists()
+        && !source.join(".mpu-completions").exists()
+}
+
+pub(super) fn get_active_multipart_source_head(source: &Path) -> Option<(Bytes, FileInfo)> {
+    let (raw, fi, head_stamp) = {
+        let map = ACTIVE_MULTIPART_SOURCE_HEADS.lock();
+        let entry = map.get(source)?;
+        (entry.raw.clone(), entry.fi.clone(), entry.head_stamp)
+    };
+    if !stamp_matches_active_disk(source, head_stamp) {
+        ACTIVE_MULTIPART_SOURCE_HEADS.lock().remove(source);
+        return None;
+    }
+    Some((raw, fi))
+}
+
+pub(super) fn has_active_multipart_source_head(source: &Path) -> bool {
+    let Some(head_stamp) = ACTIVE_MULTIPART_SOURCE_HEADS.lock().get(source).map(|entry| entry.head_stamp) else {
+        return false;
+    };
+    if !stamp_matches_active_disk(source, head_stamp) {
+        ACTIVE_MULTIPART_SOURCE_HEADS.lock().remove(source);
+        return false;
+    }
+    true
+}
+
+pub(super) fn has_clean_active_multipart_source(source: &Path) -> bool {
+    let Some((head_stamp, clean)) = ACTIVE_MULTIPART_SOURCE_HEADS
+        .lock()
+        .get(source)
+        .map(|entry| (entry.head_stamp, entry.unsettled_parts.is_empty()))
+    else {
+        return false;
+    };
+    if !stamp_matches_active_disk(source, head_stamp) {
+        ACTIVE_MULTIPART_SOURCE_HEADS.lock().remove(source);
+        return false;
+    }
+    clean
+}
+
+pub(super) fn begin_active_multipart_part_transaction(source: &Path, part_number: usize) {
+    if let Some(entry) = ACTIVE_MULTIPART_SOURCE_HEADS.lock().get_mut(source) {
+        entry.unsettled_parts.insert(part_number);
+        entry.parts.remove(&part_number);
+    }
+}
+
+pub(super) fn record_active_multipart_part_committed(source: &Path, part: rustfs_filemeta::ObjectPartInfo) {
+    let Some(data_dir) = ACTIVE_MULTIPART_SOURCE_HEADS
+        .lock()
+        .get(source)
+        .and_then(|entry| entry.fi.data_dir)
+        .filter(|dir| !dir.is_nil())
+    else {
+        return;
+    };
+    let part_meta_path = source.join(data_dir.to_string()).join(format!("part.{}.meta", part.number));
+    let Some(part_stamp) = FileStamp::from_path(&part_meta_path) else {
+        return;
+    };
+    if let Some(entry) = ACTIVE_MULTIPART_SOURCE_HEADS.lock().get_mut(source) {
+        entry.parts.insert(part.number, (part_stamp, part));
+    }
+}
+
+pub(super) fn finish_active_multipart_part_transaction(
+    source: &Path,
+    part_number: Option<usize>,
+    committed: bool,
+    retired_existing_txn: bool,
+) {
+    let mut map = ACTIVE_MULTIPART_SOURCE_HEADS.lock();
+    let Some(part_number) = part_number.filter(|_| committed) else {
+        map.remove(source);
+        return;
+    };
+    if !retired_existing_txn {
+        return;
+    }
+    if let Some(entry) = map.get_mut(source)
+        && !entry.unsettled_parts.remove(&part_number)
+    {
+        map.remove(source);
+    }
+}
+
+pub(super) fn get_active_multipart_source_part(source: &Path, part_number: usize) -> Option<rustfs_filemeta::ObjectPartInfo> {
+    let (head_stamp, data_dir, expected_stamp, part) = {
+        let map = ACTIVE_MULTIPART_SOURCE_HEADS.lock();
+        let entry = map.get(source)?;
+        if !entry.unsettled_parts.is_empty() {
+            return None;
+        }
+        let data_dir = entry.fi.data_dir.filter(|dir| !dir.is_nil())?;
+        let (expected_stamp, part) = entry.parts.get(&part_number)?;
+        (entry.head_stamp, data_dir, *expected_stamp, part.clone())
+    };
+    if !stamp_matches_active_disk(source, head_stamp) {
+        ACTIVE_MULTIPART_SOURCE_HEADS.lock().remove(source);
+        return None;
+    }
+    let part_meta_path = source.join(data_dir.to_string()).join(format!("part.{part_number}.meta"));
+    if FileStamp::from_path(&part_meta_path) != Some(expected_stamp) {
+        return None;
+    }
+    Some(part)
+}
+
+pub(super) fn invalidate_active_multipart_source_head(source: &Path) {
+    ACTIVE_MULTIPART_SOURCE_HEADS.lock().remove(source);
+}
+
+pub(super) fn invalidate_active_multipart_sources_under(prefix: &Path) {
+    ACTIVE_MULTIPART_SOURCE_HEADS.lock().retain(|key, _| !key.starts_with(prefix));
+}
+
 /// Canonical source mutations are admitted under SourceRW at the syscall,
 /// not on the earlier async caller that may have timed out.
 pub(super) fn validate_source_mutation(source: &Path, base: &Path, root: &os::PublicationRoot, initial: bool) -> Result<()> {
+    if !initial && has_active_multipart_source_head(source) {
+        return Ok(());
+    }
     let pinned = match os::pin_multipart_source_directory(source, base, root) {
         Ok(pinned) => pinned,
         Err(error) if error.kind() == std::io::ErrorKind::NotFound => {
@@ -440,6 +623,7 @@ impl SourceStore {
         terminal: Option<&Path>,
         sync_source_before: bool,
     ) -> std::io::Result<()> {
+        invalidate_active_multipart_source_head(source);
         #[cfg(unix)]
         let mut pinned = os::pin_multipart_source_directory(source, base, root)?;
         #[cfg(not(unix))]
@@ -470,7 +654,7 @@ impl SourceStore {
         let terminal = terminal_record_path(&source, state);
         let bytes = bytes.to_vec();
         os::run_blocking_namespace_operation_with_sync(self.lease.clone(), true, move || {
-            Self::write_owned_backup_std(&source, &base, &root, owner, target, &bytes)?;
+            Self::write_owned_backup_std(&source, &base, &root, owner, target, &bytes, false)?;
             Self::persist_state_std(&source, &base, &root, &raw, terminal.as_deref(), false)
         })
         .await
@@ -539,7 +723,7 @@ impl SourceStore {
         let root = self.publication_root.clone();
         let bytes = bytes.to_vec();
         os::run_blocking_namespace_operation(self.lease.clone(), move || {
-            Self::write_owned_backup_std(&source, &base, &root, owner, target, &bytes)
+            Self::write_owned_backup_std(&source, &base, &root, owner, target, &bytes, true)
         })
         .await
         .map_err(DiskError::from)
@@ -551,6 +735,7 @@ impl SourceStore {
         owner: Uuid,
         target: bool,
         bytes: &[u8],
+        sync_source_root: bool,
     ) -> std::io::Result<()> {
         let path = if target {
             target_backup_path(source, owner)
@@ -582,9 +767,10 @@ impl SourceStore {
         if !existed {
             os::replace_multipart_source_metadata(&path, bytes, base, root)?;
         }
-        os::sync_multipart_owned_record_parents(source, parent, existed, base, root)
+        os::sync_multipart_owned_record_parents_options(source, parent, existed, sync_source_root, base, root)
     }
     pub async fn replace_head(&self, bytes: &[u8]) -> Result<()> {
+        invalidate_active_multipart_source_head(&self.source);
         let path = self.source.join(STORAGE_FORMAT_FILE);
         let base = self.base.clone();
         let root = self.publication_root.clone();
@@ -1097,6 +1283,82 @@ pub(super) fn load_source(
     })
 }
 
+fn load_uninitialized_source(
+    raw: Bytes,
+    path: &str,
+    disk_id: Uuid,
+    physical_slot: usize,
+    disk_ids: &[Uuid],
+) -> Result<LoadedSource> {
+    let sha = raw_sha(&raw);
+    let length = raw.len() as u64;
+    let current = get_file_info(
+        &raw,
+        RUSTFS_META_MULTIPART_BUCKET,
+        path,
+        "",
+        FileInfoOpts {
+            data: false,
+            include_free_versions: false,
+            include_part_checksums: true,
+        },
+    )
+    .map_err(Error::other)?;
+    native_source_protocol(&current)?;
+    if source_status(&current)?.is_some()
+        || publication_identity(&current)?.is_some()
+        || SourceConditionV1::take(&mut current.clone())?.is_some()
+    {
+        return Err(DiskError::FileCorrupt);
+    }
+    let bucket = rustfs_utils::http::metadata_compat::get_consistent_str(&current.metadata, "multipart-bucket")
+        .ok_or(DiskError::FileCorrupt)?;
+    let object = rustfs_utils::http::metadata_compat::get_consistent_str(&current.metadata, "multipart-object")
+        .ok_or(DiskError::FileCorrupt)?;
+    let condition = capture_source_condition(bucket, object, path, &current, current.data_dir.ok_or(DiskError::FileCorrupt)?)?;
+    if condition.generation.epoch != 0 {
+        return Err(DiskError::FileCorrupt);
+    }
+    let slot = condition
+        .lineage
+        .geometry
+        .distribution
+        .iter()
+        .position(|index| *index == current.erasure.index)
+        .ok_or(DiskError::FileCorrupt)?;
+    if slot != physical_slot || condition.lineage.geometry.disks() != disk_ids.len() {
+        return Err(DiskError::ConditionalDeleteConflict);
+    }
+    let inventory = raw_sha(&encode(&(None::<[u8; 32]>, Some(sha), sha, None::<[u8; 32]>, None::<[u8; 32]>))?);
+    let snapshot = Some(MultipartSourceSnapshot {
+        disk_id,
+        physical_slot: slot,
+        lineage: condition.lineage,
+        generation: condition.generation,
+        revision: 0,
+        phase: MultipartSourcePhase::Active,
+        state_sha256: [0; 32],
+        head_sha256: Some(sha),
+        inventory_sha256: inventory,
+        capability_sha256: None,
+        publication: None,
+    });
+    Ok(LoadedSource {
+        root_missing: false,
+        state: None,
+        head: Some(SourceHead {
+            sha256: sha,
+            length,
+            raw: Some(raw.clone()),
+        }),
+        original: Some(raw),
+        info: Some(current),
+        snapshot,
+        head_matches_state: true,
+        head_matches_activation: false,
+    })
+}
+
 pub(super) const RETIREMENT_ANCHOR: &str = ".mpu-source-retirement";
 
 #[derive(Serialize, Deserialize)]
@@ -1323,41 +1585,6 @@ impl LocalDisk {
 }
 
 impl SourceStore {
-    /// Only native source validation may create provenance. Ordinary TMP/heal
-    /// publication can retain a valid old local anchor, never issue/rebind one.
-    pub async fn write_retirement_anchor(
-        &self,
-        state: &SourceState,
-        publication: &PublicationIdentity,
-        directory: &Path,
-        lease: Arc<os::NamespaceMutationLease>,
-    ) -> Result<()> {
-        if state.lineage != publication.lineage
-            || state.generation != publication.source_generation
-            || state.disk_id != self.disk_id
-            || state.physical_slot != self.physical_slot
-        {
-            return Err(DiskError::ConditionalDeleteConflict);
-        }
-        let record = NativeRetirementAnchor {
-            protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
-            disk_id: self.disk_id,
-            physical_slot: self.physical_slot,
-            publication: publication.clone(),
-            source_original_sha256: state.original_sha256,
-            source_validation_receipt_id: state.backup_owner,
-        };
-        let bytes = encode(&record)?;
-        let directory = directory.to_path_buf();
-        let base = self.base.clone();
-        let root = self.publication_root.clone();
-        os::run_blocking_namespace_operation(lease, move || {
-            let _parent = os::mkdir_all_below_existing_base_std(&directory, &base, &root)?;
-            os::replace_multipart_source_metadata(&directory.join(RETIREMENT_ANCHOR), &bytes, &base, &root)
-        })
-        .await
-        .map_err(DiskError::from)
-    }
     /// This caller already owns SourceRW and destination metadata ownership.
     /// An Aborting survivor can confirm a historical target only with its
     /// independent, same-disk publication anchor. Closure never creates one.
@@ -1456,6 +1683,8 @@ struct ValidatedPreparedPublication {
     length: u64,
 }
 
+type DeferredInitialPublication = (Vec<u8>, PathBuf, Vec<u8>);
+
 pub(super) struct NativeMultipartPublication {
     pub store: Arc<SourceStore>,
     pub publication: PublicationIdentity,
@@ -1463,22 +1692,49 @@ pub(super) struct NativeMultipartPublication {
     state: parking_lot::Mutex<SourceState>,
     capability: parking_lot::Mutex<Option<MultipartUndoCapability>>,
     validated_prepared: parking_lot::Mutex<Option<ValidatedPreparedPublication>>,
+    deferred_initial: parking_lot::Mutex<Option<DeferredInitialPublication>>,
     pub legacy: bool,
 }
 
 impl LocalDisk {
+    #[cfg(test)]
     pub(super) async fn prepare_native_multipart_publication(
         &self,
         path: &str,
         incoming: &mut FileInfo,
         staging: &MultipartCommitStaging,
+    ) -> Result<NativeMultipartPublication> {
+        self.prepare_native_multipart_publication_inner(path, incoming, staging, false)
+            .await
+    }
+
+    pub(super) async fn prepare_native_multipart_publication_deferred(
+        &self,
+        path: &str,
+        incoming: &mut FileInfo,
+        staging: &MultipartCommitStaging,
+    ) -> Result<NativeMultipartPublication> {
+        self.prepare_native_multipart_publication_inner(path, incoming, staging, true)
+            .await
+    }
+
+    async fn prepare_native_multipart_publication_inner(
+        &self,
+        path: &str,
+        incoming: &mut FileInfo,
+        staging: &MultipartCommitStaging,
+        defer_persist: bool,
     ) -> Result<NativeMultipartPublication> {
         // Real Complete is parts-backed, including a zero-byte last part.
         if incoming.data.is_some() || incoming.inline_data() {
             return Err(DiskError::UnsupportedDisk);
         }
         let store = Arc::new(self.multipart_store_from_staging(path, staging).await?);
-        let loaded = store.load().await?;
+        let loaded = if defer_persist && let Some(raw) = staging.fresh_source_head() {
+            load_uninitialized_source(raw, &store.path, store.disk_id, store.physical_slot, &store.disk_ids)?
+        } else {
+            store.load().await?
+        };
         let snapshot = loaded.snapshot.as_ref().ok_or(DiskError::ConditionalDeleteConflict)?;
         let original = loaded
             .head
@@ -1569,7 +1825,44 @@ impl LocalDisk {
         };
         // Copy the small source FI to an independent immutable inode. Its
         // canonical inode is never truncated to build the destination proxy.
-        store.write_owned_backup_and_persist(owner, false, original, &state).await?;
+        let directory = staging.private_directory(publication.final_data_dir);
+        let anchor_record = NativeRetirementAnchor {
+            protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
+            disk_id: store.disk_id,
+            physical_slot: store.physical_slot,
+            publication: publication.clone(),
+            source_original_sha256: state.original_sha256,
+            source_validation_receipt_id: state.backup_owner,
+        };
+        let anchor_bytes = encode(&anchor_record)?;
+        state.validate(store.disk_id, &store.path)?;
+        let original_bytes = original.to_vec();
+        let deferred_initial = if defer_persist {
+            Some((original_bytes, directory, anchor_bytes))
+        } else {
+            let state_raw = encode(&state)?;
+            let source_dir = store.source.clone();
+            let base_dir = store.base.clone();
+            let pub_root = store.publication_root.clone();
+            let terminal = terminal_record_path(&source_dir, &state);
+            #[cfg(target_os = "linux")]
+            let can_mark_presynced = has_active_multipart_source_head(&source_dir)
+                && super::effective_durability(RUSTFS_META_MULTIPART_BUCKET).syncs_data_shards();
+            os::run_blocking_namespace_operation_with_sync(staging.lease.clone(), true, move || {
+                SourceStore::write_owned_backup_std(&source_dir, &base_dir, &pub_root, owner, false, &original_bytes, false)?;
+                SourceStore::persist_state_std(&source_dir, &base_dir, &pub_root, &state_raw, terminal.as_deref(), false)?;
+                let _parent = os::mkdir_all_below_existing_base_std(&directory, &base_dir, &pub_root)?;
+                os::replace_multipart_source_metadata(&directory.join(RETIREMENT_ANCHOR), &anchor_bytes, &base_dir, &pub_root)?;
+                #[cfg(target_os = "linux")]
+                if can_mark_presynced {
+                    super::mark_presynced_tmp_data_dir(directory);
+                }
+                Ok(())
+            })
+            .await
+            .map_err(DiskError::from)?;
+            None
+        };
         let prepared_path = store.source.join(format!(".mpu-prepared-{owner}"));
         let context = NativeMultipartPublication {
             store,
@@ -1578,14 +1871,9 @@ impl LocalDisk {
             state: parking_lot::Mutex::new(state),
             capability: parking_lot::Mutex::new(Some(capability)),
             validated_prepared: parking_lot::Mutex::new(None),
+            deferred_initial: parking_lot::Mutex::new(deferred_initial),
             legacy,
         };
-        let directory = staging.private_directory(context.publication.final_data_dir);
-        let reserved = context.state.lock().clone();
-        context
-            .store
-            .write_retirement_anchor(&reserved, &context.publication, &directory, staging.lease.clone())
-            .await?;
         Ok(context)
     }
 }
@@ -1594,18 +1882,15 @@ impl NativeMultipartPublication {
     pub fn canonical_head(&self) -> PathBuf {
         self.store.source.join(STORAGE_FORMAT_FILE)
     }
+    #[cfg(test)]
     pub async fn prepare_metadata(&self, new_metadata: &[u8], old_metadata: Option<&[u8]>) -> Result<()> {
         let mut state = self.state.lock().clone();
-        let current = self.store.load().await?;
-        let current_state = current.state.as_ref().ok_or(DiskError::ConditionalDeleteConflict)?;
-        if encode(current_state)? != encode(&state)? || !current.head_matches_state {
-            return Err(DiskError::ConditionalDeleteConflict);
-        }
         // Validate the actual persisted representation once, after FileMeta's
         // canonicalization. No unvalidated input can install a source proxy.
         if !self.store.target_matches(&self.publication, new_metadata)? {
             return Err(DiskError::ConditionalDeleteConflict);
         }
+        let expected_state_raw = encode(&state)?;
         let validated = ValidatedPreparedPublication {
             publication: self.publication.clone(),
             sha256: raw_sha(new_metadata),
@@ -1618,17 +1903,297 @@ impl NativeMultipartPublication {
         state.prepared_sha256 = Some(validated.sha256);
         state.prepared_length = Some(validated.length);
         state.advance(MultipartSourcePhase::Prepared)?;
-        if let Some(previous) = old_metadata {
-            self.store
-                .write_owned_backup_and_persist(state.backup_owner, true, previous, &state)
-                .await?;
-        } else {
-            self.store.persist(&state).await?;
-        }
+        state.validate(self.store.disk_id, &self.store.path)?;
+        let next_state_raw = encode(&state)?;
+        let source = self.store.source.clone();
+        let base = self.store.base.clone();
+        let root = self.store.publication_root.clone();
+        let path = self.store.path.clone();
+        let disk_id = self.store.disk_id;
+        let physical_slot = self.store.physical_slot;
+        let disk_ids = self.store.disk_ids.clone();
+        let backup_owner = state.backup_owner;
+        let old_bytes = old_metadata.map(|b| b.to_vec());
+        let terminal = terminal_record_path(&source, &state);
+        os::run_blocking_namespace_operation_with_sync(self.store.lease.clone(), true, move || {
+            let current = load_source(&source, &base, &root, &path, disk_id, physical_slot, &disk_ids).map_err(IoError::other)?;
+            let current_state = current
+                .state
+                .as_ref()
+                .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+            let current_raw = encode(current_state).map_err(IoError::other)?;
+            if current_raw != expected_state_raw || !current.head_matches_state {
+                return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+            }
+            if let Some(previous) = old_bytes.as_deref() {
+                SourceStore::write_owned_backup_std(&source, &base, &root, backup_owner, true, previous, false)?;
+            }
+            SourceStore::persist_state_std(&source, &base, &root, &next_state_raw, terminal.as_deref(), false)
+        })
+        .await
+        .map_err(DiskError::from)?;
         *self.state.lock() = state;
         *self.validated_prepared.lock() = Some(validated);
         Ok(())
     }
+    pub(super) fn prepare_and_install_head_on_thread(
+        &self,
+        new_metadata: &[u8],
+        old_metadata: Option<&[u8]>,
+        dst_file_path: &Path,
+        directory_guard: &os::RenameCommitGuard,
+    ) -> Result<os::PreparedRenameSource> {
+        let mut state = self.state.lock().clone();
+        if !self.store.target_matches(&self.publication, new_metadata)? {
+            return Err(DiskError::ConditionalDeleteConflict);
+        }
+        let deferred_initial = self.deferred_initial.lock().take();
+        let expected_state_raw = encode(&state)?;
+        let expected_lineage = state.lineage.clone();
+        let expected_generation = state.generation;
+        let expected_revision = state.revision;
+        let validated = ValidatedPreparedPublication {
+            publication: self.publication.clone(),
+            sha256: raw_sha(new_metadata),
+            length: new_metadata.len() as u64,
+        };
+        if let Some(previous) = old_metadata {
+            state.previous_target_sha256 = Some(raw_sha(previous));
+            state.previous_target_length = Some(previous.len() as u64);
+        }
+        state.prepared_sha256 = Some(validated.sha256);
+        state.prepared_length = Some(validated.length);
+        state.advance(MultipartSourcePhase::Prepared)?;
+        state.validate(self.store.disk_id, &self.store.path)?;
+        let next_state_raw = encode(&state)?;
+        let source = &self.store.source;
+        let base = &self.store.base;
+        let root = &self.store.publication_root;
+        let path = &self.store.path;
+        let disk_id = self.store.disk_id;
+        let physical_slot = self.store.physical_slot;
+        let disk_ids = &self.store.disk_ids;
+        let backup_owner = state.backup_owner;
+        #[cfg(target_os = "linux")]
+        let can_mark_presynced = has_active_multipart_source_head(source)
+            && super::effective_durability(RUSTFS_META_MULTIPART_BUCKET).syncs_data_shards();
+        let terminal = terminal_record_path(source, &state);
+        let prepared_path = &self.prepared_path;
+        let canonical = self.canonical_head();
+        let prepared = (|| -> std::io::Result<os::PreparedRenameSource> {
+            if let Some((ref original_bytes, _, _)) = deferred_initial {
+                let current = load_source(source, base, root, path, disk_id, physical_slot, disk_ids).map_err(IoError::other)?;
+                let snapshot = current
+                    .snapshot
+                    .as_ref()
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                let current_head = current
+                    .head
+                    .as_ref()
+                    .and_then(SourceHead::raw)
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                if !current.head_matches_state
+                    || snapshot.phase != MultipartSourcePhase::Active
+                    || snapshot.capability_sha256.is_some()
+                    || snapshot.lineage != expected_lineage
+                    || snapshot.generation != expected_generation
+                    || snapshot.revision.checked_add(1) != Some(expected_revision)
+                    || current_head != original_bytes.as_slice()
+                {
+                    return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                }
+            } else {
+                let current = load_source(source, base, root, path, disk_id, physical_slot, disk_ids).map_err(IoError::other)?;
+                let current_state = current
+                    .state
+                    .as_ref()
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                let current_raw = encode(current_state).map_err(IoError::other)?;
+                if current_raw != expected_state_raw || !current.head_matches_state {
+                    return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                }
+            }
+            #[cfg(test)]
+            super::run_owned_file_before_open(prepared_path);
+            let mut prepared_metadata_source =
+                os::create_prepared_rename_source_with_commit_guard(prepared_path, dst_file_path, directory_guard)?;
+            if let Err(err) = prepared_metadata_source.write_all(new_metadata, false) {
+                drop(prepared_metadata_source);
+                let _ = std::fs::remove_file(prepared_path);
+                return Err(err);
+            }
+            let setup_res: std::io::Result<()> = (|| {
+                if let Some((original_bytes, directory, anchor_bytes)) = deferred_initial {
+                    SourceStore::write_owned_backup_std(source, base, root, backup_owner, false, &original_bytes, false)?;
+                    if let Some(previous) = old_metadata {
+                        SourceStore::write_owned_backup_std(source, base, root, backup_owner, true, previous, false)?;
+                    }
+                    SourceStore::persist_state_std(source, base, root, &next_state_raw, terminal.as_deref(), false)?;
+                    let _parent = os::mkdir_all_below_existing_base_std(&directory, base, root)?;
+                    os::replace_multipart_source_metadata(&directory.join(RETIREMENT_ANCHOR), &anchor_bytes, base, root)?;
+                    #[cfg(target_os = "linux")]
+                    if can_mark_presynced {
+                        super::mark_presynced_tmp_data_dir(directory);
+                    }
+                } else {
+                    if let Some(previous) = old_metadata {
+                        SourceStore::write_owned_backup_std(source, base, root, backup_owner, true, previous, false)?;
+                    }
+                    SourceStore::persist_state_std(source, base, root, &next_state_raw, terminal.as_deref(), false)?;
+                }
+                Ok(())
+            })();
+            if let Err(err) = setup_res {
+                drop(prepared_metadata_source);
+                let _ = std::fs::remove_file(prepared_path);
+                return Err(err);
+            }
+            if let Err(err) = prepared_metadata_source.sync_data() {
+                drop(prepared_metadata_source);
+                let _ = std::fs::remove_file(prepared_path);
+                return Err(err);
+            }
+            os::install_multipart_prepared_head(prepared_metadata_source, &canonical, base, root)
+        })()
+        .map_err(DiskError::from)?;
+        *self.state.lock() = state;
+        *self.validated_prepared.lock() = Some(validated);
+        Ok(prepared)
+    }
+
+    #[cfg(not(target_os = "linux"))]
+    pub async fn prepare_and_install_head(
+        &self,
+        new_metadata: &[u8],
+        old_metadata: Option<&[u8]>,
+        dst_file_path: &Path,
+        directory_guard: &os::RenameCommitGuard,
+        mutation_lease: Arc<os::NamespaceMutationLease>,
+    ) -> Result<os::PreparedRenameSource> {
+        let mut state = self.state.lock().clone();
+        if !self.store.target_matches(&self.publication, new_metadata)? {
+            return Err(DiskError::ConditionalDeleteConflict);
+        }
+        let deferred_initial = self.deferred_initial.lock().take();
+        let expected_state_raw = encode(&state)?;
+        let expected_lineage = state.lineage.clone();
+        let expected_generation = state.generation;
+        let expected_revision = state.revision;
+        let validated = ValidatedPreparedPublication {
+            publication: self.publication.clone(),
+            sha256: raw_sha(new_metadata),
+            length: new_metadata.len() as u64,
+        };
+        if let Some(previous) = old_metadata {
+            state.previous_target_sha256 = Some(raw_sha(previous));
+            state.previous_target_length = Some(previous.len() as u64);
+        }
+        state.prepared_sha256 = Some(validated.sha256);
+        state.prepared_length = Some(validated.length);
+        state.advance(MultipartSourcePhase::Prepared)?;
+        state.validate(self.store.disk_id, &self.store.path)?;
+        let next_state_raw = encode(&state)?;
+        let source = self.store.source.clone();
+        let base = self.store.base.clone();
+        let root = self.store.publication_root.clone();
+        let path = self.store.path.clone();
+        let disk_id = self.store.disk_id;
+        let physical_slot = self.store.physical_slot;
+        let disk_ids = self.store.disk_ids.clone();
+        let backup_owner = state.backup_owner;
+        #[cfg(target_os = "linux")]
+        let can_mark_presynced = has_active_multipart_source_head(&source)
+            && super::effective_durability(RUSTFS_META_MULTIPART_BUCKET).syncs_data_shards();
+        let old_bytes = old_metadata.map(|b| b.to_vec());
+        let new_bytes = new_metadata.to_vec();
+        let terminal = terminal_record_path(&source, &state);
+        let prepared_path = self.prepared_path.clone();
+        let canonical = self.canonical_head();
+        let dst_file_path = dst_file_path.to_path_buf();
+        let owned_directory_guard = directory_guard.clone();
+        let prepared = os::run_blocking_namespace_operation_with_sync(mutation_lease, true, move || {
+            let current = load_source(&source, &base, &root, &path, disk_id, physical_slot, &disk_ids).map_err(IoError::other)?;
+            if let Some((original_bytes, directory, anchor_bytes)) = deferred_initial {
+                let snapshot = current
+                    .snapshot
+                    .as_ref()
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                let current_head = current
+                    .head
+                    .as_ref()
+                    .and_then(SourceHead::raw)
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                if !current.head_matches_state
+                    || snapshot.phase != MultipartSourcePhase::Active
+                    || snapshot.capability_sha256.is_some()
+                    || snapshot.lineage != expected_lineage
+                    || snapshot.generation != expected_generation
+                    || snapshot.revision.checked_add(1) != Some(expected_revision)
+                    || current_head != original_bytes.as_slice()
+                {
+                    return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                }
+                SourceStore::write_owned_backup_std(&source, &base, &root, backup_owner, false, &original_bytes, false)?;
+                if let Some(previous) = old_bytes.as_deref() {
+                    SourceStore::write_owned_backup_std(&source, &base, &root, backup_owner, true, previous, false)?;
+                }
+                SourceStore::persist_state_std(&source, &base, &root, &next_state_raw, terminal.as_deref(), false)?;
+                let _parent = os::mkdir_all_below_existing_base_std(&directory, &base, &root)?;
+                os::replace_multipart_source_metadata(&directory.join(RETIREMENT_ANCHOR), &anchor_bytes, &base, &root)?;
+                #[cfg(target_os = "linux")]
+                if can_mark_presynced {
+                    super::mark_presynced_tmp_data_dir(directory);
+                }
+            } else {
+                let current_state = current
+                    .state
+                    .as_ref()
+                    .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+                let current_raw = encode(current_state).map_err(IoError::other)?;
+                if current_raw != expected_state_raw || !current.head_matches_state {
+                    return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                }
+                if let Some(previous) = old_bytes.as_deref() {
+                    SourceStore::write_owned_backup_std(&source, &base, &root, backup_owner, true, previous, false)?;
+                }
+                SourceStore::persist_state_std(&source, &base, &root, &next_state_raw, terminal.as_deref(), false)?;
+            }
+            #[cfg(test)]
+            super::run_owned_file_before_open(&prepared_path);
+            let mut prepared_metadata_source =
+                os::create_prepared_rename_source_with_commit_guard(&prepared_path, &dst_file_path, &owned_directory_guard)?;
+            prepared_metadata_source.write_all(&new_bytes, true)?;
+            os::install_multipart_prepared_head(prepared_metadata_source, &canonical, &base, &root)
+        })
+        .await
+        .map_err(DiskError::from)?;
+        *self.state.lock() = state;
+        *self.validated_prepared.lock() = Some(validated);
+        Ok(prepared)
+    }
+
+    #[cfg(target_os = "linux")]
+    pub async fn prepare_and_install_head(
+        self: &Arc<Self>,
+        new_metadata: &[u8],
+        old_metadata: Option<&[u8]>,
+        dst_file_path: &Path,
+        directory_guard: &os::RenameCommitGuard,
+        mutation_lease: Arc<os::NamespaceMutationLease>,
+    ) -> Result<os::PreparedRenameSource> {
+        let this = Arc::clone(self);
+        let new_bytes = new_metadata.to_vec();
+        let old_bytes = old_metadata.map(|b| b.to_vec());
+        let dst_file_path = dst_file_path.to_path_buf();
+        let owned_directory_guard = directory_guard.clone();
+        os::run_blocking_namespace_operation_with_sync(mutation_lease, true, move || {
+            this.prepare_and_install_head_on_thread(&new_bytes, old_bytes.as_deref(), &dst_file_path, &owned_directory_guard)
+                .map_err(IoError::other)
+        })
+        .await
+        .map_err(DiskError::from)
+    }
+    #[cfg(test)]
     pub async fn install_head(&self, prepared: os::PreparedRenameSource) -> Result<os::PreparedRenameSource> {
         {
             let guard = self.validated_prepared.lock();
@@ -1699,13 +2264,13 @@ impl NativeMultipartPublication {
         // return an error after consuming the last claimable source heads.
         self.store.restore_exact_target_with_lease(&state, lease).await
     }
-    pub async fn finish(&self, target_lease: Arc<os::NamespaceMutationLease>) -> Result<MultipartPublicationResult> {
+    pub(super) fn finish_on_thread(
+        &self,
+        target_already_synced: bool,
+        staging_cleanup: Option<(PathBuf, Arc<std::sync::atomic::AtomicBool>)>,
+    ) -> Result<MultipartPublicationResult> {
         let mut state = self.state.lock().clone();
-        let current = self.store.load().await?;
-        let current_state = current.state.as_ref().ok_or(DiskError::ConditionalDeleteConflict)?;
-        if current.head.is_some() || encode(current_state)? != encode(&state)? {
-            return Err(DiskError::ConditionalDeleteConflict);
-        }
+        let expected_state_raw = encode(&state)?;
         let (expected_sha256, expected_length) = {
             let guard = self.validated_prepared.lock();
             let validated = guard.as_ref().ok_or(DiskError::ConditionalDeleteConflict)?;
@@ -1717,16 +2282,88 @@ impl NativeMultipartPublication {
             }
             (validated.sha256, validated.length)
         };
-        self.store
-            .read_and_sync_target_with_lease(&self.publication, target_lease, move |target| {
-                if target.len() as u64 != expected_length || raw_sha(target) != expected_sha256 {
-                    return Err(DiskError::ConditionalDeleteConflict);
-                }
-                Ok(())
-            })
-            .await?;
+        let target_dir = self.store.target_path(&self.publication)?;
         state.advance(MultipartSourcePhase::ConsumedUnsealed)?;
-        let snapshot = self.store.persist_with_source_sync(&state, true).await?;
+        state.validate(self.store.disk_id, &self.store.path)?;
+        let next_state_raw = encode(&state)?;
+        let snapshot = snapshot_for_state(self.store.disk_id, &state, &next_state_raw, None)?;
+        let source = &self.store.source;
+        let base = &self.store.base;
+        let root = &self.store.publication_root;
+        let path = &self.store.path;
+        let disk_id = self.store.disk_id;
+        let physical_slot = self.store.physical_slot;
+        let disk_ids = &self.store.disk_ids;
+        let terminal = terminal_record_path(source, &state);
+        (|| -> std::io::Result<()> {
+            let current = load_source(source, base, root, path, disk_id, physical_slot, disk_ids).map_err(IoError::other)?;
+            let current_state = current
+                .state
+                .as_ref()
+                .ok_or_else(|| IoError::other(DiskError::ConditionalDeleteConflict))?;
+            let current_raw = encode(current_state).map_err(IoError::other)?;
+            if current.head.is_some() || current_raw != expected_state_raw {
+                return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+            }
+            if target_already_synced {
+                let pinned = os::pin_multipart_source_directory(&target_dir, base, root).map_err(|error| {
+                    if error.kind() == std::io::ErrorKind::NotFound {
+                        IoError::other(DiskError::ConditionalDeleteConflict)
+                    } else {
+                        error
+                    }
+                })?;
+                let target = os::read_multipart_source_metadata_in_pinned(
+                    &pinned,
+                    &target_dir,
+                    std::ffi::OsStr::new(STORAGE_FORMAT_FILE),
+                    base,
+                    root,
+                )
+                .map_err(|error| {
+                    if error.kind() == std::io::ErrorKind::NotFound {
+                        IoError::other(DiskError::ConditionalDeleteConflict)
+                    } else {
+                        error
+                    }
+                })?;
+                if target.len() as u64 != expected_length || raw_sha(&target) != expected_sha256 {
+                    return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                }
+            } else {
+                os::read_and_sync_multipart_target_metadata(
+                    &target_dir,
+                    std::ffi::OsStr::new(STORAGE_FORMAT_FILE),
+                    base,
+                    root,
+                    |target| {
+                        if target.len() as u64 != expected_length || raw_sha(target) != expected_sha256 {
+                            return Err(IoError::other(DiskError::ConditionalDeleteConflict));
+                        }
+                        Ok(())
+                    },
+                )
+                .map_err(|error| {
+                    if error.kind() == std::io::ErrorKind::NotFound {
+                        IoError::other(DiskError::ConditionalDeleteConflict)
+                    } else {
+                        error
+                    }
+                })?;
+            }
+            SourceStore::persist_state_std(source, base, root, &next_state_raw, terminal.as_deref(), true)?;
+            if let Some((io_path, cleaned)) = staging_cleanup {
+                match std::fs::remove_dir(&io_path) {
+                    Ok(()) => cleaned.store(true, std::sync::atomic::Ordering::Release),
+                    Err(err) if err.kind() == std::io::ErrorKind::NotFound => {
+                        cleaned.store(true, std::sync::atomic::Ordering::Release);
+                    }
+                    Err(_) => {}
+                }
+            }
+            Ok(())
+        })()
+        .map_err(DiskError::from)?;
         *self.state.lock() = state;
         let proof = self
             .store
@@ -1738,6 +2375,21 @@ impl NativeMultipartPublication {
             undo_capability: capability,
         })
     }
+
+    pub async fn finish(
+        self: &Arc<Self>,
+        target_lease: Arc<os::NamespaceMutationLease>,
+        target_already_synced: bool,
+        staging_cleanup: Option<(PathBuf, Arc<std::sync::atomic::AtomicBool>)>,
+    ) -> Result<MultipartPublicationResult> {
+        let this = Arc::clone(self);
+        os::run_blocking_namespace_operation_with_sync(target_lease, true, move || {
+            this.finish_on_thread(target_already_synced, staging_cleanup)
+                .map_err(IoError::other)
+        })
+        .await
+        .map_err(DiskError::from)
+    }
 }
 
 impl SourceStore {

crates/ecstore/src/disk/local/multipart_staging.rs

@@ -207,6 +207,12 @@ pub(super) struct MultipartSourceAdmission {
     initial: bool,
 }
 impl MultipartSourceAdmission {
+    pub(super) fn source(&self) -> &Path {
+        &self.source
+    }
+    pub(super) fn is_initial(&self) -> bool {
+        self.initial
+    }
     pub(super) fn precondition(&self) -> Arc<dyn Fn() -> std::io::Result<()> + Send + Sync> {
         let admission = self.clone();
         Arc::new(move || admission.validate())
@@ -250,6 +256,8 @@ pub(super) struct MultipartCommitStaging {
     logical_path: PathBuf,
     io_path: PathBuf,
     owner: Option<Arc<dyn Send + Sync>>,
+    cleaned: Arc<std::sync::atomic::AtomicBool>,
+    fresh_source_head: Option<bytes::Bytes>,
 }
 
 impl MultipartCommitStaging {
@@ -259,10 +267,19 @@ impl MultipartCommitStaging {
     pub(super) fn private_directory(&self, directory: Uuid) -> PathBuf {
         self.io_path.join(directory.to_string())
     }
+    pub(super) fn cleanup_handle(&self) -> (PathBuf, Arc<std::sync::atomic::AtomicBool>) {
+        (self.io_path.clone(), Arc::clone(&self.cleaned))
+    }
+    pub(super) fn fresh_source_head(&self) -> Option<bytes::Bytes> {
+        self.fresh_source_head.clone()
+    }
 }
 
 impl Drop for MultipartCommitStaging {
     fn drop(&mut self) {
+        if self.cleaned.load(std::sync::atomic::Ordering::Acquire) {
+            return;
+        }
         let logical_path = self.logical_path.clone();
         let io_path = self.io_path.clone();
         let owner = self.owner.clone();
@@ -285,6 +302,14 @@ impl Drop for MultipartCommitStaging {
 }
 
 impl LocalDisk {
+    pub(crate) fn has_clean_active_multipart_source(&self, upload_or_part_path: &str, part_path: bool) -> bool {
+        let root = upload_scope_root(upload_or_part_path, part_path);
+        let Ok(source) = self.io_get_object_path(RUSTFS_META_MULTIPART_BUCKET, root) else {
+            return false;
+        };
+        super::multipart_source_state::has_clean_active_multipart_source(&source)
+    }
+
     pub(super) fn multipart_source_admission(
         &self,
         volume: &str,
@@ -356,12 +381,14 @@ impl LocalDisk {
         let io_path = self.io_get_object_path(RUSTFS_META_TMP_BUCKET, &attempt)?;
         // The attempt root is exactly the namespace used by the tmp sweeper.
         let lease = os::acquire_namespace_mutation_lease_with_owner(&logical_path, owner.clone()).await;
-        let staging = MultipartCommitStaging {
+        let mut staging = MultipartCommitStaging {
             path: attempt,
             lease,
             logical_path,
             io_path,
             owner,
+            cleaned: Arc::new(std::sync::atomic::AtomicBool::new(false)),
+            fresh_source_head: None,
         };
         let canonical_metadata = self.io_get_object_path(RUSTFS_META_MULTIPART_BUCKET, &format!("{source_path}/xl.meta"))?;
         let source_object = self.io_get_object_path(RUSTFS_META_MULTIPART_BUCKET, source_path)?;
@@ -371,31 +398,37 @@ impl LocalDisk {
         let source_path = source_path.to_owned();
         let incoming = incoming.clone();
         let sync_directory = effective_durability(destination_volume).syncs_data_shards();
-        os::run_cancellable_blocking_namespace_operation(Arc::clone(&staging.lease), move |cancelled| {
+        let fresh_head = os::run_cancellable_blocking_namespace_operation(Arc::clone(&staging.lease), move |cancelled| {
             if cancelled() {
                 return Err(std::io::Error::new(std::io::ErrorKind::Interrupted, "multipart staging cancelled"));
             }
-            let raw = os::read_multipart_source_metadata(&canonical_metadata, &base, &publication_root).map_err(|error| {
-                // A retired source has no authority for this delayed completion.
-                // Preserve confinement and other I/O failures as their real errors.
-                if error.kind() == std::io::ErrorKind::NotFound {
-                    DiskError::FileNotFound.into()
-                } else {
-                    error
-                }
-            })?;
-            let source = get_file_info(
-                &raw,
-                RUSTFS_META_MULTIPART_BUCKET,
-                &source_path,
-                "",
-                FileInfoOpts {
-                    data: false,
-                    include_free_versions: false,
-                    include_part_checksums: true,
-                },
-            )
-            .map_err(std::io::Error::other)?;
+            let cached_head = super::multipart_source_state::get_active_multipart_source_head(&source_object);
+            let (raw, source, from_active_cache) = if let Some((raw, source)) = cached_head {
+                (raw, source, true)
+            } else {
+                let raw = os::read_multipart_source_metadata(&canonical_metadata, &base, &publication_root).map_err(|error| {
+                    // A retired source has no authority for this delayed completion.
+                    // Preserve confinement and other I/O failures as their real errors.
+                    if error.kind() == std::io::ErrorKind::NotFound {
+                        DiskError::FileNotFound.into()
+                    } else {
+                        error
+                    }
+                })?;
+                let source = get_file_info(
+                    &raw,
+                    RUSTFS_META_MULTIPART_BUCKET,
+                    &source_path,
+                    "",
+                    FileInfoOpts {
+                        data: false,
+                        include_free_versions: false,
+                        include_part_checksums: true,
+                    },
+                )
+                .map_err(std::io::Error::other)?;
+                (bytes::Bytes::from(raw), source, false)
+            };
             let source_dir = source
                 .data_dir
                 .filter(|dir| !dir.is_nil())
@@ -412,23 +445,31 @@ impl LocalDisk {
             let _attempt = os::mkdir_all_below_existing_base_std(&private_object, &base, &publication_root)?;
             if incoming.data.is_none() && !incoming.is_remote() {
                 let source_data_dir = source_object.join(source_dir.to_string());
-                let pinned_source_data = (!incoming.parts.is_empty())
-                    .then(|| os::pin_multipart_source_directory(&source_data_dir, &base, &publication_root))
-                    .transpose()?;
+                let mut pinned_source_data = None;
                 let mut names = Vec::with_capacity(incoming.parts.len().saturating_mul(3));
                 for part in &incoming.parts {
                     if cancelled() {
                         return Err(std::io::Error::new(std::io::ErrorKind::Interrupted, "multipart staging cancelled"));
                     }
                     let meta_name = format!("part.{}.meta", part.number);
-                    let bytes = os::read_multipart_source_metadata_in_pinned(
-                        pinned_source_data.as_ref().expect("pinned source data directory"),
-                        &source_data_dir,
-                        std::ffi::OsStr::new(&meta_name),
-                        &base,
-                        &publication_root,
-                    )?;
-                    let actual = rustfs_filemeta::ObjectPartInfo::unmarshal(&bytes).map_err(std::io::Error::other)?;
+                    let actual = if let Some(cached_part) =
+                        super::multipart_source_state::get_active_multipart_source_part(&source_object, part.number)
+                    {
+                        cached_part
+                    } else {
+                        if pinned_source_data.is_none() {
+                            pinned_source_data =
+                                Some(os::pin_multipart_source_directory(&source_data_dir, &base, &publication_root)?);
+                        }
+                        let bytes = os::read_multipart_source_metadata_in_pinned(
+                            pinned_source_data.as_ref().expect("pinned source data directory"),
+                            &source_data_dir,
+                            std::ffi::OsStr::new(&meta_name),
+                            &base,
+                            &publication_root,
+                        )?;
+                        rustfs_filemeta::ObjectPartInfo::unmarshal(&bytes).map_err(std::io::Error::other)?
+                    };
                     let normalized = i64::try_from(actual.size)
                         .map_err(std::io::Error::other)?
                         .max(actual.actual_size);
@@ -464,10 +505,15 @@ impl LocalDisk {
                     cancelled,
                 )?;
             }
-            Ok(())
+            let has_state = !from_active_cache
+                && (source_object.join(".mpu-source-state").exists()
+                    || source_object.join(".mpu-recovery").exists()
+                    || source_object.join(".mpu-completions").exists());
+            Ok((!has_state).then_some(raw))
         })
         .await
         .map_err(DiskError::from)?;
+        staging.fresh_source_head = fresh_head;
         Ok(staging)
     }
 }
@@ -1219,4 +1265,211 @@ mod tests {
             .expect("non-part empty hash cleanup must remain supported");
         assert!(!directory.exists(), "a per-upload safety guard must not strand empty hash directories");
     }
+
+    #[tokio::test]
+    async fn multipart_source_active_head_cache_populates_and_invalidates_across_part_and_delete_paths() {
+        use crate::disk::{DiskAPI as _, PartTransactionAction, ReadOptions};
+        let (_root, disk, completed, _) = fixture().await;
+        std::fs::create_dir_all(disk.io_get_bucket_path(RUSTFS_META_TMP_BUCKET).expect("TMP path")).expect("TMP bucket");
+        let upload = format!("{}/{}", "c".repeat(64), Uuid::new_v4());
+        let data_dir = Uuid::new_v4();
+        let mut init_fi = FileInfo::new(&upload, 2, 2);
+        init_fi.erasure.index = 1;
+        init_fi.data_dir = Some(data_dir);
+        init_fi.fresh = true;
+        init_fi.mod_time = Some(time::OffsetDateTime::now_utc());
+
+        disk.write_metadata("", RUSTFS_META_MULTIPART_BUCKET, &upload, init_fi.clone())
+            .await
+            .expect("fresh multipart upload metadata");
+        assert!(disk.has_clean_active_multipart_source(&upload, false));
+
+        let read_fi = disk
+            .read_version("", RUSTFS_META_MULTIPART_BUCKET, &upload, "", &ReadOptions::default())
+            .await
+            .expect("cached read_version");
+        assert_eq!(read_fi.data_dir, Some(data_dir));
+
+        let destination = format!("{upload}/{data_dir}/part.1");
+        let part_meta = bytes::Bytes::from(completed.parts[0].marshal_msg().expect("part meta"));
+        disk.write_all(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.1",
+            bytes::Bytes::from_static(b"immutable shard"),
+        )
+        .await
+        .expect("stage tmp part");
+        disk.prepare_part_transaction(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.1",
+            RUSTFS_META_MULTIPART_BUCKET,
+            &destination,
+            part_meta.clone(),
+        )
+        .await
+        .expect("prepare part transaction");
+        assert!(
+            !disk.has_clean_active_multipart_source(&upload, false),
+            "in-flight part transaction must mark active source as having unsettled part txn"
+        );
+        disk.rename_part(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.1",
+            RUSTFS_META_MULTIPART_BUCKET,
+            &destination,
+            part_meta,
+        )
+        .await
+        .expect("rename part");
+        disk.settle_part_transaction(RUSTFS_META_MULTIPART_BUCKET, &destination, PartTransactionAction::Commit)
+            .await
+            .expect("settle part transaction");
+        assert!(
+            disk.has_clean_active_multipart_source(&upload, false),
+            "settled commit must restore clean active multipart source status"
+        );
+        let read_parts = disk
+            .read_parts(RUSTFS_META_MULTIPART_BUCKET, &[format!("{destination}.meta")])
+            .await
+            .expect("cached read_parts");
+        assert_eq!(read_parts.len(), 1);
+        assert_eq!(read_parts[0].number, 1);
+
+        // Prepare part 2, then replay Commit for already-settled part 1: part 2's
+        // pending transaction must keep the upload marked as having unsettled part txns.
+        let destination2 = format!("{upload}/{data_dir}/part.2");
+        let mut part2_info = completed.parts[0].clone();
+        part2_info.number = 2;
+        let part2_meta = bytes::Bytes::from(part2_info.marshal_msg().expect("part 2 meta"));
+        disk.write_all(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.2",
+            bytes::Bytes::from_static(b"immutable shard"),
+        )
+        .await
+        .expect("stage tmp part 2");
+        disk.prepare_part_transaction(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.2",
+            RUSTFS_META_MULTIPART_BUCKET,
+            &destination2,
+            part2_meta.clone(),
+        )
+        .await
+        .expect("prepare part 2 transaction");
+        assert!(!disk.has_clean_active_multipart_source(&upload, false));
+        disk.settle_part_transaction(RUSTFS_META_MULTIPART_BUCKET, &destination, PartTransactionAction::Commit)
+            .await
+            .expect("replayed part 1 commit settlement");
+        assert!(
+            !disk.has_clean_active_multipart_source(&upload, false),
+            "replayed settlement for part 1 must not clear part 2's pending transaction"
+        );
+        disk.rename_part(
+            RUSTFS_META_TMP_BUCKET,
+            "cache-test-part/part.2",
+            RUSTFS_META_MULTIPART_BUCKET,
+            &destination2,
+            part2_meta,
+        )
+        .await
+        .expect("rename part 2");
+        disk.settle_part_transaction(RUSTFS_META_MULTIPART_BUCKET, &destination2, PartTransactionAction::Commit)
+            .await
+            .expect("settle part 2 transaction");
+        assert!(disk.has_clean_active_multipart_source(&upload, false));
+
+        // Deleting {upload}/xl.meta through delete_paths must invalidate the canonical {upload} cache key.
+        disk.delete_paths(RUSTFS_META_MULTIPART_BUCKET, &[format!("{upload}/xl.meta")])
+            .await
+            .expect("delete xl.meta via delete_paths");
+        assert!(
+            !disk.has_clean_active_multipart_source(&upload, false),
+            "per-file head deletion must invalidate the canonical upload cache key"
+        );
+        assert!(matches!(
+            disk.read_version("", RUSTFS_META_MULTIPART_BUCKET, &upload, "", &ReadOptions::default())
+                .await,
+            Err(DiskError::FileNotFound)
+        ));
+    }
+
+    #[cfg(not(windows))]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn multipart_source_cache_publication_holds_source_lock_against_concurrent_retirement() {
+        use crate::disk::{DiskAPI as _, disk_store::LocalDiskWrapper, os::prepared_publication_test_hooks as hooks};
+
+        let (_root, disk, _, _) = fixture().await;
+        std::fs::create_dir_all(disk.io_get_bucket_path(RUSTFS_META_TMP_BUCKET).expect("TMP path")).expect("TMP bucket");
+        let disk = Arc::new(disk);
+        let wrapper = LocalDiskWrapper::new(Arc::clone(&disk), false);
+        let upload = format!("{}/{}", "d".repeat(64), Uuid::new_v4());
+        let data_dir = Uuid::new_v4();
+        let mut init_fi = FileInfo::new(&upload, 2, 2);
+        init_fi.erasure.index = 1;
+        init_fi.data_dir = Some(data_dir);
+        init_fi.fresh = true;
+        init_fi.mod_time = Some(time::OffsetDateTime::now_utc());
+
+        let source_dir = disk
+            .io_get_object_path(RUSTFS_META_MULTIPART_BUCKET, &upload)
+            .expect("source dir");
+        let base_dir = disk
+            .io_get_bucket_path(RUSTFS_META_MULTIPART_BUCKET)
+            .expect("multipart bucket path");
+        let (entered_tx, entered_rx) = tokio::sync::oneshot::channel();
+        let (release_tx, release_rx) = std::sync::mpsc::channel();
+        let _hook = hooks::install_at(hooks::Stage::DirFsync, &source_dir, move || {
+            let _ = entered_tx.send(());
+            let _ = release_rx.recv();
+        });
+
+        let init_wrapper = wrapper.clone();
+        let init_upload = upload.clone();
+        let init_task = tokio::spawn(async move {
+            init_wrapper
+                .write_metadata("", RUSTFS_META_MULTIPART_BUCKET, &init_upload, init_fi)
+                .await
+        });
+
+        tokio::time::timeout(std::time::Duration::from_secs(5), entered_rx)
+            .await
+            .expect("initial write_metadata reaches parent dir fsync")
+            .expect("entered channel");
+
+        let retire_wrapper = wrapper.clone();
+        let retire_path = format!("{upload}/xl.meta");
+        let mut retire_fut = Box::pin(async move {
+            retire_wrapper
+                .delete_paths(RUSTFS_META_MULTIPART_BUCKET, &[retire_path])
+                .await
+        });
+        assert!(
+            futures::poll!(tokio::task::unconstrained(retire_fut.as_mut())).is_pending(),
+            "concurrent retirement must block on SourceRW while initial head publication is in flight"
+        );
+
+        let _ = release_tx.send(());
+        init_task.await.expect("init join").expect("initial write_metadata succeeds");
+        retire_fut
+            .await
+            .expect("concurrent retirement completes after SourceRW release");
+
+        assert!(
+            !super::super::multipart_source_state::has_active_multipart_source_head(&source_dir),
+            "retired source must not remain active in cache after interleaved initial publication"
+        );
+        assert!(
+            matches!(
+                super::super::multipart_source_state::validate_source_mutation(
+                    &source_dir,
+                    &base_dir,
+                    &disk.publication_root,
+                    false
+                ),
+                Err(DiskError::FileNotFound)
+            ),
+            "validate_source_mutation must reject non-initial mutation on retired source"
+        );
+    }
 }

crates/ecstore/src/disk/mod.rs

@@ -106,6 +106,8 @@ pub(crate) struct RenameDataGuards {
     pub(crate) scanner_publication_lease_token: Option<Uuid>,
     pub(crate) external_guard: Option<Arc<dyn Send + Sync>>,
     pub(crate) namespace_owner: Option<Arc<dyn Send + Sync>>,
+    pub(crate) metadata_reservation: Option<os::MetadataMutationReservation>,
+    pub(crate) metadata_reservation_token: Option<Uuid>,
 }
 
 /// Local preflight evidence stays outside DiskAPI and the RPC response format.
@@ -122,6 +124,13 @@ impl RenameDataObservation {
         }
     }
 
+    fn not_dispatched(error: DiskError) -> Self {
+        Self {
+            result: Err(error),
+            preflight_rejection: Some(local::LocalRenamePreflightRejection::not_dispatched()),
+        }
+    }
+
     pub(crate) fn rejected_before_publication(&self) -> bool {
         self.result.is_err() && self.preflight_rejection.is_some()
     }
@@ -311,6 +320,82 @@ pub enum Disk {
 }
 
 impl Disk {
+    /// Prepare an inert single-use reservation on this disk.
+    /// A lost remote response holds its bucket publication fence until expiration, but does not lock the object.
+    pub async fn prepare_object_metadata(
+        &self,
+        bucket: &str,
+        object: &str,
+        incarnation: Uuid,
+        timeout: std::time::Duration,
+    ) -> Result<Uuid> {
+        if timeout.is_zero() {
+            return Err(DiskError::Timeout);
+        }
+        match self {
+            Disk::Local(disk) => tokio::time::timeout(
+                timeout.min(std::time::Duration::from_secs(5)),
+                disk.prepare_object_metadata(bucket, object),
+            )
+            .await
+            .map_err(|_| DiskError::Timeout)?,
+            Disk::Remote(disk) => disk.reserve_object_metadata(bucket, object, incarnation, timeout).await,
+        }
+    }
+
+    /// Refresh an inert token while its sender stages the write.
+    pub async fn renew_object_metadata(
+        &self,
+        bucket: &str,
+        object: &str,
+        token: Uuid,
+        timeout: std::time::Duration,
+    ) -> Result<()> {
+        if timeout.is_zero() {
+            return Err(DiskError::Timeout);
+        }
+        match self {
+            Disk::Local(disk) => tokio::time::timeout(
+                timeout.min(std::time::Duration::from_secs(5)),
+                disk.renew_object_metadata(bucket, object, token),
+            )
+            .await
+            .map_err(|_| DiskError::Timeout)?,
+            Disk::Remote(disk) => disk.renew_object_metadata(bucket, object, token, timeout).await,
+        }
+    }
+
+    /// Enqueue a prepared token in object mutation order.
+    pub async fn activate_object_metadata(
+        &self,
+        bucket: &str,
+        object: &str,
+        token: Uuid,
+        timeout: std::time::Duration,
+    ) -> Result<()> {
+        if timeout.is_zero() {
+            return Err(DiskError::Timeout);
+        }
+        match self {
+            Disk::Local(disk) => tokio::time::timeout(
+                timeout.min(std::time::Duration::from_secs(5)),
+                disk.activate_object_metadata(bucket, object, token),
+            )
+            .await
+            .map_err(|_| DiskError::Timeout)?,
+            Disk::Remote(disk) => disk.activate_object_metadata(bucket, object, token, timeout).await,
+        }
+    }
+
+    /// Cancel an unclaimed reservation. False does not prove that a claim has
+    /// not already started a physical mutation.
+    pub async fn cancel_object_metadata_reservation(&self, token: Uuid) -> Result<bool> {
+        match self {
+            Disk::Local(disk) => Ok(disk.cancel_object_metadata_reservation(token)),
+            Disk::Remote(disk) => disk.cancel_object_metadata_reservation(token).await,
+        }
+    }
+
     /// Physical ownership belongs to the node hosting the local attachment.
     pub fn physical_admission(&self) -> Option<Arc<PhysicalDiskBinding>> {
         match self {
@@ -401,6 +486,13 @@ impl Disk {
         }
     }
 
+    pub(crate) fn has_clean_active_multipart_source(&self, upload_or_part_path: &str, part_path: bool) -> bool {
+        match self {
+            Self::Local(disk) => disk.has_clean_active_multipart_source(upload_or_part_path, part_path),
+            Self::Remote(_) => false,
+        }
+    }
+
     pub(crate) async fn try_publish_prepared_delete_local(
         &self,
         volume: &str,
@@ -660,13 +752,6 @@ impl DiskAPI for Disk {
         }
     }
 
-    async fn bucket_creation_witness(&self, volume: &str) -> Result<Option<os::NativeBucketCreationWitness>> {
-        match self {
-            Disk::Local(local_disk) => local_disk.bucket_creation_witness(volume).await,
-            Disk::Remote(_) => Ok(None),
-        }
-    }
-
     #[tracing::instrument(level = "trace", skip_all)]
     async fn make_volume(&self, volume: &str) -> Result<()> {
         match self {
@@ -1251,18 +1336,33 @@ impl Disk {
                     .rename_data_observed_with_guards(src_volume, src_path, fi, dst_volume, dst_path, guards)
                     .await
             }
-            Disk::Remote(remote_disk) => RenameDataObservation::unknown(
-                remote_disk
-                    .rename_data_borrowed_with_fence(
-                        src_volume,
-                        src_path,
-                        fi,
-                        dst_volume,
-                        dst_path,
-                        guards.scanner_publication_lease_token,
-                    )
-                    .await,
-            ),
+            Disk::Remote(remote_disk) => {
+                if guards.metadata_reservation.is_some() {
+                    return RenameDataObservation::not_dispatched(DiskError::InvalidPath);
+                }
+                if let Some(token) = guards.metadata_reservation_token {
+                    if guards.scanner_publication_lease_token.is_some() {
+                        return RenameDataObservation::not_dispatched(DiskError::InvalidPath);
+                    }
+                    return RenameDataObservation::unknown(
+                        remote_disk
+                            .rename_data_with_metadata_reservation(src_volume, src_path, fi, dst_volume, dst_path, token)
+                            .await,
+                    );
+                }
+                RenameDataObservation::unknown(
+                    remote_disk
+                        .rename_data_borrowed_with_fence(
+                            src_volume,
+                            src_path,
+                            fi,
+                            dst_volume,
+                            dst_path,
+                            guards.scanner_publication_lease_token,
+                        )
+                        .await,
+                )
+            }
         }
     }
 
@@ -1524,11 +1624,6 @@ pub trait DiskAPI: Debug + Send + Sync + 'static {
     // NSScanner
 
     // Volume operations.
-    /// Unknown or remote backends cannot attest process-local publication history.
-    async fn bucket_creation_witness(&self, _volume: &str) -> Result<Option<os::NativeBucketCreationWitness>> {
-        Ok(None)
-    }
-
     async fn make_volume(&self, volume: &str) -> Result<()>;
     async fn make_volumes(&self, volume: Vec<&str>) -> Result<()>;
     async fn list_volumes(&self) -> Result<Vec<VolumeInfo>>;
@@ -2304,7 +2399,7 @@ impl DeleteRecoveryJournal {
             }
             let id = request.version_id.unwrap_or_default();
             let Some(original) = originals.get(&id) else {
-                return !id.is_nil() && !locals.contains_key(&id);
+                return false;
             };
             match locals.get(&id) {
                 Some(local) => local == original,
@@ -2354,10 +2449,7 @@ impl DeleteRecoveryJournal {
             };
             match (local, remote) {
                 (Some(local), Some(remote)) => local == remote,
-                (None, None) => {
-                    (!(request.deleted && request.mark_deleted) && request.version_id.is_some_and(|id| !id.is_nil()))
-                        || (self.original_digest.is_none() && authority.original_digest.is_none())
-                }
+                (None, None) => self.original_digest.is_none() && authority.original_digest.is_none(),
                 _ => false,
             }
         })
@@ -3682,46 +3774,6 @@ mod delete_recovery_adoption_tests {
         };
         assert!(journal.matches_adoption_source(&[]), "certified non-null removal can already be absent");
         assert!(journal.matches_adoption_source(std::slice::from_ref(&source)));
-        let missing = Uuid::new_v4();
-        journal.requests.push(FileInfo {
-            version_id: Some(missing),
-            ..Default::default()
-        });
-        journal.original_digest = Some([3; 32]);
-        assert!(
-            journal.matches_prepared_authority(&journal),
-            "an explicit UUID absent from both prepared sources remains a no-op with original metadata"
-        );
-        assert!(journal.matches_adoption_source(&[]), "both non-null versions can remain absent");
-        assert!(
-            journal.matches_adoption_source(std::slice::from_ref(&source)),
-            "a missing non-null version in the batch remains a no-op"
-        );
-        let mut unexpected = source.clone();
-        unexpected.version_id = Some(missing);
-        let mut foreign_preparation = journal.clone();
-        foreign_preparation.original_versions.push(unexpected.clone());
-        assert!(
-            !journal.matches_prepared_authority(&foreign_preparation)
-                && !foreign_preparation.matches_prepared_authority(&journal),
-            "absence from only one prepared source cannot establish a generation"
-        );
-        assert!(
-            !journal.matches_adoption_source(&[source.clone(), unexpected]),
-            "a version absent from the authority cannot be deleted locally"
-        );
-        for version_id in [None, Some(Uuid::nil())] {
-            journal.requests.last_mut().expect("missing-version request").version_id = version_id;
-            assert!(
-                !journal.matches_prepared_authority(&journal),
-                "a missing null version with original metadata cannot establish its generation"
-            );
-            assert!(
-                !journal.matches_adoption_source(std::slice::from_ref(&source)),
-                "a missing null version cannot establish its generation"
-            );
-        }
-        journal.requests.pop();
         let mut foreign = source.clone();
         foreign.metadata_digest = [2; 32];
         assert!(!journal.matches_adoption_source(std::slice::from_ref(&foreign)));
@@ -3737,25 +3789,6 @@ mod delete_recovery_adoption_tests {
         assert!(journal.matches_adoption_source(std::slice::from_ref(&journal.original_versions[0])));
         journal.requests.clear();
         assert!(!journal.matches_adoption_source(&journal.original_versions));
-
-        let mut marker = journal.clone();
-        marker.requests = vec![FileInfo {
-            version_id: Some(id),
-            deleted: true,
-            mark_deleted: true,
-            ..Default::default()
-        }];
-        marker.original_versions.clear();
-        marker.original_digest = Some([3; 32]);
-        assert!(
-            !marker.matches_prepared_authority(&marker),
-            "a marker with original metadata requires a source generation"
-        );
-        marker.original_digest = None;
-        assert!(
-            marker.matches_prepared_authority(&marker),
-            "marker preparation without original metadata retains the existing allowance"
-        );
     }
 }
 

crates/ecstore/src/error/mod.rs

@@ -17,8 +17,8 @@
 use crate::bucket::error::BucketMetadataError;
 use crate::disk::error::DiskError;
 use crate::storage_api_contracts::{error::StorageErrorCode, range::HTTPRangeError};
-use rustfs_s3_types::S3ErrorCode;
 use rustfs_utils::path::decode_dir_object;
+use s3s::S3ErrorCode;
 
 pub type Error = StorageError;
 pub type Result<T> = core::result::Result<T, Error>;
@@ -1086,8 +1086,14 @@ impl From<rmp_serde::decode::Error> for Error {
     }
 }
 
-impl From<crate::bucket::utils::BucketConfigXmlError> for Error {
-    fn from(e: crate::bucket::utils::BucketConfigXmlError) -> Self {
+impl From<s3s::xml::SerError> for Error {
+    fn from(e: s3s::xml::SerError) -> Self {
+        Error::other(e)
+    }
+}
+
+impl From<s3s::xml::DeError> for Error {
+    fn from(e: s3s::xml::DeError) -> Self {
         Error::other(e)
     }
 }

crates/ecstore/src/io_support/bitrot.rs

@@ -828,13 +828,27 @@ pub async fn create_bitrot_writer(
     length: i64,
     shard_size: usize,
     checksum_algo: HashAlgorithm,
+) -> disk::error::Result<BitrotWriterWrapper> {
+    create_bitrot_writer_with_orig_volume(is_inline_buffer, disk, "", volume, path, length, shard_size, checksum_algo).await
+}
+
+#[allow(clippy::too_many_arguments)]
+pub async fn create_bitrot_writer_with_orig_volume(
+    is_inline_buffer: bool,
+    disk: Option<&DiskStore>,
+    origvolume: &str,
+    volume: &str,
+    path: &str,
+    length: i64,
+    shard_size: usize,
+    checksum_algo: HashAlgorithm,
 ) -> disk::error::Result<BitrotWriterWrapper> {
     let writer = if is_inline_buffer {
         CustomWriter::new_inline_buffer()
     } else if let Some(disk) = disk {
         let length = bitrot_create_file_size(length, shard_size, &checksum_algo);
 
-        let file = disk.create_file("", volume, path, length).await?;
+        let file = disk.create_file(origvolume, volume, path, length).await?;
         #[cfg(feature = "hotpath")]
         let file = instrument_raw_shard_writer(file, disk.is_local());
         CustomWriter::new_tokio_writer(file)

crates/ecstore/src/io_support/shard_integrity.rs

@@ -333,6 +333,34 @@ struct ProofCache {
     preferred: usize,
 }
 
+type VerifiedStripeKey = (uuid::Uuid, [u8; 32], u32);
+
+static VERIFIED_STRIPE_RECORDS: std::sync::LazyLock<parking_lot::Mutex<std::collections::HashMap<VerifiedStripeKey, Bytes>>> =
+    std::sync::LazyLock::new(|| parking_lot::Mutex::new(std::collections::HashMap::new()));
+
+fn get_verified_stripe_record(part: &PartIntegrity, stripe: u32) -> Option<Bytes> {
+    VERIFIED_STRIPE_RECORDS
+        .lock()
+        .get(&(part.generation, part.root, stripe))
+        .cloned()
+}
+
+fn put_verified_stripe_record(part: &PartIntegrity, stripe: u32, record: Bytes) {
+    let mut map = VERIFIED_STRIPE_RECORDS.lock();
+    if map.len() >= 4096
+        && let Some(evict_key) = map.keys().next().copied()
+    {
+        map.remove(&evict_key);
+    }
+    map.insert((part.generation, part.root, stripe), record);
+}
+
+pub(crate) fn invalidate_verified_stripe_records(generation: uuid::Uuid) {
+    VERIFIED_STRIPE_RECORDS
+        .lock()
+        .retain(|(stored_gen, _, _), _| *stored_gen != generation);
+}
+
 impl PartProofReader {
     pub(crate) async fn reconstruction_proof(
         &self,
@@ -410,6 +438,9 @@ impl PartProofReader {
 
     pub(crate) async fn record(&self, stripe: u32) -> io::Result<Bytes> {
         let offset = self.part.record_offset(stripe).map_err(|_| corrupt())?;
+        if let Some(bytes) = get_verified_stripe_record(&self.part, stripe) {
+            return Ok(bytes);
+        }
         let mut cached = self.cached.lock().await;
         if let Some((_, bytes)) = cached.records.iter().find(|(number, _)| *number == stripe) {
             return Ok(bytes.clone());
@@ -465,6 +496,9 @@ impl PartProofReader {
             bytes
         };
         self.part.verify_record(stripe, &record).map_err(|_| corrupt())?;
+        if self.part.stripes <= 4 {
+            put_verified_stripe_record(&self.part, stripe, record.clone());
+        }
         if cached.records.len() == 2 {
             cached.records.pop_front();
         }
@@ -863,4 +897,41 @@ mod tests {
             prepared.part.verify_shard(stripe, 15, b"x", &record).expect("spilled proof");
         }
     }
+
+    #[tokio::test]
+    async fn verified_stripe_record_cache_serves_verified_record_and_still_verifies_payload() {
+        let layout = IntegrityLayout::new(2, 2, 8, false).expect("layout");
+        let mut builder = IntegrityBuilder::new(layout, 1).expect("builder");
+        let shards = [b"ABCD".as_slice(), b"EFGH", b"IJKL", b"MNOP"];
+        builder.push(shards.into_iter()).await.expect("stripe");
+        let prepared = builder.finish(8).await.expect("index");
+        let inline_bytes = prepared.inline_bytes().expect("inline bytes");
+        let part = prepared.part;
+
+        let warm_reader = PartProofReader {
+            part: part.clone(),
+            sources: Vec::new(),
+            volume: "bucket".to_owned(),
+            inline: Some(inline_bytes),
+            cached: Mutex::new(ProofCache::default()),
+        };
+        warm_reader.verify(0, 0, shards[0]).await.expect("warm verify");
+        assert!(
+            get_verified_stripe_record(&part, 0).is_some(),
+            "small verified stripe record should populate the verified stripe cache"
+        );
+
+        let cached_reader = Arc::new(PartProofReader {
+            part,
+            sources: Vec::new(),
+            volume: "bucket".to_owned(),
+            inline: None,
+            cached: Mutex::new(ProofCache::default()),
+        });
+        cached_reader.verify(0, 0, shards[0]).await.expect("cached verify");
+        assert!(
+            cached_reader.verify(0, 0, shards[1]).await.is_err(),
+            "cached record must still reject corrupt or mismatched shard payloads"
+        );
+    }
 }

crates/ecstore/src/object_api/object_api_utils.rs

@@ -20,6 +20,7 @@
 #![allow(clippy::all)]
 
 use http::HeaderMap;
+use s3s::dto::ETag;
 use std::{collections::HashMap, io::Cursor, sync::Arc};
 use tokio::io::BufReader;
 
@@ -28,7 +29,7 @@ use crate::object_api::{GetObjectReader, ObjectInfo, ObjectOptions};
 use crate::storage_api_contracts::range::HTTPRangeSpec;
 use rustfs_filemeta::ObjectPartInfo;
 use rustfs_rio::HashReader;
-use rustfs_s3_types::S3ErrorCode;
+use s3s::S3ErrorCode;
 
 //#[derive(Clone)]
 pub struct PutObjReader {
@@ -149,6 +150,24 @@ pub fn new_getobjectreader<'a>(
     })
 }
 
+/// Convert a raw stored ETag into the strongly-typed `s3s::dto::ETag`.
+///
+/// Supports already quoted (`"abc"`), weak (`W/"abc"`), or plain (`abc`) values.
+pub fn to_s3s_etag(etag: &str) -> ETag {
+    if let Some(rest) = etag.strip_prefix("W/\"") {
+        if let Some(body) = rest.strip_suffix('"') {
+            return ETag::Weak(body.to_string());
+        }
+        return ETag::Weak(rest.to_string());
+    }
+
+    if let Some(body) = etag.strip_prefix('"').and_then(|rest| rest.strip_suffix('"')) {
+        return ETag::Strong(body.to_string());
+    }
+
+    ETag::Strong(etag.to_string())
+}
+
 pub fn get_raw_etag(metadata: &HashMap<String, String>) -> String {
     metadata
         .get("etag")
@@ -213,42 +232,28 @@ mod tests {
     }
 
     #[test]
-    fn test_new_getobjectreader_rejects_a_range_past_the_object_as_invalid_range() {
-        let oi = multipart_object_info();
-        let past_the_end = Some(HTTPRangeSpec {
-            is_suffix_length: false,
-            start: oi.size,
-            end: oi.size + 10,
-        });
+    fn test_to_s3s_etag() {
+        // Test unquoted ETag - should become strong etag
+        assert_eq!(
+            to_s3s_etag("6af8d12c0c74b78094884349f3c8a079"),
+            ETag::Strong("6af8d12c0c74b78094884349f3c8a079".to_string())
+        );
 
-        let err = new_getobjectreader(&past_the_end, &oi, &ObjectOptions::default(), &HeaderMap::new())
-            .err()
-            .expect("a range starting at the object size must be rejected");
+        assert_eq!(
+            to_s3s_etag("\"6af8d12c0c74b78094884349f3c8a079\""),
+            ETag::Strong("6af8d12c0c74b78094884349f3c8a079".to_string())
+        );
 
-        assert_eq!(err.code, S3ErrorCode::InvalidRange);
-        assert_eq!(err.code.as_str(), "InvalidRange");
-        assert_eq!(err.key, None);
-        assert_eq!(err.bucket_name, None);
-    }
+        assert_eq!(
+            to_s3s_etag("W/\"6af8d12c0c74b78094884349f3c8a079\""),
+            ETag::Weak("6af8d12c0c74b78094884349f3c8a079".to_string())
+        );
 
-    #[test]
-    fn test_new_getobjectreader_rejects_an_unknown_size_without_a_range_as_invalid_range() {
-        let oi = ObjectInfo {
-            bucket: "bucket".to_string(),
-            name: "object".to_string(),
-            size: -1,
-            ..Default::default()
-        };
+        assert_eq!(to_s3s_etag(""), ETag::Strong(String::new()));
 
-        let err = new_getobjectreader(&None, &oi, &ObjectOptions::default(), &HeaderMap::new())
-            .err()
-            .expect("an object of unknown size cannot be read without a range");
+        assert_eq!(to_s3s_etag("\"incomplete"), ETag::Strong("\"incomplete".to_string()));
 
-        assert_eq!(err.code, S3ErrorCode::InvalidRange);
-        assert_eq!(err.code.as_str(), "InvalidRange");
-        assert_eq!(err.message, "Invalid range");
-        assert_eq!(err.key.as_deref(), Some("object"));
-        assert_eq!(err.bucket_name.as_deref(), Some("bucket"));
+        assert_eq!(to_s3s_etag("incomplete\""), ETag::Strong("incomplete\"".to_string()));
     }
 
     #[test]

crates/ecstore/src/object_api/types.rs

@@ -160,6 +160,20 @@ pub struct ObjectLockConfigSnapshot {
 }
 
 impl ObjectLockConfigSnapshot {
+    pub(crate) fn owns_lifecycle_guard(&self) -> bool {
+        self._lifecycle_guard.is_some()
+    }
+
+    pub(crate) fn owns_lifecycle_guard_for(&self, bucket: &str, incarnation: Option<Uuid>) -> bool {
+        self.bucket.as_deref() == Some(bucket)
+            && incarnation.is_some()
+            && self.bucket_incarnation_id == incarnation
+            && self
+                ._lifecycle_guard
+                .as_ref()
+                .is_some_and(|guard| !guard.is_lock_lost() && !guard.is_released())
+    }
+
     pub(crate) fn new(state: crate::bucket::metadata_sys::ObjectLockConfigState) -> Self {
         Self {
             store_id: None,
@@ -1037,10 +1051,6 @@ pub struct ObjectOptions {
     /// `ecstore` must leave it unset.
     #[doc(hidden)]
     pub data_movement_transferred_tier_versions: Option<crate::data_movement::TransferredTierVersions>,
-    /// Marks the physical copy-back write of a RestoreObject operation. This
-    /// is in-memory request context and must not be copied into object metadata.
-    #[doc(hidden)]
-    pub restore_copyback: bool,
     /// Internal, in-memory protection context. Never populated from S3 metadata.
     /// None selects the rollout default only for a new write, not for a rewrite.
     #[doc(hidden)]
@@ -1547,6 +1557,8 @@ pub struct ObjectInfo {
     /// published by the same upload. Callers must not repeat quota, replication,
     /// lifecycle, or object-created side effects for that response.
     pub multipart_completion_replayed: bool,
+    /// Previous current version's size observed during `rename_data` commit.
+    pub backfilled_old_current_size: Option<crate::disk::OldCurrentSize>,
 }
 
 impl Default for ObjectInfo {
@@ -1589,6 +1601,7 @@ impl Default for ObjectInfo {
             replication_decision: String::new(),
             checksum: None,
             multipart_completion_replayed: false,
+            backfilled_old_current_size: None,
         }
     }
 }
@@ -1632,6 +1645,7 @@ impl Clone for ObjectInfo {
             replication_decision: self.replication_decision.clone(),
             checksum: self.checksum.clone(),
             multipart_completion_replayed: self.multipart_completion_replayed,
+            backfilled_old_current_size: self.backfilled_old_current_size,
             expires: self.expires,
         }
     }

crates/ecstore/src/runtime/global.rs

@@ -275,16 +275,16 @@ pub(crate) type TypeLocalDiskSetDrives = Vec<Vec<Vec<Option<DiskStore>>>>;
 ///
 /// # Returns
 /// * None
-pub fn set_global_region(region: rustfs_s3_types::Region) {
+pub fn set_global_region(region: s3s::region::Region) {
     current_ctx().set_region(region);
 }
 
 /// Get the global region
 ///
 /// # Returns
-/// * `Option<rustfs_s3_types::Region>` - The global region, if set
+/// * `Option<s3s::region::Region>` - The global region, if set
 ///
-pub fn get_global_region() -> Option<rustfs_s3_types::Region> {
+pub fn get_global_region() -> Option<s3s::region::Region> {
     current_ctx().region()
 }
 

crates/ecstore/src/runtime/instance.rs

@@ -52,7 +52,7 @@ use crate::services::event_notification::EventNotifier;
 use crate::services::tier::tier::TierConfigMgr;
 use crate::store::delete_recovery_runtime::DeleteRecoveryHints;
 use rustfs_lock::{GlobalLockManager, get_global_lock_manager};
-use rustfs_s3_types::Region;
+use s3s::region::Region;
 use std::collections::{HashMap, HashSet};
 use std::sync::{
     Arc, OnceLock,
@@ -101,6 +101,12 @@ pub(crate) struct NamespaceCommitGuard {
     counted: bool,
 }
 
+impl NamespaceCommitGuard {
+    pub(crate) fn covers_instance(&self, ctx: &Arc<InstanceContext>) -> bool {
+        Arc::ptr_eq(&self.ctx, ctx)
+    }
+}
+
 impl Drop for NamespaceCommitGuard {
     fn drop(&mut self) {
         if self.counted {
@@ -941,6 +947,32 @@ impl InstanceContext {
             .clone()
     }
 
+    /// Remove the `BucketMetadataSys` when its backing `ECStore` is gone.
+    ///
+    /// Called from `ECStore::drop`. When test fixtures share the bootstrap
+    /// context, the `BucketMetadataSys` stores a `Weak<ECStore>` to prevent a
+    /// reference cycle; once the store drops the weak is dangling and any
+    /// `object_store()` call panics. Clearing the slot here lets the next
+    /// fixture install a fresh `BucketMetadataSys` without hitting the stale one.
+    pub fn clear_bucket_metadata_sys_if_stale(&self) {
+        let mut guard = self.bucket_metadata_sys.lock().expect("bucket_metadata_sys lock poisoned");
+        if let Some(sys) = guard.as_ref() {
+            // `read()` may block briefly; the async RwLock is acceptable here
+            // because Drop is sync and the lock is held only long enough to
+            // check the weak reference.
+            let is_stale = {
+                let s = sys.try_read();
+                match s {
+                    Ok(s) => s.is_store_stale(),
+                    Err(_) => true, // contended: treat as stale-safe to clear
+                }
+            };
+            if is_stale {
+                *guard = None;
+            }
+        }
+    }
+
     /// Set this instance's background-services cancellation token (once).
     pub fn init_background_cancel_token(&self, token: CancellationToken) -> Result<(), CancellationToken> {
         self.background_cancel_token.set(token)
@@ -1125,6 +1157,8 @@ mod tests {
         other.set_scanner_publication_state(false);
         assert!(first.scanner_publication_state_allowed());
         let one = first.begin_namespace_commit();
+        assert!(one.covers_instance(&first));
+        assert!(!one.covers_instance(&other));
         let shared_owner = Arc::clone(&one);
         let two = first.begin_namespace_commit();
         assert!(first.namespace_commits_pending());

crates/ecstore/src/runtime/sources.rs

@@ -44,10 +44,10 @@ use crate::{
 };
 use rustfs_concurrency::WorkloadAdmissionSnapshotProvider;
 use rustfs_config::server_config::{Config, get_global_server_config, set_global_server_config};
-use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
 use rustfs_io_metrics::internode_metrics::global_internode_metrics;
 use rustfs_lock::client::LockClient;
-use rustfs_s3_types::Region;
+use s3s::dto::BucketLifecycleConfiguration;
+use s3s::region::Region;
 use tokio::sync::{OwnedRwLockReadGuard, RwLock};
 use tokio_util::sync::CancellationToken;
 #[cfg(test)]
@@ -371,7 +371,7 @@ pub fn global_tier_config_mgr() -> Arc<RwLock<TierConfigMgr>> {
     get_global_tier_config_mgr()
 }
 
-pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<PersistedLifecycleConfiguration> {
+pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<BucketLifecycleConfiguration> {
     GLOBAL_LIFECYCLE_SYS.get(bucket).await
 }
 

crates/ecstore/src/services/notification_sys.rs

@@ -703,28 +703,6 @@ pub async fn acquire_legacy_transition_state_reconcile_fleet_proof() -> Option<L
     .then_some(proof)
 }
 
-pub async fn require_restore_copyback_fleet_capability() -> Result<()> {
-    let Some(topology) = REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get() else {
-        #[cfg(any(test, feature = "test-util"))]
-        return Ok(());
-        #[cfg(not(any(test, feature = "test-util")))]
-        return Err(Error::other("restore copy-back capability topology is unavailable"));
-    };
-    let Some(notification_sys) = get_global_notification_sys() else {
-        #[cfg(any(test, feature = "test-util"))]
-        return Ok(());
-        #[cfg(not(any(test, feature = "test-util")))]
-        return Err(Error::other("restore copy-back capability probe is unavailable"));
-    };
-    timeout(
-        REMOTE_VERSION_STATE_PROBE_TIMEOUT,
-        notification_sys.probe_restore_copyback_fleet(topology),
-    )
-    .await
-    .map_err(|_| Error::other("restore copy-back capability fleet probe timed out"))??;
-    Ok(())
-}
-
 fn acquire_legacy_transition_state_reconcile_fleet_proof_from(
     state: &FleetCapabilityProofState,
     expected_topology: &str,
@@ -1782,30 +1760,6 @@ impl NotificationSys {
         Ok(peer_epochs)
     }
 
-    async fn probe_restore_copyback_fleet(&self, topology_fingerprint: &str) -> Result<BTreeMap<String, Uuid>> {
-        if self.peer_clients.len() != self.peer_topology_hosts.len() {
-            return Err(Error::other("restore copy-back capability fleet membership is incomplete"));
-        }
-        let probes = self.peer_clients.iter().map(|client| async {
-            let client = client
-                .as_ref()
-                .ok_or_else(|| Error::other("restore copy-back capability peer is unreachable"))?;
-            client
-                .probe_restore_copyback_capability(topology_fingerprint.to_string())
-                .await
-        });
-        let mut peer_epochs = BTreeMap::new();
-        for result in join_all(probes).await {
-            let (peer, epoch) = result?;
-            insert_remote_version_state_peer(&mut peer_epochs, peer, epoch)?;
-        }
-        let expected = self.peer_topology_hosts.iter().cloned().collect::<BTreeSet<_>>();
-        if expected.len() != self.peer_topology_hosts.len() || peer_epochs.keys().ne(expected.iter()) {
-            return Err(Error::other("restore copy-back capability fleet membership does not match topology"));
-        }
-        Ok(peer_epochs)
-    }
-
     async fn probe_transition_transaction_compaction_fleet(&self, topology_fingerprint: &str) -> Result<BTreeMap<String, Uuid>> {
         if self.peer_clients.len() != self.peer_topology_hosts.len() {
             return Err(Error::other(

crates/ecstore/src/services/rebalance/entry.rs

@@ -1032,9 +1032,8 @@ mod tests {
     use crate::storage_api_contracts::object::ObjectIO as _;
     use http::HeaderMap;
     use rustfs_filemeta::{FileInfo, FileMeta, ObjectPartInfo, TransitionVersionState};
-    use rustfs_gateway_types::persistence::{
-        PersistedLifecycleConfiguration, PersistedLifecycleExpiration, PersistedLifecycleRule,
-        PersistedNoncurrentVersionExpiration,
+    use s3s::dto::{
+        BucketLifecycleConfiguration, ExpirationStatus, LifecycleExpiration, LifecycleRule, NoncurrentVersionExpiration,
     };
     use std::time::Duration as StdDuration;
     use time::OffsetDateTime;
@@ -1078,12 +1077,12 @@ mod tests {
         }
     }
 
-    fn expired_delete_marker_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn expired_delete_marker_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     expired_object_delete_marker: Some(true),
                     ..Default::default()
                 }),
@@ -2111,11 +2110,11 @@ mod tests {
         ))
         .await
         .expect("seed the expired source version");
-        let lifecycle = PersistedLifecycleConfiguration {
+        let lifecycle = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
-                expiration: Some(PersistedLifecycleExpiration {
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
                     days: Some(1),
                     expired_object_all_versions: Some(true),
                     ..Default::default()
@@ -2364,7 +2363,7 @@ mod tests {
                         versioned,
                         version_id,
                         transition: crate::bucket::lifecycle::lifecycle::TransitionOptions {
-                            restore_request: rustfs_lifecycle::RestoreRequest {
+                            restore_request: s3s::dto::RestoreRequest {
                                 days: Some(1),
                                 ..Default::default()
                             },
@@ -2444,7 +2443,7 @@ mod tests {
         async fn run_entry_with(
             &self,
             entry: MetaCacheEntry,
-            lifecycle_config: Option<PersistedLifecycleConfiguration>,
+            lifecycle_config: Option<BucketLifecycleConfiguration>,
         ) -> RebalanceEntryOutcome {
             Box::pin(Arc::clone(&self.store).rebalance_entry(
                 RebalanceEntryTarget {
@@ -2780,17 +2779,17 @@ mod tests {
     }
 
     /// A rule that expires every noncurrent version at once.
-    fn noncurrent_expire_all_lifecycle() -> PersistedLifecycleConfiguration {
-        PersistedLifecycleConfiguration {
+    fn noncurrent_expire_all_lifecycle() -> BucketLifecycleConfiguration {
+        BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("noncurrent-expire-all".to_string()),
-                noncurrent_version_expiration: Some(PersistedNoncurrentVersionExpiration {
+                noncurrent_version_expiration: Some(NoncurrentVersionExpiration {
                     noncurrent_days: Some(0),
                     newer_noncurrent_versions: None,
                 }),

crates/ecstore/src/services/rebalance/rebalance_unit_tests.rs

@@ -67,8 +67,8 @@ use crate::error::{Error, Result};
 use crate::storage_api_contracts::range::HTTPRangeSpec;
 use rustfs_filemeta::TRANSITION_COMPLETE;
 use rustfs_filemeta::{FileInfo, MetaCacheEntry};
-use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 use rustfs_rio::Index;
+use s3s::dto::ReplicationConfiguration;
 use serde::Serialize;
 use std::io::Cursor;
 use std::sync::Arc;
@@ -2676,7 +2676,7 @@ fn test_resolve_rebalance_optional_bucket_config_result_passthrough() {
     let result = resolve_rebalance_optional_bucket_config_result(
         "bucket-a",
         "replication",
-        Ok((PersistedReplicationConfiguration::default(), OffsetDateTime::UNIX_EPOCH)),
+        Ok((ReplicationConfiguration::default(), OffsetDateTime::UNIX_EPOCH)),
     )
     .expect("bucket config should pass through");
     assert!(result.is_some());

crates/ecstore/src/services/rebalance/types.rs

@@ -34,9 +34,9 @@ pub struct RebalanceStats {
 #[derive(Debug, Default)]
 pub(super) struct RebalanceBucketConfigs {
     pub(super) bucket_incarnation_id: Option<uuid::Uuid>,
-    pub(super) lifecycle_config: Option<rustfs_gateway_types::persistence::PersistedLifecycleConfiguration>,
-    pub(super) object_lock_config: Option<rustfs_gateway_types::persistence::PersistedObjectLockConfiguration>,
-    pub(super) replication_config: Option<(rustfs_gateway_types::persistence::PersistedReplicationConfiguration, OffsetDateTime)>,
+    pub(super) lifecycle_config: Option<s3s::dto::BucketLifecycleConfiguration>,
+    pub(super) object_lock_config: Option<s3s::dto::ObjectLockConfiguration>,
+    pub(super) replication_config: Option<(s3s::dto::ReplicationConfiguration, OffsetDateTime)>,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq)]

crates/ecstore/src/services/tier/tier.rs

@@ -80,11 +80,10 @@ use crate::{
     store::ECStore,
 };
 use rustfs_filemeta::FileInfo;
-use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
 use rustfs_rio::HashReader;
 use rustfs_s3_client::{admin_handler_utils::AdminError, provider_versions::ProviderVersionCapabilities};
-use rustfs_s3_types::S3ErrorCode;
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join};
+use s3s::{S3ErrorCode, dto::BucketLifecycleConfiguration};
 
 use super::{
     tier_handlers::{ERR_TIER_BUCKET_NOT_FOUND, ERR_TIER_CONNECT_ERR, ERR_TIER_INVALID_CREDENTIALS, ERR_TIER_PERM_ERR},
@@ -1103,7 +1102,7 @@ trait TierReferenceProofStore:
         .await
     }
 
-    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>>;
+    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>>;
 
     async fn walk_bucket_for_deletion_check(
         self: Arc<Self>,
@@ -1156,7 +1155,7 @@ impl TierReferenceProofStore for ECStore {
         Ok(buckets.into_values().collect())
     }
 
-    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
+    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
         match get_lifecycle_config(bucket).await {
             Ok((config, _updated_at)) => Ok(Some(config)),
             Err(Error::ConfigNotFound) => Ok(None),
@@ -1635,7 +1634,7 @@ struct TierLifecycleReference<'a> {
 }
 
 fn lifecycle_config_target_reference<'a>(
-    config: &'a PersistedLifecycleConfiguration,
+    config: &'a BucketLifecycleConfiguration,
     targets: &[TierMutationIntentTarget],
 ) -> Option<TierLifecycleReference<'a>> {
     for rule in &config.rules {
@@ -7262,8 +7261,9 @@ mod tests {
         endpoints::{Endpoints, PoolEndpoints, SetupType},
     };
     use crate::services::tier::tier_mutation_intent::TIER_MUTATION_INTENT_RECORD_PREFIX;
-    use rustfs_gateway_types::persistence::{
-        PersistedLifecycleConfiguration, PersistedLifecycleRule, PersistedNoncurrentVersionTransition, PersistedTransition,
+    use s3s::dto::{
+        BucketLifecycleConfiguration, ExpirationStatus, LifecycleRule, NoncurrentVersionTransition, Transition,
+        TransitionStorageClass,
     };
 
     #[test]
@@ -8090,7 +8090,7 @@ mod tests {
 
     #[async_trait::async_trait]
     impl TierReferenceProofStore for LockingTierConfigStore {
-        async fn lifecycle_config_for_reference_proof(&self, _bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
+        async fn lifecycle_config_for_reference_proof(&self, _bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
             Ok(None)
         }
     }
@@ -10730,42 +10730,6 @@ mod tests {
         assert_eq!(observed.committed_config_etag.as_deref(), Some("etag-new"));
     }
 
-    #[tokio::test]
-    async fn intent_advance_retries_cas_conflict_reload_lock_timeout() {
-        let store = Arc::new(CasConfigStore::default());
-        let mutation_id = uuid::Uuid::from_u128(39);
-        let prepared = prepared_remove_intent("COLD-A", mutation_id);
-        crate::services::tier::tier_mutation_intent::save_tier_mutation_intent_record(store.clone(), &prepared)
-            .await
-            .expect("prepared intent fixture should persist");
-
-        let committed = committed_remove_intent("COLD-A", mutation_id, "etag-new");
-        let committed_bytes = committed.encode().expect("committed intent fixture should encode");
-        let object = crate::services::tier::tier_mutation_intent::tier_mutation_intent_record_object_name(mutation_id)
-            .expect("record object should build");
-        let winning_etag = format!("etag-{}", store.next_etag.load(Ordering::SeqCst) + 1);
-        store.rewrite_on_next_if_match(object.clone(), committed_bytes.clone()).await;
-        *store.read_timeout_after_if_match_race.lock().await = Some(object.clone());
-
-        let (observed, applied) = crate::services::tier::tier_mutation_intent::advance_tier_mutation_intent_record_idempotent(
-            store.clone(),
-            mutation_id,
-            TierMutationIntentState::Committed,
-            Some("etag-new".to_string()),
-        )
-        .await
-        .expect("CAS conflict reload should retry its typed lock timeout");
-
-        assert!(!applied);
-        assert_eq!(observed, committed);
-        assert!(store.read_timeout_after_if_match_race.lock().await.is_none());
-        assert_eq!(
-            store.objects.lock().await.get(&object).cloned(),
-            Some((committed_bytes, winning_etag)),
-            "retry must preserve the complete winning record and its object ETag"
-        );
-    }
-
     #[tokio::test]
     async fn intent_advance_reconciles_matching_commit_after_final_cas_race() {
         let store = Arc::new(CasConfigStore::default());
@@ -15266,7 +15230,6 @@ mod tests {
         objects: tokio::sync::Mutex<HashMap<String, (Vec<u8>, String)>>,
         legacy_state: tokio::sync::Mutex<Option<Vec<u8>>>,
         if_match_race_rewrite: tokio::sync::Mutex<std::collections::VecDeque<(String, Vec<u8>)>>,
-        read_timeout_after_if_match_race: tokio::sync::Mutex<Option<String>>,
         next_etag: AtomicUsize,
         content_derived_etags: AtomicBool,
         fail_put: AtomicBool,
@@ -15285,7 +15248,7 @@ mod tests {
         lock_manager: Arc<rustfs_lock::GlobalLockManager>,
         lock_requests: Mutex<Vec<(String, String)>>,
         listed_versions: Mutex<Vec<ObjectInfo>>,
-        lifecycle_configs: Mutex<HashMap<String, PersistedLifecycleConfiguration>>,
+        lifecycle_configs: Mutex<HashMap<String, BucketLifecycleConfiguration>>,
     }
 
     impl Default for CasConfigStore {
@@ -15296,7 +15259,6 @@ mod tests {
                 objects: tokio::sync::Mutex::new(HashMap::new()),
                 legacy_state: tokio::sync::Mutex::new(None),
                 if_match_race_rewrite: tokio::sync::Mutex::new(std::collections::VecDeque::new()),
-                read_timeout_after_if_match_race: tokio::sync::Mutex::new(None),
                 next_etag: AtomicUsize::new(0),
                 content_derived_etags: AtomicBool::new(false),
                 fail_put: AtomicBool::new(false),
@@ -15349,7 +15311,7 @@ mod tests {
                 .retain(|version| version.bucket != bucket || version.name != object);
         }
 
-        fn add_lifecycle_config(&self, bucket: &str, config: PersistedLifecycleConfiguration) {
+        fn add_lifecycle_config(&self, bucket: &str, config: BucketLifecycleConfiguration) {
             self.lifecycle_configs
                 .lock()
                 .expect("tier reference fixture should not poison")
@@ -15458,19 +15420,6 @@ mod tests {
                 crate::api::diagnostic::current_read_version_entry_source(),
                 rustfs_io_metrics::s3_http_metrics::current_s3_http_operation_name(),
             ));
-            {
-                // The queued rewrite keeps the initial read successful. Only
-                // the first target read after that CAS race consumes the fault.
-                let race_completed = self.if_match_race_rewrite.lock().await.is_empty();
-                let mut timeout = self.read_timeout_after_if_match_race.lock().await;
-                if race_completed && timeout.as_deref() == Some(object) {
-                    timeout.take();
-                    return Err(Error::Lock(rustfs_lock::LockError::Timeout {
-                        resource: format!("{bucket}/{object}@latest"),
-                        timeout: std::time::Duration::from_secs(5),
-                    }));
-                }
-            }
             let legacy_config_path = tier_config_path(TIER_CONFIG_LEGACY_FILE);
             let (data, etag) = if object == legacy_config_path {
                 let state = self.legacy_state.lock().await;
@@ -16099,7 +16048,7 @@ mod tests {
             .await
         }
 
-        async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
+        async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
             if self.reference_lifecycle_error.load(Ordering::SeqCst) {
                 return Err(Error::CorruptedFormat);
             }
@@ -16234,10 +16183,10 @@ mod tests {
         });
         store.add_lifecycle_config(
             "photos",
-            PersistedLifecycleConfiguration {
+            BucketLifecycleConfiguration {
                 expiry_updated_at: None,
-                rules: vec![PersistedLifecycleRule {
-                    status: "Enabled".to_owned(),
+                rules: vec![LifecycleRule {
+                    status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                     expiration: None,
                     abort_incomplete_multipart_upload: None,
                     del_marker_expiration: None,
@@ -16246,10 +16195,10 @@ mod tests {
                     noncurrent_version_expiration: None,
                     noncurrent_version_transitions: None,
                     prefix: None,
-                    transitions: Some(vec![PersistedTransition {
+                    transitions: Some(vec![Transition {
                         days: Some(1),
                         date: None,
-                        storage_class: Some("COLD-A".to_owned()),
+                        storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                     }]),
                 }],
             },
@@ -16902,10 +16851,10 @@ mod tests {
             .save_tiering_config_if_current(store.clone(), None)
             .await
             .expect("reference proof fixture should persist");
-        let config = PersistedLifecycleConfiguration {
+        let config = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -16914,10 +16863,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![PersistedTransition {
+                transitions: Some(vec![Transition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some("COLD-A".to_owned()),
+                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                 }]),
             }],
         };
@@ -17347,23 +17296,23 @@ mod tests {
             });
             store.add_lifecycle_config(
                 "private-bucket",
-                rustfs_gateway_types::persistence::PersistedLifecycleConfiguration {
+                BucketLifecycleConfiguration {
                     expiry_updated_at: None,
-                    rules: vec![rustfs_gateway_types::persistence::PersistedLifecycleRule {
-                        status: "Enabled".to_string(),
+                    rules: vec![LifecycleRule {
+                        status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                         id: Some("private-rule".to_string()),
                         noncurrent_version_transitions: noncurrent.then(|| {
-                            vec![rustfs_gateway_types::persistence::PersistedNoncurrentVersionTransition {
+                            vec![NoncurrentVersionTransition {
                                 noncurrent_days: Some(1),
                                 newer_noncurrent_versions: None,
-                                storage_class: Some("COLD-A".to_string()),
+                                storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                             }]
                         }),
                         transitions: (!noncurrent).then(|| {
-                            vec![rustfs_gateway_types::persistence::PersistedTransition {
+                            vec![Transition {
                                 days: Some(1),
                                 date: None,
-                                storage_class: Some("COLD-A".to_string()),
+                                storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                             }]
                         }),
                         expiration: None,
@@ -17766,10 +17715,10 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = PersistedLifecycleConfiguration {
+        let config = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -17778,10 +17727,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![PersistedTransition {
+                transitions: Some(vec![Transition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some("COLD-A".to_owned()),
+                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                 }]),
             }],
         };
@@ -17811,20 +17760,20 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = PersistedLifecycleConfiguration {
+        let config = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("move-noncurrent".to_string()),
                 noncurrent_version_expiration: None,
-                noncurrent_version_transitions: Some(vec![PersistedNoncurrentVersionTransition {
+                noncurrent_version_transitions: Some(vec![NoncurrentVersionTransition {
                     noncurrent_days: Some(1),
                     newer_noncurrent_versions: None,
-                    storage_class: Some("COLD-A".to_owned()),
+                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                 }]),
                 prefix: None,
                 transitions: None,
@@ -17860,10 +17809,10 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = PersistedLifecycleConfiguration {
+        let config = BucketLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![PersistedLifecycleRule {
-                status: "Enabled".to_owned(),
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -17872,10 +17821,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![PersistedTransition {
+                transitions: Some(vec![Transition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some("COLD-A".to_owned()),
+                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
                 }]),
             }],
         };

crates/ecstore/src/services/tier/tier_mutation_intent.rs

@@ -700,25 +700,9 @@ async fn advance_tier_mutation_intent_record_idempotent_at_prefix<S>(
 where
     S: EcstoreObjectIO,
 {
-    let mut original_intent: Option<TierMutationIntent> = None;
     for attempt in 0..TIER_MUTATION_INTENT_ADVANCE_CAS_ATTEMPTS {
         let (mut intent, current_etag) =
-            match load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await {
-                Ok(current) => current,
-                Err(Error::Lock(rustfs_lock::LockError::Timeout { .. }))
-                    if attempt + 1 < TIER_MUTATION_INTENT_ADVANCE_CAS_ATTEMPTS =>
-                {
-                    continue;
-                }
-                Err(err) => return Err(err),
-            };
-        if let Some(original) = &original_intent {
-            if !intent.same_identity_as(original) {
-                return Err(Error::PreconditionFailed);
-            }
-        } else {
-            original_intent = Some(intent.clone());
-        }
+            load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await?;
         let advanced = intent
             .advance_idempotent(next, committed_config_etag.clone())
             .map_err(tier_mutation_intent_store_error)?;
@@ -729,15 +713,7 @@ where
             Ok(()) => return Ok((intent, true)),
             Err(Error::PreconditionFailed) => {
                 let (mut current, _) =
-                    match load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await {
-                        Ok(current) => current,
-                        Err(Error::Lock(rustfs_lock::LockError::Timeout { .. }))
-                            if attempt + 1 < TIER_MUTATION_INTENT_ADVANCE_CAS_ATTEMPTS =>
-                        {
-                            continue;
-                        }
-                        Err(err) => return Err(err),
-                    };
+                    load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await?;
                 if !current.same_identity_as(&intent) {
                     return Err(Error::PreconditionFailed);
                 }
@@ -752,11 +728,6 @@ where
                 }
                 return Ok((current, false));
             }
-            Err(Error::Lock(rustfs_lock::LockError::Timeout { .. }))
-                if attempt + 1 < TIER_MUTATION_INTENT_ADVANCE_CAS_ATTEMPTS =>
-            {
-                continue;
-            }
             Err(err) => return Err(err),
         }
     }

crates/ecstore/src/services/tier/warm_backend.rs

@@ -36,9 +36,7 @@ use crate::services::tier::{
 use bytes::Bytes;
 use http::StatusCode;
 use rustfs_filemeta::metadata_keys;
-use rustfs_gateway_types::dto::{ObjectLockLegalHoldStatus, ObjectLockMode, ReplicationStatus};
 use rustfs_s3_client::credentials::{Credentials, SignatureType, Static, Value};
-use rustfs_s3_client::remote_error_code::RemoteErrorCode;
 use rustfs_s3_client::transition_api::{BucketLookupType, Options, TransitionClient, TransitionClientTimeouts, TransitionCore};
 use rustfs_s3_client::{
     admin_handler_utils::AdminError,
@@ -51,6 +49,10 @@ use rustfs_utils::egress::validate_outbound_url;
 use rustfs_utils::http::headers::{
     CACHE_CONTROL, CONTENT_DISPOSITION, CONTENT_ENCODING, CONTENT_LANGUAGE, CONTENT_TYPE, EXPIRES, HeaderExt as _,
 };
+use s3s::{
+    S3ErrorCode,
+    dto::{ObjectLockLegalHoldStatus, ObjectLockRetentionMode, ReplicationStatus},
+};
 use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Duration;
@@ -191,15 +193,12 @@ pub trait WarmBackend {
             .await
         {
             Ok(_) => Ok(TransitionCandidateProbe::VersionedPresent(remote_version_id.to_string())),
-            Err(err) if matches!(to_error_response(&err).code, RemoteErrorCode::INVALID_RANGE) => {
+            Err(err) if matches!(to_error_response(&err).code, S3ErrorCode::InvalidRange) => {
                 Ok(TransitionCandidateProbe::VersionedPresent(remote_version_id.to_string()))
             }
             Err(err)
                 if err.kind() == std::io::ErrorKind::NotFound
-                    || matches!(
-                        to_error_response(&err).code,
-                        RemoteErrorCode::NO_SUCH_KEY | RemoteErrorCode::NO_SUCH_VERSION
-                    ) =>
+                    || matches!(to_error_response(&err).code, S3ErrorCode::NoSuchKey | S3ErrorCode::NoSuchVersion) =>
             {
                 Ok(TransitionCandidateProbe::Missing)
             }
@@ -219,9 +218,9 @@ pub fn build_transition_put_options(storage_class: String, mut metadata: HashMap
     let mut opts = PutObjectOptions {
         storage_class,
         send_content_md5: true,
-        legalhold: ObjectLockLegalHoldStatus::custom(""),
+        legalhold: ObjectLockLegalHoldStatus::from_static(""),
         internal: AdvancedPutOptions {
-            replication_status: ReplicationStatus::custom(""),
+            replication_status: ReplicationStatus::from_static(""),
             ..Default::default()
         },
         ..Default::default()
@@ -252,7 +251,7 @@ pub fn build_transition_put_options(storage_class: String, mut metadata: HashMap
     }
 
     if let Some(mode) = metadata.lookup(metadata_keys::OBJECT_LOCK_MODE) {
-        opts.mode = ObjectLockMode::custom(mode.to_ascii_uppercase());
+        opts.mode = ObjectLockRetentionMode::from(mode.to_ascii_uppercase());
     }
 
     if let Some(retain_until_date) = metadata
@@ -263,7 +262,7 @@ pub fn build_transition_put_options(storage_class: String, mut metadata: HashMap
     }
 
     if let Some(legalhold) = metadata.lookup(metadata_keys::OBJECT_LOCK_LEGAL_HOLD) {
-        opts.legalhold = ObjectLockLegalHoldStatus::custom(legalhold.to_ascii_uppercase());
+        opts.legalhold = ObjectLockLegalHoldStatus::from(legalhold.to_ascii_uppercase());
     }
 
     // Promoted keys are read above through `lookup`, which accepts more than
@@ -680,7 +679,7 @@ fn is_remote_tier_auth_rejection(err: &std::io::Error) -> bool {
     matches!(response.status_code, StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN)
         || matches!(
             response.code,
-            RemoteErrorCode::ACCESS_DENIED | RemoteErrorCode::INVALID_ACCESS_KEY_ID | RemoteErrorCode::SIGNATURE_DOES_NOT_MATCH
+            S3ErrorCode::AccessDenied | S3ErrorCode::InvalidAccessKeyId | S3ErrorCode::SignatureDoesNotMatch
         )
 }
 
@@ -1305,7 +1304,7 @@ mod tests {
 
     struct AuthRejectingProbePutBackend {
         status_code: StatusCode,
-        code: RemoteErrorCode,
+        code: S3ErrorCode,
         probes: Arc<AtomicUsize>,
         removes: Arc<AtomicUsize>,
     }
@@ -1768,10 +1767,10 @@ mod tests {
     #[tokio::test]
     async fn check_warm_backend_reports_auth_rejected_probe_put_as_invalid_credentials() {
         for (status_code, code) in [
-            (StatusCode::FORBIDDEN, RemoteErrorCode::ACCESS_DENIED),
-            (StatusCode::UNAUTHORIZED, RemoteErrorCode::custom("")),
-            (StatusCode::OK, RemoteErrorCode::INVALID_ACCESS_KEY_ID),
-            (StatusCode::OK, RemoteErrorCode::SIGNATURE_DOES_NOT_MATCH),
+            (StatusCode::FORBIDDEN, S3ErrorCode::AccessDenied),
+            (StatusCode::UNAUTHORIZED, S3ErrorCode::Custom("".into())),
+            (StatusCode::OK, S3ErrorCode::InvalidAccessKeyId),
+            (StatusCode::OK, S3ErrorCode::SignatureDoesNotMatch),
         ] {
             let probes = Arc::new(AtomicUsize::new(0));
             let removes = Arc::new(AtomicUsize::new(0));
@@ -2166,31 +2165,18 @@ mod tests {
             metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(),
             ObjectLockLegalHoldStatus::ON.to_string(),
         );
-        metadata.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), ObjectLockMode::GOVERNANCE.to_string());
+        metadata.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            ObjectLockRetentionMode::GOVERNANCE.to_string(),
+        );
 
         let opts = build_transition_put_options("COLD".to_string(), metadata);
 
-        assert_eq!(opts.mode, ObjectLockMode::GOVERNANCE);
-        assert_eq!(opts.legalhold, ObjectLockLegalHoldStatus::ON);
+        assert_eq!(opts.mode.as_str(), ObjectLockRetentionMode::GOVERNANCE);
+        assert_eq!(opts.legalhold.as_str(), ObjectLockLegalHoldStatus::ON);
         assert_ne!(opts.retain_until_date, OffsetDateTime::UNIX_EPOCH);
     }
 
-    #[test]
-    fn build_transition_put_options_sends_no_object_lock_or_replication_headers_by_default() {
-        let opts = build_transition_put_options("COLD".to_string(), HashMap::new());
-        assert_eq!(opts.mode.as_str(), "");
-        assert_eq!(opts.legalhold.as_str(), "");
-        assert_eq!(opts.internal.replication_status.as_str(), "");
-        let header = opts.header();
-        for name in [
-            "x-amz-object-lock-mode",
-            "x-amz-object-lock-legal-hold",
-            "x-amz-replication-status",
-        ] {
-            assert!(header.get(name).is_none(), "{name} must not be sent: {header:?}");
-        }
-    }
-
     #[test]
     fn build_transition_put_options_filters_promoted_headers_from_user_metadata() {
         let mut metadata = HashMap::new();

crates/ecstore/src/services/tier/warm_backend_s3.rs

@@ -28,8 +28,6 @@ use crate::services::tier::{
     },
 };
 use http::HeaderMap;
-use rustfs_gateway_types::dto::Status as BucketVersioningStatus;
-use rustfs_s3_client::remote_error_code::RemoteErrorCode;
 use rustfs_s3_client::{
     api_error_response::to_error_response,
     api_get_options::GetObjectOptions,
@@ -44,6 +42,7 @@ use rustfs_s3_client::{
 };
 use rustfs_utils::egress::validate_outbound_url_allow_private;
 use rustfs_utils::path::SLASH_SEPARATOR;
+use s3s::{S3ErrorCode, dto::BucketVersioningStatus};
 
 pub struct WarmBackendS3 {
     pub client: Arc<TransitionClient>,
@@ -62,8 +61,8 @@ enum RemoteBucketVersioning {
 
 fn remote_bucket_versioning_from_status(status: Option<&str>) -> Result<RemoteBucketVersioning, std::io::Error> {
     Ok(match status {
-        Some(status) if status == BucketVersioningStatus::ENABLED.as_str() => RemoteBucketVersioning::Enabled,
-        Some(status) if status == BucketVersioningStatus::SUSPENDED.as_str() => RemoteBucketVersioning::Suspended,
+        Some(BucketVersioningStatus::ENABLED) => RemoteBucketVersioning::Enabled,
+        Some(BucketVersioningStatus::SUSPENDED) => RemoteBucketVersioning::Suspended,
         Some(status) => {
             return Err(std::io::Error::new(
                 std::io::ErrorKind::InvalidData,
@@ -238,7 +237,7 @@ impl WarmBackendS3 {
             }
             Err(err) => {
                 let response = to_error_response(&err);
-                if response.code == RemoteErrorCode::NO_SUCH_KEY {
+                if response.code == S3ErrorCode::NoSuchKey {
                     Ok(TransitionCandidateProbe::Missing)
                 } else {
                     Err(err)
@@ -325,7 +324,7 @@ impl WarmBackendS3 {
             }
             Err(err) => {
                 let response = to_error_response(&err);
-                if response.code == RemoteErrorCode::NO_SUCH_KEY {
+                if response.code == S3ErrorCode::NoSuchKey {
                     Ok(TransitionCandidateProbe::Missing)
                 } else {
                     Err(err)
@@ -689,7 +688,7 @@ mod tests {
             .probe_transition_candidate("forbidden-probe")
             .await
             .expect_err("an authorization failure must not be mistaken for a missing key");
-        assert_eq!(to_error_response(&err).code, RemoteErrorCode::ACCESS_DENIED);
+        assert_eq!(to_error_response(&err).code, S3ErrorCode::AccessDenied);
         assert_eq!(
             backend
                 .probe_transition_candidate("delete-marker-hidden")
@@ -1067,13 +1066,11 @@ mod tests {
             RemoteBucketVersioning::Disabled
         );
         assert_eq!(
-            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::ENABLED.as_str()))
-                .expect("enabled status should parse"),
+            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::ENABLED)).expect("enabled status should parse"),
             RemoteBucketVersioning::Enabled
         );
         assert_eq!(
-            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::SUSPENDED.as_str()))
-                .expect("suspended status should parse"),
+            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::SUSPENDED)).expect("suspended status should parse"),
             RemoteBucketVersioning::Suspended
         );
         let err = remote_bucket_versioning_from_status(Some("UnexpectedStatus"))

crates/ecstore/src/services/tier/warm_backend_wasabi.rs

@@ -18,6 +18,7 @@ use std::{
     sync::atomic::{AtomicBool, Ordering},
 };
 
+use s3s::header::{X_AMZ_DELETE_MARKER, X_AMZ_VERSION_ID};
 use uuid::Uuid;
 
 use crate::services::tier::{
@@ -127,7 +128,7 @@ impl WarmBackend for WarmBackendWasabi {
             self.ensure_unversioned().await?;
         }
         let (headers, reader) = self.s3.get_with_headers(object, rv, opts).await?;
-        if rv.is_empty() && (headers.contains_key("x-amz-version-id") || headers.contains_key("x-amz-delete-marker")) {
+        if rv.is_empty() && (headers.contains_key(X_AMZ_VERSION_ID) || headers.contains_key(X_AMZ_DELETE_MARKER)) {
             self.versioning_drifted.store(true, Ordering::Release);
             return Err(io::Error::other(
                 "Wasabi bucket versioning changed while reading an unversioned tier object",

crates/ecstore/src/set_disk/core/io_primitives.rs

@@ -2538,6 +2538,7 @@ pub(in crate::set_disk) struct RenameDataCommit {
     pub(in crate::set_disk) old_current_source: Option<FileInfo>,
     pub(in crate::set_disk) committed_file_info: FileInfo,
     pub(in crate::set_disk) tail_drain: Option<tokio::task::JoinHandle<Option<RenameTailOutcome>>>,
+    pub(in crate::set_disk) tail_has_metadata_reservations: bool,
 }
 
 pub(in crate::set_disk) struct RenameTailCleanup {
@@ -3468,6 +3469,8 @@ pub(in crate::set_disk) struct RenameDataFenceOptions<'a> {
     rollback_receipt: Option<RenameRollbackReceipt>,
     multipart_proof_authority: Option<crate::disk::multipart_source::ProofVerificationContext>,
     multipart_commit_owner: Option<Arc<MultipartLogicalGuards>>,
+    reservation_ctx: Option<Arc<crate::runtime::instance::InstanceContext>>,
+    prepared_reservations: Option<Vec<Option<RenameMetadataReservation>>>,
 }
 
 impl<'a> RenameDataFenceOptions<'a> {
@@ -3483,6 +3486,8 @@ impl<'a> RenameDataFenceOptions<'a> {
             rollback_receipt: None,
             multipart_proof_authority: None,
             multipart_commit_owner: None,
+            reservation_ctx: None,
+            prepared_reservations: None,
         }
     }
 
@@ -3519,6 +3524,274 @@ impl<'a> RenameDataFenceOptions<'a> {
         self.namespace_commit_guard = namespace_commit_guard;
         self
     }
+
+    pub(in crate::set_disk) fn with_reservation_context(
+        mut self,
+        ctx: Option<Arc<crate::runtime::instance::InstanceContext>>,
+        reservations: Option<Vec<Option<RenameMetadataReservation>>>,
+    ) -> Self {
+        self.reservation_ctx = ctx;
+        self.prepared_reservations = reservations;
+        self
+    }
+}
+
+#[derive(Clone)]
+struct MetadataRenewalTarget {
+    disk: DiskStore,
+    token: Uuid,
+    renewed_at: Arc<parking_lot::Mutex<tokio::time::Instant>>,
+}
+
+struct MetadataRenewalGroup {
+    targets: Arc<parking_lot::Mutex<HashMap<usize, MetadataRenewalTarget>>>,
+    task: tokio::task::JoinHandle<()>,
+}
+
+impl Drop for MetadataRenewalGroup {
+    fn drop(&mut self) {
+        self.task.abort();
+    }
+}
+
+pub(crate) struct RenameMetadataReservation {
+    disk: DiskStore,
+    token: Option<Uuid>,
+    renewed_at: Arc<parking_lot::Mutex<tokio::time::Instant>>,
+    renewal: Option<(Arc<MetadataRenewalGroup>, usize)>,
+}
+
+impl RenameMetadataReservation {
+    fn stop_renewal(&mut self) {
+        if let Some((group, index)) = self.renewal.take() {
+            let mut targets = group.targets.lock();
+            targets.remove(&index);
+            if targets.is_empty() {
+                group.task.abort();
+            }
+        }
+    }
+
+    async fn cancel(&mut self) -> disk::error::Result<()> {
+        self.stop_renewal();
+        let Some(token) = self.token else { return Ok(()) };
+        match tokio::time::timeout(Duration::from_secs(5), self.disk.cancel_object_metadata_reservation(token)).await {
+            Ok(Ok(true)) => {
+                self.token = None;
+                Ok(())
+            }
+            Ok(Ok(false)) => Err(DiskError::DiskOngoingReq),
+            Ok(Err(error)) => Err(error),
+            Err(_) => Err(DiskError::Timeout),
+        }
+    }
+}
+
+impl Drop for RenameMetadataReservation {
+    fn drop(&mut self) {
+        self.stop_renewal();
+        let Some(token) = self.token.take() else { return };
+        if let Ok(runtime) = tokio::runtime::Handle::try_current() {
+            let disk = self.disk.clone();
+            runtime.spawn(async move {
+                let _ = disk.cancel_object_metadata_reservation(token).await;
+            });
+        }
+    }
+}
+
+pub(crate) async fn prepare_rename_metadata_reservations(
+    disks: &[Option<DiskStore>],
+    bucket: &str,
+    object: &str,
+    incarnation: Uuid,
+    ctx: &Arc<crate::runtime::instance::InstanceContext>,
+    owns_bucket_fence: bool,
+) -> disk::error::Result<Option<Vec<Option<RenameMetadataReservation>>>> {
+    let prepared: Vec<disk::error::Result<Option<RenameMetadataReservation>>> = join_all(disks.iter().map(|disk| async move {
+        let Some(disk) = disk else { return Ok(None) };
+        if disk.is_local() {
+            crate::store::admit_local_disk(ctx, disk, None, false).await?;
+        }
+        let prepared_at = tokio::time::Instant::now();
+        let token = if disk.is_local() && owns_bucket_fence {
+            let store = crate::bucket::metadata_sys::object_store_if_initialized_in(ctx)
+                .await
+                .ok_or(DiskError::DiskOngoingReq)?;
+            store
+                .reserve_local_object_metadata_at_incarnation(
+                    &disk.endpoint().to_string(),
+                    bucket,
+                    object,
+                    incarnation,
+                    Duration::from_secs(5),
+                )
+                .await?
+        } else {
+            disk.prepare_object_metadata(bucket, object, incarnation, Duration::from_secs(5))
+                .await?
+        };
+        // No await separates receipt from installing the cancellation owner.
+        Ok(Some(RenameMetadataReservation {
+            disk: disk.clone(),
+            token: Some(token),
+            renewed_at: Arc::new(parking_lot::Mutex::new(prepared_at)),
+            renewal: None,
+        }))
+    }))
+    .await;
+    let mut reservations = Vec::with_capacity(prepared.len());
+    let mut failed = false;
+    for result in prepared {
+        match result {
+            Ok(reservation) => reservations.push(reservation),
+            Err(_) => {
+                failed = true;
+                reservations.push(None);
+            }
+        }
+    }
+    if failed {
+        // Prepared tokens are inert; dropping them cannot leave a queued writer.
+        return Ok(None);
+    }
+    let targets = Arc::new(parking_lot::Mutex::new(HashMap::new()));
+    for (index, reservation) in reservations.iter().enumerate() {
+        if let Some(reservation) = reservation {
+            let token = reservation.token.ok_or(DiskError::Unexpected)?;
+            targets.lock().insert(
+                index,
+                MetadataRenewalTarget {
+                    disk: reservation.disk.clone(),
+                    token,
+                    renewed_at: Arc::clone(&reservation.renewed_at),
+                },
+            );
+        }
+    }
+    if !targets.lock().is_empty() {
+        let renewal_targets = Arc::clone(&targets);
+        let renewal_ctx = Arc::clone(ctx);
+        let renewal_bucket = bucket.to_owned();
+        let renewal_object = object.to_owned();
+        // One task renews the whole fanout; short writes create no renewal RPCs.
+        let task = tokio::spawn(async move {
+            let mut delay = Duration::from_secs(20);
+            loop {
+                tokio::time::sleep(delay).await;
+                let current: Vec<_> = renewal_targets.lock().values().cloned().collect();
+                if current.is_empty() {
+                    break;
+                }
+                // Once any acknowledgment is too old, the complete fanout can
+                // no longer be safely activated. Stop retrying stale peers;
+                // their receiver-side tokens still expire on their own.
+                if current
+                    .iter()
+                    .any(|target| target.renewed_at.lock().elapsed() >= Duration::from_secs(60))
+                {
+                    break;
+                }
+                let results = join_all(current.into_iter().map(|target| {
+                    let ctx = Arc::clone(&renewal_ctx);
+                    let bucket = &renewal_bucket;
+                    let object = &renewal_object;
+                    async move {
+                        let started = tokio::time::Instant::now();
+                        let admitted = !target.disk.is_local()
+                            || crate::store::admit_local_disk(&ctx, &target.disk, None, false).await.is_ok();
+                        if admitted
+                            && target
+                                .disk
+                                .renew_object_metadata(bucket, object, target.token, Duration::from_secs(5))
+                                .await
+                                .is_ok()
+                        {
+                            *target.renewed_at.lock() = started;
+                            true
+                        } else {
+                            false
+                        }
+                    }
+                }))
+                .await;
+                delay = if results.into_iter().all(|success| success) {
+                    Duration::from_secs(20)
+                } else {
+                    Duration::from_secs(5)
+                };
+            }
+        });
+        let group = Arc::new(MetadataRenewalGroup { targets, task });
+        for (index, reservation) in reservations.iter_mut().enumerate() {
+            if let Some(reservation) = reservation {
+                reservation.renewal = Some((Arc::clone(&group), index));
+            }
+        }
+    }
+    Ok(Some(reservations))
+}
+
+pub(crate) fn align_rename_metadata_reservations(
+    mut reservations: Vec<Option<RenameMetadataReservation>>,
+    disks: &[Option<DiskStore>],
+) -> Option<Vec<Option<RenameMetadataReservation>>> {
+    if reservations.len() != disks.len() {
+        return None;
+    }
+    let mut aligned = Vec::with_capacity(disks.len());
+    for disk in disks {
+        let Some(disk) = disk else {
+            aligned.push(None);
+            continue;
+        };
+        let index = reservations
+            .iter()
+            .position(|candidate| candidate.as_ref().is_some_and(|candidate| Arc::ptr_eq(&candidate.disk, disk)))?;
+        aligned.push(reservations[index].take());
+    }
+    // Disks lost while staging need no activation; dropping their still-inert
+    // reservations schedules cancellation without blocking the commit.
+    Some(aligned)
+}
+
+async fn activate_rename_metadata_reservations(
+    mut reservations: Vec<Option<RenameMetadataReservation>>,
+    bucket: &str,
+    object: &str,
+    ctx: &Arc<crate::runtime::instance::InstanceContext>,
+) -> disk::error::Result<Option<Vec<Option<RenameMetadataReservation>>>> {
+    // Freeze renewal before activation. A failed refresh keeps the last
+    // acknowledged timestamp, so stale tokens fall back before dispatch.
+    for reservation in reservations.iter_mut().flatten() {
+        reservation.stop_renewal();
+    }
+    if reservations
+        .iter()
+        .flatten()
+        .any(|reservation| reservation.renewed_at.lock().elapsed() >= Duration::from_secs(50))
+    {
+        return Ok(None);
+    }
+    let activated = join_all(reservations.iter().flatten().map(|reservation| async move {
+        if reservation.disk.is_local() {
+            crate::store::admit_local_disk(ctx, &reservation.disk, None, false).await?;
+        }
+        let token = reservation.token.ok_or(DiskError::Unexpected)?;
+        reservation
+            .disk
+            .activate_object_metadata(bucket, object, token, Duration::from_secs(5))
+            .await
+    }))
+    .await;
+    if activated.into_iter().any(|result| result.is_err()) {
+        let cancelled = join_all(reservations.iter_mut().flatten().map(RenameMetadataReservation::cancel)).await;
+        if cancelled.into_iter().any(|result| result.is_err()) {
+            return Err(DiskError::DiskOngoingReq);
+        }
+        return Ok(None);
+    }
+    Ok(Some(reservations))
 }
 
 #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
@@ -3530,7 +3803,7 @@ type RenameDataLegacyTuple = (
     Option<OldCurrentSize>,
 );
 
-fn put_rename_early_ack_enabled() -> bool {
+pub(crate) fn put_rename_early_ack_enabled() -> bool {
     rustfs_utils::get_env_bool(ENV_RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE, true)
 }
 
@@ -3896,6 +4169,7 @@ impl SetDisks {
             old_current_source,
             committed_file_info,
             tail_drain: None,
+            tail_has_metadata_reservations: false,
         })
     }
 
@@ -3992,6 +4266,8 @@ impl SetDisks {
             rollback_receipt,
             multipart_proof_authority: _,
             multipart_commit_owner: _,
+            reservation_ctx,
+            prepared_reservations,
         } = fence_options;
         if let Some(file_info) = disks
             .iter()
@@ -4007,6 +4283,11 @@ impl SetDisks {
 
         let disk_count = disks.len();
         let fanout_disks = disks.to_vec();
+        let mut reservations = match (reservation_ctx.as_ref(), prepared_reservations) {
+            (Some(ctx), Some(prepared)) => activate_rename_metadata_reservations(prepared, dst_bucket, dst_object, ctx).await?,
+            (_, _) => None,
+        };
+        let tail_has_metadata_reservations = reservations.is_some();
         let fanout_fence_tokens = Self::scanner_publication_lease_tokens_for_disks(disks, scanner_publication_lease_tokens)?;
         let coordinator_disks = fanout_disks.clone();
         let src_bucket = Arc::new(src_bucket.to_string());
@@ -4039,6 +4320,11 @@ impl SetDisks {
                     let file_info = file_info.clone();
                     let successful_rename_completion_rank = successful_rename_completion_rank.clone();
                     let namespace_commit_guard = namespace_commit_guard.clone();
+                    let reservation = reservations
+                        .as_mut()
+                        .and_then(|entries| entries.get_mut(i))
+                        .and_then(Option::take);
+                    let reservation_ctx = reservation_ctx.clone();
                     tasks.spawn(async move {
                         let _namespace_commit_guard = namespace_commit_guard.clone();
                         let mut dispatch_state = RenameDispatchState::NotDispatched;
@@ -4049,6 +4335,17 @@ impl SetDisks {
                             let Some(disk) = disk else {
                                 return Err(DiskError::DiskNotFound);
                             };
+                            let disk_admission = if reservation.is_some() && disk.is_local() {
+                                let ctx = reservation_ctx.as_ref().ok_or(DiskError::Unexpected)?;
+                                // The fanout owner already covers this instance through
+                                // physical completion; only add a disk owner when it does not.
+                                let covered = namespace_commit_guard
+                                    .as_ref()
+                                    .is_some_and(|guard| guard.covers_instance(ctx));
+                                crate::store::admit_local_disk(ctx, &disk, None, !covered).await?
+                            } else {
+                                None
+                            };
                             let disk_wait_location_stage = if disk.is_local() {
                                 rustfs_io_metrics::PUT_STAGE_SET_DISK_RENAME_DISK_WAIT_LOCAL
                             } else {
@@ -4074,6 +4371,11 @@ impl SetDisks {
                             let disk_wait_started = rustfs_io_metrics::put_stage_timer();
                             dispatch_state = RenameDispatchState::MayHavePublished;
                             let _foreground_active_guard = rename_foreground_active_guard(i);
+                            let namespace_owner: Option<Arc<dyn Send + Sync>> = if disk_admission.is_some() {
+                                Some(Arc::new((namespace_commit_guard.clone(), disk_admission)))
+                            } else {
+                                namespace_commit_guard.clone().map(|owner| owner as Arc<dyn Send + Sync>)
+                            };
                             let observed = disk
                                 .rename_data_borrowed_with_fence_observed(
                                     &src_bucket,
@@ -4083,9 +4385,8 @@ impl SetDisks {
                                     &dst_object,
                                     disk::RenameDataGuards {
                                         scanner_publication_lease_token,
-                                        namespace_owner: namespace_commit_guard
-                                            .clone()
-                                            .map(|owner| owner as Arc<dyn Send + Sync>),
+                                        metadata_reservation_token: reservation.as_ref().and_then(|owner| owner.token),
+                                        namespace_owner,
                                         ..Default::default()
                                     },
                                 )
@@ -4363,6 +4664,7 @@ impl SetDisks {
         let attach_tail_owner_started = rustfs_io_metrics::put_stage_timer();
         commit.map(|mut commit| {
             commit.tail_drain = Some(tail_drain);
+            commit.tail_has_metadata_reservations = tail_has_metadata_reservations;
             rustfs_io_metrics::record_put_object_stage_duration_from(
                 rustfs_io_metrics::PUT_STAGE_SET_DISK_RENAME_EARLY_ACK_ATTACH_TAIL_OWNER,
                 attach_tail_owner_started,
@@ -4432,6 +4734,8 @@ impl SetDisks {
             rollback_receipt,
             multipart_proof_authority,
             multipart_commit_owner,
+            reservation_ctx: _,
+            prepared_reservations: _,
         } = fence_options;
         if let Some(file_info) = disks
             .iter()
@@ -4901,6 +5205,7 @@ impl SetDisks {
             old_current_source,
             committed_file_info,
             tail_drain: None,
+            tail_has_metadata_reservations: false,
         })
     }
 
@@ -5340,6 +5645,14 @@ impl SetDisks {
         }
 
         let disks = self.get_disks_internal().await;
+        if disks.len() == self.set_drive_count
+            && disks.iter().all(|disk| {
+                disk.as_ref()
+                    .is_some_and(|disk| disk.has_clean_active_multipart_source(dst_object, true))
+            })
+        {
+            return Ok(false);
+        }
         let transaction_path = part_transaction_path(dst_object);
         let transaction_meta_path = format!("{transaction_path}/{PART_TRANSACTION_NEW_META}");
 
@@ -5539,6 +5852,14 @@ impl SetDisks {
         write_quorum: usize,
     ) -> disk::error::Result<()> {
         let disks = self.get_disks_internal().await;
+        if disks.len() == self.set_drive_count
+            && disks.iter().all(|disk| {
+                disk.as_ref()
+                    .is_some_and(|disk| disk.has_clean_active_multipart_source(part_path, false))
+            })
+        {
+            return Ok(());
+        }
         let listings = join_all(disks.iter().map(|disk| {
             let disk = disk.clone();
             async move {
@@ -5598,10 +5919,11 @@ impl SetDisks {
         write_quorum: usize,
         quorum_context: Option<MultipartWriteQuorumContext<'_>>,
     ) -> disk::error::Result<Vec<Option<DiskStore>>> {
-        self.recover_part_transaction(dst_object, write_quorum).await?;
         let part = ObjectPartInfo::unmarshal(&meta)?;
         let integrity = part.integrity.map(Arc::new);
 
+        self.recover_part_transaction(dst_object, write_quorum).await?;
+
         let src_bucket = Arc::new(src_bucket.to_string());
         let src_object = Arc::new(src_object.to_string());
         let dst_bucket = Arc::new(dst_bucket.to_string());
@@ -5623,7 +5945,9 @@ impl SetDisks {
                 if let Err(error) = prepared {
                     return Some(Err(error));
                 }
-                if let Some(integrity) = integrity {
+                if let Some(integrity) = integrity
+                    && !disk.is_local()
+                {
                     let Some((directory, _)) = dst_object.rsplit_once('/') else {
                         return Some(Err(DiskError::FileCorrupt));
                     };
@@ -5740,7 +6064,36 @@ impl SetDisks {
             return Err(err);
         }
 
-        let committed = self.recover_part_transaction(&dst_object, write_quorum).await?;
+        let all_prepared_succeeded = errs
+            .iter()
+            .zip(prepared_disks.iter())
+            .all(|(err, disk)| err.is_none() && disk.as_ref().is_some_and(|d| d.is_local()));
+        let committed = if all_prepared_succeeded {
+            let settles = prepared_disks.iter().map(|disk| {
+                let disk = disk.clone();
+                let dst_object = dst_object.clone();
+                async move {
+                    let Some(disk) = disk else {
+                        return Err(DiskError::DiskNotFound);
+                    };
+                    disk.settle_part_transaction(RUSTFS_META_MULTIPART_BUCKET, &dst_object, PartTransactionAction::Commit)
+                        .await
+                }
+            });
+            let settle_results = join_all(settles).await;
+            if settle_results.iter().filter(|res| res.is_ok()).count() >= write_quorum
+                && settle_results
+                    .iter()
+                    .zip(prepared_disks.iter())
+                    .all(|(res, disk)| disk.is_none() || res.is_ok())
+            {
+                true
+            } else {
+                self.recover_part_transaction(&dst_object, write_quorum).await?
+            }
+        } else {
+            self.recover_part_transaction(&dst_object, write_quorum).await?
+        };
         if !committed {
             let err = DiskError::ErasureWriteQuorum;
             if let Some(context) = quorum_context {
@@ -6737,9 +7090,7 @@ impl SetDisks {
                         object,
                     );
                 }
-                let if_none_match = http_preconditions.if_none_match_value().map(str::to_owned);
-                let if_match = http_preconditions.if_match_value().map(str::to_owned);
-                if should_prevent_write(&oi, if_none_match, if_match)
+                if should_prevent_write(&oi, http_preconditions.if_none_match_value(), http_preconditions.if_match_value())
                     && !crate::data_movement::can_replace_stale_data_movement_target(&oi, opts)
                 {
                     return Some(StorageError::PreconditionFailed);
@@ -7573,6 +7924,45 @@ mod tests {
     use tempfile::TempDir;
     use tokio::io::AsyncReadExt;
 
+    #[tokio::test]
+    async fn renewal_stops_after_the_disk_loses_admission_and_token_expires() {
+        let (_dir, disk) = read_multiple_test_disk("bucket", &[]).await;
+        let ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let endpoint = disk.endpoint().to_string();
+        ctx.local_disk_map()
+            .write()
+            .await
+            .insert(endpoint.clone(), Some(disk.clone()));
+        let reservations = prepare_rename_metadata_reservations(&[Some(disk)], "bucket", "object", Uuid::new_v4(), &ctx, false)
+            .await
+            .expect("prepare")
+            .expect("complete fanout");
+        let group = Arc::clone(
+            &reservations[0]
+                .as_ref()
+                .expect("reservation")
+                .renewal
+                .as_ref()
+                .expect("renewal")
+                .0,
+        );
+        ctx.local_disk_map().write().await.remove(&endpoint);
+        tokio::task::yield_now().await;
+        tokio::time::pause();
+        tokio::time::advance(Duration::from_secs(21)).await;
+        tokio::task::yield_now().await;
+        tokio::time::advance(Duration::from_secs(41)).await;
+        for _ in 0..10 {
+            if group.task.is_finished() {
+                break;
+            }
+            tokio::task::yield_now().await;
+        }
+        tokio::time::resume();
+        assert!(group.task.is_finished(), "stale tokens must not cause indefinite renewal attempts");
+        drop(reservations);
+    }
+
     #[tokio::test]
     async fn commit_fault_schedule_keeps_multiple_disk_phases_independent() {
         use rename_fanout_barrier::{PHASE_CLEANUP, PHASE_RENAME, arm, checkpoint};

crates/ecstore/src/set_disk/mod.rs

@@ -80,7 +80,7 @@ use crate::erasure::codec::bridge::{
 use crate::erasure::coding;
 use crate::error::{Error, Result, is_err_version_not_found};
 use crate::error::{GenericError, ObjectApiError, is_err_object_not_found};
-use crate::io_support::bitrot::{create_bitrot_reader_from_bytes, create_bitrot_writer};
+use crate::io_support::bitrot::{create_bitrot_reader_from_bytes, create_bitrot_writer, create_bitrot_writer_with_orig_volume};
 use crate::object_api::ObjectOptions;
 use crate::object_api::get_object_body_cache_hook;
 use crate::object_api::object_api_utils::get_raw_etag;
@@ -120,7 +120,6 @@ use bytesize::ByteSize;
 use futures::future::join_all;
 use http::HeaderMap;
 use md5::{Digest as Md5Digest, Md5};
-use regex::Regex;
 use rustfs_config::MI_B;
 use rustfs_filemeta::metadata_keys;
 use rustfs_filemeta::{
@@ -154,8 +153,8 @@ use rustfs_utils::http::headers::{
 };
 use rustfs_utils::http::{
     SUFFIX_ACTUAL_OBJECT_SIZE_CAP, SUFFIX_ACTUAL_SIZE, SUFFIX_BUCKET_INCARNATION_ID, SUFFIX_COMPRESSION, SUFFIX_COMPRESSION_SIZE,
-    SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_RESTORE_COPYBACK, SUFFIX_RESTORE_OPERATION_ID, SUFFIX_RESTORE_WORKER_LOCK,
-    contains_key_str, get_header_map, get_str, insert_str, is_object_encryption_marker, remove_header_map,
+    SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_RESTORE_OPERATION_ID, SUFFIX_RESTORE_WORKER_LOCK, contains_key_str, get_header_map,
+    get_str, insert_str, is_object_encryption_marker, remove_header_map,
 };
 use rustfs_utils::{
     HashAlgorithm,
@@ -411,6 +410,10 @@ impl ObjectLockDiagGuard {
         self
     }
 
+    async fn downgrade_committed_tail(&mut self) -> bool {
+        self.guard.downgrade_committed_tail().await
+    }
+
     /// Whether the underlying namespace lock's heartbeat has observed a
     /// refresh-quorum loss (backlog#899 Phase 2). Callers fence their commit
     /// point on this so a stale lock holder does not race a double-write.
@@ -903,6 +906,10 @@ mod core;
 pub(crate) use core::io_primitives::disk_call_counters;
 #[cfg(test)]
 pub(crate) use core::io_primitives::{ENV_RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE, rename_fanout_barrier};
+pub(crate) use core::io_primitives::{
+    RenameMetadataReservation, align_rename_metadata_reservations, prepare_rename_metadata_reservations,
+    put_rename_early_ack_enabled,
+};
 mod ctx;
 mod metadata;
 mod ops;
@@ -8036,16 +8043,29 @@ fn parts_after_marker(part_numbers: &[usize], part_number_marker: usize) -> Opti
     Some(&part_numbers[index..])
 }
 
-pub fn canonicalize_etag(etag: &str) -> String {
-    let re = Regex::new("\"*?([^\"]*?)\"*?$").unwrap();
-    re.replace_all(etag, "$1").to_string()
+fn canonicalized_etag(etag: &str) -> Cow<'_, str> {
+    // Preserve the legacy suffix-matching behavior for embedded quotes.
+    let trimmed = etag.trim_end_matches('"');
+    let Some(last_quote) = trimmed.rfind('"') else {
+        return Cow::Borrowed(trimmed);
+    };
+    let prefix = trimmed[..last_quote].trim_end_matches('"');
+    let suffix = &trimmed[last_quote + 1..];
+    if prefix.is_empty() {
+        Cow::Borrowed(suffix)
+    } else {
+        let mut result = String::with_capacity(prefix.len() + suffix.len());
+        result.push_str(prefix);
+        result.push_str(suffix);
+        Cow::Owned(result)
+    }
 }
 
 pub fn e_tag_matches(etag: &str, condition: &str) -> bool {
     if condition.trim() == "*" {
         return true;
     }
-    canonicalize_etag(etag) == canonicalize_etag(condition)
+    canonicalized_etag(etag) == canonicalized_etag(condition)
 }
 
 fn weak_etag_matches(etag: &str, condition: &str) -> bool {
@@ -8060,15 +8080,12 @@ fn weak_etag_matches(etag: &str, condition: &str) -> bool {
             .or_else(|| trimmed.strip_prefix("w/"))
             .unwrap_or(trimmed)
     }
-    canonicalize_etag(strip_weak(etag)) == canonicalize_etag(strip_weak(condition))
+    canonicalized_etag(strip_weak(etag)) == canonicalized_etag(strip_weak(condition))
 }
 
-pub fn should_prevent_write(oi: &ObjectInfo, if_none_match: Option<String>, if_match: Option<String>) -> bool {
-    let if_none_match = if_none_match
-        .as_deref()
-        .map(str::trim)
-        .filter(|condition| !condition.is_empty());
-    let if_match = if_match.as_deref().map(str::trim).filter(|condition| !condition.is_empty());
+pub fn should_prevent_write(oi: &ObjectInfo, if_none_match: Option<&str>, if_match: Option<&str>) -> bool {
+    let if_none_match = if_none_match.map(str::trim).filter(|condition| !condition.is_empty());
+    let if_match = if_match.map(str::trim).filter(|condition| !condition.is_empty());
 
     match &oi.etag {
         Some(etag) => {
@@ -13772,6 +13789,22 @@ mod tests {
         assert!(result2.iter().all(|d| d.is_none()));
     }
 
+    #[test]
+    fn canonicalized_etag_preserves_legacy_regex_semantics() {
+        let regex = regex::Regex::new("\"*?([^\"]*?)\"*?$").expect("valid legacy ETag pattern");
+        let alphabet = ['a', '"', '\n', 'é', '\0'];
+        for length in 0..=6_u32 {
+            for mut number in 0..alphabet.len().pow(length) {
+                let mut input = String::new();
+                for _ in 0..length {
+                    input.push(alphabet[number % alphabet.len()]);
+                    number /= alphabet.len();
+                }
+                assert_eq!(canonicalized_etag(&input), regex.replace_all(&input, "$1"), "input: {input:?}");
+            }
+        }
+    }
+
     #[test]
     fn test_etag_matches() {
         assert!(e_tag_matches("abc", "abc"));
@@ -13980,7 +14013,10 @@ mod tests {
         let requested_until = now + Duration::from_secs(60 * 60 * 24);
 
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             existing_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -13992,7 +14028,7 @@ mod tests {
         };
         let opts = ObjectOptions {
             object_lock_retention: Some(crate::storage_api_contracts::object::ObjectLockRetentionOptions {
-                mode: Some("COMPLIANCE".to_string()),
+                mode: Some(s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string()),
                 retain_until: Some(requested_until),
                 bypass_governance: true,
             }),
@@ -14012,7 +14048,10 @@ mod tests {
         let requested_until = now + Duration::from_secs(60 * 60 * 24);
 
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "GOVERNANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             existing_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14024,7 +14063,7 @@ mod tests {
         };
         let opts = ObjectOptions {
             object_lock_retention: Some(crate::storage_api_contracts::object::ObjectLockRetentionOptions {
-                mode: Some("GOVERNANCE".to_string()),
+                mode: Some(s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string()),
                 retain_until: Some(requested_until),
                 bypass_governance: true,
             }),
@@ -14039,7 +14078,10 @@ mod tests {
     async fn test_check_object_lock_delete_blocks_compliance_version_delete() {
         let retain_until = OffsetDateTime::now_utc() + Duration::from_secs(60 * 60 * 24 * 60);
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14081,7 +14123,10 @@ mod tests {
             },
             restore_expires: Some(restore_expiry),
             user_defined: Arc::new(HashMap::from([
-                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
+                (
+                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
+                    s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+                ),
                 (
                     metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
                     retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14152,7 +14197,10 @@ mod tests {
     async fn test_check_object_lock_delete_allows_versioned_delete_marker_creation() {
         let retain_until = OffsetDateTime::now_utc() + Duration::from_secs(60 * 60 * 24 * 60);
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14177,7 +14225,10 @@ mod tests {
     fn governance_retained_obj_info() -> ObjectInfo {
         let retain_until = OffsetDateTime::now_utc() + Duration::from_secs(60 * 60 * 24 * 60);
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "GOVERNANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14227,7 +14278,10 @@ mod tests {
     async fn test_check_object_lock_delete_blocks_replicated_compliance_version_purge() {
         let retain_until = OffsetDateTime::now_utc() + Duration::from_secs(60 * 60 * 24 * 60);
         let mut user_defined = HashMap::new();
-        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string());
+        user_defined.insert(
+            metadata_keys::OBJECT_LOCK_MODE.to_string(),
+            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
+        );
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14282,11 +14336,11 @@ mod tests {
 
     #[test]
     fn test_object_lock_delete_check_required_keeps_object_lock_config_buckets() {
-        use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
+        use s3s::dto::{ObjectLockConfiguration, ObjectLockEnabled};
 
         let mut bm = crate::bucket::metadata::BucketMetadata::new("lock-config-bucket");
         bm.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some("Enabled".to_owned()),
+            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
             ..Default::default()
         });
         assert!(object_lock_delete_check_required(Some(&bm)));
@@ -14326,61 +14380,61 @@ mod tests {
             etag: Some("abc".to_string()),
             ..Default::default()
         };
-        let if_none_match = Some("abc".to_string());
+        let if_none_match = Some("abc");
         let if_match = None;
         assert!(should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some("*".to_string());
+        let if_none_match = Some("*");
         let if_match = None;
         assert!(should_prevent_write(&oi, if_none_match, if_match));
 
         let if_none_match = None;
-        let if_match = Some("def".to_string());
+        let if_match = Some("def");
         assert!(should_prevent_write(&oi, if_none_match, if_match));
 
         let if_none_match = None;
-        let if_match = Some("*".to_string());
+        let if_match = Some("*");
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some("def".to_string());
+        let if_none_match = Some("def");
         let if_match = None;
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some("def".to_string());
-        let if_match = Some("*".to_string());
+        let if_none_match = Some("def");
+        let if_match = Some("*");
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some("def".to_string());
-        let if_match = Some("\"abc\"".to_string());
+        let if_none_match = Some("def");
+        let if_match = Some("\"abc\"");
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some("*".to_string());
-        let if_match = Some("\"abc\"".to_string());
+        let if_none_match = Some("*");
+        let if_match = Some("\"abc\"");
         assert!(should_prevent_write(&oi, if_none_match, if_match));
 
         let oi = ObjectInfo {
             etag: None,
             ..Default::default()
         };
-        let if_none_match = Some("*".to_string());
-        let if_match = Some("\"abc\"".to_string());
+        let if_none_match = Some("*");
+        let if_match = Some("\"abc\"");
         assert!(should_prevent_write(&oi, if_none_match, if_match));
 
         let if_none_match = None;
         let if_match = None;
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
-        let if_none_match = Some(String::new());
-        let if_match = Some(" ".to_string());
+        let if_none_match = Some("");
+        let if_match = Some(" ");
         assert!(!should_prevent_write(&oi, if_none_match, if_match));
 
         let oi = ObjectInfo {
             etag: Some("abc".to_string()),
             ..Default::default()
         };
-        assert!(should_prevent_write(&oi, Some("W/\"abc\"".to_string()), None));
-        assert!(!should_prevent_write(&oi, Some("W/\"def\"".to_string()), None));
-        assert!(should_prevent_write(&oi, None, Some("W/\"abc\"".to_string())));
+        assert!(should_prevent_write(&oi, Some("W/\"abc\""), None));
+        assert!(!should_prevent_write(&oi, Some("W/\"def\""), None));
+        assert!(should_prevent_write(&oi, None, Some("W/\"abc\"")));
     }
 
     #[test]
@@ -17183,8 +17237,8 @@ mod tests {
         // otherwise `from_file_info` never synthesizes the null version id and
         // the branch under test is not reached.
         let suspended = crate::bucket::replication::DeleteReplicationConfigSnapshot::from_configs_for_test(
-            rustfs_gateway_types::persistence::PersistedVersioningConfiguration {
-                status: Some("Suspended".to_owned()),
+            s3s::dto::VersioningConfiguration {
+                status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::SUSPENDED)),
                 ..Default::default()
             },
             None,

crates/ecstore/src/set_disk/ops/delete_recovery.rs

@@ -2374,86 +2374,6 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn delete_recovery_batch_with_missing_version_recovers_after_preparation_failure() {
-        use crate::disk::RUSTFS_META_TMP_BUCKET;
-        use crate::storage_api_contracts::object::ObjectToDelete;
-
-        let bucket = "recover-batch-missing-version";
-        let (dirs, store, disks) = fixture(bucket).await;
-        let deleted = put(&store, bucket, &vec![48; 4096], true).await.expect("version to delete");
-        let remaining = vec![49; 4096];
-        let retained = put(&store, bucket, &remaining, true).await.expect("version to retain");
-        let missing = Uuid::new_v4();
-        let temporary = dirs[3].path().join(RUSTFS_META_TMP_BUCKET);
-        let saved = dirs[3].path().join("saved-delete-tmp");
-        tokio::fs::rename(&temporary, &saved)
-            .await
-            .expect("save returning disk's staging directory");
-        tokio::fs::write(&temporary, b"block preparation")
-            .await
-            .expect("fail staging after the recovery preflight");
-
-        let (objects, errors) = store
-            .delete_objects(
-                bucket,
-                [deleted, missing]
-                    .into_iter()
-                    .map(|version| ObjectToDelete {
-                        object_name: "object".into(),
-                        version_id: Some(version),
-                        ..Default::default()
-                    })
-                    .collect(),
-                ObjectOptions {
-                    versioned: true,
-                    ..Default::default()
-                },
-            )
-            .await;
-        assert_eq!(objects.len(), 2);
-        assert!(errors.iter().all(Option::is_none), "batch deletion succeeds at quorum: {errors:?}");
-        assert!(
-            disks[3]
-                .discover_delete_transactions(bucket, Some("object"), None, 100)
-                .await
-                .expect("inspect failed preparation")
-                .transactions
-                .is_empty(),
-            "returning disk must adopt the quorum decision without a local preparation"
-        );
-        tokio::fs::remove_file(&temporary).await.expect("remove preparation failure");
-        tokio::fs::rename(&saved, &temporary)
-            .await
-            .expect("restore returning disk staging");
-
-        let result = store
-            .recover_delete_transactions(bucket, &HealOpts::default())
-            .await
-            .expect("recover batch containing a missing non-null version");
-        assert!(result.detail.contains("blocked=0"), "{}", result.detail);
-        assert_eq!(body(&store, bucket, Some(retained)).await, remaining);
-        for version in [deleted, missing] {
-            assert_eq!(
-                disks[3]
-                    .read_version(bucket, bucket, "object", &version.to_string(), &ReadOptions::default())
-                    .await
-                    .expect_err("returning disk must not retain or create a deleted version"),
-                DiskError::FileVersionNotFound
-            );
-        }
-        for disk in &disks {
-            assert!(
-                disk.discover_delete_transactions(bucket, Some("object"), None, 100)
-                    .await
-                    .expect("inspect converged batch")
-                    .transactions
-                    .is_empty(),
-                "every participant must converge and release the decision"
-            );
-        }
-    }
-
     #[tokio::test]
     async fn delete_recovery_dry_run_matches_adoption_of_already_absent_non_null_version() {
         let bucket = "recover-preview-non-null";
@@ -2820,140 +2740,6 @@ mod tests {
             .expect("no retained version decision may keep the empty bucket nonempty");
     }
 
-    #[tokio::test]
-    async fn delete_recovery_partial_batch_receipt_handles_missing_versions() {
-        use crate::disk::RUSTFS_META_TMP_BUCKET;
-        use crate::set_disk::ops::object::DeleteCleanupBarrier;
-        use crate::storage_api_contracts::object::ObjectToDelete;
-
-        for include_missing in [false, true] {
-            let bucket = "recover-prepared-batch-missing-versions";
-            let (dirs, store, disks) = fixture(bucket).await;
-            metadata_sys::update_in(
-                &store.ctx,
-                bucket,
-                crate::bucket::metadata::BUCKET_VERSIONING_CONFIG,
-                b"<VersioningConfiguration xmlns=\"http://s3.amazonaws.com/doc/2006-03-01/\"><Status>Enabled</Status></VersioningConfiguration>".to_vec(),
-            )
-            .await
-            .expect("enable supported versioned bucket");
-            let first = put(&store, bucket, &vec![81; 1024 * 1024], true)
-                .await
-                .expect("first version");
-            let second = put(&store, bucket, &vec![82; 4096], true).await.expect("retained version");
-            let missing = Uuid::new_v4();
-            let barrier = DeleteCleanupBarrier::install(bucket);
-            let deleting = store.clone();
-            let task = tokio::spawn(async move {
-                let mut versions = vec![first];
-                if include_missing {
-                    versions.push(missing);
-                }
-                deleting
-                    .delete_objects(
-                        bucket,
-                        versions
-                            .into_iter()
-                            .map(|version| ObjectToDelete {
-                                object_name: "object".into(),
-                                version_id: Some(version),
-                                ..Default::default()
-                            })
-                            .collect(),
-                        ObjectOptions {
-                            versioned: true,
-                            ..Default::default()
-                        },
-                    )
-                    .await
-            });
-            barrier.wait_until_paused().await;
-            let transactions = disks[0]
-                .discover_delete_transactions(bucket, Some("object"), None, 100)
-                .await
-                .expect("discover batch preparation")
-                .transactions;
-            assert_eq!(transactions.len(), 1);
-            let transaction = transactions[0].transaction;
-            for disk in &disks {
-                let state = disk
-                    .inspect_delete_transaction(bucket, "object", transaction)
-                    .await
-                    .expect("inspect prepared batch");
-                assert_eq!(state.state, DeleteRecoveryState::NewPrepared);
-                let journal = state.journal.expect("prepared journal");
-                assert!(journal.original_digest.is_some());
-                assert!(journal.original_versions.iter().all(|v| v.version_id != Some(missing)));
-            }
-            // Fail durable receipt staging on two participants after metadata quorum.
-            for dir in dirs.iter().skip(2) {
-                tokio::fs::rename(dir.path().join(RUSTFS_META_TMP_BUCKET), dir.path().join("saved-receipt-staging"))
-                    .await
-                    .expect("save receipt staging directory");
-                tokio::fs::write(dir.path().join(RUSTFS_META_TMP_BUCKET), b"receipt staging I/O failure")
-                    .await
-                    .expect("inject receipt staging failure");
-            }
-            barrier.release();
-            let (_, errors) = task.await.expect("batch deletion must join");
-            drop(barrier);
-            assert_eq!(errors.len(), if include_missing { 2 } else { 1 });
-            assert!(errors.iter().all(Option::is_some), "partial receipts are indeterminate: {errors:?}");
-            let mut receipts = 0;
-            let mut prepared = 0;
-            for disk in &disks {
-                match disk
-                    .inspect_delete_transaction(bucket, "object", transaction)
-                    .await
-                    .expect("partial decision")
-                    .state
-                {
-                    DeleteRecoveryState::QuorumCommitted => receipts += 1,
-                    DeleteRecoveryState::NewPrepared => prepared += 1,
-                    state => panic!("unexpected post-publication state: {state:?}"),
-                }
-            }
-            assert_eq!((receipts, prepared), (2, 2));
-            for dir in dirs.iter().skip(2) {
-                tokio::fs::remove_file(dir.path().join(RUSTFS_META_TMP_BUCKET))
-                    .await
-                    .expect("remove injected staging failure");
-                tokio::fs::rename(dir.path().join("saved-receipt-staging"), dir.path().join(RUSTFS_META_TMP_BUCKET))
-                    .await
-                    .expect("restore receipt staging");
-            }
-            store
-                .delete_object(
-                    bucket,
-                    "object",
-                    ObjectOptions {
-                        versioned: true,
-                        version_id: Some(second.to_string()),
-                        ..Default::default()
-                    },
-                )
-                .await
-                .expect("later version deletion succeeds");
-            store
-                .recover_delete_transactions(bucket, &HealOpts::default())
-                .await
-                .expect("certified batch must recover after later version deletion");
-            for disk in &disks {
-                assert!(
-                    disk.discover_delete_transactions(bucket, Some("object"), None, 100)
-                        .await
-                        .expect("settled transaction inventory")
-                        .transactions
-                        .is_empty()
-                );
-            }
-            store
-                .delete_bucket(bucket, &DeleteBucketOptions::default())
-                .await
-                .expect("settled receipts must not block removal of the empty bucket");
-        }
-    }
-
     #[tokio::test]
     async fn delete_recovery_partial_receipt_converges_after_later_version_absence() {
         let bucket = "recover-partial-chained-versions";

crates/ecstore/src/set_disk/ops/heal.rs

@@ -34,6 +34,7 @@ mod heal_lock_wait_tests;
 #[derive(Clone, Copy)]
 struct ExplicitVersionHeal<'a> {
     opts: &'a HealOpts,
+    staging_guard: Option<&'a Arc<tokio::sync::Mutex<rustfs_lock::NamespaceLockGuard>>>,
     allow_regeneration: bool,
     preserve_unrepairable: bool,
     commit_fence: Option<&'a (dyn Fn() -> bool + Sync)>,
@@ -41,6 +42,58 @@ struct ExplicitVersionHeal<'a> {
     retirement: Option<&'a MarkerRetirementContext<'a>>,
 }
 
+async fn restore_heal_exclusive_guard(
+    guard: &Arc<tokio::sync::Mutex<rustfs_lock::NamespaceLockGuard>>,
+) -> disk::error::Result<()> {
+    tokio::time::timeout(get_lock_acquire_timeout(), async {
+        loop {
+            {
+                let mut guard = guard.lock().await;
+                if guard.is_lock_lost() {
+                    return Err(DiskError::DiskOngoingReq);
+                }
+                let waiting = tokio::time::timeout(std::time::Duration::from_secs(1), guard.waiting_writer()).await;
+                if !matches!(waiting, Ok(Some(false))) {
+                    return Err(DiskError::Timeout);
+                }
+                if tokio::time::timeout(std::time::Duration::from_secs(1), guard.try_upgrade_committed_tail())
+                    .await
+                    .unwrap_or(false)
+                {
+                    // A writer may have waited on a peer that upgraded while
+                    // another peer was still waiting for its readers to drain.
+                    let waiting = tokio::time::timeout(std::time::Duration::from_secs(1), guard.waiting_writer()).await;
+                    return if matches!(waiting, Ok(Some(false))) {
+                        Ok(())
+                    } else {
+                        Err(DiskError::Timeout)
+                    };
+                }
+            }
+            tokio::time::sleep(std::time::Duration::from_millis(10)).await;
+        }
+    })
+    .await
+    .map_err(|_| DiskError::Timeout)?
+}
+
+async fn wait_for_heal_writer(guard: &Arc<tokio::sync::Mutex<rustfs_lock::NamespaceLockGuard>>) {
+    loop {
+        tokio::time::sleep(std::time::Duration::from_millis(250)).await;
+        let waiting =
+            tokio::time::timeout(std::time::Duration::from_secs(1), async { guard.lock().await.waiting_writer().await }).await;
+        if !matches!(waiting, Ok(Some(false))) {
+            return;
+        }
+    }
+}
+
+fn cleanup_heal_staging_without_waiting(set: SetDisks, tmp_id: String) {
+    drop(tokio::spawn(async move {
+        let _ = set.delete_all(RUSTFS_META_TMP_BUCKET, &tmp_id).await;
+    }));
+}
+
 // Keep the large heal state machine and its construction temporary out of
 // each dispatch caller's poll frame. Ordinary object writes do not use this
 // heal-only allocation boundary.
@@ -844,6 +897,7 @@ impl SetDisks {
             version_id,
             ExplicitVersionHeal {
                 opts,
+                staging_guard: None,
                 allow_regeneration: true,
                 preserve_unrepairable: false,
                 commit_fence: None,
@@ -994,6 +1048,7 @@ impl SetDisks {
     ) -> disk::error::Result<(HealResultItem, Option<DiskError>)> {
         let ExplicitVersionHeal {
             opts,
+            staging_guard,
             allow_regeneration: allow_explicit_version_regen,
             preserve_unrepairable,
             commit_fence,
@@ -1701,7 +1756,29 @@ impl SetDisks {
                         let dst_data_dir = data_dir;
 
                         if !latest_meta.deleted && !latest_meta.is_remote() {
-                            let stage_result: disk::error::Result<Option<DiskError>> = async {
+                            let mut shared_staging = false;
+                            if let Some(guard) = staging_guard {
+                                let supported = tokio::time::timeout(std::time::Duration::from_secs(1), async {
+                                    guard.lock().await.supports_heal_lock_transition().await
+                                })
+                                .await
+                                .unwrap_or(false);
+                                if supported {
+                                    let converted = tokio::time::timeout(std::time::Duration::from_secs(1), async {
+                                        guard.lock().await.downgrade_committed_tail().await
+                                    })
+                                    .await
+                                    .unwrap_or(false);
+                                    if converted {
+                                        shared_staging = true;
+                                    } else {
+                                        // Some peers may have converted. Abort without committing;
+                                        // dropping the guard releases each retained lease.
+                                        return Err(DiskError::Timeout);
+                                    }
+                                }
+                            }
+                            let stage = async {
                                 let erasure_info = latest_meta.erasure.clone();
                                 let mut writer_failures = HealWriterFailures::new(disks_to_heal_count);
 
@@ -1926,17 +2003,43 @@ impl SetDisks {
 
                                 writer_failures.warn(bucket, object, version_id, "partial_targets_unavailable");
                                 Ok(None)
-                            }
-                            .await;
+                            };
+                            let stage_result: disk::error::Result<Option<DiskError>> = if shared_staging {
+                                if let Some(guard) = staging_guard {
+                                    tokio::select! {
+                                        result = stage => result,
+                                        _ = wait_for_heal_writer(guard) => {
+                                            // The UUID prefix is private to this attempt. Cleanup must not
+                                            // retain the namespace lock while a foreground writer waits;
+                                            // the stale-tmp sweeper also covers a cancelled disk operation.
+                                            cleanup_heal_staging_without_waiting(self.clone(), tmp_id.clone());
+                                            return Err(DiskError::Timeout);
+                                        }
+                                    }
+                                } else {
+                                    stage.await
+                                }
+                            } else {
+                                stage.await
+                            };
                             match stage_result {
                                 Ok(None) => {}
                                 failure => {
-                                    // Release readers and writers before removing all parts
-                                    // staged under this heal's exclusively owned prefix.
-                                    let _ = self.delete_all(RUSTFS_META_TMP_BUCKET, &tmp_id).await;
+                                    if shared_staging {
+                                        cleanup_heal_staging_without_waiting(self.clone(), tmp_id.clone());
+                                    } else {
+                                        let _ = self.delete_all(RUSTFS_META_TMP_BUCKET, &tmp_id).await;
+                                    }
                                     return failure.map(|error| (result, error));
                                 }
                             }
+                            if shared_staging
+                                && let Some(guard) = staging_guard
+                                && let Err(error) = restore_heal_exclusive_guard(guard).await
+                            {
+                                cleanup_heal_staging_without_waiting(self.clone(), tmp_id.clone());
+                                return Err(error);
+                            }
                         }
                         let _read_repair_write_lock_guard = if let Some(expected_fingerprint) = read_repair_commit_fingerprint {
                             if _read_repair_read_lock_guard
@@ -3616,6 +3719,7 @@ impl SetDisks {
             &version,
             ExplicitVersionHeal {
                 opts: &opts,
+                staging_guard: None,
                 allow_regeneration: false,
                 preserve_unrepairable: true,
                 commit_fence: Some(can_commit),
@@ -3676,6 +3780,7 @@ impl SetDisks {
                         version_id,
                         ExplicitVersionHeal {
                             opts: &read_opts,
+                            staging_guard: None,
                             allow_regeneration: false,
                             preserve_unrepairable: true,
                             commit_fence: Some(&can_commit),
@@ -3702,7 +3807,7 @@ impl SetDisks {
                 .new_ns_lock(bucket, object)
                 .await
                 .map_err(|e| e.narrow_to_disk().unwrap_or_else(DiskError::other))?;
-            Some(Arc::new(
+            Some(Arc::new(tokio::sync::Mutex::new(
                 measure_heal_object_lock(
                     "heal_object",
                     HealObjectLockKind::Write,
@@ -3714,7 +3819,7 @@ impl SetDisks {
                         .narrow_to_disk()
                         .unwrap_or_else(DiskError::other)
                 })?,
-            ))
+            )))
         } else {
             None
         };
@@ -3732,10 +3837,18 @@ impl SetDisks {
             } else {
                 None
             };
+        let write_lock_loss = if let Some(guard) = &_write_lock_guard {
+            guard.lock().await.lock_lost_signal()
+        } else {
+            None
+        };
         let locks_lost = || {
-            _write_lock_guard.as_deref().is_some_and(|guard| guard.is_lock_lost())
+            write_lock_loss.as_ref().is_some_and(|signal| signal.is_lost())
                 || transaction_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
         };
+        // Recovery cleanup can outlive its future. Never permit reads until
+        // every physical operation holding its recovery owner has finished.
+        let recovery_tail = Arc::new(());
         let mut recovery_removed_version = false;
         if !opts.read_repair
             && _write_lock_guard.is_some()
@@ -3762,6 +3875,7 @@ impl SetDisks {
                             Arc::new((
                                 scope.clone(),
                                 self.ctx.begin_namespace_commit(),
+                                Arc::clone(&recovery_tail),
                                 _write_lock_guard.clone(),
                                 transaction_guard.clone(),
                             )) as Arc<dyn Send + Sync>
@@ -3885,6 +3999,9 @@ impl SetDisks {
                 version_id,
                 ExplicitVersionHeal {
                     opts: &inner_opts,
+                    staging_guard: (Arc::strong_count(&recovery_tail) == 1)
+                        .then_some(_write_lock_guard.as_ref())
+                        .flatten(),
                     allow_regeneration: true,
                     preserve_unrepairable: false,
                     commit_fence: Some(&commit_fence),
@@ -3909,6 +4026,9 @@ impl SetDisks {
                             version_id,
                             ExplicitVersionHeal {
                                 opts: &inner_opts,
+                                staging_guard: (Arc::strong_count(&recovery_tail) == 1)
+                                    .then_some(_write_lock_guard.as_ref())
+                                    .flatten(),
                                 allow_regeneration: true,
                                 preserve_unrepairable: false,
                                 commit_fence: Some(&commit_fence),
@@ -6047,6 +6167,7 @@ mod heal_result_report_tests {
                 "",
                 ExplicitVersionHeal {
                     opts: &opts,
+                    staging_guard: None,
                     allow_regeneration: true,
                     preserve_unrepairable: false,
                     commit_fence: Some(&can_commit),
@@ -10637,6 +10758,369 @@ mod heal_result_report_tests {
         assert_eq!(actual, expected);
     }
 
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn background_heal_storage_entry_keeps_readers_available_during_reconstruction() {
+        use crate::object_api::WriteCompletion;
+        use std::time::Duration;
+        use tokio::io::AsyncReadExt;
+
+        let (dirs, disks, set) = hermetic_set_disks_isolated(4).await;
+        let bucket = "background-heal-storage-entry";
+        let object = "object.bin";
+        let expected = vec![0x37; 1024 * 1024];
+        set.make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create bucket");
+        set.put_object(
+            bucket,
+            object,
+            &mut PutObjReader::from_vec(expected.clone()),
+            &ObjectOptions {
+                write_completion: WriteCompletion::TailDrained,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("write object");
+        let metadata = disks[0]
+            .read_version("", bucket, object, "", &ReadOptions::default())
+            .await
+            .expect("read metadata");
+        let missing = dirs[0]
+            .path()
+            .join(bucket)
+            .join(object)
+            .join(metadata.data_dir.expect("external object data").to_string())
+            .join("part.1");
+        tokio::fs::remove_file(&missing).await.expect("remove shard");
+
+        let (_pause, arrived, release) =
+            ReadRepairPauseScope::install_at(bucket, object, super::ReadRepairPausePhase::PartStaging(1));
+        let heal_set = set.clone();
+        let heal = tokio::spawn(async move {
+            let mut absence = None;
+            let result = heal_set
+                .heal_object_with_retirement(
+                    bucket,
+                    object,
+                    "",
+                    &HealOpts {
+                        scan_mode: HealScanMode::Deep,
+                        read_repair: false,
+                        ..Default::default()
+                    },
+                    &mut absence,
+                    AbsenceProofRequest {
+                        retirement: None,
+                        allow_unversioned: false,
+                    },
+                )
+                .await;
+            (result, absence)
+        });
+        tokio::time::timeout(Duration::from_secs(10), arrived.notified())
+            .await
+            .expect("read-repair reached shard reconstruction");
+        let mut reader = tokio::time::timeout(
+            Duration::from_millis(200),
+            set.get_object_reader(
+                bucket,
+                object,
+                None,
+                http::HeaderMap::new(),
+                &ObjectOptions {
+                    suppress_read_repair: true,
+                    ..Default::default()
+                },
+            ),
+        )
+        .await
+        .expect("background heal reconstruction should allow concurrent readers")
+        .expect("open object while reconstruction is paused");
+        let mut during_heal = Vec::new();
+        tokio::time::timeout(Duration::from_secs(10), reader.stream.read_to_end(&mut during_heal))
+            .await
+            .expect("read completes while reconstruction is paused")
+            .expect("read object during reconstruction");
+        assert_eq!(during_heal, expected);
+        drop(reader);
+        release.notify_one();
+        let (result, absence) = tokio::time::timeout(Duration::from_secs(20), heal)
+            .await
+            .expect("repair finishes")
+            .expect("repair task joins");
+        let (_, error) = result.expect("repair operation succeeds");
+        assert!(error.is_none(), "repair failed: {error:?}");
+        assert!(absence.is_none());
+        assert!(missing.exists(), "missing shard was not restored");
+        let mut reader = set
+            .get_object_reader(bucket, object, None, http::HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("read repaired object");
+        let mut actual = Vec::new();
+        reader.stream.read_to_end(&mut actual).await.expect("read full object");
+        assert_eq!(actual, expected);
+    }
+
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn background_heal_does_not_timeout_foreground_put_during_long_reconstruction() {
+        use crate::object_api::WriteCompletion;
+        use std::time::Duration;
+        use tokio::io::AsyncReadExt;
+
+        let (dirs, disks, set) = hermetic_set_disks_isolated(4).await;
+        let bucket = "background-heal-long-put";
+        let object = "object.bin";
+        set.make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create bucket");
+        set.put_object(
+            bucket,
+            object,
+            &mut PutObjReader::from_vec(vec![0x37; 1024 * 1024]),
+            &ObjectOptions {
+                write_completion: WriteCompletion::TailDrained,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("write original object");
+        let metadata = disks[0]
+            .read_version("", bucket, object, "", &ReadOptions::default())
+            .await
+            .expect("read metadata");
+        let missing = dirs[0]
+            .path()
+            .join(bucket)
+            .join(object)
+            .join(metadata.data_dir.expect("external data").to_string())
+            .join("part.1");
+        tokio::fs::remove_file(missing).await.expect("remove shard");
+        let (_pause, arrived, release) =
+            ReadRepairPauseScope::install_at(bucket, object, super::ReadRepairPausePhase::PartStaging(1));
+        let heal_set = set.clone();
+        let heal = tokio::spawn(async move {
+            let mut absence = None;
+            heal_set
+                .heal_object_with_retirement(
+                    bucket,
+                    object,
+                    "",
+                    &HealOpts {
+                        scan_mode: HealScanMode::Deep,
+                        ..Default::default()
+                    },
+                    &mut absence,
+                    AbsenceProofRequest {
+                        retirement: None,
+                        allow_unversioned: false,
+                    },
+                )
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(10), arrived.notified())
+            .await
+            .expect("reconstruction started");
+        let put = tokio::time::timeout(
+            Duration::from_secs(7),
+            set.put_object(bucket, object, &mut PutObjReader::from_vec(vec![0x52; 1024]), &ObjectOptions::default()),
+        )
+        .await;
+        release.notify_one();
+        let heal_result = tokio::time::timeout(Duration::from_secs(20), heal)
+            .await
+            .expect("heal finishes after releasing reconstruction")
+            .expect("heal task joins");
+        put.expect("foreground PUT finishes during reconstruction")
+            .expect("foreground PUT must not time out");
+        assert!(
+            matches!(heal_result, Err(Error::Timeout)),
+            "interrupted heal must be retryable: {heal_result:?}"
+        );
+        drop(_pause);
+        let mut absence = None;
+        let (_, retry_error) = set
+            .heal_object_with_retirement(
+                bucket,
+                object,
+                "",
+                &HealOpts {
+                    scan_mode: HealScanMode::Deep,
+                    ..Default::default()
+                },
+                &mut absence,
+                AbsenceProofRequest {
+                    retirement: None,
+                    allow_unversioned: false,
+                },
+            )
+            .await
+            .expect("a later heal must retry the current object");
+        assert!(retry_error.is_none(), "retry must complete: {retry_error:?}");
+        let mut reader = set
+            .get_object_reader(bucket, object, None, http::HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("read foreground write after heal");
+        let mut actual = Vec::new();
+        reader.stream.read_to_end(&mut actual).await.expect("read full object");
+        assert_eq!(actual, vec![0x52; 1024], "stale heal must not overwrite the foreground write");
+    }
+
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn background_heal_yields_to_version_delete_without_resurrecting_it() {
+        use crate::object_api::WriteCompletion;
+        use std::time::Duration;
+        use tokio::io::AsyncReadExt;
+
+        let (dirs, disks, set) = hermetic_set_disks_isolated(4).await;
+        let bucket = "background-heal-version-delete";
+        let object = "object.bin";
+        set.make_bucket(
+            bucket,
+            &MakeBucketOptions {
+                versioning_enabled: true,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("create versioned bucket");
+        let original = set
+            .put_object(
+                bucket,
+                object,
+                &mut PutObjReader::from_vec(vec![0x37; 1024 * 1024]),
+                &ObjectOptions {
+                    versioned: true,
+                    write_completion: WriteCompletion::TailDrained,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("write original version");
+        let version = original.version_id.expect("version id").to_string();
+        let survivor = set
+            .put_object(
+                bucket,
+                object,
+                &mut PutObjReader::from_vec(vec![0x52; 1024]),
+                &ObjectOptions {
+                    versioned: true,
+                    write_completion: WriteCompletion::TailDrained,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("write surviving version")
+            .version_id
+            .expect("surviving version id");
+        let metadata = disks[0]
+            .read_version("", bucket, object, &version, &ReadOptions::default())
+            .await
+            .expect("read original metadata");
+        let missing = dirs[0]
+            .path()
+            .join(bucket)
+            .join(object)
+            .join(metadata.data_dir.expect("external data").to_string())
+            .join("part.1");
+        tokio::fs::remove_file(missing).await.expect("remove shard");
+        let (_pause, arrived, release) =
+            ReadRepairPauseScope::install_at(bucket, object, super::ReadRepairPausePhase::PartStaging(1));
+        let heal_set = set.clone();
+        let heal_version = version.clone();
+        let heal = tokio::spawn(async move {
+            heal_set
+                .heal_object_with_retirement(
+                    bucket,
+                    object,
+                    &heal_version,
+                    &HealOpts {
+                        scan_mode: HealScanMode::Deep,
+                        ..Default::default()
+                    },
+                    &mut None,
+                    AbsenceProofRequest {
+                        retirement: None,
+                        allow_unversioned: false,
+                    },
+                )
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(10), arrived.notified())
+            .await
+            .expect("reconstruction started");
+        let delete = tokio::time::timeout(
+            Duration::from_secs(7),
+            set.delete_object(
+                bucket,
+                object,
+                ObjectOptions {
+                    versioned: true,
+                    version_id: Some(version.clone()),
+                    object_lock_config_snapshot: Some(Arc::new(crate::set_disk::ObjectLockConfigSnapshot::new(
+                        crate::bucket::metadata_sys::ObjectLockConfigState::ConfirmedAbsent,
+                    ))),
+                    ..Default::default()
+                },
+            ),
+        )
+        .await;
+        release.notify_one();
+        let heal_result = tokio::time::timeout(Duration::from_secs(20), heal)
+            .await
+            .expect("heal finishes")
+            .expect("heal task joins");
+        delete
+            .expect("foreground delete finishes")
+            .expect("foreground delete succeeds");
+        assert!(
+            matches!(heal_result, Err(Error::Timeout)),
+            "interrupted heal must be retryable: {heal_result:?}"
+        );
+        let result = set
+            .get_object_info(
+                bucket,
+                object,
+                &ObjectOptions {
+                    versioned: true,
+                    version_id: Some(version.clone()),
+                    ..Default::default()
+                },
+            )
+            .await;
+        assert!(
+            matches!(&result, Err(Error::VersionNotFound(b, o, v)) if b == bucket && o == object && v == &version),
+            "stale heal must not resurrect the deleted version: {result:?}"
+        );
+        let mut survivor = set
+            .get_object_reader(
+                bucket,
+                object,
+                None,
+                http::HeaderMap::new(),
+                &ObjectOptions {
+                    versioned: true,
+                    version_id: Some(survivor.to_string()),
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("surviving version remains readable");
+        let mut body = Vec::new();
+        survivor.stream.read_to_end(&mut body).await.expect("read surviving version");
+        assert_eq!(body, vec![0x52; 1024]);
+        for disk in &disks {
+            let physical = disk.read_version("", bucket, object, &version, &ReadOptions::default()).await;
+            assert!(
+                matches!(physical, Err(DiskError::FileNotFound | DiskError::FileVersionNotFound)),
+                "deleted version must remain absent on each disk: {physical:?}"
+            );
+        }
+    }
+
     #[tokio::test]
     async fn read_repair_storage_entry_does_not_prove_missing_versions_absent() {
         let (_dirs, _disks, set) = hermetic_set_disks_isolated(4).await;

crates/ecstore/src/set_disk/ops/multipart.rs

@@ -38,12 +38,12 @@ use super::super::{
     ObjectLockDiagGuard, ObjectOptions, ObjectPartInfo, OffsetDateTime, PartInfo, PutObjReader, RUSTFS_META_MULTIPART_BUCKET,
     RUSTFS_META_TMP_BUCKET, RUSTFS_MULTIPART_BUCKET_KEY, RUSTFS_MULTIPART_OBJECT_KEY, Result, SLASH_SEPARATOR,
     SUFFIX_ACTUAL_OBJECT_SIZE_CAP, SUFFIX_ACTUAL_SIZE, SUFFIX_BUCKET_INCARNATION_ID, SUFFIX_COMPRESSION_SIZE,
-    SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_RESTORE_COPYBACK, SUFFIX_RESTORE_OPERATION_ID, SUFFIX_RESTORE_WORKER_LOCK, SetDisks,
-    SmallWritePath, StorageError, Uuid, WriteLayout, check_object_lock_for_deletion_with_state,
-    classify_multipart_part_write_path, coding, complete_multipart_part_error, complete_multipart_part_error_result,
-    complete_part_checksum, contains_key_str, create_bitrot_writer, debug, disk, error, file_info_is_valid_for_metadata,
-    get_complete_multipart_md5, get_header_map, get_str, insert_str, is_err_object_not_found, is_err_version_not_found,
-    is_min_allowed_part_size, log_multipart_write_quorum_failure, multipart_part_logical_size, parts_after_marker, path_join_buf,
+    SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_RESTORE_OPERATION_ID, SUFFIX_RESTORE_WORKER_LOCK, SetDisks, SmallWritePath, StorageError,
+    Uuid, WriteLayout, check_object_lock_for_deletion_with_state, classify_multipart_part_write_path, coding,
+    complete_multipart_part_error, complete_multipart_part_error_result, complete_part_checksum, contains_key_str,
+    create_bitrot_writer_with_orig_volume, debug, disk, error, file_info_is_valid_for_metadata, get_complete_multipart_md5,
+    get_header_map, get_str, insert_str, is_err_object_not_found, is_err_version_not_found, is_min_allowed_part_size,
+    log_multipart_write_quorum_failure, multipart_part_logical_size, parts_after_marker, path_join_buf,
     prepare_complete_multipart_parts, record_compression_total_memory, record_multipart_write_quorum_failure,
     reduce_read_quorum_errs, reduce_write_quorum_errs, remove_header_map, resolve_write_layout,
     restore_commit_operation_id_from_metadata, should_persist_encryption_original_size, strip_internal_multipart_metadata,
@@ -124,9 +124,10 @@ async fn create_part_writers(
         let Some(disk) = disk else {
             return (None, Some(DiskError::DiskNotFound));
         };
-        match create_bitrot_writer(
+        match create_bitrot_writer_with_orig_volume(
             false,
             Some(disk),
+            RUSTFS_META_MULTIPART_BUCKET,
             RUSTFS_META_TMP_BUCKET,
             path,
             length,
@@ -2436,7 +2437,16 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             }
 
             let part_integrity = if let Some(integrity) = integrity {
-                Some(integrity.write(&mut shuffle_disks, bucket, RUSTFS_META_TMP_BUCKET, &tmp_part).await?)
+                Some(
+                    integrity
+                        .write(
+                            &mut shuffle_disks,
+                            RUSTFS_META_MULTIPART_BUCKET,
+                            RUSTFS_META_TMP_BUCKET,
+                            &tmp_part,
+                        )
+                        .await?,
+                )
             } else { None };
             if shuffle_disks.iter().filter(|disk| disk.is_some()).count() < write_quorum {
                 return Err(Error::ErasureWriteQuorum);
@@ -3433,14 +3443,7 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             crate::object_api::ENCRYPTED_PART_LAYOUT_CANDIDATE_SUFFIX,
         );
         rustfs_utils::http::metadata_compat::remove_str(&mut fi.metadata, crate::object_api::ENCRYPTED_PART_LAYOUT_QUORUM_SUFFIX);
-        // Pending copy-back keeps its recovery identity until finalization.
-        if expected_restore_operation_id.is_some()
-            && !fi
-                .metadata
-                .get(metadata_keys::RESTORE)
-                .and_then(|value| rustfs_filemeta::parse_restore_obj_status(value).ok())
-                .is_some_and(|status| status.is_restore_in_progress == Some(true))
-        {
+        if expected_restore_operation_id.is_some() {
             rustfs_utils::http::metadata_compat::remove_str(&mut fi.metadata, SUFFIX_RESTORE_OPERATION_ID);
             rustfs_utils::http::metadata_compat::remove_str(&mut fi.metadata, SUFFIX_RESTORE_WORKER_LOCK);
         }
@@ -3478,10 +3481,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
         bind_multipart_proof_snapshot(&disks, &self.multipart_proof_verification_context())?;
 
         let part_path = format!("{}/{}/", upload_id_path, fi.data_dir.unwrap_or(Uuid::nil()));
-        self.recover_part_transactions(&part_path, read_quorum, write_quorum)
-            .await
-            .map_err(|err| to_object_err(err.into(), vec![bucket, object]))?;
-
         let part_meta_paths = uploaded_parts
             .iter()
             .map(|v| format!("{part_path}part.{0}.meta", v.part_num))
@@ -3489,6 +3488,10 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
 
         let part_numbers = uploaded_parts.iter().map(|v| v.part_num).collect::<Vec<usize>>();
 
+        self.recover_part_transactions(&part_path, read_quorum, write_quorum)
+            .await
+            .map_err(|err| to_object_err(err.into(), vec![bucket, object]))?;
+
         let object_parts = Self::read_parts(&disks, RUSTFS_META_MULTIPART_BUCKET, &part_meta_paths, &part_numbers, read_quorum)
             .await
             .map_err(|err| to_object_err(err.into(), vec![bucket, object]))?;
@@ -3825,13 +3828,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
         fi.metadata.remove(rustfs_rio::RUSTFS_MULTIPART_CHECKSUM);
         fi.metadata.remove(rustfs_rio::RUSTFS_MULTIPART_CHECKSUM_TYPE);
         strip_internal_multipart_metadata(&mut fi.metadata);
-        if opts.restore_copyback {
-            insert_str(
-                &mut fi.metadata,
-                SUFFIX_RESTORE_COPYBACK,
-                rustfs_utils::http::metadata_compat::RESTORE_COPYBACK_PROTOCOL_V1.to_string(),
-            );
-        }
 
         // A previous coordinator can stop after a minority publication without
         // undo. A fresh final directory keeps this attempt separate from old
@@ -4403,6 +4399,7 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             let needs_immediate_heal = rename_commit.needs_immediate_heal();
             let op_old_dir = rename_commit.data_dir;
             let cleanup_disks = rename_commit.cleanup_disks;
+            let old_current_size = rename_commit.old_current_size;
             let committed_file_info = rename_commit.committed_file_info;
 
             if needs_immediate_heal {
@@ -4471,7 +4468,13 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             // parts; deleting them before the commit would strand the upload
             // permanently. This mirrors the "clean up only after commit" pattern
             // already used for the old data-dir GC and the upload-dir delete_all below.
-            commit_set.cleanup_multipart_path(&parts).await;
+            let can_skip_part_cleanup = !needs_upgrade
+                && commit_disks
+                    .iter()
+                    .all(|disk| disk.as_ref().is_some_and(|disk| disk.is_local()));
+            if !can_skip_part_cleanup {
+                commit_set.cleanup_multipart_path(&parts).await;
+            }
 
             if let Some(old_dir) = op_old_dir {
                 // backlog#898: best-effort reclaim of the dereferenced old data dir.
@@ -4517,7 +4520,9 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
 
             drop(_upload_guard.take());
 
-            Ok(ObjectInfo::from_file_info(&fi, &commit_bucket, &commit_object, commit_is_versioned))
+            let mut obj_info = ObjectInfo::from_file_info(&fi, &commit_bucket, &commit_object, commit_is_versioned);
+            obj_info.backfilled_old_current_size = old_current_size;
+            Ok(obj_info)
         };
 
         let result = if detach_commit_owner {

crates/ecstore/src/set_disk/replication.rs

@@ -27,7 +27,6 @@ use std::sync::Arc;
 struct RestoreFinalizeBarrierState {
     bucket: String,
     object: String,
-    before_lock: bool,
     arrived: tokio::sync::Notify,
     release: tokio::sync::Notify,
 }
@@ -44,18 +43,9 @@ pub(in crate::set_disk) struct RestoreFinalizeBarrier {
 #[cfg(all(test, feature = "test-util"))]
 impl RestoreFinalizeBarrier {
     pub(in crate::set_disk) fn install(bucket: &str, object: &str) -> Self {
-        Self::install_at(bucket, object, false)
-    }
-
-    pub(in crate::set_disk) fn install_before_lock(bucket: &str, object: &str) -> Self {
-        Self::install_at(bucket, object, true)
-    }
-
-    fn install_at(bucket: &str, object: &str, before_lock: bool) -> Self {
         let state = Arc::new(RestoreFinalizeBarrierState {
             bucket: bucket.to_string(),
             object: object.to_string(),
-            before_lock,
             arrived: tokio::sync::Notify::new(),
             release: tokio::sync::Notify::new(),
         });
@@ -91,13 +81,13 @@ impl Drop for RestoreFinalizeBarrier {
 }
 
 #[cfg(all(test, feature = "test-util"))]
-async fn maybe_pause_restore_finalize(bucket: &str, object: &str, before_lock: bool) {
+async fn maybe_pause_restore_finalize(bucket: &str, object: &str) {
     let barrier = RESTORE_FINALIZE_BARRIER
         .get_or_init(|| std::sync::Mutex::new(None))
         .lock()
         .expect("restore finalize barrier mutex should not poison")
         .as_ref()
-        .filter(|barrier| barrier.bucket == bucket && barrier.object == object && barrier.before_lock == before_lock)
+        .filter(|barrier| barrier.bucket == bucket && barrier.object == object)
         .cloned();
     if let Some(barrier) = barrier {
         barrier.arrived.notify_one();
@@ -160,8 +150,6 @@ impl SetDisks {
         obj_info: &ObjectInfo,
         opts: &ObjectOptions,
     ) -> Result<ObjectInfo> {
-        #[cfg(all(test, feature = "test-util"))]
-        maybe_pause_restore_finalize(bucket, object, true).await;
         let expected = RestoreCleanupIdentity::from_object_info(obj_info);
         let expected_operation_id = restore_operation_id_from_metadata(&opts.user_defined)?;
         let expected_etag = obj_info
@@ -220,7 +208,7 @@ impl SetDisks {
             return Err(Error::other("restored object changed before restore metadata finalization"));
         }
         #[cfg(all(test, feature = "test-util"))]
-        maybe_pause_restore_finalize(bucket, object, false).await;
+        maybe_pause_restore_finalize(bucket, object).await;
         let restore_expiry =
             lifecycle::expected_expiry_time(OffsetDateTime::now_utc(), opts.transition.restore_request.days.unwrap_or(1));
         fi.metadata.insert(
@@ -231,14 +219,7 @@ impl SetDisks {
             }
             .to_string(),
         );
-        fi.transition_status = lifecycle::TRANSITION_COMPLETE.to_string();
-        rustfs_utils::http::metadata_compat::insert_str(
-            &mut fi.metadata,
-            rustfs_utils::http::metadata_compat::SUFFIX_TRANSITION_STATUS,
-            lifecycle::TRANSITION_COMPLETE.to_string(),
-        );
         for suffix in [
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
             rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID,
             rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK,
         ] {
@@ -334,20 +315,10 @@ impl SetDisks {
         if !expected.matches_file_info(&fi, &expected_etag) {
             return Ok(());
         }
-        let copyback_data_dir = rustfs_utils::http::metadata_compat::get_consistent_str(
-            &fi.metadata,
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
-        )
-        .filter(|value| *value == rustfs_utils::http::metadata_compat::RESTORE_COPYBACK_PROTOCOL_V1)
-        .and_then(|_| fi.data_dir.filter(|data_dir| !data_dir.is_nil()));
         ensure_restore_metadata_lock_held(bucket, object, opts, "restore_cleanup_metadata")?;
         fi.metadata.remove(metadata_keys::RESTORE);
         fi.metadata.remove(metadata_keys::RESTORE_EXPIRY_DAYS);
         fi.metadata.remove(metadata_keys::RESTORE_REQUEST_DATE);
-        rustfs_utils::http::metadata_compat::remove_str(
-            &mut fi.metadata,
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
-        );
         rustfs_utils::http::metadata_compat::remove_str(
             &mut fi.metadata,
             rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID,
@@ -377,20 +348,6 @@ impl SetDisks {
         )
         .await?;
         self.invalidate_get_object_metadata_cache(bucket, object).await;
-        if let Some(data_dir) = copyback_data_dir {
-            let cleanup = self
-                .commit_rename_data_dir_and_mark_capacity(
-                    &disks,
-                    bucket,
-                    object,
-                    &data_dir.to_string(),
-                    "",
-                    self.default_write_quorum(),
-                )
-                .await;
-            self.report_old_data_dir_cleanup(bucket, object, &data_dir.to_string(), &cleanup)
-                .await;
-        }
         Ok(())
     }
 }

crates/ecstore/src/set_disk/transition_matrix_tests.rs

@@ -274,452 +274,3 @@ async fn prepared_snapshot_transition_duplicate_and_late_get_use_committed_remot
         "late GET must route to the remote generation instead of reopening the released local source"
     );
 }
-
-#[tokio::test]
-#[serial_test::serial]
-async fn single_part_restore_copyback_stays_pending_before_finalization() {
-    restore_copyback_stays_pending_before_finalization(false).await;
-}
-
-#[tokio::test]
-#[serial_test::serial]
-async fn multipart_restore_copyback_stays_pending_before_finalization() {
-    restore_copyback_stays_pending_before_finalization(true).await;
-}
-
-async fn restore_copyback_stays_pending_before_finalization(multipart: bool) {
-    use crate::bucket::lifecycle::bucket_lifecycle_ops::LifecycleOps as _;
-    use crate::set_disk::replication::RestoreFinalizeBarrier;
-    use crate::storage_api_contracts::multipart::{CompletePart, MultipartOperations as _};
-    use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
-    use rustfs_lifecycle::Lifecycle as _;
-
-    let (dirs, set_disks) = make_local_set_disks(4, 2).await;
-    let bucket = "restore-before-finalization-bucket";
-    let object = "object.bin";
-    let payload = vec![0x5au8; 6 * 1024 * 1024];
-    set_disks
-        .make_bucket(bucket, &MakeBucketOptions::default())
-        .await
-        .expect("bucket should be created");
-    let write_opts = ObjectOptions {
-        write_completion: WriteCompletion::TailDrained,
-        ..Default::default()
-    };
-    let original = if multipart {
-        let upload = set_disks
-            .new_multipart_upload(bucket, object, &write_opts)
-            .await
-            .expect("source multipart upload should be created");
-        let mut parts = Vec::new();
-        for (index, bytes) in payload.chunks(5 * 1024 * 1024).enumerate() {
-            let mut reader = PutObjReader::from_vec(bytes.to_vec());
-            let part = set_disks
-                .put_object_part(bucket, object, &upload.upload_id, index + 1, &mut reader, &write_opts)
-                .await
-                .expect("source part should be written");
-            parts.push(CompletePart {
-                part_num: part.part_num,
-                etag: part.etag,
-                ..Default::default()
-            });
-        }
-        set_disks
-            .clone()
-            .complete_multipart_upload(bucket, object, &upload.upload_id, parts, &write_opts)
-            .await
-            .expect("source multipart upload should complete")
-    } else {
-        let mut reader = PutObjReader::from_vec(payload.clone());
-        set_disks
-            .put_object(bucket, object, &mut reader, &write_opts)
-            .await
-            .expect("source object should be written")
-    };
-    let tier_name = format!("COLDTIER{}", &Uuid::new_v4().simple().to_string()[..8]).to_uppercase();
-    let backend = register_mock_tier(&set_disks.instance_ctx().tier_config_mgr(), &tier_name).await;
-    set_disks
-        .transition_object(
-            bucket,
-            object,
-            &ObjectOptions {
-                transition: TransitionOptions {
-                    status: TRANSITION_PENDING.to_string(),
-                    tier: tier_name.clone(),
-                    etag: original.etag.clone().expect("source ETag should be present"),
-                    ..Default::default()
-                },
-                version_id: original.version_id.map(|version| version.to_string()),
-                mod_time: original.mod_time,
-                ..Default::default()
-            },
-        )
-        .await
-        .expect("source should transition before restore");
-
-    let operation_id = Uuid::new_v4().to_string();
-    let mut operation_metadata = std::collections::HashMap::new();
-    rustfs_utils::http::metadata_compat::insert_str(
-        &mut operation_metadata,
-        rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID,
-        operation_id.clone(),
-    );
-    rustfs_utils::http::metadata_compat::insert_str(
-        &mut operation_metadata,
-        rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK,
-        rustfs_utils::http::metadata_compat::RESTORE_WORKER_LOCK_PROTOCOL_V1.to_string(),
-    );
-    operation_metadata.insert(
-        rustfs_filemeta::metadata_keys::RESTORE.to_string(),
-        "ongoing-request=\"true\"".to_string(),
-    );
-    set_disks
-        .put_object_metadata(
-            bucket,
-            object,
-            &ObjectOptions {
-                eval_metadata: Some(operation_metadata.clone()),
-                ..Default::default()
-            },
-        )
-        .await
-        .expect("the restore operation generation should be installed");
-    let barrier = RestoreFinalizeBarrier::install_before_lock(bucket, object);
-    let restore_set = Arc::clone(&set_disks);
-    let restore = tokio::spawn(async move {
-        let mut opts = ObjectOptions {
-            user_defined: operation_metadata,
-            ..Default::default()
-        };
-        opts.transition.restore_request.days = Some(1);
-        restore_set.restore_transitioned_object(bucket, object, &opts).await
-    });
-    tokio::time::timeout(Duration::from_secs(30), barrier.wait_until_paused())
-        .await
-        .expect("copy-back must reach the gap before finalization acquires its lock");
-    let pending = set_disks
-        .get_object_info(bucket, object, &ObjectOptions::default())
-        .await
-        .expect("HEAD must remain available before finalization");
-    let pending_data_dir = pending
-        .data_dir
-        .expect("pending copy-back must retain its local data directory");
-    for dir in &dirs {
-        assert!(
-            dir.path()
-                .join(bucket)
-                .join(object)
-                .join(pending_data_dir.to_string())
-                .exists(),
-            "pending copy-back must retain its committed local data directory"
-        );
-    }
-    let status = parse_restore_obj_status(
-        pending
-            .user_defined
-            .get(rustfs_filemeta::metadata_keys::RESTORE)
-            .expect("copy-back must retain restore metadata"),
-    )
-    .expect("pending restore status should parse");
-    assert_eq!(
-        pending.transitioned_object.status, TRANSITION_COMPLETE,
-        "copy-back must retain the remote identity needed by recovery and cleanup"
-    );
-    backend.set_unreachable(true).await;
-    let mut restored_before_finalization = Vec::new();
-    set_disks
-        .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
-        .await
-        .expect("pending copy-back must be readable locally before metadata finalization")
-        .stream
-        .read_to_end(&mut restored_before_finalization)
-        .await
-        .expect("pending local body should drain");
-    assert_eq!(restored_before_finalization, payload, "pending local copy must preserve every byte");
-    set_disks
-        .transition_object(
-            bucket,
-            object,
-            &ObjectOptions {
-                transition: TransitionOptions {
-                    status: TRANSITION_PENDING.to_string(),
-                    tier: pending.transitioned_object.tier.clone(),
-                    etag: pending.etag.clone().expect("pending ETag should be present"),
-                    ..Default::default()
-                },
-                version_id: pending.version_id.map(|version| version.to_string()),
-                mod_time: pending.mod_time,
-                ..Default::default()
-            },
-        )
-        .await
-        .expect("a queued transition must skip the pending restore copy-back");
-    let after_queued_transition = set_disks
-        .get_object_info(bucket, object, &ObjectOptions::default())
-        .await
-        .expect("restore metadata should remain readable after a queued transition");
-    assert_eq!(
-        after_queued_transition.transitioned_object.status, TRANSITION_COMPLETE,
-        "queued transition must not remove local copy-back data before finalization"
-    );
-    // Evaluate beyond any pre-copy deadline without sleeping or advancing the
-    // process clock. Pending copy-back must never become a restore-expiry candidate.
-    let lifecycle = PersistedLifecycleConfiguration::default();
-    let expiry_evaluation_time = OffsetDateTime::now_utc() + time::Duration::days(2);
-    let action = lifecycle
-        .eval_inner(&pending.to_lifecycle_opts(), expiry_evaluation_time, 0)
-        .await
-        .action;
-
-    barrier.release();
-    tokio::time::timeout(Duration::from_secs(30), restore)
-        .await
-        .expect("released restore should complete promptly")
-        .expect("restore task should join")
-        .expect("copy-back should finalize successfully");
-
-    assert!(status.on_going(), "copy-back must not advertise completion before metadata finalization");
-    assert!(status.expiry().is_none(), "pending copy-back must not expose a completed expiry");
-    assert!(
-        !action.delete_restored(),
-        "scanner must not reclaim copy-back before metadata finalization"
-    );
-
-    for (suffix, expected) in [
-        (rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID, operation_id.as_str()),
-        (
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK,
-            rustfs_utils::http::metadata_compat::RESTORE_WORKER_LOCK_PROTOCOL_V1,
-        ),
-    ] {
-        for prefix in [
-            rustfs_utils::http::RUSTFS_INTERNAL_PREFIX,
-            rustfs_utils::http::MINIO_INTERNAL_PREFIX,
-        ] {
-            assert_eq!(
-                pending.user_defined.get(&format!("{prefix}{suffix}")).map(String::as_str),
-                Some(expected),
-                "pending copy-back must retain each recovery marker alias"
-            );
-        }
-    }
-
-    let completed = set_disks
-        .get_object_info(bucket, object, &ObjectOptions::default())
-        .await
-        .expect("finalized metadata should be readable");
-    assert_eq!(
-        completed.data_dir,
-        Some(pending_data_dir),
-        "finalization must retain the local copy-back data directory"
-    );
-    assert_eq!(
-        completed.transitioned_object.status, TRANSITION_COMPLETE,
-        "finalization must restore the completed transition state"
-    );
-    let completed_status = parse_restore_obj_status(
-        completed
-            .user_defined
-            .get(rustfs_filemeta::metadata_keys::RESTORE)
-            .expect("finalized restore status should be present"),
-    )
-    .expect("finalized restore status should parse");
-    assert!(!completed_status.on_going(), "finalization must publish completion");
-    assert!(completed_status.expiry().is_some(), "finalization must publish the completed expiry");
-    for suffix in [
-        rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID,
-        rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK,
-    ] {
-        for prefix in [
-            rustfs_utils::http::RUSTFS_INTERNAL_PREFIX,
-            rustfs_utils::http::MINIO_INTERNAL_PREFIX,
-        ] {
-            assert!(
-                !completed.user_defined.contains_key(&format!("{prefix}{suffix}")),
-                "finalization must remove every recovery marker alias"
-            );
-        }
-    }
-    let completed_action = lifecycle
-        .eval_inner(&completed.to_lifecycle_opts(), expiry_evaluation_time, 0)
-        .await
-        .action;
-    assert!(
-        completed_action.delete_restored(),
-        "the same lifecycle evaluator must expire a completed restore beyond its deadline"
-    );
-    backend.set_unreachable(true).await;
-    let mut restored = Vec::new();
-    set_disks
-        .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
-        .await
-        .expect("completed restore must be readable with the remote tier unavailable")
-        .stream
-        .read_to_end(&mut restored)
-        .await
-        .expect("local restored body should drain");
-    assert_eq!(restored, payload, "finalized local copy must preserve every byte");
-    for dir in &dirs {
-        assert!(
-            dir.path()
-                .join(bucket)
-                .join(object)
-                .join(pending_data_dir.to_string())
-                .exists(),
-            "completed restore must retain its local data directory"
-        );
-    }
-}
-
-#[tokio::test]
-#[serial_test::serial]
-async fn restore_copyback_overwrite_preserves_remote_cleanup_owner() {
-    assert_transition_overwrite_preserves_remote_cleanup_owner(true).await;
-}
-
-#[tokio::test]
-#[serial_test::serial]
-async fn completed_transition_overwrite_preserves_remote_cleanup_owner() {
-    assert_transition_overwrite_preserves_remote_cleanup_owner(false).await;
-}
-
-async fn assert_transition_overwrite_preserves_remote_cleanup_owner(during_copyback: bool) {
-    use crate::set_disk::replication::RestoreFinalizeBarrier;
-
-    let (dirs, set_disks) = make_local_set_disks(4, 2).await;
-    let bucket = "restore-copyback-cleanup-owner";
-    let object = "object.bin";
-    let payload = vec![0x37u8; 6 * 1024 * 1024];
-    let write_opts = ObjectOptions {
-        write_completion: WriteCompletion::TailDrained,
-        ..Default::default()
-    };
-    set_disks
-        .make_bucket(bucket, &MakeBucketOptions::default())
-        .await
-        .expect("cleanup probe bucket should be created");
-    let mut reader = PutObjReader::from_vec(payload);
-    let original = set_disks
-        .put_object(bucket, object, &mut reader, &write_opts)
-        .await
-        .expect("cleanup probe source should be written");
-    let tier_name = format!("COLDTIER{}", &Uuid::new_v4().simple().to_string()[..8]).to_uppercase();
-    let backend = register_mock_tier(&set_disks.instance_ctx().tier_config_mgr(), &tier_name).await;
-    set_disks
-        .transition_object(
-            bucket,
-            object,
-            &ObjectOptions {
-                transition: TransitionOptions {
-                    status: TRANSITION_PENDING.to_string(),
-                    tier: tier_name.clone(),
-                    etag: original.etag.clone().expect("source ETag should be present"),
-                    ..Default::default()
-                },
-                mod_time: original.mod_time,
-                ..Default::default()
-            },
-        )
-        .await
-        .expect("cleanup probe source should transition");
-    let source = set_disks
-        .get_object_info(bucket, object, &ObjectOptions::default())
-        .await
-        .expect("transitioned cleanup source should resolve");
-    assert_eq!(backend.object_count().await, 1, "transition must retain the real remote cleanup source");
-    let mut barrier = None;
-    let mut restore = None;
-    if during_copyback {
-        let operation_id = Uuid::new_v4();
-        let mut metadata = std::collections::HashMap::new();
-        rustfs_utils::http::metadata_compat::insert_str(
-            &mut metadata,
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID,
-            operation_id.to_string(),
-        );
-        rustfs_utils::http::metadata_compat::insert_str(
-            &mut metadata,
-            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK,
-            rustfs_utils::http::metadata_compat::RESTORE_WORKER_LOCK_PROTOCOL_V1.to_string(),
-        );
-        metadata.insert(
-            rustfs_filemeta::metadata_keys::RESTORE.to_string(),
-            "ongoing-request=\"true\"".to_string(),
-        );
-        set_disks
-            .put_object_metadata(
-                bucket,
-                object,
-                &ObjectOptions {
-                    eval_metadata: Some(metadata.clone()),
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("cleanup probe restore generation should be installed");
-        let installed = RestoreFinalizeBarrier::install_before_lock(bucket, object);
-        let restore_set = Arc::clone(&set_disks);
-        restore = Some(tokio::spawn(async move {
-            let mut opts = ObjectOptions {
-                user_defined: metadata,
-                ..Default::default()
-            };
-            opts.transition.restore_request.days = Some(1);
-            restore_set.restore_transitioned_object(bucket, object, &opts).await
-        }));
-        tokio::time::timeout(Duration::from_secs(30), installed.wait_until_paused())
-            .await
-            .expect("real restore should publish local bytes before finalization");
-        let pending = set_disks
-            .get_object_info(bucket, object, &ObjectOptions::default())
-            .await
-            .expect("copy-back metadata should resolve");
-        assert!(pending.restore_ongoing, "probe must hit an unfinished real restore");
-        barrier = Some(installed);
-    }
-    let mut replacement = PutObjReader::from_vec(vec![0x69u8; 6 * 1024 * 1024]);
-    set_disks
-        .put_object(bucket, object, &mut replacement, &write_opts)
-        .await
-        .expect("ordinary overwrite must succeed while restore finalization is paused");
-    if let Some(barrier) = barrier {
-        barrier.release();
-    }
-    if let Some(restore) = restore {
-        tokio::time::timeout(Duration::from_secs(30), restore)
-            .await
-            .expect("superseded restore must finish")
-            .expect("superseded restore must not panic")
-            .expect_err("restore finalization must not revive the overwritten generation");
-    }
-    let mut cleanup_owners = Vec::new();
-    for dir in &dirs {
-        let encoded = match tokio::fs::read(dir.path().join(bucket).join(object).join(crate::disk::STORAGE_FORMAT_FILE)).await {
-            Ok(encoded) => encoded,
-            Err(err) if err.kind() == std::io::ErrorKind::NotFound => {
-                cleanup_owners.push(0);
-                continue;
-            }
-            Err(err) => panic!("cleanup source metadata should be readable: {err}"),
-        };
-        let versions = rustfs_filemeta::FileMeta::load(&encoded)
-            .expect("cleanup source metadata should decode")
-            .get_file_info_versions(bucket, object, false)
-            .expect("cleanup source versions should decode");
-        let count = versions
-            .free_versions
-            .iter()
-            .filter(|version| {
-                version.transition_tier == source.transitioned_object.tier
-                    && version.transitioned_objname == source.transitioned_object.name
-            })
-            .count();
-        cleanup_owners.push(count);
-    }
-    let remote_objects = backend.object_count().await;
-    println!("restore_cleanup_probe during_copyback={during_copyback} owners={cleanup_owners:?} remote_objects={remote_objects}");
-    assert!(
-        remote_objects == 0 || cleanup_owners.iter().filter(|count| **count > 0).count() >= 3,
-        "ordinary overwrite must remove the remote object or retain its cleanup owner at write quorum: during_copyback={during_copyback}, owners={cleanup_owners:?}, remote_objects={remote_objects}"
-    );
-}

crates/ecstore/src/store/bucket_fence.rs

@@ -45,7 +45,8 @@
 use crate::error::{Result, StorageError};
 use std::collections::HashMap;
 use std::future::Future;
-use std::sync::{Arc, Mutex};
+use std::sync::{Arc, Mutex, Weak};
+use std::time::Duration;
 
 use rustfs_lock::NamespaceLockGuard;
 use rustfs_lock::distributed_lock::LockLostSignal;
@@ -98,9 +99,74 @@ pub(super) struct FenceRegistration {
 #[derive(Default)]
 pub(crate) struct BucketFenceRegistry {
     entries: Mutex<HashMap<String, FenceEntry>>,
+    reservation_slots: Mutex<HashMap<(String, Uuid), Weak<ReservationFenceSlot>>>,
+}
+
+#[derive(Default)]
+pub(super) struct ReservationFenceSlot {
+    pub(super) owner: tokio::sync::Mutex<Weak<ReservationFenceGuard>>,
+}
+
+/// Shared by short bursts of reservations, but never kept alive by the cache.
+pub(super) struct ReservationFenceGuard {
+    publication: NamespaceLockGuard,
+    lifecycle: BucketIncarnationFenceGuard,
+    created_at: tokio::time::Instant,
+    _slot: Arc<ReservationFenceSlot>,
+}
+
+impl ReservationFenceGuard {
+    pub(super) fn new(
+        publication: NamespaceLockGuard,
+        lifecycle: BucketIncarnationFenceGuard,
+        slot: Arc<ReservationFenceSlot>,
+    ) -> Self {
+        Self {
+            publication,
+            lifecycle,
+            created_at: tokio::time::Instant::now(),
+            _slot: slot,
+        }
+    }
+
+    pub(super) fn is_reusable(&self) -> bool {
+        // Stop admitting new reservations through this shared reader so a
+        // queued bucket writer can progress after the existing tails drain.
+        self.created_at.elapsed() < Duration::from_secs(1) && self.is_held()
+    }
+
+    fn is_held(&self) -> bool {
+        !self.publication.is_lock_lost()
+            && !self.publication.is_released()
+            && !self.lifecycle.is_lock_lost()
+            && self
+                .lifecycle
+                .namespace_lock_guard()
+                .is_some_and(|guard| !guard.is_released())
+    }
+}
+
+impl crate::disk::os::MetadataReservationFence for ReservationFenceGuard {
+    fn is_valid(&self) -> bool {
+        self.is_held()
+    }
 }
 
 impl BucketFenceRegistry {
+    pub(super) fn reservation_slot(&self, bucket: &str, incarnation: Uuid) -> Arc<ReservationFenceSlot> {
+        let mut slots = self.reservation_slots.lock().expect("reservation fence registry poisoned");
+        let key = (bucket.to_owned(), incarnation);
+        if let Some(slot) = slots.get(&key).and_then(Weak::upgrade) {
+            return slot;
+        }
+        if slots.len() >= 1024 {
+            slots.retain(|_, slot| slot.strong_count() != 0);
+        }
+        let slot = Arc::new(ReservationFenceSlot::default());
+        slots.insert(key, Arc::downgrade(&slot));
+        slot
+    }
+
     /// Register a new live guard for `bucket` and return the memoized
     /// incarnation id if one is valid for the current coverage window.
     fn enter(&self, bucket: &str, loss_probe: LockLossProbe) -> FenceRegistration {