chore(gateway): update test-count baseline after main sync

Main #9241 removed the node pool-metadata readiness cache with 8 tests
and added 6, so the rustfs row drops from 6440 to 6438. The ecstore and
s3select rows rise for tests added by #9237, #9239 and #9241.

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=219ac11e720b3f54dfee4d5d5e95171785ba159b2fe01db3037a8f5061c1ad32
-sha256-linux=2d34698cc8ec52d5b6c37ac612919a1d21f36cd56a1af539a7522a4eea518e26
+sha256-darwin=0374e6b9b061dd5d622686301386232a4aa1cb22eedc18af70d2a7411bc66f8d
+sha256-linux=f2cee240f7a744ae991971820f48a6c9c28d19f525e8b2345fcb76eeaaf1e382

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

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

.config/e2e-smoke-selection.txt

@@ -1 +1 @@
-sha256=5bdd4c2eadbc41625ae3184fff34062e000d328108436259b2b7c4fe2531ecaf
+sha256=fbabc660e6e01fd8ff6b9d1de7d68e58242d60c942f85aed5074b0ec2918c154

.config/make/lint-fmt.mak

@@ -70,6 +70,11 @@ 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 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 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
 	@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 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 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
 	@echo "✅ All pre-PR checks passed!"
 
 .PHONY: dev-check

.config/make/tests.mak

@@ -52,8 +52,11 @@ 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_download_release_asset.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_inspect_btf_prototypes.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_release_build_workflow.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_release_glibc.py

.config/nextest.toml

@@ -110,7 +110,7 @@ threads-required = "num-test-threads"
 # Reserve capacity so sibling disk-heavy tests cannot consume its 2s
 # registration budget; retain the queued backup and deadlock assertions.
 [[profile.default.overrides]]
-filter = 'package(rustfs) & test(=admin::handlers::table_catalog::tests::cold_strong_registration_completes_while_other_bucket_backup_waits)'
+filter = 'package(rustfs) & test(=admin::handlers::table_catalog::tests::cold_object_registration_completes_while_other_bucket_backup_waits)'
 threads-required = "num-test-threads"
 
 [[profile.default.overrides]]
@@ -356,6 +356,12 @@ 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`)
 # ---------------------------------------------------------------------------
@@ -410,7 +416,7 @@ threads-required = "num-test-threads"
 # Reserve capacity so sibling disk-heavy tests cannot consume its 2s
 # registration budget; retain the queued backup and deadlock assertions.
 [[profile.ci.overrides]]
-filter = 'package(rustfs) & test(=admin::handlers::table_catalog::tests::cold_strong_registration_completes_while_other_bucket_backup_waits)'
+filter = 'package(rustfs) & test(=admin::handlers::table_catalog::tests::cold_object_registration_completes_while_other_bucket_backup_waits)'
 threads-required = "num-test-threads"
 
 [[profile.ci.overrides]]
@@ -424,6 +430,12 @@ 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)'
@@ -698,6 +710,14 @@ 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) & (
@@ -706,6 +726,7 @@ 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)$/)
     )

.config/s3s-edge-allowlist.txt

@@ -0,0 +1,37 @@
+# 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

@@ -0,0 +1,17 @@
+# 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=219
+ecstore_files=0

.config/test-count-baseline.txt

@@ -0,0 +1,64 @@
+# 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	1026	20
+rustfs	6438	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	58	0
+rustfs-ecstore	7850	2
+rustfs-extension-schema	23	0
+rustfs-filemeta	361	0
+rustfs-heal	1128	3
+rustfs-heal-contracts	8	0
+rustfs-iam	415	0
+rustfs-io-core	52	0
+rustfs-io-metrics	157	0
+rustfs-keystone	31	0
+rustfs-kms	716	25
+rustfs-license	2	0
+rustfs-lifecycle	156	0
+rustfs-lock	182	0
+rustfs-log-analyzer	91	0
+rustfs-madmin	210	0
+rustfs-notify	140	0
+rustfs-object-capacity	92	0
+rustfs-object-data-cache	141	0
+rustfs-obs	387	0
+rustfs-policy	758	0
+rustfs-protocols	770	1
+rustfs-protos	75	0
+rustfs-replication	209	0
+rustfs-rio	223	0
+rustfs-rio-v2	34	6
+rustfs-s3-client	71	0
+rustfs-s3-ops	8	0
+rustfs-s3-types	105	0
+rustfs-s3select-api	232	0
+rustfs-s3select-query	235	2
+rustfs-scanner	1386	20
+rustfs-scanner-metrics	90	0
+rustfs-security-governance	20	0
+rustfs-signer	34	0
+rustfs-storage-api	34	0
+rustfs-targets	612	26
+rustfs-test-utils	1	0
+rustfs-tls-runtime	44	0
+rustfs-trusted-proxies	72	0
+rustfs-utils	251	0
+rustfs-zip	30	0
+zstd	11	0

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

@@ -28,11 +28,12 @@ runs:
 
     - name: Install actionlint
       shell: bash
+      env:
+        GH_TOKEN: ${{ github.token }}
+        ASSET_DOWNLOADER: ${{ github.action_path }}/../setup/download_release_asset.py
       run: |
         actionlint_dir="$(mktemp -d "${RUNNER_TEMP}/actionlint.XXXXXX")"
-        curl --fail --location --silent --show-error \
-          --output "$actionlint_dir/actionlint.tar.gz" \
-          https://github.com/rhysd/actionlint/releases/download/v1.7.12/actionlint_1.7.12_linux_amd64.tar.gz
+        python3 "$ASSET_DOWNLOADER" --repository rhysd/actionlint --tag v1.7.12 --asset actionlint_1.7.12_linux_amd64.tar.gz --output "$actionlint_dir/actionlint.tar.gz" --sha256 8aca8db96f1b94770f1b0d72b6dddcb1ebb8123cb3712530b08cc387b349a3d8
         echo "8aca8db96f1b94770f1b0d72b6dddcb1ebb8123cb3712530b08cc387b349a3d8  $actionlint_dir/actionlint.tar.gz" | sha256sum --check --status
         tar -xzf "$actionlint_dir/actionlint.tar.gz" -C "$actionlint_dir" actionlint
         rm "$actionlint_dir/actionlint.tar.gz"
@@ -96,6 +97,10 @@ 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/actions/setup/action.yml

@@ -66,10 +66,7 @@ inputs:
 runs:
   using: "composite"
   steps:
-    # protobuf-compiler is deliberately absent: the setup-protoc step below
-    # installs 35.1 into the tool cache and prepends it to PATH, so the apt
-    # build (older, and never version-matched) was shadowed on every run and
-    # simply never used.
+    # Use the pinned code generators rather than Ubuntu's older packages.
     - name: Install system dependencies (Ubuntu)
       if: runner.os == 'Linux'
       shell: bash
@@ -79,7 +76,9 @@ runs:
           build-essential \
           pkg-config \
           libssl-dev \
-          ripgrep
+          ripgrep \
+          python3 \
+          gh
 
     # musl-gcc is needed by the native musl release leg, and zip/unzip by the
     # release packaging steps. No CI test lane touches any of them.
@@ -88,13 +87,32 @@ runs:
       shell: bash
       run: sudo apt-get install -y musl-tools zip unzip
 
+    - name: Install pinned code generators through GitHub API (Linux x64)
+      if: runner.os == 'Linux' && runner.arch == 'X64'
+      shell: bash
+      env:
+        GH_TOKEN: ${{ github.token }}
+        ASSET_DOWNLOADER: ${{ github.action_path }}/download_release_asset.py
+      run: |
+        python3 "$ASSET_DOWNLOADER" --repository protocolbuffers/protobuf --tag v35.1 --asset protoc-35.1-linux-x86_64.zip --output "$RUNNER_TEMP/protoc.zip"
+        python3 "$ASSET_DOWNLOADER" --repository google/flatbuffers --tag v25.12.19 --asset 'Linux.flatc.binary.g++-13.zip' --output "$RUNNER_TEMP/flatc.zip"
+        python3 -m zipfile -e "$RUNNER_TEMP/protoc.zip" "$RUNNER_TEMP/protoc"
+        python3 -m zipfile -e "$RUNNER_TEMP/flatc.zip" "$RUNNER_TEMP/flatc"
+        chmod +x "$RUNNER_TEMP/protoc/bin/protoc" "$RUNNER_TEMP/flatc/flatc"
+        "$RUNNER_TEMP/protoc/bin/protoc" --version
+        "$RUNNER_TEMP/flatc/flatc" --version
+        echo "$RUNNER_TEMP/protoc/bin" >> "$GITHUB_PATH"
+        echo "$RUNNER_TEMP/flatc" >> "$GITHUB_PATH"
+
     - name: Install protoc
+      if: runner.os != 'Linux' || runner.arch != 'X64'
       uses: rustfs/setup-protoc@a3705324d8f9bf5b6c3573fb6cf8ae421db55dd6 # v3.0.1
       with:
         version: "35.1"
         repo-token: ${{ github.token }}
 
     - name: Install flatc
+      if: runner.os != 'Linux' || runner.arch != 'X64'
       uses: Nugine/setup-flatc@698800de72a96bfb22cf60431dc21a2ff9a7e07b # v1
       with:
         version: "25.12.19"

.github/actions/setup/download_release_asset.py

@@ -0,0 +1,84 @@
+#!/usr/bin/env python3
+"""Download a named GitHub release asset through its API and verify its digest."""
+
+import argparse
+import hashlib
+import json
+from pathlib import Path
+import re
+import subprocess
+import sys
+import tempfile
+from urllib.parse import quote
+
+
+def download(repository, tag, name, output, expected_sha256=None):
+    if not re.fullmatch(r"[A-Za-z0-9][A-Za-z0-9_.-]*/[A-Za-z0-9][A-Za-z0-9_.-]*", repository):
+        raise ValueError("invalid release repository")
+    if not tag or not name or "/" in name or "\\" in name:
+        raise ValueError("invalid release tag or asset name")
+    if expected_sha256 is not None and not re.fullmatch(r"[0-9a-f]{64}", expected_sha256):
+        raise ValueError("invalid pinned asset digest")
+    result = subprocess.run(
+        ["gh", "api", "--hostname", "github.com", f"repos/{repository}/releases/tags/{quote(tag, safe='')}"],
+        check=True, capture_output=True, text=True, timeout=60,
+    )
+    release = json.loads(result.stdout)
+    if not isinstance(release, dict) or release.get("tag_name") != tag:
+        raise ValueError("release metadata does not match the requested tag")
+    assets = release.get("assets")
+    if not isinstance(assets, list) or any(not isinstance(asset, dict) for asset in assets):
+        raise ValueError("invalid release asset inventory")
+    matches = [asset for asset in assets if asset.get("name") == name]
+    if len(matches) != 1:
+        raise ValueError("release must contain exactly one named asset")
+    asset = matches[0]
+    asset_id, size, digest = asset.get("id"), asset.get("size"), asset.get("digest")
+    if type(asset_id) is not int or asset_id <= 0 or type(size) is not int or not 0 < size <= 512 * 1024 * 1024:
+        raise ValueError("invalid release asset identity or size")
+    if not isinstance(digest, str) or not re.fullmatch(r"sha256:[0-9a-f]{64}", digest):
+        raise ValueError("release asset has no valid SHA-256 digest")
+    sha256 = digest.removeprefix("sha256:")
+    if expected_sha256 is not None and expected_sha256 != sha256:
+        raise ValueError("release digest differs from the pinned digest")
+    output = Path(output)
+    output.parent.mkdir(parents=True, exist_ok=True)
+    temporary = None
+    try:
+        with tempfile.NamedTemporaryFile(dir=output.parent, prefix=".release-asset-", delete=False) as stream:
+            temporary = Path(stream.name)
+            # gh authenticates to the API; its HTTP client strips credentials on
+            # cross-host redirects. No browser URL or token is supplied in argv.
+            subprocess.run(
+                ["gh", "api", "--hostname", "github.com", "--header", "Accept: application/octet-stream", f"repos/{repository}/releases/assets/{asset_id}"],
+                check=True, stdout=stream, timeout=300,
+            )
+        with temporary.open("rb") as stream:
+            actual = hashlib.file_digest(stream, "sha256").hexdigest()
+        if temporary.stat().st_size != size or actual != sha256:
+            raise ValueError("downloaded release asset failed size or SHA-256 verification")
+        temporary.replace(output)
+        print(f"Verified release asset: {repository} {tag} {name} sha256={actual}")
+    finally:
+        if temporary is not None:
+            temporary.unlink(missing_ok=True)
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--repository", required=True)
+    parser.add_argument("--tag", required=True)
+    parser.add_argument("--asset", required=True)
+    parser.add_argument("--output", required=True, type=Path)
+    parser.add_argument("--sha256")
+    args = parser.parse_args()
+    try:
+        download(args.repository, args.tag, args.asset, args.output, args.sha256)
+    except (ValueError, KeyError, TypeError, json.JSONDecodeError, subprocess.SubprocessError, OSError) as error:
+        print(f"Release asset download failed: {error}", file=sys.stderr)
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

.github/workflows/ci.yml

@@ -16,7 +16,9 @@ name: Continuous Integration
 
 on:
   push:
-    branches: [ main ]
+    # 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 ]
     paths-ignore:
       - "**.md"
       - "docs/**"
@@ -142,15 +144,23 @@ jobs:
       - uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
         with:
           persist-credentials: false
+      - name: Install pinned typos through GitHub API
+        env:
+          GH_TOKEN: ${{ github.token }}
+          TYPOS_VERSION: 1.50.2
+        run: |
+          python3 .github/actions/setup/download_release_asset.py --repository crate-ci/typos --tag "v${TYPOS_VERSION}" --asset "typos-v${TYPOS_VERSION}-x86_64-unknown-linux-musl.tar.gz" --output "$RUNNER_TEMP/typos.tar.gz"
+          mkdir -p "$RUNNER_TEMP/typos"
+          tar -xzf "$RUNNER_TEMP/typos.tar.gz" -C "$RUNNER_TEMP/typos"
       - name: Typos check with custom config file
-        uses: crate-ci/typos@512fc24f32f44ab01972217aaaf3dc86ec234d53 # v1.50.2
+        run: '"$RUNNER_TEMP/typos/typos" .'
 
   # Fail early with compile-free checks for every pull request.
   quick-checks:
     name: Quick Checks
     if: github.event_name != 'pull_request' || github.event.action != 'closed'
     runs-on: ubuntu-latest
-    timeout-minutes: 10
+    timeout-minutes: 20
     steps:
       - name: Checkout repository
         uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
@@ -957,6 +967,8 @@ 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
@@ -1077,18 +1089,122 @@ 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 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 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).
     if: >-
       github.event_name == 'workflow_dispatch' ||
       github.event_name == 'merge_group' ||
-      (github.event_name == 'push' && github.ref == 'refs/heads/main')
+      (github.event_name == 'push' && (github.ref == 'refs/heads/main' || github.ref == 'refs/heads/gateway/integration'))
     needs: [ build-rustfs-debug-binary ]
     # Retain the producer image and fixture paths until the runtime pilot passes.
     runs-on: sm-standard-4
@@ -1137,11 +1253,16 @@ jobs:
 
       - name: Install mc
         env:
+          GH_TOKEN: ${{ github.token }}
           MC_VERSION: RELEASE.2025-08-13T08-35-41Z
           MC_SHA256: 01f866e9c5f9b87c2b09116fa5d7c06695b106242d829a8bb32990c00312e891
         run: |
+          if ! command -v gh >/dev/null 2>&1; then
+            sudo apt-get update
+            sudo apt-get install -y gh
+          fi
           MC_BINARY="mc.linux-amd64.${MC_VERSION}"
-          curl -fsSLo "$RUNNER_TEMP/mc" "https://github.com/minio/mc/releases/download/${MC_VERSION}/${MC_BINARY}"
+          python3 .github/actions/setup/download_release_asset.py --repository minio/mc --tag "$MC_VERSION" --asset "$MC_BINARY" --output "$RUNNER_TEMP/mc" --sha256 "$MC_SHA256"
           echo "${MC_SHA256}  $RUNNER_TEMP/mc" | sha256sum --check --status
           chmod +x "$RUNNER_TEMP/mc"
           echo "$RUNNER_TEMP" >> "$GITHUB_PATH"

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`
-     (`S3S_ECSTORE_FILES_BASELINE`; the `object_lock` module was converted to
+     (the `ecstore_files` line of `.config/s3s-footprint-baseline.txt`; 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

@@ -8697,6 +8697,15 @@ 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"
@@ -9581,7 +9590,7 @@ dependencies = [
  "proptest",
  "prost 0.14.4",
  "pyroscope",
- "quick-xml",
+ "quick-xml 0.42.0",
  "quinn",
  "rand 0.10.3",
  "rcgen",
@@ -9600,6 +9609,7 @@ dependencies = [
  "rustfs-ecstore",
  "rustfs-extension-schema",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-iam",
@@ -9837,6 +9847,7 @@ dependencies = [
  "google-cloud-auth",
  "google-cloud-storage",
  "h2",
+ "hashbrown 0.17.1",
  "hex-simd",
  "hmac 0.13.0",
  "hostname",
@@ -9863,7 +9874,7 @@ dependencies = [
  "path-absolutize",
  "pin-project-lite",
  "proptest",
- "quick-xml",
+ "quick-xml 0.42.0",
  "rand 0.10.3",
  "ratelimit",
  "rcgen",
@@ -9880,6 +9891,7 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-erasure-codec",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal-contracts",
  "rustfs-io-metrics",
  "rustfs-lifecycle",
@@ -9987,6 +9999,48 @@ 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"
@@ -10258,10 +10312,10 @@ dependencies = [
  "metrics-util",
  "proptest",
  "rustfs-config",
+ "rustfs-gateway-types",
  "rustfs-replication",
  "rustfs-scanner-metrics",
  "rustfs-storage-api",
- "s3s",
  "serial_test",
  "temp-env",
  "time",
@@ -10325,7 +10379,6 @@ dependencies = [
  "reqwest",
  "rmp-serde",
  "rustfs-signer",
- "s3s",
  "serde",
  "serde_json",
  "sysinfo",
@@ -10365,7 +10418,7 @@ dependencies = [
  "jiff",
  "metrics",
  "percent-encoding",
- "quick-xml",
+ "quick-xml 0.42.0",
  "rayon",
  "rustc-hash",
  "rustfs-config",
@@ -10547,10 +10600,13 @@ 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",
@@ -10558,13 +10614,13 @@ 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",
@@ -10614,7 +10670,7 @@ dependencies = [
  "regex",
  "rmp",
  "rmp-serde",
- "s3s",
+ "rustfs-gateway-types",
  "serde",
  "time",
  "tokio",
@@ -10709,11 +10765,13 @@ dependencies = [
  "hyper-util",
  "lazy_static",
  "md-5 0.11.0",
- "quick-xml",
+ "quick-xml 0.42.0",
  "rand 0.10.3",
  "rustfs-checksums",
  "rustfs-config",
+ "rustfs-gateway-types",
  "rustfs-rio",
+ "rustfs-s3-types",
  "rustfs-signer",
  "rustfs-storage-api",
  "rustfs-tls-runtime",
@@ -10747,9 +10805,19 @@ 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]]
@@ -10759,23 +10827,29 @@ dependencies = [
  "arc-swap",
  "async-compression",
  "async-trait",
+ "base64-simd",
+ "brotli 9.1.1",
  "bytes",
  "chrono",
  "crc-fast",
  "datafusion",
+ "flatbuffers",
  "flate2",
  "futures",
  "futures-core",
  "hotpath",
  "http 1.5.0",
+ "lz4",
+ "lz4_flex",
  "memchr",
  "metrics",
  "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",
@@ -10803,7 +10877,6 @@ dependencies = [
  "parking_lot 0.13.0",
  "rustfs-s3select-api",
  "rustfs-test-utils",
- "s3s",
  "serde",
  "serde_json",
  "tokio",
@@ -10835,6 +10908,7 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-ecstore",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-lifecycle",
@@ -10892,9 +10966,10 @@ dependencies = [
  "bytes",
  "hotpath",
  "http 1.5.0",
+ "http-body",
  "hyper",
+ "rustfs-s3-types",
  "rustfs-utils",
- "s3s",
  "serde_urlencoded",
  "thiserror 2.0.21",
  "time",
@@ -10953,7 +11028,6 @@ dependencies = [
  "rustfs-utils",
  "rustls",
  "rustls-native-certs",
- "s3s",
  "serde",
  "serde_json",
  "sha2 0.11.1",
@@ -11345,7 +11419,7 @@ dependencies = [
  "numeric_cast",
  "phf 0.14.0",
  "pin-project-lite",
- "quick-xml",
+ "quick-xml 0.42.0",
  "regex",
  "s3s-multipart",
  "s3s-rfc2047",

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 type definitions
+    "crates/s3-types", # S3 event and error type definitions
     "crates/s3-ops", # S3 operation definitions and mapping
     "crates/s3select-api", # S3 Select API interface
     "crates/s3select-query", # S3 Select query engine
@@ -143,6 +143,13 @@ 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" }
@@ -296,6 +303,7 @@ local-ip-address = "0.6.13"
 log = "0.4"
 memmap2 = "0.9.12"
 lz4 = "1.28.1"
+lz4_flex = { version = "=0.14.0", default-features = false, features = ["std", "frame"] }
 matchit = "0.9.2"
 md-5 = "0.11.0"
 mime_guess = "2.0.5"
@@ -328,6 +336,10 @@ 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"
@@ -336,6 +348,7 @@ 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 = "0.40.1"
 temp-env = "0.3.6"
 tempfile = "3.27.0"

crates/config/src/constants/internode.rs

@@ -184,15 +184,27 @@ const _: () = assert!(!DEFAULT_INTERNODE_RPC_REPLAY_SCOPE_STRICT);
 /// Capacity (distinct nonces) of the process-local internode RPC replay cache that enforces
 /// one-time consumption of authenticated RPC signatures.
 ///
-/// The cache retains each nonce for the ~10-minute signature freshness envelope. Once peers use
-/// replay-scoped v3 authentication, every authenticated RPC consumes one entry, so the steady
-/// state holds roughly `authenticated RPC RPS x 601s` entries. This default is the minimum floor:
-/// explicit operator values and resource-aware auto sizing both clamp upward to at least this
-/// value. Overflow fails closed — legitimate signed traffic is the only thing that can fill the
-/// cache (replays are rejected before insertion, and an attacker cannot mint valid nonces without
-/// the shared secret) — and increments
-/// `rustfs_system_network_internode_replay_cache_overflow_total`, so a sustained non-zero overflow
-/// counter means this capacity is undersized for the node's peak authenticated RPC rate.
+/// Nonces remain protected until their signed timestamp is outside the observed wall-clock
+/// validity window, including after clock rollback. Current-time signatures live about five
+/// minutes; the maximum accepted forward clock skew can extend that to ten minutes. A body-bound
+/// v3 request consumes both its v3 and compatibility-v2 nonces, so size for receiver nonce intake,
+/// not S3 request rate or CPU count.
+///
+/// Automatic admission budgets 13% of effective process memory for signed-time
+/// hash tables, allocator overhead and simultaneous old/new tables during growth.
+/// The reported count is an absolute UUID payload ceiling, not guaranteed throughput:
+/// actual byte admission can reject earlier depending on timestamp distribution.
+/// Tables grow with demand and are released together when their signed time expires.
+/// CPU count, a historical minimum, and a fixed 32M ceiling do not constrain this budget.
+/// If memory detection fails, the historical default remains a bounded count fallback.
+///
+/// Explicit values remain count ceilings with the historical minimum of 1,048,576
+/// for configuration compatibility, and bypass automatic byte admission. That
+/// compatibility floor does not apply to the automatically detected memory budget.
+/// Zero never disables replay protection.
+/// Overflow rejects the request without evicting a still-valid identity, increments
+/// `rustfs_system_network_internode_replay_cache_overflow_total`, and returns resource
+/// exhaustion rather than invalid credentials. Measure headroom before overriding.
 pub const ENV_INTERNODE_RPC_REPLAY_CACHE_CAPACITY: &str = "RUSTFS_INTERNODE_RPC_REPLAY_CACHE_CAPACITY";
 pub const DEFAULT_INTERNODE_RPC_REPLAY_CACHE_CAPACITY: usize = 1_048_576;
 

crates/config/src/constants/object.rs

@@ -32,8 +32,9 @@ pub const ENV_OBJECT_MEDIUM_CONCURRENCY_THRESHOLD: &str = "RUSTFS_OBJECT_MEDIUM_
 
 /// Environment variable name for maximum concurrent disk reads for object operations.
 /// - Purpose: Limit the number of concurrent disk read operations for object reads to prevent I/O saturation.
-/// - Unit: request count (usize).
-/// - Semantics: Throttling disk reads helps maintain overall system responsiveness under load.
+/// - Unit: request count (usize) for GET; estimated drive fanout for native LIST.
+/// - Semantics: GET bodies and native LIST walks use separate per-instance budgets of this size.
+///   A LIST consumes up to one budget, including its disk-producer cleanup. `0` disables both caps.
 /// - Example: `export RUSTFS_OBJECT_MAX_CONCURRENT_DISK_READS=16`
 /// - Note: This setting may interact with OS-level I/O scheduling and should be tuned based on hardware capabilities.
 pub const ENV_OBJECT_MAX_CONCURRENT_DISK_READS: &str = "RUSTFS_OBJECT_MAX_CONCURRENT_DISK_READS";

crates/e2e_test/src/anonymous_access_test.rs

@@ -17,7 +17,8 @@
 //! when PublicAccessBlock configuration is missing or explicitly set.
 
 use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
-use aws_sdk_s3::types::PublicAccessBlockConfiguration;
+use crate::sts_query_compat_test::{OPA_AUTH_TOKEN, OpaMock};
+use aws_sdk_s3::types::{PublicAccessBlockConfiguration, Tag, Tagging};
 use tracing::info;
 
 async fn setup_public_bucket(
@@ -69,6 +70,110 @@ 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 tag keys are ambiguous, so PutBucketTagging rejects them
+    // before they can reach stored metadata or the OPA tag-loading path.
+    let duplicate = 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
+        .expect_err("duplicate bucket tag keys must be rejected");
+    assert_eq!(duplicate.into_service_error().meta().code(), Some("InvalidTag"));
+    client
+        .put_bucket_tagging()
+        .bucket(bucket)
+        .tagging(
+            Tagging::builder()
+                .tag_set(Tag::builder().key("department").value("finance").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/dual_stack_differential/compare.rs

@@ -0,0 +1,512 @@
+// 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

@@ -0,0 +1,199 @@
+// 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

@@ -31,6 +31,12 @@ 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

@@ -32,7 +32,9 @@
 //! 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};
+use crate::common::{
+    AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via, signed_s3_request,
+};
 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;
@@ -774,6 +776,81 @@ 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/table_location_authorization_test.rs

@@ -114,10 +114,10 @@ fn staged_commit(metadata: &Value, location: &str) -> Value {
     })
 }
 
-async fn location_claim_preserves_other_users_objects_after_drop_and_restart(backing: &str) -> TestResult {
+async fn location_claim_preserves_other_users_objects_after_drop_and_restart() -> TestResult {
     init_logging();
     let mut env = RustFSTestEnvironment::new().await?;
-    let server_env = [("RUSTFS_TABLE_CATALOG_BACKING", backing)];
+    let server_env = [("RUSTFS_TABLE_CATALOG_BACKING", "object")];
     env.start_rustfs_server_without_cleanup_with_env(&[("RUSTFS_TABLE_CATALOG_BACKING", "object")])
         .await?;
     env.wait_for_server_ready().await?;
@@ -147,39 +147,6 @@ async fn location_claim_preserves_other_users_objects_after_drop_and_restart(bac
         StatusCode::OK,
     )
     .await?;
-    if backing == "durable-strong" {
-        catalog_request(
-            &env,
-            &root,
-            Method::POST,
-            &format!("/iceberg/v1/{bucket}/catalog/warehouse-index/backfill"),
-            None,
-            StatusCode::NO_CONTENT,
-        )
-        .await?;
-        let preflight = catalog_request(
-            &env,
-            &root,
-            Method::GET,
-            &format!("/iceberg/v1/{bucket}/catalog/migration"),
-            None,
-            StatusCode::OK,
-        )
-        .await?;
-        assert_eq!(preflight["blockers"], json!([]), "{preflight}");
-        let migration = catalog_request(
-            &env,
-            &root,
-            Method::POST,
-            &format!("/iceberg/v1/{bucket}/catalog/migration"),
-            None,
-            StatusCode::OK,
-        )
-        .await?;
-        assert_eq!(migration["ready_to_enable_durable_strong"], json!(true), "{migration}");
-        env.restart_server_preserving_data(vec![], &server_env).await?;
-        env.wait_for_server_ready().await?;
-    }
     admin_create_user(&env, &creator.0, &creator.1).await?;
     admin_create_user(&env, &reader.0, &reader.1).await?;
     let create_statement = json!({
@@ -341,7 +308,7 @@ async fn location_claim_preserves_other_users_objects_after_drop_and_restart(bac
         StatusCode::FORBIDDEN,
     )
     .await?;
-    if backing == "object" {
+    {
         catalog_request(
             &env,
             &creator,
@@ -408,10 +375,5 @@ async fn location_claim_preserves_other_users_objects_after_drop_and_restart(bac
 
 #[tokio::test]
 async fn object_catalog_location_claim_preserves_objects_after_drop_and_restart() -> TestResult {
-    location_claim_preserves_other_users_objects_after_drop_and_restart("object").await
-}
-
-#[tokio::test]
-async fn durable_strong_catalog_location_claim_preserves_objects_after_drop_and_restart() -> TestResult {
-    location_claim_preserves_other_users_objects_after_drop_and_restart("durable-strong").await
+    location_claim_preserves_other_users_objects_after_drop_and_restart().await
 }

crates/ecstore/Cargo.toml

@@ -121,6 +121,7 @@ test-util = []
 e2e-test-hooks = []
 
 [dependencies]
+hashbrown.workspace = true
 hotpath.workspace = true
 rustfs-filemeta.workspace = true
 rustfs-utils = { workspace = true, features = ["full"] }
@@ -140,6 +141,7 @@ 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
@@ -158,7 +160,6 @@ 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 }
@@ -252,6 +253,7 @@ 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

@@ -66,6 +66,8 @@ 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,
@@ -137,7 +139,7 @@ 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,
-                object_opts_from_object_info,
+                format_lifecycle_date, object_opts_from_object_info, parse_lifecycle_date,
             };
         }
 
@@ -240,7 +242,7 @@ pub mod bucket {
     }
 
     pub mod object_lock {
-        pub use crate::bucket::object_lock::{ObjectLockApi, ObjectLockStatusExt};
+        pub use crate::bucket::object_lock::ObjectLockApi;
 
         pub mod types {
             pub use crate::bucket::object_lock::types::{
@@ -335,9 +337,9 @@ pub mod bucket {
 
     pub mod utils {
         pub use crate::bucket::utils::{
-            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, has_xml_invalid_control_characters,
-            is_meta_bucketname, is_valid_object_prefix, serialize,
+            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,
+            has_xml_invalid_control_characters, is_meta_bucketname, is_valid_object_prefix, serialize,
         };
     }
 
@@ -364,7 +366,7 @@ pub mod capacity {
 }
 
 pub mod object_api_utils {
-    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader, to_s3s_etag};
+    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader};
 }
 
 pub mod cluster {
@@ -640,17 +642,17 @@ pub mod rpc {
     pub use crate::cluster::rpc::{
         AuthenticatedChannel, KMS_SIGNAL_SUBSYSTEM, LocalPeerS3Client, MAX_NETWORK_PROBE_BYTES, MAX_NETWORK_PROBE_DURATION,
         NetworkPeerProbeClient, NetworkPeerProbeError, NetworkPeerProbeMeasurement, NetworkPeerTarget, PEER_RESTDRY_RUN,
-        PEER_RESTSIGNAL, PEER_RESTSUB_SYS, PeerRestClient, PeerS3Client, S3PeerSys, SERVICE_SIGNAL_REFRESH_CONFIG,
-        SERVICE_SIGNAL_RELOAD_DYNAMIC, ScannerBucketListing, ScannerDirtyUsageAcknowledgement, ScannerPeerActivity,
-        ScannerPeerDirtyUsageBucket, ScannerPeerDirtyUsageSnapshot, ScannerPublicationLease, ScannerScopedDirtyUsageAckEntry,
-        TONIC_RPC_PREFIX, TonicInterceptor, build_put_file_auth_trailer, check_and_record_signed_rpc_nonce,
-        decode_heal_bucket_rpc_options, encode_heal_bucket_rpc_options, gen_signature_headers, gen_tonic_replay_scope_headers,
-        gen_tonic_signature_headers, gen_tonic_signature_interceptor, node_service_time_out_client,
-        node_service_time_out_client_no_auth, normalize_tonic_rpc_audience, rpc_background_admission,
-        set_tonic_canonical_body_digest, sign_ns_scanner_capability, sign_ns_scanner_capability_with_checkpoint_coverage,
-        sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability, sign_tonic_rpc_response_proof,
-        tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason,
-        verify_ns_scanner_capability, verify_ns_scanner_capability_with_checkpoint_coverage,
+        PEER_RESTSIGNAL, PEER_RESTSUB_SYS, PeerRestClient, PeerS3Client, RpcReplayCacheAdmissionError, S3PeerSys,
+        SERVICE_SIGNAL_REFRESH_CONFIG, SERVICE_SIGNAL_RELOAD_DYNAMIC, ScannerBucketListing, ScannerDirtyUsageAcknowledgement,
+        ScannerPeerActivity, ScannerPeerDirtyUsageBucket, ScannerPeerDirtyUsageSnapshot, ScannerPublicationLease,
+        ScannerScopedDirtyUsageAckEntry, TONIC_RPC_PREFIX, TonicInterceptor, build_put_file_auth_trailer,
+        check_and_record_signed_rpc_nonce, decode_heal_bucket_rpc_options, encode_heal_bucket_rpc_options, gen_signature_headers,
+        gen_tonic_replay_scope_headers, gen_tonic_signature_headers, gen_tonic_signature_interceptor,
+        node_service_time_out_client, node_service_time_out_client_no_auth, normalize_tonic_rpc_audience,
+        rpc_auth_is_resource_exhausted, rpc_background_admission, set_tonic_canonical_body_digest, sign_ns_scanner_capability,
+        sign_ns_scanner_capability_with_checkpoint_coverage, sign_ns_scanner_capability_with_tier_registry_generation,
+        sign_put_file_capability, sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers,
+        tonic_rpc_auth_failure_reason, verify_ns_scanner_capability, verify_ns_scanner_capability_with_checkpoint_coverage,
         verify_ns_scanner_capability_with_tier_registry_generation, verify_put_file_auth_trailer, verify_put_file_capability,
         verify_rpc_signature, verify_tonic_boot_epoch_response, verify_tonic_canonical_body_digest,
         verify_tonic_mutation_body_digest, verify_tonic_mutation_body_digest_reject_unsigned, verify_tonic_rpc_response_proof,

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

@@ -88,12 +88,14 @@ 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};
@@ -300,7 +302,7 @@ impl LifecycleSys {
         Arc::new(Self)
     }
 
-    pub async fn get(&self, bucket: &str) -> Option<BucketLifecycleConfiguration> {
+    pub async fn get(&self, bucket: &str) -> Option<PersistedLifecycleConfiguration> {
         match metadata_boundary::get_lifecycle_config(bucket).await {
             Ok((lc, _)) => Some(lc),
             Err(Error::ConfigNotFound) => None,
@@ -3557,7 +3559,7 @@ async fn stale_upload_lifecycle_due(
         Err(_) => return None,
     };
 
-    let has_size_filter = |filter: &s3s::dto::LifecycleRuleFilter| {
+    let has_size_filter = |filter: &PersistedLifecycleFilter| {
         filter.object_size_greater_than.is_some()
             || filter.object_size_less_than.is_some()
             || filter
@@ -3566,7 +3568,7 @@ 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() != ExpirationStatus::DISABLED
+        rule.status.as_str() != "Disabled"
             && rule.abort_incomplete_multipart_upload.is_some()
             && rule.filter.as_ref().is_some_and(has_size_filter)
     });
@@ -4000,7 +4002,7 @@ pub fn init_background_stale_multipart_upload_cleanup(api: Arc<ECStore>) {
     });
 }
 
-pub async fn validate_transition_tier(lc: &BucketLifecycleConfiguration) -> Result<(), std::io::Error> {
+pub async fn validate_transition_tier(lc: &PersistedLifecycleConfiguration) -> Result<(), std::io::Error> {
     for rule in &lc.rules {
         if let Some(transitions) = &rule.transitions {
             for transition in transitions {
@@ -4117,8 +4119,9 @@ 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(lock_config.clone())
+        .with_lock_retention(evaluator_lock_config)
         .eval(&object_opts)
         .await
     {
@@ -4598,19 +4601,26 @@ pub async fn enqueue_transition_for_existing_objects_scoped(
     }
 }
 
-fn lifecycle_rule_has_date_expiration(lc: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
+fn lifecycle_rule_has_date_expiration(lc: &PersistedLifecycleConfiguration, rule_id: &str) -> bool {
     lc.rules.iter().any(|rule| {
-        rule.status == ExpirationStatus::from_static(ExpirationStatus::ENABLED)
+        rule.status == "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: &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
-    })
+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
+            }
+        })
 }
 
 async fn apply_existing_object_expiry(
@@ -4630,7 +4640,7 @@ async fn apply_existing_object_expiry(
 struct ExistingObjectExpiryContext<'a> {
     api: Arc<ECStore>,
     bucket: &'a str,
-    lc: Arc<BucketLifecycleConfiguration>,
+    lc: Arc<PersistedLifecycleConfiguration>,
     lock_config: Option<Arc<ObjectLockConfiguration>>,
     bucket_incarnation_id: Uuid,
     src: &'a LcEventSrc,
@@ -4650,8 +4660,9 @@ 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(context.lock_config.clone())
+        .with_lock_retention(evaluator_lock_config)
         .eval(&object_opts)
         .await
     {
@@ -4845,7 +4856,7 @@ fn manual_transition_version_marker(oi: &ObjectInfo) -> String {
 async fn enqueue_transition_with_lifecycle_report(
     api: Option<Arc<ECStore>>,
     oi: &ObjectInfo,
-    lc: &BucketLifecycleConfiguration,
+    lc: &PersistedLifecycleConfiguration,
     src: &LcEventSrc,
     options: &ManualTransitionRunOptions,
     report: &mut ManualTransitionRunReport,
@@ -4911,7 +4922,7 @@ async fn enqueue_transition_with_lifecycle_report(
     false
 }
 
-async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &BucketLifecycleConfiguration, src: &LcEventSrc) -> bool {
+async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &PersistedLifecycleConfiguration, src: &LcEventSrc) -> bool {
     let event = lc.eval(&oi.to_lifecycle_opts()).await;
     match event.action {
         IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
@@ -5425,12 +5436,8 @@ pub async fn post_restore_opts(version_id: &str, bucket: &str, object: &str) ->
     })
 }
 
-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)
-    {
+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") {
         return Ok(None);
     }
     let output_location = rreq
@@ -5472,21 +5479,20 @@ 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() == RestoreRequestType::SELECT)
-    {
+    if rreq.type_.as_ref().is_some_and(|type_| type_.as_str() == "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.
     if let Some(expires) = oi.expires {
         meta.insert(
             "expires".to_string(),
@@ -5495,12 +5501,11 @@ pub async fn put_restore_opts(
                 .map_err(std::io::Error::other)?,
         );
     }
-    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(false),
-            restore_expiry_date: Some(restore_expiry),
+            is_restore_in_progress: Some(true),
+            restore_expiry_date: None,
         }
         .to_string(),
     );
@@ -5514,6 +5519,7 @@ 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,
         ..Default::default()
     })
 }
@@ -5532,6 +5538,18 @@ 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)
     }
 }
@@ -5543,25 +5561,25 @@ 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() != RestoreRequestType::SELECT) && self.select_parameters.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != "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() == RestoreRequestType::SELECT
+            && type_.as_str() == "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() != RestoreRequestType::SELECT) && self.output_location.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != "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() == RestoreRequestType::SELECT
+            && type_.as_str() == "SELECT"
             && self.days.is_some_and(|d| d > 0)
         {
             return Err(std::io::Error::other("Days cannot be specified with SELECT restore request"));
@@ -5603,7 +5621,7 @@ pub fn lifecycle_version_delete_target(oi: &ObjectInfo) -> Option<ObjectToDelete
 }
 
 pub async fn eval_action_from_lifecycle(
-    lc: &BucketLifecycleConfiguration,
+    lc: &PersistedLifecycleConfiguration,
     lock_config: Option<&ObjectLockConfiguration>,
     oi: &ObjectInfo,
 ) -> lifecycle::Event {
@@ -6308,14 +6326,19 @@ 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;
@@ -9341,7 +9364,7 @@ mod tests {
             output_location,
             select_parameters: None,
             tier: None,
-            type_: Some(RestoreRequestType::from_static(RestoreRequestType::SELECT)),
+            type_: Some("SELECT".to_owned()),
         }
     }
 
@@ -9658,12 +9681,12 @@ mod tests {
 
     #[test]
     fn lifecycle_rule_has_date_expiration_detects_enabled_date_rule() {
-        let lc = BucketLifecycleConfiguration {
+        let lc = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
-                    date: Some(Timestamp::from(OffsetDateTime::now_utc())),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
+                    date: Some(lifecycle::format_lifecycle_date(OffsetDateTime::now_utc()).unwrap()),
                     ..Default::default()
                 }),
                 id: Some("rule-date".to_string()),
@@ -10180,14 +10203,15 @@ mod tests {
     #[test]
     fn should_defer_date_expiry_for_recent_config_update_respects_grace_window() {
         let now = OffsetDateTime::now_utc();
-        let recent = BucketLifecycleConfiguration {
-            expiry_updated_at: Some(Timestamp::from(now - time::Duration::seconds(1))),
+        let recent = PersistedLifecycleConfiguration {
+            expiry_updated_at: Some(lifecycle::format_lifecycle_date(now - time::Duration::seconds(1)).unwrap()),
             rules: Vec::new(),
         };
-        let stale = BucketLifecycleConfiguration {
-            expiry_updated_at: Some(Timestamp::from(
-                now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1),
-            )),
+        let stale = PersistedLifecycleConfiguration {
+            expiry_updated_at: Some(
+                lifecycle::format_lifecycle_date(now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1))
+                    .unwrap(),
+            ),
             rules: Vec::new(),
         };
 
@@ -10315,12 +10339,12 @@ mod tests {
         assert_eq!(deleted.version_id, None);
     }
 
-    fn expired_delete_marker_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn expired_delete_marker_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     expired_object_delete_marker: Some(true),
                     ..Default::default()
                 }),
@@ -10336,12 +10360,12 @@ mod tests {
         }
     }
 
-    fn latest_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn latest_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     ..Default::default()
                 }),
@@ -10357,12 +10381,12 @@ mod tests {
         }
     }
 
-    fn current_and_noncurrent_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn current_and_noncurrent_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(30),
                     ..Default::default()
                 }),
@@ -10370,7 +10394,7 @@ mod tests {
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("expire-current-and-noncurrent".to_string()),
-                noncurrent_version_expiration: Some(NoncurrentVersionExpiration {
+                noncurrent_version_expiration: Some(PersistedNoncurrentVersionExpiration {
                     noncurrent_days: Some(1),
                     newer_noncurrent_versions: None,
                 }),
@@ -10381,12 +10405,12 @@ mod tests {
         }
     }
 
-    fn all_versions_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn all_versions_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     expired_object_all_versions: Some(true),
                     ..Default::default()
@@ -10405,29 +10429,29 @@ mod tests {
 
     fn lock_enabled_without_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         }
     }
 
     fn lock_enabled_with_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(30),
-                    mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
+                    mode: Some("COMPLIANCE".to_owned()),
                     years: None,
                 }),
             }),
         }
     }
 
-    fn latest_transition_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn latest_transition_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -10436,10 +10460,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![Transition {
+                transitions: Some(vec![PersistedTransition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some(TransitionStorageClass::from_static("WARM")),
+                    storage_class: Some("WARM".to_owned()),
                 }]),
             }],
         }
@@ -12976,8 +13000,9 @@ 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(lock_config.clone().map(Arc::new))
+                .with_lock_retention(evaluator_lock_config)
                 .eval(&[object_opts])
                 .await
                 .expect("batch lifecycle evaluation should succeed")
@@ -14234,10 +14259,7 @@ mod tests {
         let object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (
-                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                    s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                ),
+                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14249,7 +14271,7 @@ mod tests {
 
     #[tokio::test]
     async fn restored_copy_expiry_is_not_blocked_by_retention() {
-        let lifecycle = BucketLifecycleConfiguration {
+        let lifecycle = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
             rules: Vec::new(),
         };
@@ -14259,10 +14281,7 @@ mod tests {
         let mut object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (
-                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                    ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                ),
+                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14479,7 +14498,7 @@ mod tests {
                 "removed" => {
                     metadata_sys::delete(&bucket, BUCKET_LIFECYCLE_CONFIG).await.unwrap();
                 }
-                "disabled" => changed.rules[0].status = ExpirationStatus::from_static(ExpirationStatus::DISABLED),
+                "disabled" => changed.rules[0].status = "Disabled".to_owned(),
                 "longer-days" => {
                     changed.rules[0]
                         .noncurrent_version_expiration
@@ -14595,7 +14614,7 @@ mod tests {
         );
     }
 
-    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<BucketLifecycleConfiguration> {
+    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<PersistedLifecycleConfiguration> {
         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
@@ -15288,7 +15307,11 @@ mod tests {
     #[tokio::test]
     #[serial]
     async fn tier_free_version_recovery_production_entrypoint_enqueues_seeded_item() {
-        let (disk_paths, ecstore) = setup_test_env().await;
+        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 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,
+    expected_expiry_time, expiration_action_has_valid_target, format_lifecycle_date, parse_lifecycle_date,
 };
 
 pub fn object_opts_from_object_info(oi: &ObjectInfo) -> ObjectOpts {

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

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

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

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

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

@@ -14,7 +14,7 @@
 
 use std::sync::Arc;
 
-use s3s::dto::BucketLifecycleConfiguration;
+use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
 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<BucketLifecycleConfiguration> {
+pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<PersistedLifecycleConfiguration> {
     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, ObjectLockStatusExt};
+use super::object_lock::ObjectLockApi;
 use super::versioning::VersioningApi;
 use super::{quota::BucketQuota, target::BucketTargets};
 use crate::bucket::replication::invalid_replication_config_status_field;
@@ -25,12 +25,17 @@ use crate::object_api::WriteCompletion;
 use crate::runtime::sources as runtime_sources;
 use crate::store::ECStore;
 use byteorder::{BigEndian, ByteOrder, LittleEndian};
-use rustfs_policy::policy::BucketPolicy;
-use s3s::dto::{
-    AccelerateConfiguration, BucketLifecycleConfiguration, BucketLoggingStatus, BucketVersioningStatus, CORSConfiguration,
-    NotificationConfiguration, ObjectLockConfiguration, PublicAccessBlockConfiguration, ReplicationConfiguration,
-    RequestPaymentConfiguration, ServerSideEncryptionConfiguration, Tagging, VersioningConfiguration, WebsiteConfiguration,
+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 serde::Serializer;
 use sha2::{Digest, Sha256};
 use std::collections::HashMap;
@@ -40,23 +45,11 @@ use time::{Date, OffsetDateTime, PrimitiveDateTime, Time as CivilTime, UtcOffset
 use tracing::error;
 use uuid::Uuid;
 
-// 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).
+// The persisted Object Lock configuration answers the storage-level trait
+// here because this module owns the persisted `ObjectLockConfiguration`.
 impl ObjectLockApi for ObjectLockConfiguration {
     fn enabled(&self) -> bool {
-        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
-        )
+        self.object_lock_enabled.as_deref() == Some("Enabled")
     }
 }
 
@@ -257,12 +250,14 @@ pub const BUCKET_NOTIFICATION_CONFIG: &str = "notification.xml";
 pub const BUCKET_LIFECYCLE_CONFIG: &str = "lifecycle.xml";
 pub const BUCKET_SSECONFIG: &str = "bucket-encryption.xml";
 /// Validate tag entries supplied to the S3 API.
-pub fn validate_tag_set_entries(tag_set: &[s3s::dto::Tag]) -> std::result::Result<(), &'static str> {
+pub fn validate_tag_set_entries(
+    tag_set: &[rustfs_gateway_types::cors_tagging::PersistedTag],
+) -> std::result::Result<(), &'static str> {
     validate_tag_entries(tag_set, validate_s3_tag_entry)
 }
 
 fn validate_tag_entries(
-    tag_set: &[s3s::dto::Tag],
+    tag_set: &[rustfs_gateway_types::cors_tagging::PersistedTag],
     mut validate_entry: impl FnMut(&str, &str) -> std::result::Result<(), &'static str>,
 ) -> std::result::Result<(), &'static str> {
     let mut keys = std::collections::HashSet::with_capacity(tag_set.len());
@@ -323,7 +318,9 @@ pub fn is_swift_bucket_metadata_tag_key(key: &str) -> bool {
         || key.starts_with(SWIFT_CONTAINER_META_TAG_PREFIX)
 }
 
-fn validate_persisted_bucket_tags(tags: &[s3s::dto::Tag]) -> std::result::Result<(), &'static str> {
+fn validate_persisted_bucket_tags(
+    tags: &[rustfs_gateway_types::cors_tagging::PersistedTag],
+) -> std::result::Result<(), &'static str> {
     const SWIFT_ACL_MAX_BYTES: usize = 8192;
     const SWIFT_META_NAME_MAX_BYTES: usize = 128;
     const MAX_PERSISTED_TAGS: usize = MAX_BUCKET_TAGS + 2 * MAX_SWIFT_METADATA_COUNT + 4;
@@ -561,6 +558,8 @@ 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>,
@@ -609,13 +608,13 @@ pub struct BucketMetadata {
 
     pub policy_config: Option<BucketPolicy>,
     pub notification_config: Option<NotificationConfiguration>,
-    pub lifecycle_config: Option<BucketLifecycleConfiguration>,
+    pub lifecycle_config: Option<PersistedLifecycleConfiguration>,
     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<ReplicationConfiguration>,
+    pub replication_config: Option<PersistedReplicationConfiguration>,
     pub bucket_target_config: Option<BucketTargets>,
     pub bucket_target_config_meta: Option<HashMap<String, String>>,
     pub cors_config: Option<CORSConfiguration>,
@@ -636,6 +635,7 @@ 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(),
@@ -874,7 +874,10 @@ 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)?,
+                "BucketCreationCommitted" => {
+                    self.bucket_creation_committed = read_msgp_bool(rd)?;
+                    self.bucket_creation_commit_record_present = true;
+                }
                 other => {
                     tracing::debug!(field = %other, "BucketMetadata decode_from: skipping unknown field");
                     skip_msgp_value(rd)?;
@@ -887,8 +890,10 @@ impl BucketMetadata {
 
     /// Encode to msgp bytes. Field order follows MinIO BucketMetadata for compatibility.
     pub fn encode_to<W: Write>(&self, wr: &mut W) -> Result<()> {
-        // Map size: MinIO fields (25) + RustFS extensions (22)
-        let map_len: u32 = 47;
+        // 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);
         rmp::encode::write_map_len(wr, map_len)?;
 
         // MinIO field order (same as Go struct)
@@ -969,8 +974,10 @@ 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)?;
-        rmp::encode::write_str(wr, "BucketCreationCommitted")?;
-        rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
+        if commit_record_present {
+            rmp::encode::write_str(wr, "BucketCreationCommitted")?;
+            rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
+        }
 
         Ok(())
     }
@@ -1147,9 +1154,11 @@ impl BucketMetadata {
                     None
                 } else {
                     let config = deserialize::<VersioningConfiguration>(&data)?;
-                    if config.status.as_ref().is_some_and(|status| {
-                        !matches!(status.as_str(), BucketVersioningStatus::ENABLED | BucketVersioningStatus::SUSPENDED)
-                    }) {
+                    if config
+                        .status
+                        .as_ref()
+                        .is_some_and(|status| !matches!(status.as_str(), "Enabled" | "Suspended"))
+                    {
                         return Err(Error::other("bucket versioning configuration has an invalid status"));
                     }
                     Some(config)
@@ -1162,7 +1171,7 @@ impl BucketMetadata {
                 let config = if data.is_empty() {
                     None
                 } else {
-                    let config = deserialize::<ReplicationConfiguration>(&data)?;
+                    let config = deserialize::<PersistedReplicationConfiguration>(&data)?;
                     if let Some(field) = invalid_replication_config_status_field(&config) {
                         return Err(Error::other(format!("replication field {field} has an invalid status")));
                     }
@@ -1356,8 +1365,8 @@ impl BucketMetadata {
         }
         self.lifecycle_config = None;
         if !self.lifecycle_config_xml.is_empty()
-            && let Err(e) =
-                deserialize::<BucketLifecycleConfiguration>(&self.lifecycle_config_xml).map(|c| self.lifecycle_config = Some(c))
+            && let Err(e) = deserialize::<PersistedLifecycleConfiguration>(&self.lifecycle_config_xml)
+                .map(|c| self.lifecycle_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1443,8 +1452,8 @@ impl BucketMetadata {
         }
         self.replication_config = None;
         if !self.replication_config_xml.is_empty()
-            && let Err(e) =
-                deserialize::<ReplicationConfiguration>(&self.replication_config_xml).map(|c| self.replication_config = Some(c))
+            && let Err(e) = deserialize::<PersistedReplicationConfiguration>(&self.replication_config_xml)
+                .map(|c| self.replication_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1824,6 +1833,13 @@ 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());
     }
@@ -1841,6 +1857,7 @@ 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);
     }
@@ -1912,7 +1929,7 @@ mod test {
 
     #[test]
     fn persisted_bucket_tags_keep_separate_s3_and_swift_limits() {
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let tag = |key: String, value: &str| Tag {
             key: Some(key),
@@ -2004,17 +2021,15 @@ 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(s3s::dto::BucketVersioningStatus::from_static("Enabled")),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         });
         assert!(!bm.object_locking());
 
         bm.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             ..Default::default()
         });
         assert!(bm.object_locking());

crates/ecstore/src/bucket/metadata_sys.rs

@@ -31,14 +31,18 @@ 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::{
@@ -169,7 +173,7 @@ 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(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+                object_lock_enabled: Some("Enabled".to_owned()),
                 rule: None,
             },
             updated_at: bm.object_lock_config_updated_at,
@@ -213,17 +217,10 @@ pub(crate) fn configured_object_lock_state_for_tests(
 ) -> ObjectLockConfigState {
     ObjectLockConfigState::Configured {
         config: ObjectLockConfiguration {
-            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
-                        }
-                    })),
+            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()),
                     days: Some(days),
                     years: None,
                 }),
@@ -234,7 +231,7 @@ pub(crate) fn configured_object_lock_state_for_tests(
 }
 
 fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -> Result<()> {
-    if config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str) != Some(ObjectLockEnabled::ENABLED) {
+    if config.object_lock_enabled.as_deref() != Some("Enabled") {
         return Err(Error::other("persisted bucket Object Lock enabled state is invalid"));
     }
 
@@ -247,7 +244,7 @@ fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -
     if !retention
         .mode
         .as_ref()
-        .is_some_and(|mode| matches!(mode.as_str(), ObjectLockRetentionMode::COMPLIANCE | ObjectLockRetentionMode::GOVERNANCE))
+        .is_some_and(|mode| matches!(mode.as_str(), "COMPLIANCE" | "GOVERNANCE"))
     {
         return Err(Error::other("persisted bucket Object Lock retention mode is invalid"));
     }
@@ -1641,7 +1638,7 @@ 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<(BucketLifecycleConfiguration, OffsetDateTime)> {
+pub async fn get_lifecycle_config(bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1917,7 +1914,7 @@ 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<(ReplicationConfiguration, OffsetDateTime)> {
+pub async fn get_replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1927,7 +1924,7 @@ pub async fn get_replication_config(bucket: &str) -> Result<(ReplicationConfigur
 pub(crate) async fn get_replication_config_in(
     ctx: &crate::runtime::instance::InstanceContext,
     bucket: &str,
-) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = bucket_metadata_sys_of(ctx)?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -2065,9 +2062,27 @@ 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.
@@ -2112,6 +2127,7 @@ 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()),
@@ -2143,6 +2159,81 @@ 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()
     }
@@ -2342,6 +2433,7 @@ 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,
@@ -2425,6 +2517,7 @@ 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")
@@ -2486,6 +2579,7 @@ 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);
@@ -2509,6 +2603,7 @@ 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;
@@ -2526,6 +2621,7 @@ 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();
@@ -2665,6 +2761,7 @@ 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;
@@ -2673,6 +2770,7 @@ 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?;
@@ -2682,7 +2780,9 @@ 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;
@@ -2692,6 +2792,7 @@ 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
@@ -2740,6 +2841,7 @@ 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")
@@ -2796,6 +2898,7 @@ 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)]
@@ -3121,6 +3224,7 @@ 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_guard = acquire_metadata_transaction_guard(
             self.object_store(),
             bucket,
@@ -3198,6 +3302,7 @@ 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}")));
@@ -3420,7 +3525,7 @@ impl BucketMetadataSys {
         }
     }
 
-    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
+    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         match ConfigState::of(&bm.lifecycle_config_xml, &bm.lifecycle_config).require(bucket, BUCKET_LIFECYCLE_CONFIG)? {
@@ -3543,7 +3648,7 @@ impl BucketMetadataSys {
         }))
     }
 
-    pub async fn get_replication_config(&self, bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+    pub async fn get_replication_config(&self, bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         if !bm.replication_config_xml.is_empty() && bm.replication_config.is_none() {
@@ -3645,6 +3750,7 @@ 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;
@@ -3699,6 +3805,180 @@ 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;
@@ -3887,7 +4167,7 @@ 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(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         });
         sys.set("authoritative-versioned".to_string(), Arc::new(versioned)).await;
@@ -4010,25 +4290,24 @@ mod tests {
         else {
             panic!("legacy lock-enabled metadata must remain configured");
         };
-        assert_eq!(
-            config.object_lock_enabled.as_ref().map(|value| value.as_str()),
-            Some(ObjectLockEnabled::ENABLED)
-        );
+        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
         assert!(config.rule.is_none());
     }
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_retention_periods() {
-        use s3s::dto::{DefaultRetention, ObjectLockRule};
+        use rustfs_gateway_types::persistence::{
+            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as 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(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+                object_lock_enabled: Some("Enabled".to_owned()),
                 rule: Some(ObjectLockRule {
                     default_retention: Some(DefaultRetention {
                         days,
-                        mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
+                        mode: Some("COMPLIANCE".to_owned()),
                         years,
                     }),
                 }),
@@ -4043,9 +4322,11 @@ mod tests {
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_enabled_mode_and_rule() {
-        use s3s::dto::{DefaultRetention, ObjectLockRule};
+        use rustfs_gateway_types::persistence::{
+            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as ObjectLockRule,
+        };
 
-        for enabled in [None, Some(ObjectLockEnabled::from("Disabled".to_string()))] {
+        for enabled in [None, Some("Disabled".to_string())] {
             let mut metadata = BucketMetadata::new("invalid-object-lock-enabled");
             metadata.lock_enabled = true;
             metadata.object_lock_config = Some(ObjectLockConfiguration {
@@ -4057,11 +4338,11 @@ mod tests {
 
         let mut invalid_mode = BucketMetadata::new("invalid-object-lock-mode");
         invalid_mode.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(1),
-                    mode: Some(ObjectLockRetentionMode::from("invalid".to_string())),
+                    mode: Some("invalid".to_string()),
                     years: None,
                 }),
             }),
@@ -4070,7 +4351,7 @@ mod tests {
 
         let mut missing_retention = BucketMetadata::new("missing-object-lock-retention");
         missing_retention.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule { default_retention: None }),
         });
         assert!(object_lock_config_state_from_authoritative_metadata(&missing_retention).is_err());
@@ -4080,7 +4361,7 @@ 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(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         metadata
@@ -4098,10 +4379,7 @@ mod tests {
         else {
             panic!("an ever-enabled bucket must remain Object Lock configured");
         };
-        assert_eq!(
-            config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str),
-            Some(ObjectLockEnabled::ENABLED)
-        );
+        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
     }
 
     #[tokio::test]
@@ -4117,7 +4395,7 @@ mod tests {
 
         let node = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore.clone())));
         let config = ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         update_with_sys(
@@ -5231,7 +5509,7 @@ mod tests {
     #[tokio::test]
     async fn update_config_with_persists_tagging_rewrite_across_disk_reload() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
 
@@ -5395,7 +5673,7 @@ mod tests {
     #[tokio::test]
     async fn concurrent_update_config_with_calls_do_not_lose_writes() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
         let sys = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore)));
@@ -5495,7 +5773,7 @@ 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(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         persisted
@@ -5829,7 +6107,7 @@ mod tests {
     #[serial]
     async fn bucket_delete_waits_for_config_mutation_fence() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as 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

@@ -17,15 +17,8 @@ pub mod objectlock_sys;
 pub mod types;
 
 /// Whether a bucket Object Lock configuration has locking enabled. The
-/// 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).
+/// persisted `ObjectLockConfiguration` implements this in the bucket-metadata
+/// module, which owns the persisted configuration type.
 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

@@ -12,6 +12,7 @@
 // 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 s3s::dto::ReplicationConfiguration;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 
 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: ReplicationConfiguration) -> ReplicationLifecycleConfig {
+    pub(crate) fn new_config(config: PersistedReplicationConfiguration) -> ReplicationLifecycleConfig {
         ReplicationConfig::new(Some(config), None)
     }
 
@@ -118,16 +118,17 @@ impl ReplicationLifecycleBridge {
 
 #[cfg(test)]
 mod tests {
-    use s3s::dto::{Destination, ReplicationRule, ReplicationRuleStatus};
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::{PersistedReplicationDestination, PersistedReplicationRule};
 
     use super::super::replication_filemeta_boundary::{ReplicateTargetDecision, VersionPurgeStatusType};
     use super::*;
 
-    fn replication_rule() -> ReplicationRule {
-        ReplicationRule {
+    fn replication_rule() -> PersistedReplicationRule {
+        PersistedReplicationRule {
             delete_marker_replication: None,
             delete_replication: None,
-            destination: Destination {
+            destination: PersistedReplicationDestination {
                 bucket: "arn:aws:s3:::target-bucket".to_string(),
                 ..Default::default()
             },
@@ -137,13 +138,13 @@ mod tests {
             prefix: Some(String::new()),
             priority: Some(1),
             source_selection_criteria: None,
-            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+            status: Status::ENABLED.to_string(),
         }
     }
 
     #[test]
     fn has_pending_version_purge_preserves_replication_active_rule_behavior() {
-        let config = ReplicationLifecycleBridge::new_config(ReplicationConfiguration {
+        let config = ReplicationLifecycleBridge::new_config(PersistedReplicationConfiguration {
             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};
@@ -37,11 +37,11 @@ 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<(ReplicationConfiguration, OffsetDateTime)> {
+    pub(crate) async fn replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
         metadata_sys::get_replication_config(bucket).await
     }
 
-    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<ReplicationConfiguration>> {
+    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<PersistedReplicationConfiguration>> {
         let config = match Self::replication_config(bucket).await {
             Ok((config, _)) => Some(config),
             Err(err) => {

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

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

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

@@ -5074,9 +5074,10 @@ mod tests {
     #[tokio::test]
     async fn heal_queue_marks_missing_versioning_state_as_missed() {
         use super::super::replication_target_boundary::BucketTargets;
-        use s3s::dto::{
-            DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule,
-            ReplicationRuleStatus,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::{
+            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationRule,
+            PersistedReplicationStatus,
         };
 
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
@@ -5091,14 +5092,14 @@ mod tests {
                 ..Default::default()
             },
             ReplicationConfig::new(
-                Some(ReplicationConfiguration {
+                Some(PersistedReplicationConfiguration {
                     role: String::new(),
-                    rules: vec![ReplicationRule {
+                    rules: vec![PersistedReplicationRule {
                         delete_marker_replication: None,
-                        delete_replication: Some(DeleteReplication {
-                            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                        delete_replication: Some(PersistedReplicationStatus {
+                            status: Status::ENABLED.to_string(),
                         }),
-                        destination: Destination {
+                        destination: PersistedReplicationDestination {
                             bucket: arn.to_string(),
                             ..Default::default()
                         },
@@ -5108,7 +5109,7 @@ mod tests {
                         prefix: Some(String::new()),
                         priority: Some(1),
                         source_selection_criteria: None,
-                        status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                        status: Status::ENABLED.to_string(),
                     }],
                 }),
                 Some(BucketTargets::default()),

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

@@ -85,6 +85,8 @@ 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,
@@ -92,8 +94,6 @@ 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;
@@ -6579,9 +6579,10 @@ mod tests {
 
     use super::super::replication_target_boundary::{BucketTarget, BucketTargets};
     use super::*;
-    use s3s::dto::{
-        BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ExcludedPrefix, ReplicationRule,
-        ReplicationRuleStatus, VersioningConfiguration,
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+    use rustfs_gateway_types::persistence::{
+        PersistedReplicationDestination, PersistedReplicationRule, PersistedReplicationStatus,
     };
     use std::collections::HashMap;
     use time::OffsetDateTime;
@@ -8693,7 +8694,7 @@ mod tests {
             ..Default::default()
         };
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: role.to_string(),
                 rules: vec![],
             }),
@@ -8719,7 +8720,7 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
         );
@@ -8752,22 +8753,20 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
-                excluded_prefixes: Some(vec![ExcludedPrefix {
-                    prefix: Some("archive/".to_string()),
-                }]),
+                status: Some("Enabled".to_owned()),
+                excluded_prefixes: Some(vec![Some("archive/".to_string())]),
                 ..Default::default()
             },
         );
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::ENABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
@@ -8777,7 +8776,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
             Some(BucketTargets {
@@ -8813,14 +8812,14 @@ 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(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: current_role.to_string(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::DISABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: current_role.to_string(),
                         ..Default::default()
                     },
@@ -8830,7 +8829,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
             Some(BucketTargets {
@@ -8878,12 +8877,12 @@ mod tests {
     #[tokio::test]
     async fn heal_rejects_semantically_invalid_replication_config() {
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
                     delete_replication: None,
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: "arn:rustfs:replication:us-east-1:target:bucket".to_string(),
                         ..Default::default()
                     },
@@ -8893,7 +8892,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static("Enabld"),
+                    status: "Enabld".to_owned(),
                 }],
             }),
             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 s3s::dto::ReplicationConfiguration;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 
 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: &ReplicationConfiguration, arn: &str) -> bool {
+    pub(crate) fn target_is_used_by_rules(config: &PersistedReplicationConfiguration, 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 = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             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 s3s::dto::VersioningConfiguration;
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as 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 s3s::dto::Tag;
+use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 use std::cmp::Ordering;
 use std::collections::HashMap;
 use url::form_urlencoded;

crates/ecstore/src/bucket/utils.rs

@@ -15,9 +15,23 @@
 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 {
@@ -74,37 +88,173 @@ pub fn check_valid_bucket_name_strict(bucket_name: &str) -> Result<()> {
     check_bucket_name_common(bucket_name, true)
 }
 
-// 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)
+/// 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>;
 }
 
-#[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())
+/// 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),
 }
 
-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)?;
+/// 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))
     }
-    Ok(buf)
 }
 
 pub fn has_bad_path_component(path: &str) -> bool {
@@ -446,6 +596,40 @@ 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 == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED))
+        self.status.as_deref() == Some("Enabled")
     }
     fn prefix_enabled(&self, prefix: &str) -> bool {
-        if self.status != Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
+        if self.status.as_deref() != Some("Enabled") {
             return false;
         }
 
@@ -44,12 +44,10 @@ impl VersioningApi for VersioningConfiguration {
         }
 
         if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-            for p in excluded_prefixes.iter() {
-                if let Some(ref sprefix) = p.prefix {
-                    let pattern = format!("{sprefix}*");
-                    if match_simple(&pattern, prefix) {
-                        return false;
-                    }
+            for sprefix in excluded_prefixes.iter().flatten() {
+                let pattern = format!("{sprefix}*");
+                if match_simple(&pattern, prefix) {
+                    return false;
                 }
             }
         }
@@ -58,11 +56,11 @@ impl VersioningApi for VersioningConfiguration {
     }
 
     fn prefix_suspended(&self, prefix: &str) -> bool {
-        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)) {
+        if self.status.as_deref() == Some("Suspended") {
             return true;
         }
 
-        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
+        if self.status.as_deref() == Some("Enabled") {
             if prefix.is_empty() {
                 return false;
             }
@@ -75,12 +73,10 @@ impl VersioningApi for VersioningConfiguration {
             }
 
             if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-                for p in excluded_prefixes.iter() {
-                    if let Some(ref sprefix) = p.prefix {
-                        let pattern = format!("{sprefix}*");
-                        if match_simple(&pattern, prefix) {
-                            return true;
-                        }
+                for sprefix in excluded_prefixes.iter().flatten() {
+                    let pattern = format!("{sprefix}*");
+                    if match_simple(&pattern, prefix) {
+                        return true;
                     }
                 }
             }
@@ -92,14 +88,12 @@ impl VersioningApi for VersioningConfiguration {
         self.prefix_enabled(prefix) || self.prefix_suspended(prefix)
     }
     fn suspended(&self) -> bool {
-        self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED))
+        self.status.as_deref() == Some("Suspended")
     }
 }
 
 #[cfg(test)]
 mod tests {
-    use s3s::dto::{BucketVersioningStatus, ExcludedPrefix};
-
     use super::*;
 
     struct LegacyVersioning;
@@ -134,10 +128,8 @@ mod tests {
     #[test]
     fn delete_state_treats_excluded_prefixes_as_unversioned() {
         let config = VersioningConfiguration {
-            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
-            excluded_prefixes: Some(vec![ExcludedPrefix {
-                prefix: Some("archive/".to_string()),
-            }]),
+            status: Some("Enabled".to_owned()),
+            excluded_prefixes: Some(vec![Some("archive/".to_string())]),
             ..Default::default()
         };
 
@@ -146,7 +138,7 @@ mod tests {
         assert_eq!(config.delete_state("live/object"), (true, false));
 
         let suspended = VersioningConfiguration {
-            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)),
+            status: Some("Suspended".to_owned()),
             ..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 s3s::dto::VersioningConfiguration;
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 use tracing::{error, warn};
 
 pub struct BucketVersioningSys {}
@@ -163,7 +163,7 @@ mod tests {
 
     fn enabled() -> VersioningConfiguration {
         VersioningConfiguration {
-            status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         }
     }

crates/ecstore/src/cache_value/metacache_set.rs

@@ -217,6 +217,8 @@ impl Drop for NotifyOnDrop {
 
 #[derive(Default)]
 pub struct ListPathRawOptions {
+    /// Retain native listing admission through producer cancellation/draining.
+    pub read_budget: Option<Arc<tokio::sync::OwnedSemaphorePermit>>,
     pub disks: Vec<Option<DiskStore>>,
     pub fallback_disks: Vec<Option<DiskStore>>,
     pub bucket: String,
@@ -227,6 +229,7 @@ pub struct ListPathRawOptions {
     pub hidden_entry_budget: usize,
     pub pending_prefix: Option<String>,
     pub filter_prefix: Option<String>,
+    pub identifier_tree_leaf: Option<String>,
     pub forward_to: Option<String>,
     pub min_disks: usize,
     /// Deliver every replica to the partial callback, including matching headers.
@@ -255,6 +258,7 @@ pub struct ListPathRawOptions {
 impl Clone for ListPathRawOptions {
     fn clone(&self) -> Self {
         Self {
+            read_budget: self.read_budget.clone(),
             disks: self.disks.clone(),
             fallback_disks: self.fallback_disks.clone(),
             bucket: self.bucket.clone(),
@@ -265,6 +269,7 @@ impl Clone for ListPathRawOptions {
             hidden_entry_budget: self.hidden_entry_budget,
             pending_prefix: self.pending_prefix.clone(),
             filter_prefix: self.filter_prefix.clone(),
+            identifier_tree_leaf: self.identifier_tree_leaf.clone(),
             forward_to: self.forward_to.clone(),
             min_disks: self.min_disks,
             preserve_replica_metadata: self.preserve_replica_metadata,
@@ -297,6 +302,7 @@ fn walk_dir_options(opts: &ListPathRawOptions) -> WalkDirOptions {
         pending_prefix: opts.pending_prefix.clone(),
         report_notfound: opts.report_not_found,
         filter_prefix: opts.filter_prefix.clone(),
+        identifier_tree_leaf: opts.identifier_tree_leaf.clone(),
         forward_to: opts.forward_to.clone(),
         limit: opts.per_disk_limit,
         skip_total_timeout: opts.skip_walkdir_total_timeout,
@@ -389,7 +395,9 @@ async fn list_path_raw_inner(
         let disk_io_context_clone = disk_io_context.clone();
         let (rd, wr) = tokio::io::duplex(64 * 1024);
         readers.push(MetacacheReader::new(rd));
+        let read_budget = opts.read_budget.clone();
         let producer = with_optional_disk_io_context(disk_io_context_clone, async move {
+            let _read_budget = read_budget;
             let mut wr = PublishedBytesWriter::new(wr);
             #[cfg(test)]
             if let Some(observer) = &test_disk_io_context_observed {
@@ -1228,6 +1236,21 @@ mod tests {
         assert!(!options.skip_total_timeout);
     }
 
+    #[test]
+    fn identifier_tree_selection_survives_job_clone_and_disk_request() {
+        let request = ListPathRawOptions {
+            path: "catalog/namespaces/".to_string(),
+            identifier_tree_leaf: Some("table-entry.json".to_string()),
+            recursive: true,
+            ..Default::default()
+        };
+        let cloned = request.clone();
+        let walk = walk_dir_options(&cloned);
+        assert_eq!(walk.identifier_tree_leaf, request.identifier_tree_leaf);
+        assert_eq!(walk.base_dir, request.path);
+        assert!(walk.recursive);
+    }
+
     #[test]
     fn cloned_reader_options_preserve_hidden_scan_budget_and_pending_prefix() {
         let request = ListPathRawOptions {
@@ -1897,6 +1920,42 @@ mod tests {
         assert_eq!(seen.lock().expect("seen mutex poisoned").as_slice(), &["bucket/object".to_string()]);
     }
 
+    #[tokio::test]
+    async fn list_read_budget_is_retained_by_raw_producer_until_cancellation() {
+        let semaphore = Arc::new(tokio::sync::Semaphore::new(4));
+        let grant = Arc::new(semaphore.clone().acquire_many_owned(4).await.expect("listing grant"));
+        let weak_grant = Arc::downgrade(&grant);
+        let entered = Arc::new(tokio::sync::Notify::new());
+        let stopped = Arc::new(tokio::sync::Notify::new());
+        let worker = tokio::spawn(list_path_raw(
+            CancellationToken::new(),
+            ListPathRawOptions {
+                read_budget: Some(grant),
+                disks: vec![None],
+                min_disks: 1,
+                test_reader_behaviors: vec![TestReaderBehavior::StallWithSignals {
+                    entered: entered.clone(),
+                    stopped: stopped.clone(),
+                }],
+                ..Default::default()
+            },
+        ));
+        timeout(Duration::from_secs(5), entered.notified())
+            .await
+            .expect("producer entered");
+        assert!(weak_grant.strong_count() >= 2, "producer must retain a grant independently of the parent");
+        assert_eq!(semaphore.available_permits(), 0);
+        worker.abort();
+        assert!(worker.await.expect_err("parent cancelled").is_cancelled());
+        timeout(Duration::from_secs(5), stopped.notified())
+            .await
+            .expect("producer stopped");
+        let _next = timeout(Duration::from_secs(5), semaphore.acquire_many(4))
+            .await
+            .expect("cancelled listing releases all permits")
+            .expect("budget stays open");
+    }
+
     #[tokio::test]
     async fn list_path_raw_aborts_unresponsive_producer_after_quorum_eof() {
         let result = timeout(

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

@@ -35,17 +35,17 @@ pub use client::{
 // Re-exported through `api::rpc`; not every item is consumed inside this crate.
 #[allow(unused_imports)]
 pub use http_auth::{
-    RPC_BACKGROUND_ADMISSION_HEADER, TONIC_RPC_PREFIX, build_auth_headers, build_put_file_auth_trailer,
-    check_and_record_signed_rpc_nonce, gen_signature_headers, gen_tonic_replay_scope_headers, gen_tonic_signature_headers,
-    normalize_tonic_rpc_audience, rpc_background_admission, set_tonic_canonical_body_digest, set_tonic_mutation_body_digest,
-    set_tonic_rolling_canonical_body_digest, set_tonic_rolling_mutation_body_digest, sign_ns_scanner_capability,
-    sign_ns_scanner_capability_with_checkpoint_coverage, sign_ns_scanner_capability_with_tier_registry_generation,
-    sign_put_file_capability, sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers,
-    tonic_rpc_auth_failure_reason, verify_ns_scanner_capability, verify_ns_scanner_capability_with_checkpoint_coverage,
-    verify_ns_scanner_capability_with_tier_registry_generation, verify_put_file_auth_trailer, verify_put_file_capability,
-    verify_rpc_signature, verify_tonic_boot_epoch_response, verify_tonic_canonical_body_digest,
-    verify_tonic_mutation_body_digest, verify_tonic_mutation_body_digest_reject_unsigned, verify_tonic_rpc_response_proof,
-    verify_tonic_rpc_signature, verify_tonic_rpc_signature_with_bootstrap,
+    RPC_BACKGROUND_ADMISSION_HEADER, RpcReplayCacheAdmissionError, TONIC_RPC_PREFIX, build_auth_headers,
+    build_put_file_auth_trailer, check_and_record_signed_rpc_nonce, gen_signature_headers, gen_tonic_replay_scope_headers,
+    gen_tonic_signature_headers, normalize_tonic_rpc_audience, rpc_auth_is_resource_exhausted, rpc_background_admission,
+    set_tonic_canonical_body_digest, set_tonic_mutation_body_digest, set_tonic_rolling_canonical_body_digest,
+    set_tonic_rolling_mutation_body_digest, sign_ns_scanner_capability, sign_ns_scanner_capability_with_checkpoint_coverage,
+    sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability, sign_tonic_rpc_response_proof,
+    tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason, verify_ns_scanner_capability,
+    verify_ns_scanner_capability_with_checkpoint_coverage, verify_ns_scanner_capability_with_tier_registry_generation,
+    verify_put_file_auth_trailer, verify_put_file_capability, verify_rpc_signature, verify_tonic_boot_epoch_response,
+    verify_tonic_canonical_body_digest, verify_tonic_mutation_body_digest, verify_tonic_mutation_body_digest_reject_unsigned,
+    verify_tonic_rpc_response_proof, verify_tonic_rpc_signature, verify_tonic_rpc_signature_with_bootstrap,
 };
 #[cfg(test)]
 pub(crate) use internode_data_transport::TcpHttpInternodeDataTransport;

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

@@ -1965,6 +1965,15 @@ 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/peer_s3_client.rs

@@ -18,7 +18,7 @@ use crate::cluster::rpc::client::{
     node_service_time_out_client,
 };
 use crate::cluster::rpc::set_tonic_mutation_body_digest;
-use crate::core::pools::{PoolMeta, PoolMetaWriteState};
+use crate::core::pools::{PoolMeta, PoolMetaSaveGate, PoolMetaSaveGateWriteGuard};
 use crate::disk::error::DiskError;
 use crate::disk::error::{Error, Result};
 use crate::disk::error_reduce::{BUCKET_OP_IGNORED_ERRS, is_all_buckets_not_found, reduce_write_quorum_errs};
@@ -798,12 +798,12 @@ fn try_acquire_bucket_heal_movement_guard<'a>(
 }
 
 async fn acquire_bucket_heal_write_guard<'a>(
-    gate: Option<&'a tokio::sync::Mutex<PoolMetaWriteState>>,
-) -> Result<Option<tokio::sync::MutexGuard<'a, PoolMetaWriteState>>> {
+    gate: Option<&'a PoolMetaSaveGate>,
+) -> Result<Option<PoolMetaSaveGateWriteGuard<'a>>> {
     let Some(gate) = gate else {
         return Ok(None);
     };
-    let guard = gate.lock().await;
+    let guard = gate.write().await;
     guard
         .ensure_write_safe("bucket heal cannot run while pool metadata requires recovery")
         .map_err(|e| e.narrow_to_disk().unwrap_or_else(Error::other))?;
@@ -2388,8 +2388,8 @@ mod tests {
 
     #[tokio::test]
     async fn local_bucket_heal_refuses_receiver_with_unsafe_pool_metadata() {
-        let gate = tokio::sync::Mutex::new(PoolMetaWriteState::default());
-        gate.lock().await.observe_replicas(PoolMetaReplicaState {
+        let gate = PoolMetaSaveGate::default();
+        gate.write().await.observe_replicas(PoolMetaReplicaState {
             needs_repair: true,
             repair_write_safe: false,
         });

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

@@ -376,6 +376,21 @@ struct WalkCopyError {
     source: io::Error,
 }
 
+impl WalkCopyError {
+    fn into_disk_error(self) -> DiskError {
+        // Only a local output write/flush proves the metacache consumer closed.
+        // An identical I/O kind from the remote input remains a transport failure.
+        if matches!(self.reason, RemoteWalkFailureReason::Write | RemoteWalkFailureReason::Flush)
+            && self.source.kind() == io::ErrorKind::BrokenPipe
+            && self.source.get_ref().is_none()
+        {
+            DiskError::metacache_output_stream_closed()
+        } else {
+            DiskError::Io(self.source)
+        }
+    }
+}
+
 async fn copy_stream_with_buffer<R, W>(
     reader: &mut R,
     writer: &mut W,
@@ -2096,6 +2111,7 @@ impl RemoteDisk {
         failure_health_action: FailureHealthAction,
     ) {
         if let Err(err) = operation_result
+            && !err.is_metacache_output_stream_closed()
             && is_network_like_disk_error(err)
         {
             counter!(
@@ -4056,7 +4072,7 @@ impl DiskAPI for RemoteDisk {
                             Ok(_) => return Ok(()),
                             Err(error) => {
                                 failure_reason = Some(error.reason);
-                                return Err(DiskError::Io(error.source));
+                                return Err(error.into_disk_error());
                             }
                         }
                     }
@@ -9687,6 +9703,43 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    async fn remote_walk_preserves_local_metacache_consumer_close() {
+        let transport =
+            RetryingWalkDirInternodeDataTransport::with_steps(vec![WalkDirTestStep::Data(b"metacache payload".to_vec())]);
+        let disk = new_remote_disk_with_transport(Arc::new(transport.clone())).await;
+        let (mut writer, reader) = tokio::io::duplex(1);
+        drop(reader);
+        let error = disk
+            .walk_dir(WalkDirOptions::default(), &mut writer)
+            .await
+            .expect_err("a dropped metacache consumer must stop the producer");
+        assert!(
+            error.is_metacache_output_stream_closed(),
+            "preserve the benign completion marker: {error:?}"
+        );
+        assert_eq!(transport.calls().len(), 1, "local consumer closure is not a remote open retry");
+        assert!(!disk.health.is_faulty());
+    }
+
+    #[tokio::test]
+    async fn remote_walk_keeps_remote_read_broken_pipe_as_transport_failure() {
+        let transport = RetryingWalkDirInternodeDataTransport::with_steps(vec![WalkDirTestStep::PartialDataThenError {
+            data: b"partial metacache".to_vec(),
+            error: io::ErrorKind::BrokenPipe.into(),
+        }]);
+        let disk = new_remote_disk_with_transport(Arc::new(transport.clone())).await;
+        let mut writer = Vec::new();
+        let error = disk
+            .walk_dir(WalkDirOptions::default(), &mut writer)
+            .await
+            .expect_err("an upstream close after partial output is a failed walk");
+        assert!(!error.is_metacache_output_stream_closed());
+        assert!(error.contains_io_error_kind(io::ErrorKind::BrokenPipe));
+        assert_eq!(writer, b"partial metacache");
+        assert_eq!(transport.calls().len(), 1, "partial output must not be replayed");
+    }
+
     #[tokio::test]
     async fn walk_stream_backpressure_limits_read_ahead_and_is_cancellable() {
         let mut reader = Cursor::new(vec![5; 4 * DEFAULT_READ_BUFFER_SIZE]);

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

@@ -423,11 +423,15 @@ impl RemoteClient {
     /// peer toward the offline threshold. See issue #4567.
     fn is_transport_failure(status: &tonic::Status) -> bool {
         use tonic::Code;
+        // A valid stale-epoch request is rejected before execution. Its signed
+        // response refreshes the peer epoch on the existing healthy channel.
+        let epoch_rejection = status.code() == Code::Unavailable && status.message() == "RPC boot epoch changed";
         std::error::Error::source(status).is_some()
-            || matches!(
-                status.code(),
-                Code::Unavailable | Code::DeadlineExceeded | Code::Unknown | Code::Cancelled
-            )
+            || (!epoch_rejection
+                && matches!(
+                    status.code(),
+                    Code::Unavailable | Code::DeadlineExceeded | Code::Unknown | Code::Cancelled
+                ))
     }
 
     async fn evict_connection(&self, op: &'static str, reason: &str, resource_summary: &str) {
@@ -2827,6 +2831,37 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    async fn stale_boot_epoch_rejections_do_not_open_lock_request_breaker() {
+        let addr = "http://stale-boot-epoch-lock-breaker:9000";
+        let client = RemoteClient::new(addr.to_string());
+        reset_lock_peer_health_for_test(addr);
+
+        for _ in 0..LOCK_RPC_REQUEST_BACKOFF_THRESHOLD {
+            let error = client
+                .execute_rpc::<(), _>(
+                    "lock",
+                    "bucket/new-object",
+                    Duration::from_secs(1),
+                    async { Err(tonic::Status::unavailable("RPC boot epoch changed")) },
+                    RpcCleanup::new(None, None, None),
+                )
+                .await
+                .expect_err("a stale epoch still rejects the operation before execution");
+            assert!(error.to_string().contains("RPC boot epoch changed"));
+        }
+
+        let health = lock_peer_health_for_test(addr);
+        assert_eq!(health.request_consecutive_timeouts, 0);
+        assert!(health.request_backoff_until.is_none());
+        assert!(health.last_eviction.is_none());
+        let admission = client
+            .admit_lock_request("lock")
+            .expect("authenticated epoch refresh may proceed immediately");
+        client.release_lock_request("lock", Some(admission));
+        reset_lock_peer_health_for_test(addr);
+    }
+
     #[test]
     #[serial_test::serial]
     fn test_remote_client_rpc_timeout_honors_configured_deadline() {

crates/ecstore/src/core/pool_meta_native_integration_tests.rs

@@ -628,7 +628,7 @@ async fn pool_meta_native_physical_runtime_fork_after_apply_latches_original_gat
     let fork = identity_epoch_fork(&fixture).await;
     let since = OffsetDateTime::from_unix_timestamp(1_700_000_000).expect("fixed original failure timestamp should be valid");
     {
-        let mut state = fixture.store.pool_meta_save_gate.lock().await;
+        let mut state = fixture.store.pool_meta_save_gate.write().await;
         assert_eq!(state.active_transactions.load(Ordering::SeqCst), 0);
         state.write_blocked = false;
         state.block_context = None;
@@ -685,7 +685,7 @@ async fn pool_meta_native_physical_runtime_fork_after_apply_latches_original_gat
     );
     assert_committed(&fixture).await;
     {
-        let state = fixture.store.pool_meta_save_gate.lock().await;
+        let state = fixture.store.pool_meta_save_gate.write().await;
         assert!(
             state.write_blocked,
             "candidate validation must propagate its permanent latch to the original state"
@@ -706,7 +706,7 @@ async fn pool_meta_native_physical_runtime_fork_after_apply_latches_original_gat
             .expect("permanently blocked recovery should return without mutation"),
         "ordinary-readable repaired heads must not clear the original known-fork latch"
     );
-    let state = fixture.store.pool_meta_save_gate.lock().await;
+    let state = fixture.store.pool_meta_save_gate.write().await;
     assert!(state.write_blocked);
     assert_eq!(
         state
@@ -979,7 +979,7 @@ async fn pool_meta_native_heal_v1_with_all_upgrade_gates_disabled_keeps_exact_ra
         async {
             let (fixture, raw) = legacy_native_owner_fixture(POOL_META_V1_VERSION, false).await;
             let guard = owner(&fixture.store).await;
-            let mut state = fixture.store.pool_meta_save_gate.lock().await;
+            let mut state = fixture.store.pool_meta_save_gate.write().await;
             let result = tokio::time::timeout(
                 std::time::Duration::from_secs(30),
                 fixture.store.heal_pool_meta_replica_under_fence(0, guard, &mut state),
@@ -1036,7 +1036,7 @@ async fn pool_meta_native_heal_rejects_legacy_source_with_trailing_bytes_without
                 let before = snapshots(&fixture.store).await;
                 let identities = owner_repair_identity_heads(&fixture.store).await;
                 let guard = owner(&fixture.store).await;
-                let mut state = fixture.store.pool_meta_save_gate.lock().await;
+                let mut state = fixture.store.pool_meta_save_gate.write().await;
                 let result = tokio::time::timeout(
                     std::time::Duration::from_secs(30),
                     fixture.store.heal_pool_meta_replica_under_fence(1, guard, &mut state),
@@ -1110,7 +1110,7 @@ async fn pool_meta_native_heal_mixed_v1_v2_preserves_observed_format_floor() {
                     assert_eq!(selection.revision.version, POOL_META_VERSION, "the observed V2 format floor must be retained");
                     assert_eq!(selection.canonical.as_deref(), Some(semantic_v2.as_slice()));
                     let guard = owner(&fixture.store).await;
-                    let mut state = fixture.store.pool_meta_save_gate.lock().await;
+                    let mut state = fixture.store.pool_meta_save_gate.write().await;
                     tokio::time::timeout(std::time::Duration::from_secs(30),
                         fixture.store.heal_pool_meta_replica_under_fence(target_pool, guard, &mut state))
                         .await.expect("mixed legacy heal must terminate")
@@ -1199,7 +1199,7 @@ async fn pool_meta_native_heal_rejects_v1_and_v3_prepare_in_either_pool_order()
                 let before = snapshots(&fixture.store).await;
                 let identities = owner_repair_identity_heads(&fixture.store).await;
                 let guard = owner(&fixture.store).await;
-                let mut state = fixture.store.pool_meta_save_gate.lock().await;
+                let mut state = fixture.store.pool_meta_save_gate.write().await;
                 let result = tokio::time::timeout(
                     std::time::Duration::from_secs(30),
                     fixture
@@ -1301,7 +1301,7 @@ async fn pool_meta_native_heal_rejects_v1_legacy_dont_save_true_without_writes()
             let before = snapshots(&fixture.store).await;
             let identities = owner_repair_identity_heads(&fixture.store).await;
             let guard = owner(&fixture.store).await;
-            let mut state = fixture.store.pool_meta_save_gate.lock().await;
+            let mut state = fixture.store.pool_meta_save_gate.write().await;
             let result = tokio::time::timeout(
                 std::time::Duration::from_secs(30),
                 fixture.store.heal_pool_meta_replica_under_fence(1, guard, &mut state),
@@ -1410,7 +1410,7 @@ async fn pool_meta_native_heal_rejects_target_v1_dont_save_true_without_writes()
             let before = snapshots(&fixture.store).await;
             let identities = owner_repair_identity_heads(&fixture.store).await;
             let guard = owner(&fixture.store).await;
-            let mut state = fixture.store.pool_meta_save_gate.lock().await;
+            let mut state = fixture.store.pool_meta_save_gate.write().await;
             let result = tokio::time::timeout(
                 std::time::Duration::from_secs(30),
                 fixture.store.heal_pool_meta_replica_under_fence(1, guard, &mut state),

crates/ecstore/src/core/pools_test.rs

@@ -527,8 +527,8 @@ mod decommission_lock_order_tests {
             rebalance_meta: tokio::sync::RwLock::new(other_store.rebalance_meta.read().await.clone()),
             decommission_cancelers: tokio::sync::RwLock::new(vec![None; other_store.pools.len()]),
             start_gate: tokio::sync::Mutex::new(()),
-            pool_meta_save_gate: tokio::sync::Mutex::new(
-                other_store.pool_meta_save_gate.lock().await.independent_clone_for_test(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::new(
+                other_store.pool_meta_save_gate.write().await.independent_clone_for_test(),
             ),
             ctx,
             bucket_fence_registry: Arc::default(),
@@ -1616,7 +1616,7 @@ mod decommission_lock_order_tests {
             run_routine_to_a_deferral_pause(
                 &store,
                 1,
-                "target pool 1 no longer fits the selected mutation: source pool 0 reservation estimate is exhausted",
+                "target pool 1 no longer fits the selected mutation: target pool 1 reservation requires",
             )
             .await;
         });
@@ -1647,7 +1647,7 @@ mod decommission_lock_order_tests {
             run_routine_to_a_deferral_pause(
                 &store,
                 1,
-                "target pool 1 no longer fits the selected mutation: source pool 0 reservation estimate is exhausted",
+                "target pool 1 no longer fits the selected mutation: target pool 1 reservation requires",
             )
             .await;
         });
@@ -6062,6 +6062,12 @@ mod decommission_lock_order_tests {
             )
             .await
             .expect("read the next source object for the multipart capacity oracle");
+        // Reconciled bytes remain charged. With no physical room for a new
+        // reservation, a distinct mutation must fail instead of reusing them.
+        let mut exhausted_capacity = capacity_snapshot();
+        exhausted_capacity[2] = DecommissionPoolCapacityInfo::for_test(2, layout, 0, target_total, target_total);
+        set_decommission_capacity_info_overrides_for_test(other_store.id, vec![exhausted_capacity]);
+
         let next_result = tokio::time::timeout(
             Duration::from_secs(30),
             data_movement::migrate_decommission_object(
@@ -7827,6 +7833,12 @@ mod decommission_lock_order_tests {
             .expect("same-mutation retry should publish a readable target");
         assert_eq!(retry_info.size, i64::try_from(body.len()).expect("test body size fits i64"));
 
+        // Reconciled bytes remain charged. With no physical room for a new
+        // reservation, a distinct mutation must fail instead of reusing them.
+        let mut exhausted_capacity = capacity_snapshot();
+        exhausted_capacity[2] = DecommissionPoolCapacityInfo::for_test(2, layout, 0, target_total, target_total);
+        set_decommission_capacity_info_overrides_for_test(other_store.id, vec![exhausted_capacity]);
+
         let next_object = "data-movement-put-after-retry.bin";
         let next_version_id = uuid::Uuid::new_v4().to_string();
         let mut next_opts = ObjectOptions {
@@ -7841,7 +7853,11 @@ mod decommission_lock_order_tests {
         let next_result = other_store
             .put_object_for_data_movement(RUSTFS_META_BUCKET, next_object, &mut next_data, &next_opts, None)
             .await;
-        assert!(next_result.is_err(), "a different mutation must not reuse A's recovered capacity");
+        let next_error = match next_result {
+            Err(error) | Ok((_, Err(error))) => error,
+            Ok((_, Ok(_))) => panic!("a different mutation must not reuse A's recovered capacity"),
+        };
+        assert!(next_error.to_string().contains("decommission_capacity_blocked"), "{next_error}");
         let next_target_err = other_store.pools[2]
             .get_object_info(
                 RUSTFS_META_BUCKET,
@@ -9929,6 +9945,12 @@ mod decommission_lock_order_tests {
             )
             .await
             .expect("read the next source object for the capacity oracle");
+        // Reconciled bytes remain charged. With no physical room for a new
+        // reservation, a distinct mutation must fail instead of reusing them.
+        let mut exhausted_capacity = capacity_snapshot();
+        exhausted_capacity[2] = DecommissionPoolCapacityInfo::for_test(2, layout, 0, target_total, target_total);
+        set_decommission_capacity_info_overrides_for_test(other_store.id, vec![exhausted_capacity]);
+
         let next_result = tokio::time::timeout(
             Duration::from_secs(30),
             data_movement::migrate_decommission_object(
@@ -10190,11 +10212,20 @@ mod decommission_lock_order_tests {
         assert_eq!(reconciled_reservation.pending_target_physical_bytes, 0);
         assert_eq!(reconciled_reservation.consumed_target_physical_bytes, 1);
 
+        // Reconciled bytes remain charged. With no physical room for a new
+        // reservation, a distinct mutation must fail instead of reusing them.
+        let mut exhausted_capacity = capacity_snapshot();
+        exhausted_capacity[2] = DecommissionPoolCapacityInfo::for_test(2, layout, 1, 2, 1);
+        set_decommission_capacity_info_overrides_for_test(other_store.id, vec![exhausted_capacity]);
+
         let next_after_reconcile = other_store
             .delete_object(&bucket, next_object, next_opts)
             .await
             .expect_err("the consumed exact-fit capacity must reject the next marker");
-        assert!(next_after_reconcile.to_string().contains("decommission_capacity_blocked"));
+        assert!(
+            next_after_reconcile.to_string().contains("decommission_capacity_blocked"),
+            "{next_after_reconcile}"
+        );
         let next_target_err = other_store.pools[2]
             .get_object_info(
                 &bucket,

crates/ecstore/src/core/sets.rs

@@ -65,6 +65,38 @@ 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)
 }
@@ -266,6 +298,14 @@ 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)),

crates/ecstore/src/data_usage/mod.rs

@@ -3436,7 +3436,7 @@ mod tests {
             rebalance_meta: RwLock::new(None),
             decommission_cancelers: RwLock::new(Vec::new()),
             start_gate: TokioMutex::new(()),
-            pool_meta_save_gate: TokioMutex::default(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
             ctx,
             bucket_fence_registry: Arc::default(),
         })

crates/ecstore/src/diagnostics/admin_server_info.rs

@@ -682,15 +682,11 @@ async fn get_pools_info(all_disks: &[Disk]) -> Result<HashMap<i32, HashMap<i32,
         return Err(Error::other("ServerNotInitialized"));
     };
 
-    let mut pools_info: HashMap<i32, HashMap<i32, ErasureSetInfo>> = HashMap::new();
-    for d in all_disks {
-        let pool_info = pools_info.entry(d.pool_index).or_default();
-        let erasure_set = pool_info.entry(d.set_index).or_default();
-
-        if erasure_set.id == 0 {
-            erasure_set.id = d.set_index;
+    collect_pools_info(all_disks, |pool_index, set_index| {
+        let store = store.clone();
+        async move {
             let load = load_data_usage_cache(
-                store.pools[d.pool_index as usize].disk_set[d.set_index as usize].as_ref(),
+                store.pools[pool_index as usize].disk_set[set_index as usize].as_ref(),
                 DATA_USAGE_CACHE_NAME,
             );
             #[cfg(feature = "diagnostic-read-version-caller")]
@@ -698,11 +694,34 @@ async fn get_pools_info(all_disks: &[Disk]) -> Result<HashMap<i32, HashMap<i32,
                 rustfs_io_metrics::internode_metrics::ReadVersionEntrySource::ServerInfoPools,
                 load,
             );
-            match load.await {
-                Ok(cache) => apply_erasure_set_usage(&cache, erasure_set),
-                Err(_) => erasure_set.usage_error = Some(DATA_USAGE_UNAVAILABLE_ERROR.to_string()),
-            }
+            load.await
         }
+    })
+    .await
+}
+
+async fn collect_pools_info<F, Fut>(all_disks: &[Disk], mut load_usage: F) -> Result<HashMap<i32, HashMap<i32, ErasureSetInfo>>>
+where
+    F: FnMut(i32, i32) -> Fut,
+    Fut: std::future::Future<Output = Result<rustfs_data_usage::DataUsageCache>>,
+{
+    let mut pools_info: HashMap<i32, HashMap<i32, ErasureSetInfo>> = HashMap::new();
+    for d in all_disks {
+        let pool_info = pools_info.entry(d.pool_index).or_default();
+        let erasure_set = match pool_info.entry(d.set_index) {
+            std::collections::hash_map::Entry::Occupied(entry) => entry.into_mut(),
+            std::collections::hash_map::Entry::Vacant(entry) => {
+                let erasure_set = entry.insert(ErasureSetInfo {
+                    id: d.set_index,
+                    ..Default::default()
+                });
+                match load_usage(d.pool_index, d.set_index).await {
+                    Ok(cache) => apply_erasure_set_usage(&cache, erasure_set),
+                    Err(_) => erasure_set.usage_error = Some(DATA_USAGE_UNAVAILABLE_ERROR.to_string()),
+                }
+                erasure_set
+            }
+        };
 
         erasure_set.raw_capacity += d.total_space;
         erasure_set.raw_usage += d.used_space;
@@ -1241,6 +1260,59 @@ mod tests {
         assert_eq!(info.deployment_id, expected_deployment_id);
     }
 
+    #[tokio::test]
+    async fn pool_usage_is_loaded_once_per_set_including_set_zero() {
+        let mut disks = Vec::new();
+        for (pool, set, count) in [(0, 0, 16), (1, 0, 16), (1, 1, 4)] {
+            disks.extend((0..count).map(|_| Disk {
+                pool_index: pool,
+                set_index: set,
+                used_space: 10,
+                total_space: 100,
+                ..Default::default()
+            }));
+        }
+        let mut loads = Vec::new();
+        let info = super::collect_pools_info(&disks, |pool, set| {
+            loads.push((pool, set));
+            async move {
+                if pool == 1 && set == 0 {
+                    return Err(crate::error::Error::other("usage unavailable"));
+                }
+                let mut cache = rustfs_data_usage::DataUsageCache::default();
+                cache.info.name = DATA_USAGE_ROOT.to_owned();
+                cache.info.last_update = Some(std::time::SystemTime::UNIX_EPOCH);
+                cache.info.snapshot_complete = true;
+                cache.replace(
+                    DATA_USAGE_ROOT,
+                    "",
+                    rustfs_data_usage::DataUsageEntry {
+                        size: 25,
+                        objects: 2,
+                        ..Default::default()
+                    },
+                );
+                Ok(cache)
+            }
+        })
+        .await
+        .expect("collect mixed successful and unavailable usage");
+        assert_eq!(loads, [(0, 0), (1, 0), (1, 1)]);
+        for (pool, set, count) in [(0, 0, 16), (1, 0, 16), (1, 1, 4)] {
+            let set_info = &info[&pool][&set];
+            assert_eq!(set_info.id, set);
+            assert_eq!(set_info.raw_usage, count * 10);
+            assert_eq!(set_info.raw_capacity, count * 100);
+            if pool == 1 && set == 0 {
+                assert_eq!(set_info.usage_error.as_deref(), Some(DATA_USAGE_UNAVAILABLE_ERROR));
+            } else {
+                assert_eq!(set_info.usage, 25, "logical usage must not multiply by drive count");
+                assert_eq!(set_info.objects_count, 2);
+                assert!(set_info.usage_error.is_none());
+            }
+        }
+    }
+
     #[serial]
     #[tokio::test]
     async fn local_server_property_includes_runtime_stats_without_endpoint_pools() {

crates/ecstore/src/disk/disk_store.rs

@@ -3053,6 +3053,15 @@ 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() {

crates/ecstore/src/disk/fs.rs

@@ -92,6 +92,9 @@ 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();
@@ -144,6 +147,7 @@ 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
 }
@@ -216,11 +220,13 @@ 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
@@ -228,6 +234,7 @@ 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

@@ -15,8 +15,6 @@
 //! Single-disk object rename publication and rollback. The shared execution core
 //! retains instrumentation, mutation leases, and commit guards through the syscall.
 
-#[cfg(target_os = "linux")]
-use super::has_rename_data_test_failpoints;
 #[cfg(all(test, windows))]
 use super::run_destination_commit_directory_preparation;
 #[cfg(any(not(windows), test))]
@@ -24,8 +22,8 @@ use super::should_fail_local_inline_rollback_hardlink;
 use super::{
     EVENT_DISK_LOCAL_ACCESS_FAILED, EVENT_DISK_LOCAL_HEAL_PURGE_FAILED, EVENT_DISK_LOCAL_RENAME_REJECTED, LOG_COMPONENT_ECSTORE,
     LOG_SUBSYSTEM_DISK_LOCAL, LocalDisk, SyncMode, effective_durability, has_inline_commit_test_failpoints,
-    inline_metadata_rollback_dir, observe_old_current, remove_dir_all_if_exists, remove_dst_base_before_commit,
-    remove_file_if_exists, rename_data_versions_signature, run_inline_preparation_before_backup,
+    has_rename_data_test_failpoints, inline_metadata_rollback_dir, observe_old_current, remove_dir_all_if_exists,
+    remove_dst_base_before_commit, remove_file_if_exists, rename_data_versions_signature, run_inline_preparation_before_backup,
     should_fail_after_metadata_commit, should_fail_before_old_metadata_backup, should_fail_commit_rename,
     should_remove_staged_meta_before_commit, skip_access_checks, tmp_staging_mutation_path,
 };
@@ -172,6 +170,44 @@ impl RenameCommitContext<'_> {
     }
 }
 
+#[cfg(any(not(windows), test))]
+fn inline_rollback_backup_path(staging_file_path: &Path) -> 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 = if staging_file_path.file_name().and_then(|n| n.to_str()) == Some(STORAGE_FORMAT_FILE) {
+        staging_parent.join(STORAGE_FORMAT_FILE_BACKUP)
+    } else {
+        staging_file_path.with_extension("meta.bkp")
+    };
+    remove_file_if_exists(&backup_path)?;
+    Ok(backup_path)
+}
+
+#[cfg(any(not(windows), test))]
+fn write_local_inline_rollback_backup_from_bytes(
+    staging_file_path: &Path,
+    old_metadata: &[u8],
+    sync: bool,
+) -> std::io::Result<PathBuf> {
+    let backup_path = inline_rollback_backup_path(staging_file_path)?;
+    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);
+    }
+    Ok(backup_path)
+}
+
 #[cfg(any(not(windows), test))]
 pub(super) fn create_local_inline_rollback_backup(
     dst_file_path: &Path,
@@ -179,27 +215,12 @@ pub(super) fn create_local_inline_rollback_backup(
     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)?;
+    let backup_path = inline_rollback_backup_path(staging_file_path)?;
     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() {
+        return write_local_inline_rollback_backup_from_bytes(staging_file_path, old_metadata, sync);
+    } else if cfg!(test) && sync && !old_metadata.is_empty() {
+        // Skip redundant fdatasync on hardlinked rollback backups in production;
+        // keep a Windows-safe sync open in tests.
         os::open_file_for_sync_std(&backup_path)?.sync_data()?;
     }
     Ok(backup_path)
@@ -658,7 +679,14 @@ impl LocalDisk {
                 })
                 .collect()
         };
-        let src_volume_dir = self.io_get_bucket_path(src_volume)?;
+        let src_volume_dir = if cfg!(target_os = "linux")
+            && src_volume == super::super::RUSTFS_META_TMP_BUCKET
+            && skip_access_checks(src_volume)
+        {
+            self.io_root().join(src_volume)
+        } else {
+            self.io_get_bucket_path(src_volume)?
+        };
         if !skip_access_checks(src_volume)
             && let Err(e) = super::super::fs::access_std(&src_volume_dir)
         {
@@ -702,26 +730,53 @@ impl LocalDisk {
         } else {
             STORAGE_FORMAT_FILE
         };
+        let flat_inline_tmp = cfg!(target_os = "linux")
+            && !no_inline
+            && src_volume == super::super::RUSTFS_META_TMP_BUCKET
+            && !preserve_upload_metadata
+            && multipart_publication.is_none()
+            && pool_meta_condition.is_none()
+            && !fi_healing
+            && !os::dst_dir_fsync_group_commit_enabled()
+            && !os::file_fdatasync_group_commit_enabled()
+            && !has_rename_data_test_failpoints(dst_path);
         let src_file_path = match multipart_publication.as_ref() {
             Some(publication) => publication.canonical_head(),
+            None if flat_inline_tmp => {
+                self.io_get_object_open_path(src_volume, format!("{src_path}.{source_metadata_name}").as_str())?
+            }
+            None if cfg!(target_os = "linux") => {
+                self.io_get_object_open_path(src_volume, format!("{src_path}/{source_metadata_name}").as_str())?
+            }
             None => self.io_get_object_path(src_volume, format!("{src_path}/{source_metadata_name}").as_str())?,
         };
         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() {
+        let has_data_dir_path = if no_inline && (!fi.is_remote() || pending_multipart_restore) {
             if let Some(data_dir) = fi
                 .data_dir
                 .map(|dir| rustfs_utils::path::retain_slash(dir.to_string().as_str()))
             {
-                let src_data_path = self.io_get_object_path(
-                    src_volume,
-                    rustfs_utils::path::retain_slash(format!("{}/{}", src_path, data_dir).as_str()).as_str(),
-                )?;
-                let dst_data_path = self.io_get_object_path(
-                    dst_volume,
-                    rustfs_utils::path::retain_slash(format!("{}/{}", dst_path, data_dir).as_str()).as_str(),
-                )?;
+                let src_rel = rustfs_utils::path::retain_slash(format!("{}/{}", src_path, data_dir).as_str());
+                let src_data_path = if cfg!(target_os = "linux") && multipart_publication.is_none() && !fi_healing {
+                    self.io_get_object_open_path(src_volume, src_rel.as_str())?
+                } else {
+                    self.io_get_object_path(src_volume, src_rel.as_str())?
+                };
+                let dst_data_path = destination_object_path.join(data_dir.trim_matches('/'));
 
                 Some((src_data_path, dst_data_path))
             } else {
@@ -804,7 +859,7 @@ impl LocalDisk {
 
             let fused_attempt = move |is_cancelled: &dyn Fn() -> bool| -> std::io::Result<FusedCommitAttempt> {
                 let (guard, has_dst_buf) = prepare_rename_commit_and_destination_metadata_std(
-                    !no_inline,
+                    !no_inline && !flat_inline_tmp,
                     &src_parent_owned,
                     &dst_parent_owned,
                     &dst_file_path_owned,
@@ -876,7 +931,16 @@ impl LocalDisk {
                 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, false)?;
+                let cleanup_flat_tmp = || {
+                    if flat_inline_tmp {
+                        let _ = remove_file_if_exists(&src_file_path_owned);
+                    }
+                };
+                if let Err(err) = prepared_metadata_source.write_all(&new_dst_buf, false) {
+                    drop(prepared_metadata_source);
+                    cleanup_flat_tmp();
+                    return Err(err);
+                }
                 if !no_inline {
                     run_inline_preparation_before_backup(&dst_path_str);
                 }
@@ -884,14 +948,21 @@ impl LocalDisk {
                 let mut staged_rollback_path = None;
                 if (!no_inline || rollback_data_dir.is_none())
                     && let Some(ref old_metadata) = has_dst_buf
-                    && (rollback_data_dir.is_some() || sync_commit_metadata || cfg!(test))
+                    && (rollback_data_dir.is_some() || cfg!(test))
                 {
-                    let backup_path = create_local_inline_rollback_backup(
+                    let backup_path = match create_local_inline_rollback_backup(
                         &dst_file_path_owned,
                         &src_file_path_owned,
                         old_metadata,
                         sync_commit_metadata,
-                    )?;
+                    ) {
+                        Ok(path) => path,
+                        Err(err) => {
+                            drop(prepared_metadata_source);
+                            cleanup_flat_tmp();
+                            return Err(err);
+                        }
+                    };
                     staged_rollback_path = Some(StagedRollbackBackup(backup_path));
                 }
 
@@ -920,6 +991,8 @@ impl LocalDisk {
                         return Err(err);
                     }
                 } else if sync_commit_metadata && let Err(err) = prepared_metadata_source.sync_data() {
+                    drop(prepared_metadata_source);
+                    cleanup_flat_tmp();
                     if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
                         let _ = remove_file_if_exists(backup_path);
                     }
@@ -928,6 +1001,7 @@ impl LocalDisk {
 
                 if is_cancelled() {
                     drop(prepared_metadata_source);
+                    cleanup_flat_tmp();
                     if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
                         let _ = remove_file_if_exists(backup_path);
                     }
@@ -944,6 +1018,8 @@ impl LocalDisk {
                 ) {
                     Ok(guard) => guard,
                     Err(err) => {
+                        drop(prepared_metadata_source);
+                        cleanup_flat_tmp();
                         if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
                             let _ = remove_file_if_exists(backup_path);
                         }
@@ -1027,6 +1103,8 @@ impl LocalDisk {
                         None,
                         sync_commit_metadata,
                     ) {
+                        drop(prepared_metadata_source);
+                        cleanup_flat_tmp();
                         let _ = remove_file_if_exists(staged_backup);
                         return Ok(FusedCommitAttempt::CommitErr(to_file_error(err).into()));
                     }
@@ -1047,6 +1125,7 @@ impl LocalDisk {
                     )
                 };
                 if let Err(err) = rename_res {
+                    cleanup_flat_tmp();
                     if let Some(backup_path) = staged_rollback_path.as_ref().map(StagedRollbackBackup::path) {
                         let _ = remove_file_if_exists(backup_path);
                     }
@@ -1075,8 +1154,38 @@ impl LocalDisk {
                     return Ok(FusedCommitAttempt::CommitErr(DiskError::Unexpected));
                 }
 
-                if sync_commit_metadata
-                    && let Err(err) = fsync_inline_commit_std(
+                if sync_commit_metadata {
+                    if staged_rollback_path.is_none()
+                        && rollback_data_dir.is_none()
+                        && let Some(ref old_metadata) = has_dst_buf
+                        && (!no_inline || rollback_data_dir.is_none())
+                    {
+                        let fsync_res = (|| -> std::io::Result<()> {
+                            if let Some(dst_parent) = dst_file_path_owned.parent() {
+                                os::fsync_commit_directory_limited_std(dst_parent, &guard)?;
+                            }
+                            Ok(())
+                        })();
+                        if let Err(err) = fsync_res {
+                            let rollback_ok =
+                                write_local_inline_rollback_backup_from_bytes(&src_file_path_owned, old_metadata, true)
+                                    .and_then(|backup_path| {
+                                        let backup = StagedRollbackBackup(backup_path);
+                                        let res =
+                                            rollback_inline_metadata_std(&dst_file_path_owned, rollback_data_dir, Some(&backup));
+                                        let _ = remove_file_if_exists(backup.path());
+                                        res
+                                    })
+                                    .is_ok();
+                            if no_inline {
+                                if rollback_ok && let Some((_, new_data_path)) = has_data_dir_path_owned.as_ref() {
+                                    let _ = remove_dir_all_if_exists(new_data_path);
+                                }
+                                return Ok(FusedCommitAttempt::CommitErr(to_file_error(err).into()));
+                            }
+                            return Ok(FusedCommitAttempt::CommitErr(DiskError::from(err)));
+                        }
+                    } else if let Err(err) = fsync_inline_commit_std(
                         &dst_file_path_owned,
                         &dst_volume_dir_owned,
                         destination_was_absent,
@@ -1087,19 +1196,19 @@ impl LocalDisk {
                             .filter(|_| no_inline)
                             .map(|(_, new_data_path)| new_data_path.as_path()),
                         &guard,
-                    )
-                {
-                    if no_inline {
-                        return Ok(FusedCommitAttempt::CommitErr(to_file_error(err).into()));
+                    ) {
+                        if no_inline {
+                            return Ok(FusedCommitAttempt::CommitErr(to_file_error(err).into()));
+                        }
+                        return Ok(FusedCommitAttempt::CommitErr(DiskError::from(err)));
                     }
-                    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() {
+                if !flat_inline_tmp && let Some(parent) = src_file_path_owned.parent() {
                     let _ = std::fs::remove_dir(parent);
                 }
 
@@ -1364,7 +1473,7 @@ impl LocalDisk {
                             }
                         }
                     }
-                    match publication_owned.finish_on_thread(sync_commit_metadata, staging_cleanup) {
+                    match publication_owned.finish_on_thread(sync_commit_metadata, !cfg!(test), staging_cleanup) {
                         Ok(result) => Ok(MultipartFusedCommitOutcome::Committed(Box::new(RenameDataResp {
                             multipart_source: Some(Box::new(result)),
                             old_data_dir: has_old_data_dir,

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

@@ -856,6 +856,12 @@ pub(super) fn inspection_with_markers_bounded(
     let prepared_digest = prepared.as_deref().map(digest);
     let target_digest = target.as_deref().map(digest);
     let receipt_digest = receipt_bytes.as_deref().map(digest);
+    let finished = finished
+        || (committed
+            && prepared.is_none()
+            && before.is_none()
+            && target.is_none()
+            && current_digest == journal.as_ref().and_then(|j| j.target_digest));
     if let Some(journal) = &journal {
         if (!finished
             && journal.target_digest != target_digest
@@ -1113,13 +1119,34 @@ pub(super) fn inspection_with_markers_bounded(
     Ok((inspection, has_physical_data_dirs_or_markers))
 }
 
-type PreparedDeleteReceiptEntry = (DeleteRecoveryJournal, bool);
+#[derive(Clone)]
+struct PreparedDeleteReceiptEntry {
+    object_path: PathBuf,
+    bucket_path: PathBuf,
+    file_path: PathBuf,
+    staging_root: PathBuf,
+    journal_raw: Vec<u8>,
+    journal: DeleteRecoveryJournal,
+    sole_transaction: bool,
+}
+
+#[derive(Clone)]
+struct PublishedDeleteReceiptEntry {
+    object_path: PathBuf,
+    bucket_path: PathBuf,
+    file_path: PathBuf,
+    staging_root: PathBuf,
+    journal_raw: Vec<u8>,
+}
 
 const PREPARED_DELETE_RECEIPTS_PER_SHARD: usize = 64;
 
 static PREPARED_DELETE_RECEIPTS: std::sync::LazyLock<os::ShardedPathCache<PreparedDeleteReceiptEntry>> =
     std::sync::LazyLock::new(|| os::ShardedPathCache::new(PREPARED_DELETE_RECEIPTS_PER_SHARD));
 
+static PUBLISHED_DELETE_RECEIPTS: std::sync::LazyLock<os::ShardedPathCache<PublishedDeleteReceiptEntry>> =
+    std::sync::LazyLock::new(|| os::ShardedPathCache::new(PREPARED_DELETE_RECEIPTS_PER_SHARD));
+
 fn put_prepared_delete_receipt(rollback: impl AsRef<Path>, entry: PreparedDeleteReceiptEntry) {
     PREPARED_DELETE_RECEIPTS.insert(rollback.as_ref(), entry);
 }
@@ -1128,6 +1155,27 @@ fn take_prepared_delete_receipt(rollback: &Path) -> Option<PreparedDeleteReceipt
     PREPARED_DELETE_RECEIPTS.remove(rollback)
 }
 
+fn put_published_delete_receipt(rollback: impl AsRef<Path>, entry: PublishedDeleteReceiptEntry) {
+    PUBLISHED_DELETE_RECEIPTS.insert(rollback.as_ref(), entry);
+}
+
+fn take_published_delete_receipt(rollback: &Path) -> Option<PublishedDeleteReceiptEntry> {
+    PUBLISHED_DELETE_RECEIPTS.remove(rollback)
+}
+
+fn format_commit_receipt_bytes(journal_raw: &[u8], finished: bool) -> Vec<u8> {
+    let suffix: &[u8] = if finished {
+        b",\"finished\":true}"
+    } else {
+        b",\"finished\":false}"
+    };
+    let mut out = Vec::with_capacity(11 + journal_raw.len() + suffix.len());
+    out.extend_from_slice(b"{\"journal\":");
+    out.extend_from_slice(journal_raw);
+    out.extend_from_slice(suffix);
+    out
+}
+
 impl LocalDisk {
     // Native MPU source closure has its own conditional protocol. Generic
     // object recovery must never restore its head or acquire SourceRW under
@@ -2376,6 +2424,7 @@ impl LocalDisk {
         barrier(volume, object, transaction, stage, &self.root).await
     }
 
+    #[cfg(test)]
     pub(super) async fn try_fused_recoverable_delete_version(
         &self,
         volume: &str,
@@ -2384,6 +2433,21 @@ impl LocalDisk {
         opts: &DeleteOptions,
         transaction: Uuid,
         owner: Option<Arc<dyn Send + Sync>>,
+    ) -> Result<Option<()>> {
+        self.try_fused_recoverable_delete_version_with_paths(volume, object, fi, opts, transaction, owner, None)
+            .await
+    }
+
+    #[expect(clippy::too_many_arguments)]
+    pub(super) async fn try_fused_recoverable_delete_version_with_paths(
+        &self,
+        volume: &str,
+        object: &str,
+        fi: &FileInfo,
+        opts: &DeleteOptions,
+        transaction: Uuid,
+        owner: Option<Arc<dyn Send + Sync>>,
+        prevalidated_paths: Option<(PathBuf, PathBuf, PathBuf)>,
     ) -> Result<Option<()>> {
         if !opts.recoverable_delete
             || opts.undo_write
@@ -2396,11 +2460,22 @@ impl LocalDisk {
         self.reject_multipart_delete_recovery_mutation(volume, object)?;
         self.wait_for_startup_cleanup().await;
 
-        let (_, file_path) = self.io_get_bucket_and_object_path(volume, object)?;
-        check_path_length(file_path.to_string_lossy().as_ref())?;
+        let (object_path, bucket_path, file_path) = match prevalidated_paths {
+            Some(paths) => paths,
+            None => {
+                let obj = self.get_object_path(volume, object)?;
+                let (b, f) = self.io_get_bucket_and_object_path(volume, object)?;
+                check_path_length(f.to_string_lossy().as_ref())?;
+                (obj, b, f)
+            }
+        };
         let xl_path = path_join(&[file_path.as_path(), Path::new(STORAGE_FORMAT_FILE)]);
         let destination = file_path.join(transaction.to_string());
-        let staging_root = self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?;
+        let staging_root = if cfg!(target_os = "linux") {
+            self.io_root().join(RUSTFS_META_TMP_BUCKET)
+        } else {
+            self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?
+        };
         let sync = effective_durability(volume).syncs_commit_metadata();
         let incarnation = opts.delete_recovery_bucket_incarnation;
         let lease = os::acquire_namespace_mutation_lease_with_owner(&destination, owner).await;
@@ -2578,6 +2653,7 @@ impl LocalDisk {
                     Err(err) => return Err(err),
                 }
                 let staged_dir = staged.keep();
+                os::record_native_publication(&destination);
                 if let Err(err) = std::fs::rename(&staged_dir, &destination) {
                     let _ = std::fs::remove_dir_all(&staged_dir);
                     return Err(err);
@@ -2587,6 +2663,8 @@ impl LocalDisk {
                     let _ = std::fs::remove_dir_all(&destination);
                     return Err(err);
                 }
+                #[cfg(all(test, unix))]
+                fused_prepare_checkpoint::run(&file_path);
                 let mut reserved_data_path = None;
                 if let Some(old_path) = old_data_path {
                     let reserved_marker = old_path.join(format!("{RESERVED_DELETE_DATA_DIR_MARKER_PREFIX}{transaction}"));
@@ -2606,9 +2684,11 @@ impl LocalDisk {
                     reserved_data_path = Some((old_path, reserved_marker));
                 }
                 if let Some(ref target_bytes) = target {
+                    // A provisional head must stay outside the published recovery transaction.
                     let target_in_rollback = destination.join(TARGET);
                     let linked_xl = staging_root.join(format!("delete-xl-{transaction}"));
                     let _ = std::fs::remove_file(&linked_xl);
+                    os::record_native_publication(&xl_path);
                     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() {
@@ -2653,7 +2733,18 @@ impl LocalDisk {
                         Err(err) => return Err(err),
                     }
                 }
-                put_prepared_delete_receipt(destination, (journal, sole_transaction));
+                put_prepared_delete_receipt(
+                    destination,
+                    PreparedDeleteReceiptEntry {
+                        object_path,
+                        bucket_path,
+                        file_path,
+                        staging_root,
+                        journal_raw,
+                        journal,
+                        sole_transaction,
+                    },
+                );
                 Ok(Ok(Some(old_dir)))
             },
         )
@@ -2690,19 +2781,45 @@ impl LocalDisk {
         namespace_owner: Option<Arc<dyn Send + Sync>>,
         work_budget: Option<Arc<crate::object_api::DeleteRecoveryWorkBudget>>,
     ) -> Result<Option<(DeleteRecoveryJournal, bool, bool)>> {
-        self.reject_multipart_delete_recovery_mutation(volume, object)?;
-        if !safe_path(object, false) || transaction.is_nil() {
+        if transaction.is_nil() {
             return Err(DiskError::InvalidPath);
         }
-        let object_path = self.get_object_path(volume, object)?;
+        let rollback_open_path = if volume != RUSTFS_META_MULTIPART_BUCKET {
+            self.io_get_object_open_path(volume, object)
+                .ok()
+                .map(|p| p.join(transaction.to_string()))
+        } else {
+            None
+        };
+        let cached_prepared = rollback_open_path.as_deref().and_then(take_prepared_delete_receipt);
+        let (object_path, bucket_path, root, staging_root, cached_receipt) = match cached_prepared {
+            Some(entry) => (
+                entry.object_path,
+                Some(entry.bucket_path),
+                entry.file_path,
+                entry.staging_root,
+                Some((entry.journal_raw, entry.journal, entry.sole_transaction)),
+            ),
+            None => {
+                self.reject_multipart_delete_recovery_mutation(volume, object)?;
+                if !safe_path(object, false) {
+                    return Err(DiskError::InvalidPath);
+                }
+                let obj_path = self.get_object_path(volume, object)?;
+                (
+                    obj_path,
+                    None,
+                    self.io_get_object_path(volume, object)?,
+                    self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?,
+                    None,
+                )
+            }
+        };
         let metadata_owner = os::acquire_metadata_mutation_lease(&object_path, namespace_owner).await;
         let volume_guard = os::disk_volume_mutation_lock(&self.root, volume).read_owned().await;
         let owner: Option<Arc<dyn Send + Sync>> = Some(Arc::new((metadata_owner, volume_guard)));
-
-        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 staging_root = self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?;
         let sync = effective_durability(volume).syncs_commit_metadata();
         let lease = os::acquire_namespace_mutation_lease_with_owner(&receipt_path, owner).await;
         let volume_str = volume.to_owned();
@@ -2712,23 +2829,39 @@ impl LocalDisk {
             lease,
             sync,
             move || -> std::io::Result<Result<Option<(DeleteRecoveryJournal, bool, bool)>>> {
-                if let Some((journal, sole_transaction)) = take_prepared_delete_receipt(&rollback)
+                if let Some((journal_raw, journal, sole_transaction)) = cached_receipt
                     && 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()?;
+                    let tmp_receipt = staging_root.join(format!("receipt-{transaction}.tmp"));
+                    let receipt_bytes = format_commit_receipt_bytes(&journal_raw, false);
+                    let write_res = (|| -> std::io::Result<()> {
+                        let mut file = std::fs::File::create(&tmp_receipt)?;
+                        std::io::Write::write_all(&mut file, &receipt_bytes)?;
+                        if sync && !receipt_bytes.is_empty() {
+                            file.sync_data()?;
+                        }
+                        os::record_native_publication(&receipt_path);
+                        std::fs::rename(&tmp_receipt, &receipt_path)?;
+                        if sync {
+                            os::fsync_dir_std(&rollback)?;
+                        }
+                        Ok(())
+                    })();
+                    if let Err(err) = write_res {
+                        let _ = std::fs::remove_file(&tmp_receipt);
+                        return Err(err);
                     }
-                    std::fs::rename(staged.path(), &receipt_path)?;
-                    if sync {
-                        os::fsync_dir_std(&rollback)?;
+                    if let Some(bucket_path) = bucket_path {
+                        put_published_delete_receipt(
+                            rollback,
+                            PublishedDeleteReceiptEntry {
+                                object_path,
+                                bucket_path,
+                                file_path: root,
+                                staging_root,
+                                journal_raw,
+                            },
+                        );
                     }
                     return Ok(Ok(Some((journal, sole_transaction, true))));
                 }
@@ -2804,6 +2937,7 @@ 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)?;
@@ -2851,14 +2985,46 @@ impl LocalDisk {
                 Err(err) => Err(err),
             }
         }
-        self.reject_multipart_delete_recovery_mutation(volume, object)?;
-        if !safe_path(object, false) || transaction.is_nil() {
+        if transaction.is_nil() {
             return Err(DiskError::InvalidPath);
         }
-        let object_path = self.get_object_path(volume, object)?;
+        let rollback_open_path = if volume != RUSTFS_META_MULTIPART_BUCKET {
+            self.io_get_object_open_path(volume, object)
+                .ok()
+                .map(|p| p.join(transaction.to_string()))
+        } else {
+            None
+        };
+        let cached_published = rollback_open_path.as_deref().and_then(take_published_delete_receipt);
+        let (object_path, bucket_path, staging_root, root, cached_journal_raw) = match cached_published {
+            Some(entry) => (
+                entry.object_path,
+                entry.bucket_path,
+                entry.staging_root,
+                entry.file_path,
+                Some(entry.journal_raw),
+            ),
+            None => {
+                self.reject_multipart_delete_recovery_mutation(volume, object)?;
+                if !safe_path(object, false) {
+                    return Err(DiskError::InvalidPath);
+                }
+                let obj_path = self.get_object_path(volume, object)?;
+                (
+                    obj_path,
+                    self.io_get_bucket_path(volume)?,
+                    self.io_get_bucket_path(RUSTFS_META_TMP_BUCKET)?,
+                    self.io_get_object_path(volume, object)?,
+                    None,
+                )
+            }
+        };
         let metadata_owner = os::acquire_metadata_mutation_lease(&object_path, namespace_owner).await;
         let volume_guard = os::disk_volume_mutation_lock(&self.root, volume).read_owned().await;
         let owner: Option<Arc<dyn Send + Sync>> = Some(Arc::new((metadata_owner, volume_guard)));
+        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 mut claims: Vec<SnapshotLeaseDeleteClaim> = Vec::with_capacity(journal.data_dirs.len());
         for data_dir in &journal.data_dirs {
@@ -2880,15 +3046,9 @@ impl LocalDisk {
             claims.push(claim);
         }
 
-        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 data_dirs = journal.data_dirs.clone();
         let has_target = journal.target_digest.is_some();
-        let journal_for_receipt = journal.clone();
+        let journal_for_receipt = cached_journal_raw.is_none().then(|| journal.clone());
         let lease = os::acquire_namespace_mutation_lease_with_owner(&receipt_path, owner).await;
         let prune_paths: Vec<_> = rollback
             .ancestors()
@@ -2897,6 +3057,32 @@ impl LocalDisk {
             .collect();
 
         let res = os::run_blocking_namespace_operation_with_sync(lease, sync, move || -> std::io::Result<()> {
+            let persist_receipt = |finished: bool| -> std::io::Result<()> {
+                let receipt_bytes = match &cached_journal_raw {
+                    Some(raw) => format_commit_receipt_bytes(raw, finished),
+                    None => {
+                        let receipt = CommitReceipt {
+                            journal: journal_for_receipt
+                                .clone()
+                                .ok_or_else(|| std::io::Error::other("missing journal for receipt"))?,
+                            finished,
+                        };
+                        serde_json::to_vec(&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()?;
+                }
+                os::record_native_publication(&receipt_path);
+                std::fs::rename(staged_receipt.path(), &receipt_path)?;
+                if sync {
+                    let _ = os::fsync_dir_std(&rollback);
+                }
+                Ok(())
+            };
             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}");
@@ -2951,38 +3137,38 @@ impl LocalDisk {
                 os::fsync_dir_std(&root)?;
             }
             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)?;
+                if let Err(err) = ignore_not_found(std::fs::remove_file(rollback.join(name))) {
+                    if has_target || had_physical_data_dirs {
+                        let _ = persist_receipt(true);
+                    }
+                    return Err(err);
                 }
             }
-            for name in [TARGET, PREPARED, FINISHED] {
+            for name in [PREPARED, TARGET, 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 let Err(err) = ignore_not_found(std::fs::remove_file(path)) {
+                    if has_target || had_physical_data_dirs {
+                        let _ = persist_receipt(true);
+                    }
+                    return Err(err);
+                }
             }
-            if sync {
-                os::fsync_dir_std(&rollback)?;
+            if sync && let Err(err) = os::fsync_dir_std(&rollback) {
+                if has_target || had_physical_data_dirs {
+                    let _ = persist_receipt(true);
+                }
+                return Err(err);
             }
             #[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))?;
+            if let Err(err) = ignore_not_found(std::fs::remove_file(&receipt_path)) {
+                if has_target || had_physical_data_dirs {
+                    let _ = persist_receipt(true);
+                }
+                return Err(err);
+            }
             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);
@@ -3105,6 +3291,42 @@ pub(super) async fn restore_barrier(volume: &str, object_dir: &Path, transaction
     Ok(())
 }
 
+#[cfg(all(test, unix))]
+mod fused_prepare_checkpoint {
+    use std::collections::HashMap;
+    use std::path::{Path, PathBuf};
+    use std::sync::{LazyLock, Mutex};
+
+    type Hook = Box<dyn FnOnce() + Send>;
+    static HOOKS: LazyLock<Mutex<HashMap<PathBuf, Hook>>> = LazyLock::new(|| Mutex::new(HashMap::new()));
+
+    pub(super) struct Guard(PathBuf);
+
+    impl Drop for Guard {
+        fn drop(&mut self) {
+            HOOKS.lock().expect("checkpoint hooks").remove(&self.0);
+        }
+    }
+
+    pub(super) fn install(root: &Path, hook: impl FnOnce() + Send + 'static) -> Guard {
+        assert!(
+            HOOKS
+                .lock()
+                .expect("checkpoint hooks")
+                .insert(root.to_path_buf(), Box::new(hook))
+                .is_none()
+        );
+        Guard(root.to_path_buf())
+    }
+
+    pub(super) fn run(root: &Path) {
+        let hook = HOOKS.lock().expect("checkpoint hooks").remove(root);
+        if let Some(hook) = hook {
+            hook();
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -7020,6 +7242,121 @@ mod tests {
         marker.deleted = false;
     }
 
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn delete_recovery_fused_prepare_checkpoint_reopens_before_head_publication() {
+        use std::os::unix::fs::MetadataExt;
+
+        let _mode = durability_mode_override::set(DurabilityMode::Strict);
+        let (_temp, disk, mut first, _) = fixture().await;
+        first.version_id = Some(Uuid::new_v4());
+        let mut second = first.clone();
+        second.version_id = Some(Uuid::new_v4());
+        let mut meta = FileMeta::new();
+        meta.add_version(first).expect("add surviving version");
+        meta.add_version(second.clone()).expect("add deleted version");
+        let original = meta.marshal_msg().expect("encode original versions");
+        let (_, root) = disk.io_get_bucket_and_object_path("recovery", "object").expect("object path");
+        fs::write(root.join(STORAGE_FORMAT_FILE), &original)
+            .await
+            .expect("write versions");
+
+        let (checkpoint_temp, checkpoint_disk, _, _) = fixture().await;
+        let checkpoint_root = checkpoint_temp.path().join("recovery/object");
+        drop(checkpoint_disk);
+        let transaction = Uuid::new_v4();
+        let source_root = root.clone();
+        let copy_root = checkpoint_root.clone();
+        let _capture = fused_prepare_checkpoint::install(&root, move || {
+            let source = source_root.join(transaction.to_string());
+            let destination = copy_root.join(transaction.to_string());
+            std::fs::create_dir(&destination).expect("create checkpoint transaction");
+            let source_head = source_root.join(STORAGE_FORMAT_FILE);
+            let source_head_meta = std::fs::metadata(&source_head).expect("stat source head");
+            let copy_head = copy_root.join(STORAGE_FORMAT_FILE);
+            std::fs::copy(&source_head, &copy_head).expect("capture head before publication");
+            for entry in std::fs::read_dir(source).expect("read published preparation") {
+                let entry = entry.expect("prepared file");
+                assert!(entry.file_type().expect("prepared file type").is_file());
+                let dst = destination.join(entry.file_name());
+                if entry.file_name() == STORAGE_FORMAT_FILE_BACKUP {
+                    let backup_meta = entry.metadata().expect("stat prepared backup");
+                    assert_eq!(
+                        (backup_meta.dev(), backup_meta.ino()),
+                        (source_head_meta.dev(), source_head_meta.ino()),
+                        "fused preparation must hardlink backup to live head"
+                    );
+                    std::fs::hard_link(&copy_head, &dst).expect("preserve hardlinked backup");
+                } else {
+                    std::fs::copy(entry.path(), dst).expect("capture prepared file");
+                }
+            }
+        });
+        let opts = DeleteOptions {
+            recoverable_delete: true,
+            ..Default::default()
+        };
+        assert_eq!(
+            disk.try_fused_recoverable_delete_version("recovery", "object", &second, &opts, transaction, None)
+                .await
+                .expect("finish original preparation"),
+            Some(())
+        );
+        assert!(checkpoint_root.join(transaction.to_string()).join(PREPARED).is_file());
+        let endpoint =
+            Endpoint::try_from(checkpoint_temp.path().to_str().expect("checkpoint path")).expect("checkpoint endpoint");
+        let reopened = LocalDisk::new(&endpoint, false)
+            .await
+            .expect("reopen checkpoint as a fresh disk");
+        let inspection = reopened
+            .delete_recovery_inspect("recovery", "object", transaction)
+            .await
+            .expect("actual preparation checkpoint must remain recoverable");
+        assert_eq!(inspection.state, DeleteRecoveryState::NewPrepared);
+        let reopened_tx = reopened
+            .io_get_object_path("recovery", &format!("object/{transaction}"))
+            .expect("reopened transaction path");
+        os::fsync_dir_recorder::set_failure(&reopened_tx, ErrorKind::Other);
+        assert!(
+            reopened
+                .settle_delete_transaction(
+                    "recovery",
+                    "object",
+                    transaction,
+                    inspection.fingerprint,
+                    DeleteRecoveryAction::Rollback
+                )
+                .await
+                .is_err(),
+            "failed transaction directory sync after backup removal must retain journal"
+        );
+        assert!(reopened_tx.join(PREPARED).is_file());
+        assert!(!reopened_tx.join(STORAGE_FORMAT_FILE_BACKUP).exists());
+        let retry_inspection = reopened
+            .delete_recovery_inspect("recovery", "object", transaction)
+            .await
+            .expect("inspect interrupted rollback");
+        assert_eq!(retry_inspection.state, DeleteRecoveryState::NewPrepared);
+        let settled = reopened
+            .settle_delete_transaction(
+                "recovery",
+                "object",
+                transaction,
+                retry_inspection.fingerprint,
+                DeleteRecoveryAction::Rollback,
+            )
+            .await
+            .expect("roll back interrupted preparation");
+        assert_eq!(settled.state, DeleteRecoveryState::Absent, "same-inode backup must not survive rollback");
+        assert!(!checkpoint_root.join(transaction.to_string()).exists());
+        assert_eq!(
+            fs::read(checkpoint_root.join(STORAGE_FORMAT_FILE))
+                .await
+                .expect("read rolled-back versions"),
+            original
+        );
+    }
+
     #[tokio::test]
     async fn delete_recovery_fused_versioned_delete_with_surviving_version_and_interrupted_target_cleanup() {
         let (_temp, disk, mut v1, _) = fixture().await;

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

@@ -372,29 +372,32 @@ impl IndexOwner {
         #[cfg(unix)]
         {
             use rustix::fs::{Mode, OFlags, mkdirat, openat};
-            let mut directories = vec![self._identity_pin.try_clone()?];
-            for component in [".rustfs.sys", "multipart-index-v1", &sha(bucket)] {
-                let parent = directories.last().ok_or_else(|| io::Error::other("index parent missing"))?;
-                match mkdirat(parent, component, Mode::RWXU | Mode::RWXG | Mode::RWXO) {
-                    Ok(()) | Err(rustix::io::Errno::EXIST) => {}
-                    Err(error) => return Err(error.into()),
+            let open_dir = |parent: rustix::fd::BorrowedFd<'_>, component: &str| -> io::Result<rustix::fd::OwnedFd> {
+                let flags = OFlags::RDONLY | OFlags::DIRECTORY | OFlags::NOFOLLOW | OFlags::CLOEXEC;
+                match openat(parent, component, flags, Mode::empty()) {
+                    Ok(dir) => Ok(dir),
+                    Err(rustix::io::Errno::NOENT) => {
+                        match mkdirat(parent, component, Mode::RWXU | Mode::RWXG | Mode::RWXO) {
+                            Ok(()) | Err(rustix::io::Errno::EXIST) => {}
+                            Err(error) => return Err(error.into()),
+                        }
+                        openat(parent, component, flags, Mode::empty()).map_err(Into::into)
+                    }
+                    Err(error) => Err(error.into()),
                 }
-                let directory = openat(
-                    parent,
-                    component,
-                    OFlags::RDONLY | OFlags::DIRECTORY | OFlags::NOFOLLOW | OFlags::CLOEXEC,
-                    Mode::empty(),
-                )?;
-                directories.push(std::fs::File::from(directory));
-            }
-            let parent = directories.last().ok_or_else(|| io::Error::other("index parent missing"))?;
+            };
+            use std::os::fd::AsFd;
+            let bucket_sha = sha(bucket);
+            let sys_dir = open_dir(self._identity_pin.as_fd(), ".rustfs.sys")?;
+            let index_dir = open_dir(sys_dir.as_fd(), "multipart-index-v1")?;
+            let parent = open_dir(index_dir.as_fd(), &bucket_sha)?;
             // The index is boot-local: every start rescans the staging area
             // before serving it, so a marker lost to a crash is recreated before
             // it can be consulted. Creation is enough; no fsync of the marker or
             // its ancestors. An existing marker is touched so pruning sees the
             // new upload's age, not the previous one's.
             match openat(
-                parent,
+                &parent,
                 source_root,
                 OFlags::CREATE | OFlags::EXCL | OFlags::WRONLY | OFlags::NOFOLLOW | OFlags::CLOEXEC,
                 Mode::RUSR | Mode::WUSR,
@@ -409,7 +412,7 @@ impl IndexOwner {
                     // the previous upload's age, which only shortens the prune
                     // grace for this key; index completeness does not depend on it.
                     if let Ok(marker) =
-                        openat(parent, source_root, OFlags::WRONLY | OFlags::NOFOLLOW | OFlags::CLOEXEC, Mode::empty())
+                        openat(&parent, source_root, OFlags::WRONLY | OFlags::NOFOLLOW | OFlags::CLOEXEC, Mode::empty())
                     {
                         let _ = std::fs::File::from(marker).set_modified(std::time::SystemTime::now());
                     }
@@ -430,6 +433,18 @@ impl IndexOwner {
     }
 
     fn ensure_current_root(&self) -> io::Result<()> {
+        #[cfg(unix)]
+        let current = {
+            use std::os::unix::fs::MetadataExt;
+            let metadata = std::fs::symlink_metadata(&*self.root_path)?;
+            if !metadata.is_dir() {
+                return Err(io::Error::other("multipart index physical root is not a directory"));
+            }
+            let mut inode = [0; 16];
+            inode[..8].copy_from_slice(&metadata.ino().to_le_bytes());
+            (metadata.dev(), inode)
+        };
+        #[cfg(not(unix))]
         let current = os::PublicationRoot::new(&self.root_path)?.physical_identity()?;
         if current != self.physical {
             self.disabled.store(true, Ordering::Release);

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

@@ -1979,25 +1979,28 @@ impl NativeMultipartPublication {
         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));
+                if cfg!(test) || !can_mark_presynced {
+                    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)?;
@@ -2050,7 +2053,7 @@ impl NativeMultipartPublication {
                 let _ = std::fs::remove_file(prepared_path);
                 return Err(err);
             }
-            os::install_multipart_prepared_head(prepared_metadata_source, &canonical, base, root)
+            os::install_multipart_prepared_head(prepared_metadata_source, &canonical, base, root, Some(directory_guard))
         })()
         .map_err(DiskError::from)?;
         *self.state.lock() = state;
@@ -2160,7 +2163,7 @@ impl NativeMultipartPublication {
             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)
+            os::install_multipart_prepared_head(prepared_metadata_source, &canonical, &base, &root, Some(&owned_directory_guard))
         })
         .await
         .map_err(DiskError::from)?;
@@ -2207,7 +2210,7 @@ impl NativeMultipartPublication {
         let base = self.store.base.clone();
         let root = self.store.publication_root.clone();
         os::run_blocking_namespace_operation(self.store.lease.clone(), move || {
-            os::install_multipart_prepared_head(prepared, &canonical, &base, &root)
+            os::install_multipart_prepared_head(prepared, &canonical, &base, &root, None)
         })
         .await
         .map_err(DiskError::from)
@@ -2264,6 +2267,7 @@ impl NativeMultipartPublication {
     pub(super) fn finish_on_thread(
         &self,
         target_already_synced: bool,
+        fused_in_place: bool,
         staging_cleanup: Option<(PathBuf, Arc<std::sync::atomic::AtomicBool>)>,
     ) -> Result<MultipartPublicationResult> {
         let mut state = self.state.lock().clone();
@@ -2293,62 +2297,64 @@ impl NativeMultipartPublication {
         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 {
+            if !fused_in_place {
+                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));
                 }
-            } 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));
+                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
                         }
-                        Ok(())
-                    },
-                )
-                .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)?;
+            SourceStore::persist_state_std(source, base, root, &next_state_raw, terminal.as_deref(), !fused_in_place)?;
             if let Some((io_path, cleaned)) = staging_cleanup {
                 match std::fs::remove_dir(&io_path) {
                     Ok(()) => cleaned.store(true, std::sync::atomic::Ordering::Release),
@@ -2381,7 +2387,7 @@ impl NativeMultipartPublication {
     ) -> 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)
+            this.finish_on_thread(target_already_synced, false, staging_cleanup)
                 .map_err(IoError::other)
         })
         .await

crates/ecstore/src/disk/mod.rs

@@ -732,6 +732,13 @@ 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 {
@@ -1598,6 +1605,11 @@ 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>>;
@@ -2222,6 +2234,36 @@ pub struct WalkDirOptions {
     // Override the remote stream stall timeout for long background walks.
     #[serde(default)]
     pub stall_timeout_ms: Option<u64>,
+
+    /// Select this leaf below identifier-like directory components. The root is
+    /// `base_dir`; unrelated object metadata must not be read or returned.
+    /// Older peers ignore this additive field, so consumers must also filter.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub identifier_tree_leaf: Option<String>,
+}
+
+pub(crate) fn is_identifier_tree_component(value: &str) -> bool {
+    !value.is_empty()
+        && value
+            .bytes()
+            .all(|byte| byte.is_ascii_alphanumeric() || matches!(byte, b'_' | b'-'))
+}
+
+pub(crate) fn is_identifier_tree_leaf(value: &str) -> bool {
+    value.len() <= 255
+        && value.contains('.')
+        && !matches!(value, "." | "..")
+        && value
+            .bytes()
+            .all(|byte| byte.is_ascii_alphanumeric() || matches!(byte, b'_' | b'-' | b'.'))
+}
+
+pub(crate) fn identifier_tree_leaf_matches(prefix: &str, leaf: &str, object: &str) -> bool {
+    let Some(relative) = object.strip_prefix(prefix) else {
+        return false;
+    };
+    let mut components = relative.split('/');
+    components.next_back() == Some(leaf) && components.all(is_identifier_tree_component)
 }
 
 impl WalkDirOptions {
@@ -2423,7 +2465,7 @@ impl DeleteRecoveryJournal {
             }
             let id = request.version_id.unwrap_or_default();
             let Some(original) = originals.get(&id) else {
-                return false;
+                return !id.is_nil() && !locals.contains_key(&id);
             };
             match locals.get(&id) {
                 Some(local) => local == original,
@@ -2473,7 +2515,10 @@ impl DeleteRecoveryJournal {
             };
             match (local, remote) {
                 (Some(local), Some(remote)) => local == remote,
-                (None, None) => self.original_digest.is_none() && authority.original_digest.is_none(),
+                (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())
+                }
                 _ => false,
             }
         })
@@ -3157,6 +3202,7 @@ mod tests {
             skip_total_timeout: false,
             timeout_ms: Some(10_000),
             stall_timeout_ms: Some(20_000),
+            identifier_tree_leaf: None,
         };
 
         assert_eq!(opts.bucket, "test-bucket");
@@ -3191,6 +3237,69 @@ mod tests {
         assert!(!decoded.skip_hidden_prefix_check);
     }
 
+    #[test]
+    fn identifier_tree_leaf_selection_preserves_nested_identifier_directories() {
+        let prefix = "catalog/namespaces/";
+        let leaf = "table-entry.json";
+        for relative in [
+            "sales/tables/orders/table-entry.json",
+            "sales/tables/orders/maintenance/tables/owner/table-entry.json",
+            "maintenance/tables/tables/table-entry.json",
+        ] {
+            assert!(identifier_tree_leaf_matches(prefix, leaf, &format!("{prefix}{relative}")));
+        }
+        for relative in [
+            "sales/tables/orders/not-table-entry.json",
+            "sales/tables/orders/table-entry.json/child",
+            "sales/report.json/tables/owner/table-entry.json",
+            "sales//tables/owner/table-entry.json",
+            "sales/../tables/owner/table-entry.json",
+        ] {
+            assert!(!identifier_tree_leaf_matches(prefix, leaf, &format!("{prefix}{relative}")));
+        }
+        assert!(!identifier_tree_leaf_matches(prefix, leaf, "other/sales/tables/orders/table-entry.json"));
+        assert!(is_identifier_tree_leaf(leaf));
+        for invalid in ["", ".", "..", "table-entry", "table/entry.json", "table\\entry.json"] {
+            assert!(!is_identifier_tree_leaf(invalid), "invalid selector {invalid}");
+        }
+    }
+
+    #[test]
+    fn identifier_tree_walk_options_are_additive_on_json_and_messagepack_maps() {
+        #[derive(serde::Deserialize)]
+        struct LegacyView {
+            bucket: String,
+            base_dir: String,
+            recursive: bool,
+        }
+        let options = WalkDirOptions {
+            bucket: "catalog".to_string(),
+            base_dir: "namespaces/".to_string(),
+            recursive: true,
+            identifier_tree_leaf: Some("table-entry.json".to_string()),
+            ..Default::default()
+        };
+        let json = serde_json::to_vec(&options).expect("encode current HTTP options");
+        let messagepack = rmp_serde::to_vec_named(&options).expect("encode legacy RPC map");
+        for old in [
+            serde_json::from_slice::<LegacyView>(&json).expect("old JSON readers ignore the additive field"),
+            rmp_serde::from_slice::<LegacyView>(&messagepack).expect("old MessagePack readers ignore the additive field"),
+        ] {
+            assert_eq!(old.bucket, options.bucket);
+            assert_eq!(old.base_dir, options.base_dir);
+            assert!(old.recursive);
+        }
+        let mut old = serde_json::to_value(&options).expect("encode legacy options");
+        old.as_object_mut().expect("options are a map").remove("identifier_tree_leaf");
+        let from_json: WalkDirOptions = serde_json::from_value(old.clone()).expect("decode old HTTP options");
+        let from_messagepack: WalkDirOptions =
+            rmp_serde::from_slice(&rmp_serde::to_vec_named(&old).expect("encode old RPC map")).expect("decode old RPC map");
+        assert!(from_json.identifier_tree_leaf.is_none());
+        assert!(from_messagepack.identifier_tree_leaf.is_none());
+        assert_eq!(from_json.base_dir, options.base_dir);
+        assert_eq!(from_messagepack.base_dir, options.base_dir);
+    }
+
     /// Test DeleteOptions structure
     #[test]
     fn test_delete_options() {
@@ -3887,6 +3996,46 @@ 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)));
@@ -3902,6 +4051,25 @@ 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/disk/os.rs

@@ -21,7 +21,7 @@ use futures::TryStreamExt;
 use parking_lot::Mutex;
 use rustfs_utils::path::SLASH_SEPARATOR;
 use std::{
-    collections::{HashMap, VecDeque},
+    collections::{HashMap, HashSet, VecDeque},
     io,
     path::{Component, Path, PathBuf},
     sync::{Arc, LazyLock, Weak},
@@ -350,6 +350,14 @@ pub(crate) mod prepared_publication_test_hooks {
         Rollback,
         #[cfg(test)]
         DirFsync,
+        #[cfg(test)]
+        CreateDirectory,
+        #[cfg(test)]
+        NativeRootIdentity,
+        #[cfg(all(test, target_os = "linux"))]
+        DirectWriteOpen,
+        #[cfg(all(test, target_os = "linux"))]
+        BeforeDataUsageLayout,
     }
 
     type Hook = Box<dyn FnOnce() + Send>;
@@ -2161,6 +2169,226 @@ fn fsync_spawn_blocking<T: Send + 'static>(f: impl FnOnce() -> T + Send + 'stati
     })
 }
 static DISK_VOLUME_MUTATION_LOCKS: LazyLock<ShardedPathLockRegistry<RwLock<()>>> = LazyLock::new(ShardedPathLockRegistry::new);
+
+#[cfg(windows)]
+type NativeRootIdentity = (u64, [u8; 16]);
+#[cfg(not(windows))]
+type NativeRootIdentity = (u64, u64);
+
+// Publication history outlives disk handles, cleanup, and metadata snapshots.
+// Reusing a pathname or descriptor for another physical root is not new proof.
+#[derive(Debug, Default)]
+struct NativePublicationHistory {
+    roots: HashMap<NativeRootIdentity, NativeRootHistory>,
+    aliases: HashMap<PathBuf, HashSet<NativeRootIdentity>>,
+}
+
+#[derive(Debug, Default)]
+struct NativeRootHistory {
+    unknown: bool,
+    started: HashSet<String>,
+}
+
+static NATIVE_PUBLICATION_HISTORY: LazyLock<Mutex<NativePublicationHistory>> =
+    LazyLock::new(|| Mutex::new(NativePublicationHistory::default()));
+
+#[derive(Clone, Debug)]
+pub(crate) struct NativePublicationRegistration {
+    identity: NativeRootIdentity,
+    root: PathBuf,
+    io_root: PathBuf,
+}
+
+/// Opaque, process-local observation. Unsupported disks cannot construct it.
+#[derive(Clone, Debug)]
+pub struct NativeBucketCreationWitness {
+    registration: NativePublicationRegistration,
+    backend: Arc<dyn super::local::LocalIoBackend>,
+    bucket: String,
+    disk_id: uuid::Uuid,
+}
+
+fn native_root_identity(path: &Path) -> Option<NativeRootIdentity> {
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::MetadataExt;
+        #[cfg(test)]
+        prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::NativeRootIdentity, path);
+        let metadata = std::fs::metadata(path).ok()?;
+        metadata.is_dir().then(|| (metadata.dev(), metadata.ino()))
+    }
+    #[cfg(windows)]
+    {
+        // Resolve and validate the same directory anchor used for publication;
+        // keep its volume and complete 128-bit file ID in the history key.
+        let (_, directory) = open_windows_publication_root(path).ok()?;
+        windows_file_identity(directory.handle.as_ref()).ok()
+    }
+    #[cfg(all(not(unix), not(windows)))]
+    {
+        let _ = path;
+        None
+    }
+}
+
+pub(crate) fn register_native_publication_root(root: &Path, io_root: &Path) -> Option<NativePublicationRegistration> {
+    let identity = native_root_identity(root)?;
+    if native_root_identity(io_root) != Some(identity) {
+        return None;
+    }
+    let mut history = NATIVE_PUBLICATION_HISTORY.lock();
+    history.roots.entry(identity).or_default();
+    for alias in [root, io_root] {
+        let alias: PathBuf = alias.components().collect();
+        let previous = history.aliases.entry(alias).or_default();
+        let collisions: Vec<_> = previous.iter().copied().filter(|old| *old != identity).collect();
+        previous.insert(identity);
+        if !collisions.is_empty() {
+            history.roots.get_mut(&identity)?.unknown = true;
+            for old in collisions {
+                history.roots.get_mut(&old)?.unknown = true;
+            }
+        }
+    }
+    Some(NativePublicationRegistration {
+        identity,
+        root: root.to_path_buf(),
+        io_root: io_root.to_path_buf(),
+    })
+}
+
+/// Record before queueing filesystem work. Cancellation never removes Started.
+pub(crate) fn record_native_publication(path: &Path) {
+    let aliases: Vec<_> = {
+        let history = NATIVE_PUBLICATION_HISTORY.lock();
+        path.ancestors()
+            .filter(|alias| history.aliases.contains_key(*alias))
+            .map(Path::to_path_buf)
+            .collect()
+    };
+    // Physical lookups must never hold the process-wide publication mutex.
+    let observations: Vec<_> = aliases
+        .into_iter()
+        .filter_map(|alias| {
+            let relative = path.strip_prefix(&alias).ok()?;
+            let mut components = relative.components().filter(|part| !matches!(part, Component::CurDir));
+            let first = components.next()?;
+            let bucket = match first {
+                Component::Normal(bucket) if components.all(|part| matches!(part, Component::Normal(_))) => {
+                    bucket.to_str().map(str::to_owned)
+                }
+                _ => None,
+            };
+            let physical = native_root_identity(&alias);
+            Some((alias, physical, bucket))
+        })
+        .collect();
+    let mut history = NATIVE_PUBLICATION_HISTORY.lock();
+    for (alias, physical, bucket) in observations {
+        // Aliases only accumulate identities. Include registration concurrent
+        // with the stat, rather than using a stale captured identity list.
+        let identities = history.aliases.get(&alias).cloned().unwrap_or_default();
+        if physical.is_none() || identities.iter().any(|identity| Some(*identity) != physical) {
+            for identity in identities {
+                history.roots.entry(identity).or_default().unknown = true;
+            }
+            if let Some(identity) = physical {
+                history.roots.entry(identity).or_default().unknown = true;
+            }
+        } else if let Some(identity) = physical {
+            let root = history.roots.entry(identity).or_default();
+            match bucket {
+                Some(bucket) => {
+                    root.started.insert(bucket);
+                }
+                None => root.unknown = true,
+            }
+        }
+    }
+}
+
+impl NativePublicationRegistration {
+    pub(crate) fn invalidate(&self) {
+        if let Some(root) = NATIVE_PUBLICATION_HISTORY.lock().roots.get_mut(&self.identity) {
+            root.unknown = true;
+        }
+    }
+
+    pub(crate) fn witness(
+        &self,
+        bucket: &str,
+        disk_id: uuid::Uuid,
+        backend: Arc<dyn super::local::LocalIoBackend>,
+    ) -> Option<NativeBucketCreationWitness> {
+        if disk_id.is_nil()
+            || !matches!(Path::new(bucket).components().next(), Some(Component::Normal(_)))
+            || Path::new(bucket).components().count() != 1
+        {
+            return None;
+        }
+        let witness = NativeBucketCreationWitness {
+            registration: self.clone(),
+            backend,
+            bucket: bucket.to_owned(),
+            disk_id,
+        };
+        validate_native_creation_witnesses(std::slice::from_ref(&witness)).then_some(witness)
+    }
+}
+
+impl NativeBucketCreationWitness {
+    pub(crate) fn matches(&self, other: &Self) -> bool {
+        self.registration.identity == other.registration.identity
+            && Arc::ptr_eq(&self.backend, &other.backend)
+            && self.disk_id == other.disk_id
+            && self.bucket == other.bucket
+    }
+
+    fn physical_root_is_current(&self) -> bool {
+        let root = native_root_identity(&self.registration.root);
+        let io_root = native_root_identity(&self.registration.io_root);
+        if root == Some(self.registration.identity) && io_root == Some(self.registration.identity) {
+            return true;
+        }
+        let mut history = NATIVE_PUBLICATION_HISTORY.lock();
+        history.roots.entry(self.registration.identity).or_default().unknown = true;
+        for identity in [root, io_root].into_iter().flatten() {
+            history.roots.entry(identity).or_default().unknown = true;
+        }
+        false
+    }
+
+    fn valid_in(&self, history: &NativePublicationHistory) -> bool {
+        history
+            .roots
+            .get(&self.registration.identity)
+            .is_some_and(|root| !root.unknown && !root.started.contains(&self.bucket))
+    }
+}
+
+pub(crate) fn validate_native_creation_witnesses(witnesses: &[NativeBucketCreationWitness]) -> bool {
+    if witnesses.is_empty() || !witnesses.iter().all(NativeBucketCreationWitness::physical_root_is_current) {
+        return false;
+    }
+    let history = NATIVE_PUBLICATION_HISTORY.lock();
+    witnesses.iter().all(|witness| witness.valid_in(&history))
+}
+
+pub(crate) fn consume_native_creation_witnesses(witnesses: &[NativeBucketCreationWitness]) -> bool {
+    if witnesses.is_empty() || !witnesses.iter().all(NativeBucketCreationWitness::physical_root_is_current) {
+        return false;
+    }
+    let mut history = NATIVE_PUBLICATION_HISTORY.lock();
+    if !witnesses.iter().all(|witness| witness.valid_in(&history)) {
+        return false;
+    }
+    for witness in witnesses {
+        if let Some(root) = history.roots.get_mut(&witness.registration.identity) {
+            root.started.insert(witness.bucket.clone());
+        }
+    }
+    true
+}
 type NamespaceMutationLock = AsyncMutex<()>;
 static DISK_NAMESPACE_MUTATION_LOCKS: LazyLock<ShardedPathLockRegistry<NamespaceMutationLock>> =
     LazyLock::new(ShardedPathLockRegistry::new);
@@ -2200,6 +2428,7 @@ pub(crate) const MAX_FSYNC_WORKERS_PER_DISK: usize = 8;
 // Return busy workers to the shared pool so other disks can acquire a slot.
 const MAX_BLOCKING_JOBS_PER_WORKER: usize = 64;
 const WORKER_IDLE_SPIN_ITERS: usize = 256;
+const FSYNC_WORKER_IDLE_SPIN_ITERS: usize = 1024;
 const FALLBACK_NAMESPACE_QUEUE_SHARDS: usize = 8;
 
 #[derive(Clone, Copy, PartialEq, Eq)]
@@ -2408,7 +2637,7 @@ impl DiskBlockingQueue {
                         .ready
                         .store(!state.jobs.is_empty(), std::sync::atomic::Ordering::Relaxed);
                     if job.is_none() {
-                        if state.idle_workers > 0 {
+                        if lane != DiskBlockingLane::Fsync && state.idle_workers > 0 {
                             state.active_workers = state.active_workers.saturating_sub(1);
                             guard.state = DiskBlockingWorkerState::Retired;
                             return;
@@ -2426,7 +2655,12 @@ impl DiskBlockingQueue {
                     }
                     idle_polled = false;
                 } else {
-                    for _ in 0..WORKER_IDLE_SPIN_ITERS {
+                    let spin_iters = if lane == DiskBlockingLane::Fsync {
+                        FSYNC_WORKER_IDLE_SPIN_ITERS
+                    } else {
+                        WORKER_IDLE_SPIN_ITERS
+                    };
+                    for _ in 0..spin_iters {
                         if queue.lane(lane).ready.load(std::sync::atomic::Ordering::Relaxed) {
                             break;
                         }
@@ -3689,6 +3923,7 @@ pub(in crate::disk) async fn rename_with_namespace_owner(
     dst: &Path,
     namespace_owner: Option<Arc<dyn Send + Sync>>,
 ) -> io::Result<()> {
+    record_native_publication(dst);
     if namespace_owner.is_none() {
         return tokio::fs::rename(src, dst).await;
     }
@@ -3710,12 +3945,19 @@ pub(in crate::disk) async fn create_dir_all_with_namespace_owner(
     path: &Path,
     namespace_owner: Option<Arc<dyn Send + Sync>>,
 ) -> io::Result<()> {
+    record_native_publication(path);
     if namespace_owner.is_none() {
         return tokio::fs::create_dir_all(path).await;
     }
     let path = path.to_path_buf();
     let lease = acquire_namespace_mutation_lease_with_owner(&path, namespace_owner).await;
-    run_blocking_namespace_operation(lease, move || std::fs::create_dir_all(path)).await
+    run_blocking_namespace_operation(lease, move || {
+        #[cfg(all(test, not(windows)))]
+        prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::CreateDirectory, &path);
+        record_native_publication(&path);
+        std::fs::create_dir_all(path)
+    })
+    .await
 }
 
 #[tracing::instrument(name = "rename_all", level = "debug", skip_all)]
@@ -3726,6 +3968,7 @@ pub(crate) async fn rename_all_with_owner(
     publication_root: &PublicationRoot,
     namespace_owner: Option<Arc<dyn Send + Sync>>,
 ) -> Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     let lease = acquire_namespace_mutation_lease_with_owner(dst_file_path.as_ref(), namespace_owner).await;
     rename_all_with_lease(src_file_path, dst_file_path, base_dir, publication_root, lease).await
 }
@@ -3737,6 +3980,7 @@ pub(crate) async fn rename_all_with_lease(
     publication_root: &PublicationRoot,
     lease: Arc<NamespaceMutationLease>,
 ) -> Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     reliable_rename_inner_with_lease(
         src_file_path.as_ref().to_path_buf(),
         dst_file_path.as_ref().to_path_buf(),
@@ -3837,6 +4081,7 @@ pub(crate) async fn rename_reconciled_metadata(
     owner: Option<Arc<dyn Send + Sync>>,
     authority: Arc<crate::bucket::lifecycle::legacy_transition_state_reconcile::TransitionStateReconcileAuthority>,
 ) -> Result<()> {
+    record_native_publication(&destination);
     let lease = acquire_namespace_mutation_lease_with_owner(&destination, owner).await;
     run_blocking_namespace_operation_with_sync(lease, true, move || {
         let preparation = prepare_rename_with_retry(&source, &destination, &base_dir, &publication_root)?;
@@ -4492,6 +4737,7 @@ pub(crate) async fn rename_all_with_prepared_source(
     commit_guard: &RenameCommitGuard,
     lease: Arc<NamespaceMutationLease>,
 ) -> Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     let src_file_path = src_file_path.as_ref().to_path_buf();
     let dst_file_path = dst_file_path.as_ref().to_path_buf();
     let base_dir = base_dir.as_ref().to_path_buf();
@@ -4538,6 +4784,7 @@ pub(crate) async fn rename_all_with_commit_guard(
     commit_guard: &RenameCommitGuard,
     lease: Arc<NamespaceMutationLease>,
 ) -> Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     #[cfg(target_os = "linux")]
     {
         let source = src_file_path.as_ref().to_path_buf();
@@ -4591,6 +4838,7 @@ pub(crate) async fn rename_all_ignore_missing_source_with_owner(
     publication_root: &PublicationRoot,
     namespace_owner: Option<Arc<dyn Send + Sync>>,
 ) -> Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     let src_file_path = src_file_path.as_ref();
     let lease = acquire_namespace_mutation_lease_with_owner(dst_file_path.as_ref(), namespace_owner).await;
     match reliable_rename_inner_with_lease(
@@ -4645,6 +4893,7 @@ async fn reliable_rename_inner(
     publication_root: &PublicationRoot,
     warn_on_missing_source: bool,
 ) -> io::Result<()> {
+    record_native_publication(dst_file_path.as_ref());
     let src_file_path = src_file_path.as_ref().to_path_buf();
     let dst_file_path = dst_file_path.as_ref().to_path_buf();
     let base_dir = base_dir.as_ref().to_path_buf();
@@ -4668,6 +4917,7 @@ async fn reliable_rename_inner_with_lease(
     warn_on_missing_source: bool,
     lease: Arc<NamespaceMutationLease>,
 ) -> io::Result<()> {
+    record_native_publication(&dst_file_path);
     let operation = {
         let src_file_path = src_file_path.clone();
         let dst_file_path = dst_file_path.clone();
@@ -5160,6 +5410,7 @@ fn rename_into_existing_parent(
     parent_guard: Option<&ExistingBaseDirectoryGuard>,
     #[cfg(target_os = "linux")] source_parent: &std::fs::File,
 ) -> io::Result<()> {
+    record_native_publication(dst_file_path);
     use rustix::fs::renameat;
 
     let Some(parent_guard) = parent_guard else {
@@ -5213,6 +5464,7 @@ fn rename_into_existing_parent(
     parent_guard: Option<&ExistingBaseDirectoryGuard>,
     source: &winapi_util::Handle,
 ) -> io::Result<()> {
+    record_native_publication(dst_file_path);
     use std::{
         mem::size_of,
         os::windows::{ffi::OsStrExt, io::AsRawHandle},
@@ -5502,6 +5754,7 @@ impl PublicationRoot {
     }
 
     pub(crate) fn create_truncate(&self, path: &Path) -> io::Result<std::fs::File> {
+        record_native_publication(path);
         #[cfg(target_os = "linux")]
         {
             use rustix::fs::{Mode, OFlags};
@@ -5512,6 +5765,7 @@ impl PublicationRoot {
     }
 
     pub(crate) fn create_truncate_with_parent(&self, path: &Path) -> io::Result<(std::fs::File, Option<std::fs::File>)> {
+        record_native_publication(path);
         #[cfg(target_os = "linux")]
         {
             use rustix::fs::{Mode, OFlags};
@@ -5784,6 +6038,7 @@ impl LinuxCommitDirectories {
     }
 
     fn rename(&self, source: &Path, destination: &Path, prepared: Option<&PreparedRenameSource>) -> io::Result<()> {
+        record_native_publication(destination);
         let source_name = self.file_name(source, true)?;
         let destination_name = self.file_name(destination, false)?;
         let destination_parent = self.prepare_destination()?;
@@ -7443,6 +7698,7 @@ pub(crate) fn install_multipart_prepared_head(
     canonical_head: &Path,
     base: &Path,
     publication_root: &PublicationRoot,
+    directory_guard: Option<&RenameCommitGuard>,
 ) -> io::Result<PreparedRenameSource> {
     let original_path = prepared.path.clone();
     let parent = canonical_head
@@ -7451,9 +7707,37 @@ pub(crate) fn install_multipart_prepared_head(
     if original_path.parent() != Some(parent) || original_path == canonical_head {
         return Err(io::Error::new(io::ErrorKind::InvalidInput, "MPU proxy must be a private sibling"));
     }
+    #[cfg(target_os = "linux")]
+    let guard = if let Some(directory_guard) = directory_guard.filter(|g| g.directories.source_path == parent) {
+        let dir = &directory_guard.directories.source;
+        RenameCommitGuard {
+            directories: Arc::new(LinuxCommitDirectories {
+                root: directory_guard.directories.root.clone(),
+                source_path: parent.to_path_buf(),
+                destination_path: parent.to_path_buf(),
+                source: LinuxCommitDirectory {
+                    file: Arc::clone(&dir.file),
+                    device: dir.device,
+                    inode: dir.inode,
+                },
+                destination: std::sync::OnceLock::from(LinuxCommitDirectory {
+                    file: Arc::clone(&dir.file),
+                    device: dir.device,
+                    inode: dir.inode,
+                }),
+                destination_base: base.to_path_buf(),
+            }),
+        }
+    } else {
+        prepare_rename_commit_guard(parent, parent, base, publication_root)?
+    };
     #[cfg(not(target_os = "linux"))]
     let _pinned = pin_multipart_source_directory(parent, base, publication_root)?;
-    let guard = prepare_rename_commit_guard(parent, parent, base, publication_root)?;
+    #[cfg(not(target_os = "linux"))]
+    let guard = {
+        let _ = directory_guard;
+        prepare_rename_commit_guard(parent, parent, base, publication_root)?
+    };
     rename_prepared_source_std(&prepared, &original_path, canonical_head, base, publication_root, &guard)?;
     prepared.path = canonical_head.to_path_buf();
     fsync_commit_directory_limited_std(parent, &guard)?;
@@ -8195,6 +8479,7 @@ pub(crate) fn mkdir_all_below_existing_base_std(
     base_dir: &Path,
     publication_root: &PublicationRoot,
 ) -> io::Result<ExistingBaseDirectoryGuard> {
+    record_native_publication(dir_path);
     let relative = existing_base_directory_suffix(dir_path, base_dir)?;
 
     #[cfg(unix)]
@@ -9985,6 +10270,77 @@ mod tests {
         assert_eq!(err.raw_os_error(), Some(disk_full));
     }
 
+    #[cfg(windows)]
+    #[test]
+    fn windows_native_root_identity_matches_directory_handles() {
+        let temp = tempdir().expect("create native identity fixture");
+        let first = temp.path().join("first");
+        let second = temp.path().join("second");
+        std::fs::create_dir(&first).expect("create first root");
+        std::fs::create_dir(&second).expect("create second root");
+        let (_, first_anchor) = open_windows_publication_root(&first).expect("open first directory anchor");
+        let (_, second_anchor) = open_windows_publication_root(&second).expect("open second directory anchor");
+        let first_identity = windows_file_identity(first_anchor.handle.as_ref()).expect("query full first identity");
+        let second_identity = windows_file_identity(second_anchor.handle.as_ref()).expect("query full second identity");
+
+        assert_eq!(native_root_identity(&first), Some(first_identity));
+        assert_eq!(native_root_identity(&first.join(".")), Some(first_identity));
+        assert_eq!(native_root_identity(&second), Some(second_identity));
+        assert_ne!(first_identity, second_identity, "distinct physical roots must have distinct identities");
+    }
+
+    #[cfg(windows)]
+    #[tokio::test]
+    async fn windows_native_creation_witness_retains_started_history() {
+        let temp = tempdir().expect("create native witness fixture");
+        let root = std::fs::canonicalize(temp.path()).expect("canonical native root");
+        let _publication_root = PublicationRoot::new(&root).expect("retain native directory anchor");
+        let registration = register_native_publication_root(&root, &root.join(".")).expect("register same physical root");
+        let backend: Arc<dyn super::super::local::LocalIoBackend> = Arc::new(super::super::local::StdBackend::new(root.clone()));
+        let disk_id = uuid::Uuid::new_v4();
+        let witness = registration
+            .witness("published", disk_id, Arc::clone(&backend))
+            .expect("fresh native witness");
+        let bucket = root.join("published");
+
+        super::super::fs::mkdir(&bucket)
+            .await
+            .expect("publish through the actual native entry");
+        assert!(bucket.is_dir(), "the production entry must create the directory");
+        assert!(!validate_native_creation_witnesses(std::slice::from_ref(&witness)));
+        std::fs::remove_dir(&bucket).expect("remove published directory");
+        assert!(registration.witness("published", disk_id, Arc::clone(&backend)).is_none());
+        let reconnected = register_native_publication_root(&root, &root).expect("reconnect same physical root");
+        assert!(reconnected.witness("published", disk_id, Arc::clone(&backend)).is_none());
+
+        let unstarted = reconnected
+            .witness("unstarted", disk_id, Arc::clone(&backend))
+            .expect("untouched bucket keeps its witness");
+        assert!(consume_native_creation_witnesses(std::slice::from_ref(&unstarted)));
+        assert!(!validate_native_creation_witnesses(std::slice::from_ref(&unstarted)));
+        assert!(reconnected.witness("unstarted", disk_id, backend).is_none());
+    }
+
+    #[cfg(windows)]
+    #[test]
+    fn windows_native_creation_witness_rejects_unknown_roots() {
+        let temp = tempdir().expect("create unknown identity fixture");
+        let root = temp.path().join("root");
+        let other = temp.path().join("other");
+        let file = temp.path().join("file");
+        std::fs::create_dir(&root).expect("create native root");
+        std::fs::create_dir(&other).expect("create distinct native root");
+        std::fs::write(&file, b"not a directory").expect("create regular file");
+
+        assert!(native_root_identity(&temp.path().join("missing")).is_none());
+        assert!(native_root_identity(&file).is_none());
+        assert!(register_native_publication_root(&root, &other).is_none());
+        let registration = register_native_publication_root(&root, &root).expect("register actual native root");
+        let backend: Arc<dyn super::super::local::LocalIoBackend> = Arc::new(super::super::local::StdBackend::new(root));
+        registration.invalidate();
+        assert!(registration.witness("unstarted", uuid::Uuid::new_v4(), backend).is_none());
+    }
+
     #[cfg(windows)]
     #[test]
     fn windows_rename_all_legacy_file_identity_distinguishes_sources() {
@@ -13034,6 +13390,328 @@ mod tests {
     }
 }
 
+#[cfg(all(test, unix))]
+mod native_creation_history_tests {
+    use super::*;
+
+    struct NativeFixture {
+        _temp: tempfile::TempDir,
+        root: PathBuf,
+        registration: NativePublicationRegistration,
+        backend: Arc<dyn super::super::local::LocalIoBackend>,
+        disk_id: uuid::Uuid,
+    }
+
+    impl NativeFixture {
+        fn new() -> Self {
+            let temp = tempfile::tempdir().expect("native root");
+            let root = std::fs::canonicalize(temp.path()).expect("canonical root");
+            let registration = register_native_publication_root(&root, &root).expect("native registration");
+            let backend = Arc::new(super::super::local::StdBackend::new(root.clone()));
+            Self {
+                _temp: temp,
+                root,
+                registration,
+                backend,
+                disk_id: uuid::Uuid::new_v4(),
+            }
+        }
+
+        fn witness(&self, bucket: &str) -> Option<NativeBucketCreationWitness> {
+            self.registration.witness(bucket, self.disk_id, Arc::clone(&self.backend))
+        }
+    }
+
+    #[tokio::test]
+    async fn every_native_directory_publication_entry_invalidates_retry() {
+        let fixture = NativeFixture::new();
+        for operation in [
+            "recursive",
+            "single",
+            "open",
+            "rename",
+            "rename_std",
+            "fallback",
+            "owned_mkdir",
+            "owned_rename",
+            "handle_relative",
+            "root_create",
+            "root_create_with_parent",
+            "guarded_rename",
+            "guarded_prepared_rename",
+        ] {
+            let bucket = format!("native-{operation}-{}", uuid::Uuid::new_v4());
+            let path = fixture.root.join(&bucket);
+            let witness = fixture.witness(&bucket).expect("fresh native witness");
+            match operation {
+                "recursive" => super::super::fs::make_dir_all(path.join("object/child"))
+                    .await
+                    .expect("recursive mkdir"),
+                "single" => super::super::fs::mkdir(&path).await.expect("single mkdir"),
+                "open" => {
+                    let file = super::super::fs::open_file(&path, super::super::fs::O_CREATE | super::super::fs::O_WRONLY)
+                        .await
+                        .expect("create file");
+                    drop(file);
+                }
+                "rename" | "rename_std" | "owned_rename" => {
+                    let source = fixture.root.join(format!("source-{}", uuid::Uuid::new_v4()));
+                    std::fs::create_dir(&source).expect("rename source");
+                    match operation {
+                        "rename" => super::super::fs::rename(&source, &path).await.expect("async rename"),
+                        "rename_std" => super::super::fs::rename_std(&source, &path).expect("sync rename"),
+                        _ => rename_with_namespace_owner(&source, &path, Some(Arc::new(())))
+                            .await
+                            .expect("owned rename"),
+                    }
+                }
+                "fallback" => make_dir_all(path.join("object/child"), &path)
+                    .await
+                    .expect("implicit volume fallback"),
+                "owned_mkdir" => create_dir_all_with_namespace_owner(&path.join("child"), Some(Arc::new(())))
+                    .await
+                    .expect("owned recursive mkdir"),
+                "handle_relative" => {
+                    let publication_root = PublicationRoot::new(&fixture.root).expect("publication root");
+                    let guard = mkdir_all_below_existing_base_std(&path.join("child"), &fixture.root, &publication_root)
+                        .expect("first-level handle-relative mkdir");
+                    drop(guard);
+                }
+                "root_create" | "root_create_with_parent" => {
+                    let publication_root = PublicationRoot::new(&fixture.root).expect("publication root");
+                    if operation == "root_create" {
+                        drop(publication_root.create_truncate(&path).expect("root-relative creation"));
+                    } else {
+                        drop(
+                            publication_root
+                                .create_truncate_with_parent(&path)
+                                .expect("root-relative creation with parent"),
+                        );
+                    }
+                }
+                "guarded_rename" | "guarded_prepared_rename" => {
+                    let publication_root = PublicationRoot::new(&fixture.root).expect("publication root");
+                    let source_parent = fixture.root.join(format!("source-{}", uuid::Uuid::new_v4()));
+                    std::fs::create_dir(&source_parent).expect("source directory");
+                    std::fs::create_dir(&path).expect("existing destination directory");
+                    let guard = prepare_rename_commit_guard(&source_parent, &path, &fixture.root, &publication_root)
+                        .expect("existing commit directories");
+                    let source = source_parent.join("xl.meta");
+                    let destination = path.join("xl.meta");
+                    assert!(
+                        validate_native_creation_witnesses(std::slice::from_ref(&witness)),
+                        "preparing existing directories must not mask the publication entry"
+                    );
+                    let lease = acquire_namespace_mutation_lease(&destination).await;
+                    if operation == "guarded_rename" {
+                        std::fs::write(&source, b"metadata").expect("source metadata");
+                        rename_all_with_commit_guard(&source, &destination, &fixture.root, &publication_root, &guard, lease)
+                            .await
+                            .expect("guarded publication");
+                    } else {
+                        let mut staged = create_prepared_rename_source_with_commit_guard(&source, &destination, &guard)
+                            .expect("prepared source");
+                        staged.write_all(b"metadata", false).expect("source metadata");
+                        rename_all_with_prepared_source(
+                            staged,
+                            &source,
+                            &destination,
+                            &fixture.root,
+                            &publication_root,
+                            &guard,
+                            lease,
+                        )
+                        .await
+                        .expect("prepared guarded publication");
+                    }
+                    assert_eq!(std::fs::read(&destination).expect("published metadata"), b"metadata");
+                }
+                _ => unreachable!(),
+            }
+            assert!(
+                !validate_native_creation_witnesses(std::slice::from_ref(&witness)),
+                "{operation}: old proof remains live"
+            );
+            if path.is_dir() {
+                std::fs::remove_dir_all(&path).expect("remove published directory");
+            } else {
+                std::fs::remove_file(&path).expect("remove published file");
+            }
+            assert!(fixture.witness(&bucket).is_none(), "{operation}: removal must not regrant proof");
+        }
+    }
+
+    #[tokio::test]
+    async fn cancelled_namespace_waiter_cannot_restore_native_proof() {
+        let fixture = NativeFixture::new();
+        let bucket = format!("native-cancel-{}", uuid::Uuid::new_v4());
+        let path = fixture.root.join(&bucket);
+        let witness = fixture.witness(&bucket).expect("fresh witness");
+        let lease = acquire_namespace_mutation_lease(&path).await;
+        let mut attempt = Box::pin(create_dir_all_with_namespace_owner(&path, Some(Arc::new(()))));
+        assert!(matches!(futures::poll!(&mut attempt), std::task::Poll::Pending));
+        assert!(
+            !validate_native_creation_witnesses(std::slice::from_ref(&witness)),
+            "Started must precede admission wait"
+        );
+        assert!(!path.exists(), "the namespace wait must still precede the syscall");
+        drop(attempt);
+        drop(lease);
+        assert!(fixture.witness(&bucket).is_none(), "cancellation cannot undo an admitted attempt");
+    }
+
+    #[tokio::test]
+    async fn queued_mkdir_cancellation_retains_lease_and_consumed_history() {
+        use futures::FutureExt;
+        use std::sync::mpsc;
+
+        let fixture = NativeFixture::new();
+        let bucket = format!("native-queued-cancel-{}", uuid::Uuid::new_v4());
+        let path = fixture.root.join(&bucket);
+        let witness = fixture.witness(&bucket).expect("fresh witness");
+        let (entered_tx, entered_rx) = mpsc::channel();
+        let (release_tx, release_rx) = mpsc::channel();
+        let _hook = prepared_publication_test_hooks::install_at(
+            prepared_publication_test_hooks::Stage::CreateDirectory,
+            &path,
+            move || {
+                entered_tx.send(()).expect("signal real executor entry");
+                let _ = release_rx.recv();
+            },
+        );
+        let operation_path = path.clone();
+        let operation =
+            tokio::spawn(async move { create_dir_all_with_namespace_owner(&operation_path, Some(Arc::new(()))).await });
+        tokio::task::spawn_blocking(move || entered_rx.recv_timeout(std::time::Duration::from_secs(30)))
+            .await
+            .expect("entry waiter")
+            .expect("the real blocking executor must enter");
+        operation.abort();
+        assert!(operation.await.expect_err("cancel the async waiter").is_cancelled());
+        assert!(
+            !validate_native_creation_witnesses(std::slice::from_ref(&witness)),
+            "queued publication is permanently Started"
+        );
+        assert!(
+            acquire_namespace_mutation_lease(&path).now_or_never().is_none(),
+            "the actual blocked executor must still own the namespace lease"
+        );
+        release_tx.send(()).expect("release the syscall");
+        let drained = tokio::time::timeout(std::time::Duration::from_secs(30), acquire_namespace_mutation_lease(&path))
+            .await
+            .expect("the lease must drain with the syscall");
+        drop(drained);
+        assert!(path.is_dir(), "the cancelled waiter cannot discard the already queued syscall");
+        std::fs::remove_dir_all(&path).expect("remove the published directory");
+        assert!(fixture.witness(&bucket).is_none(), "drain and removal cannot restore proof");
+    }
+
+    #[tokio::test]
+    async fn publication_detects_retargeted_physical_alias_without_registration() {
+        let source = NativeFixture::new();
+        let destination = NativeFixture::new();
+        let bucket = format!("native-retarget-{}", uuid::Uuid::new_v4());
+        let witness = destination.witness(&bucket).expect("destination witness");
+        let retired = source.root.with_extension("retired");
+        std::fs::rename(&source.root, &retired).expect("move the original physical root");
+        std::os::unix::fs::symlink(&destination.root, &source.root).expect("retarget the registered alias");
+        super::super::fs::mkdir(source.root.join(&bucket))
+            .await
+            .expect("publish through the changed alias");
+        assert!(
+            !validate_native_creation_witnesses(std::slice::from_ref(&witness)),
+            "publication-time alias drift must invalidate the actual destination root"
+        );
+        std::fs::remove_dir(destination.root.join(&bucket)).expect("remove the published bucket");
+        assert!(destination.witness(&bucket).is_none(), "removal must not undo alias uncertainty");
+        std::fs::remove_file(&source.root).expect("remove the symlink");
+        std::fs::rename(retired, &source.root).expect("restore the test root");
+    }
+
+    #[test]
+    fn blocked_physical_identity_lookup_does_not_block_other_publication() {
+        use std::sync::mpsc;
+        let stalled = NativeFixture::new();
+        let healthy = NativeFixture::new();
+        let witness = stalled.witness("unpublished").expect("stalled root witness");
+        let (entered_tx, entered_rx) = mpsc::channel();
+        let (release_tx, release_rx) = mpsc::channel();
+        let _hook = prepared_publication_test_hooks::install_at(
+            prepared_publication_test_hooks::Stage::NativeRootIdentity,
+            &stalled.root,
+            move || {
+                entered_tx.send(()).expect("signal physical lookup");
+                let _ = release_rx.recv();
+            },
+        );
+        let validator = std::thread::spawn(move || validate_native_creation_witnesses(&[witness]));
+        entered_rx
+            .recv_timeout(std::time::Duration::from_secs(30))
+            .expect("enter the actual identity lookup");
+        let (published_tx, published_rx) = mpsc::channel();
+        let path = healthy.root.join("published");
+        let publisher = std::thread::spawn(move || {
+            record_native_publication(&path);
+            published_tx.send(()).expect("signal independent publication");
+        });
+        let published = published_rx.recv_timeout(std::time::Duration::from_secs(5));
+        release_tx.send(()).expect("release the identity lookup");
+        validator.join().expect("identity lookup thread");
+        publisher.join().expect("publication thread");
+        assert!(
+            published.is_ok(),
+            "a slow physical lookup must not retain the process-wide publication mutex"
+        );
+    }
+
+    #[test]
+    fn native_consume_is_atomic_and_history_survives_reconnect() {
+        let fixture = NativeFixture::new();
+        let first = format!("native-first-{}", uuid::Uuid::new_v4());
+        let second = format!("native-second-{}", uuid::Uuid::new_v4());
+        let first_witness = fixture.witness(&first).expect("first witness");
+        let second_witness = fixture.witness(&second).expect("second witness");
+        record_native_publication(&fixture.root.join(&second));
+        assert!(!consume_native_creation_witnesses(&[first_witness.clone(), second_witness]));
+        assert!(
+            validate_native_creation_witnesses(std::slice::from_ref(&first_witness)),
+            "failed cohort consume must be atomic"
+        );
+        assert!(consume_native_creation_witnesses(std::slice::from_ref(&first_witness)));
+        assert!(!validate_native_creation_witnesses(std::slice::from_ref(&first_witness)));
+        let reconnected = register_native_publication_root(&fixture.root, &fixture.root).expect("reconnected registration");
+        let backend: Arc<dyn super::super::local::LocalIoBackend> =
+            Arc::new(super::super::local::StdBackend::new(fixture.root.clone()));
+        assert!(
+            reconnected.witness(&first, fixture.disk_id, backend).is_none(),
+            "reconnect must retain Started"
+        );
+    }
+
+    #[test]
+    fn physical_alias_reuse_is_unknown_and_backend_identity_is_real() {
+        let temp = tempfile::tempdir().expect("alias parent");
+        let root = temp.path().join("root");
+        std::fs::create_dir(&root).expect("first physical root");
+        let old = register_native_publication_root(&root, &root).expect("first registration");
+        let backend: Arc<dyn super::super::local::LocalIoBackend> = Arc::new(super::super::local::StdBackend::new(root.clone()));
+        let id = uuid::Uuid::new_v4();
+        let witness = old.witness("same-name", id, Arc::clone(&backend)).expect("old witness");
+        let other_backend: Arc<dyn super::super::local::LocalIoBackend> =
+            Arc::new(super::super::local::StdBackend::new(root.clone()));
+        let other = old.witness("same-name", id, other_backend).expect("new backend witness");
+        assert!(!witness.matches(&other), "a different actual backend cannot reuse a pending observation");
+        std::fs::rename(&root, temp.path().join("retired-root")).expect("retire physical root");
+        std::fs::create_dir(&root).expect("replacement root at the same alias");
+        let replacement = register_native_publication_root(&root, &root).expect("replacement registration");
+        assert!(replacement.witness("same-name", id, backend).is_none(), "alias reuse must remain Unknown");
+        assert!(
+            !validate_native_creation_witnesses(&[witness]),
+            "old physical identity must not validate at a new root"
+        );
+    }
+}
+
 #[cfg(all(test, target_os = "linux"))]
 mod fd_relative_tests {
     use super::*;

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>;
@@ -1096,14 +1096,8 @@ impl From<rmp_serde::decode::Error> for Error {
     }
 }
 
-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 {
+impl From<crate::bucket::utils::BucketConfigXmlError> for Error {
+    fn from(e: crate::bucket::utils::BucketConfigXmlError) -> Self {
         Error::other(e)
     }
 }

crates/ecstore/src/object_api/object_api_utils.rs

@@ -20,7 +20,6 @@
 #![allow(clippy::all)]
 
 use http::HeaderMap;
-use s3s::dto::ETag;
 use std::{collections::HashMap, io::Cursor, sync::Arc};
 use tokio::io::BufReader;
 
@@ -29,7 +28,7 @@ use crate::object_api::{GetObjectReader, ObjectInfo, ObjectOptions};
 use crate::storage_api_contracts::range::HTTPRangeSpec;
 use rustfs_filemeta::ObjectPartInfo;
 use rustfs_rio::HashReader;
-use s3s::S3ErrorCode;
+use rustfs_s3_types::S3ErrorCode;
 
 //#[derive(Clone)]
 pub struct PutObjReader {
@@ -150,24 +149,6 @@ 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")
@@ -232,28 +213,42 @@ mod tests {
     }
 
     #[test]
-    fn test_to_s3s_etag() {
-        // Test unquoted ETag - should become strong etag
-        assert_eq!(
-            to_s3s_etag("6af8d12c0c74b78094884349f3c8a079"),
-            ETag::Strong("6af8d12c0c74b78094884349f3c8a079".to_string())
-        );
+    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,
+        });
 
-        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("W/\"6af8d12c0c74b78094884349f3c8a079\""),
-            ETag::Weak("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(""), ETag::Strong(String::new()));
+    #[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("\"incomplete"), ETag::Strong("\"incomplete".to_string()));
+        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"));
     }
 
     #[test]

crates/ecstore/src/object_api/types.rs

@@ -1128,6 +1128,10 @@ 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)]

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: s3s::region::Region) {
+pub fn set_global_region(region: rustfs_s3_types::Region) {
     current_ctx().set_region(region);
 }
 
 /// Get the global region
 ///
 /// # Returns
-/// * `Option<s3s::region::Region>` - The global region, if set
+/// * `Option<rustfs_s3_types::Region>` - The global region, if set
 ///
-pub fn get_global_region() -> Option<s3s::region::Region> {
+pub fn get_global_region() -> Option<rustfs_s3_types::Region> {
     current_ctx().region()
 }
 

crates/ecstore/src/runtime/instance.rs

@@ -52,13 +52,13 @@ 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 s3s::region::Region;
+use rustfs_s3_types::Region;
 use std::collections::{HashMap, HashSet};
 use std::sync::{
     Arc, OnceLock,
     atomic::{AtomicBool, AtomicU8, AtomicU64, Ordering},
 };
-use tokio::sync::{Mutex, Notify, OnceCell, OwnedRwLockReadGuard, RwLock};
+use tokio::sync::{Mutex, Notify, OnceCell, OwnedRwLockReadGuard, OwnedSemaphorePermit, RwLock, Semaphore};
 use tokio::time::Instant;
 use tokio_util::sync::CancellationToken;
 use tokio_util::task::TaskTracker;
@@ -187,6 +187,38 @@ impl DecommissionCountLedger {
     }
 }
 
+#[derive(Debug)]
+struct ListReadBudget {
+    permits: Arc<Semaphore>,
+    capacity: usize,
+}
+
+impl ListReadBudget {
+    fn new(capacity: usize) -> Self {
+        let capacity = capacity
+            .min(Semaphore::MAX_PERMITS)
+            .min(usize::try_from(u32::MAX).unwrap_or(usize::MAX));
+        Self {
+            permits: Arc::new(Semaphore::new(capacity)),
+            capacity,
+        }
+    }
+
+    async fn acquire(&self, fanout: usize) -> Result<Option<Arc<OwnedSemaphorePermit>>, tokio::sync::AcquireError> {
+        if self.capacity == 0 {
+            return Ok(None);
+        }
+        // A single walk must remain admissible even when the topology exceeds
+        // the configured budget. Tokio accepts at most u32::MAX in one grant.
+        let weight = u32::try_from(fanout.max(1).min(self.capacity)).unwrap_or(u32::MAX);
+        self.permits
+            .clone()
+            .acquire_many_owned(weight)
+            .await
+            .map(|permit| Some(Arc::new(permit)))
+    }
+}
+
 /// Runtime state owned by a single `ECStore` instance.
 ///
 /// This is intentionally minimal in the first migration slice; subsequent
@@ -201,6 +233,7 @@ pub struct InstanceContext {
     /// drift out of sync). Stored as one value so the three predicates can
     /// never observe a torn intermediate state.
     erasure_kind: RwLock<SetupType>,
+    list_read_budget: OnceLock<ListReadBudget>,
     /// This instance's namespace lock manager (Phase 5 Slice 3, backlog#939).
     ///
     /// Owned per-instance so two instances no longer share a lock namespace
@@ -354,6 +387,24 @@ impl InstanceContext {
         Self::with_lock_manager(Arc::new(GlobalLockManager::new()))
     }
 
+    /// Bound native listing fanout independently of GET body permits. A walk
+    /// keeps its shared grant until all of its disk producers have terminated.
+    /// Acquire before creating the merge channels or taking object locks.
+    pub(crate) async fn acquire_list_read_budget(
+        &self,
+        fanout: usize,
+    ) -> Result<Option<Arc<OwnedSemaphorePermit>>, tokio::sync::AcquireError> {
+        self.list_read_budget
+            .get_or_init(|| {
+                ListReadBudget::new(rustfs_utils::get_env_usize(
+                    rustfs_config::ENV_OBJECT_MAX_CONCURRENT_DISK_READS,
+                    rustfs_config::DEFAULT_OBJECT_MAX_CONCURRENT_DISK_READS,
+                ))
+            })
+            .acquire(fanout)
+            .await
+    }
+
     /// Build a context bound to a specific lock manager.
     ///
     /// [`bootstrap_ctx`] uses this to alias the process-global lock manager so
@@ -362,6 +413,7 @@ impl InstanceContext {
     fn with_lock_manager(lock_manager: Arc<GlobalLockManager>) -> Self {
         Self {
             erasure_kind: RwLock::new(SetupType::Unknown),
+            list_read_budget: OnceLock::new(),
             lock_manager,
             region: OnceLock::new(),
             deployment_id: OnceLock::new(),
@@ -1111,6 +1163,85 @@ pub fn bootstrap_ctx() -> Arc<InstanceContext> {
 mod tests {
     use super::*;
 
+    #[tokio::test]
+    async fn list_read_budget_weights_and_clamps_fanout() {
+        let budget = ListReadBudget::new(8);
+        let first = budget.acquire(3).await.expect("first walk").expect("bounded grant");
+        assert_eq!(budget.permits.available_permits(), 5);
+        let mut blocked = std::pin::pin!(budget.acquire(6));
+        assert!(futures::poll!(&mut blocked).is_pending());
+        drop(first);
+        let second = blocked.await.expect("queued walk").expect("bounded grant");
+        assert_eq!(budget.permits.available_permits(), 2);
+        drop(second);
+        let oversized = budget
+            .acquire(usize::MAX)
+            .await
+            .expect("oversized walk")
+            .expect("bounded grant");
+        assert_eq!(budget.permits.available_permits(), 0);
+        drop(oversized);
+        assert_eq!(budget.permits.available_permits(), 8);
+    }
+
+    #[tokio::test]
+    async fn list_read_budget_zero_preserves_unbounded_configuration() {
+        let budget = ListReadBudget::new(0);
+        assert!(budget.acquire(usize::MAX).await.expect("disabled budget").is_none());
+    }
+
+    #[tokio::test]
+    async fn list_read_budget_cancelled_waiter_returns_partial_grant() {
+        let budget = ListReadBudget::new(4);
+        let held = budget.acquire(3).await.expect("first walk");
+        {
+            let mut waiter = std::pin::pin!(budget.acquire(4));
+            assert!(futures::poll!(&mut waiter).is_pending());
+        }
+        assert_eq!(budget.permits.available_permits(), 1);
+        drop(held);
+        assert!(budget.acquire(4).await.expect("next walk").is_some());
+    }
+
+    #[tokio::test]
+    async fn list_read_budget_shared_grant_survives_parent_drop() {
+        let budget = ListReadBudget::new(4);
+        let parent = budget.acquire(4).await.expect("first walk").expect("bounded grant");
+        let producer = parent.clone();
+        drop(parent);
+        assert_eq!(budget.permits.available_permits(), 0);
+        drop(producer);
+        assert_eq!(budget.permits.available_permits(), 4);
+    }
+
+    #[tokio::test]
+    async fn list_read_budget_is_shared_by_context_and_isolated_between_instances() {
+        let first = Arc::new(InstanceContext::new());
+        first.list_read_budget.set(ListReadBudget::new(4)).expect("fresh context");
+        let same = first.clone();
+        let other = InstanceContext::new();
+        other.list_read_budget.set(ListReadBudget::new(4)).expect("fresh context");
+        let held = first.acquire_list_read_budget(4).await.expect("first context");
+        let mut waiter = std::pin::pin!(same.acquire_list_read_budget(1));
+        assert!(futures::poll!(&mut waiter).is_pending());
+        assert!(
+            other
+                .acquire_list_read_budget(4)
+                .await
+                .expect("independent context")
+                .is_some()
+        );
+        drop(held);
+        assert!(waiter.await.expect("shared context resumed").is_some());
+    }
+
+    #[test]
+    fn list_read_budget_extreme_configuration_is_bounded_without_panicking() {
+        let budget = ListReadBudget::new(usize::MAX);
+        assert!(budget.capacity <= Semaphore::MAX_PERMITS);
+        assert!(u32::try_from(budget.capacity).is_ok());
+    }
+
     #[test]
     fn decommission_count_ledger_bounds_retained_entries() {
         let mut ledger = super::DecommissionCountLedger::default();

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 s3s::dto::BucketLifecycleConfiguration;
-use s3s::region::Region;
+use rustfs_s3_types::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<BucketLifecycleConfiguration> {
+pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<PersistedLifecycleConfiguration> {
     GLOBAL_LIFECYCLE_SYS.get(bucket).await
 }
 

crates/ecstore/src/services/notification_sys.rs

@@ -703,6 +703,28 @@ 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,
@@ -1760,6 +1782,30 @@ 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/control.rs

@@ -593,7 +593,7 @@ impl ECStore {
         #[cfg(test)]
         crate::core::pools::observe_pool_activation_start_attempt(crate::core::pools::PoolActivationStartKind::Rebalance);
         let fleet_proof = acquire_pool_activation_fleet_proof(&self.ctx).await?;
-        let mut pool_meta_guard = self.pool_meta_save_gate.lock().await;
+        let mut pool_meta_guard = self.pool_meta_save_gate.write().await;
         pool_meta_guard.ensure_write_safe(stage)?;
         let activation_fence = acquire_pool_rebalance_activation_locks(pool.clone(), fleet_proof).await?;
         let pool_meta = self
@@ -625,7 +625,7 @@ impl ECStore {
     {
         // Lock order: pool_meta_save_gate -> rebalance.bin -> pool.bin. These
         // shared fences admit a peer without waiting for the run's own entries.
-        let mut pool_meta_guard = self.pool_meta_save_gate.lock().await;
+        let mut pool_meta_guard = self.pool_meta_save_gate.write().await;
         pool_meta_guard.ensure_write_safe("rebalance worker activation")?;
         let mut worker_fence = acquire_pool_rebalance_worker_locks(pool.clone()).await?;
         let pool_meta = self
@@ -654,7 +654,7 @@ impl ECStore {
     {
         // The admission read fences must already be dropped. Reacquire in the
         // original order and revalidate every durable boundary across that gap.
-        let mut pool_meta_guard = self.pool_meta_save_gate.lock().await;
+        let mut pool_meta_guard = self.pool_meta_save_gate.write().await;
         pool_meta_guard.ensure_write_safe("rebalance worker metadata update")?;
         let mut activation_fence = acquire_pool_rebalance_activation_locks(pool.clone(), None).await?;
         let pool_meta = self
@@ -1714,7 +1714,7 @@ mod tests {
         persist_pool_meta_identity_for_startup(store.pools.clone(), &mut write_state, true)
             .await
             .expect("initialized pool metadata identity should persist");
-        *store.pool_meta_save_gate.lock().await = write_state;
+        *store.pool_meta_save_gate.write().await = write_state;
         // Simulate physical loss; raw deletion of authoritative pool.bin is
         // deliberately rejected by the native metadata mutation guard.
         for pool in &store.pools {

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

@@ -1043,8 +1043,9 @@ mod tests {
     use crate::storage_api_contracts::object::ObjectIO as _;
     use http::HeaderMap;
     use rustfs_filemeta::{FileInfo, FileMeta, ObjectPartInfo, TransitionVersionState};
-    use s3s::dto::{
-        BucketLifecycleConfiguration, ExpirationStatus, LifecycleExpiration, LifecycleRule, NoncurrentVersionExpiration,
+    use rustfs_gateway_types::persistence::{
+        PersistedLifecycleConfiguration, PersistedLifecycleExpiration, PersistedLifecycleRule,
+        PersistedNoncurrentVersionExpiration,
     };
     use std::time::Duration as StdDuration;
     use time::OffsetDateTime;
@@ -1088,12 +1089,12 @@ mod tests {
         }
     }
 
-    fn expired_delete_marker_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn expired_delete_marker_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     expired_object_delete_marker: Some(true),
                     ..Default::default()
                 }),
@@ -2125,11 +2126,11 @@ mod tests {
         ))
         .await
         .expect("seed the expired source version");
-        let lifecycle = BucketLifecycleConfiguration {
+        let lifecycle = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     expired_object_all_versions: Some(true),
                     ..Default::default()
@@ -2567,7 +2568,7 @@ mod tests {
                         versioned,
                         version_id,
                         transition: crate::bucket::lifecycle::lifecycle::TransitionOptions {
-                            restore_request: s3s::dto::RestoreRequest {
+                            restore_request: rustfs_lifecycle::RestoreRequest {
                                 days: Some(1),
                                 ..Default::default()
                             },
@@ -2647,7 +2648,7 @@ mod tests {
         async fn run_entry_with(
             &self,
             entry: MetaCacheEntry,
-            lifecycle_config: Option<BucketLifecycleConfiguration>,
+            lifecycle_config: Option<PersistedLifecycleConfiguration>,
         ) -> RebalanceEntryOutcome {
             Box::pin(Arc::clone(&self.store).rebalance_entry(
                 RebalanceEntryTarget {
@@ -2983,17 +2984,17 @@ mod tests {
     }
 
     /// A rule that expires every noncurrent version at once.
-    fn noncurrent_expire_all_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn noncurrent_expire_all_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("noncurrent-expire-all".to_string()),
-                noncurrent_version_expiration: Some(NoncurrentVersionExpiration {
+                noncurrent_version_expiration: Some(PersistedNoncurrentVersionExpiration {
                     noncurrent_days: Some(0),
                     newer_noncurrent_versions: None,
                 }),

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

@@ -100,7 +100,7 @@ pub async fn test_store_with_persisted_rebalance_meta(
         rebalance_meta: tokio::sync::RwLock::new(Some(meta)),
         decommission_cancelers: tokio::sync::RwLock::new(vec![None]),
         start_gate: tokio::sync::Mutex::new(()),
-        pool_meta_save_gate: tokio::sync::Mutex::default(),
+        pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
         ctx,
         bucket_fence_registry: std::sync::Arc::default(),
     });
@@ -335,7 +335,7 @@ async fn test_pool_stores_with_contexts(
             rebalance_meta: tokio::sync::RwLock::new(rebalance_meta.clone()),
             decommission_cancelers: tokio::sync::RwLock::new(vec![None; pool_count]),
             start_gate: tokio::sync::Mutex::new(()),
-            pool_meta_save_gate: tokio::sync::Mutex::new(pool_meta_write_state.independent_clone_for_test()),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::new(pool_meta_write_state.independent_clone_for_test()),
             ctx: store_ctx,
             bucket_fence_registry: std::sync::Arc::default(),
         })

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((ReplicationConfiguration::default(), OffsetDateTime::UNIX_EPOCH)),
+        Ok((PersistedReplicationConfiguration::default(), OffsetDateTime::UNIX_EPOCH)),
     )
     .expect("bucket config should pass through");
     assert!(result.is_some());
@@ -3412,7 +3412,7 @@ fn test_store_with_rebalance_meta(meta: RebalanceMeta) -> Arc<crate::store::ECSt
         rebalance_meta: tokio::sync::RwLock::new(Some(meta)),
         decommission_cancelers: tokio::sync::RwLock::new(Vec::new()),
         start_gate: tokio::sync::Mutex::new(()),
-        pool_meta_save_gate: tokio::sync::Mutex::default(),
+        pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
         ctx: crate::runtime::instance::bootstrap_ctx(),
         bucket_fence_registry: std::sync::Arc::default(),
     })

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<s3s::dto::BucketLifecycleConfiguration>,
-    pub(super) object_lock_config: Option<s3s::dto::ObjectLockConfiguration>,
-    pub(super) replication_config: Option<(s3s::dto::ReplicationConfiguration, OffsetDateTime)>,
+    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)>,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq)]

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

@@ -79,10 +79,11 @@ 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},
@@ -1102,7 +1103,7 @@ trait TierReferenceProofStore:
         .await
     }
 
-    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>>;
+    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>>;
 
     async fn walk_bucket_for_deletion_check(
         self: Arc<Self>,
@@ -1155,7 +1156,7 @@ impl TierReferenceProofStore for ECStore {
         Ok(buckets.into_values().collect())
     }
 
-    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
+    async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
         match get_lifecycle_config(bucket).await {
             Ok((config, _updated_at)) => Ok(Some(config)),
             Err(Error::ConfigNotFound) => Ok(None),
@@ -1634,7 +1635,7 @@ struct TierLifecycleReference<'a> {
 }
 
 fn lifecycle_config_target_reference<'a>(
-    config: &'a BucketLifecycleConfiguration,
+    config: &'a PersistedLifecycleConfiguration,
     targets: &[TierMutationIntentTarget],
 ) -> Option<TierLifecycleReference<'a>> {
     for rule in &config.rules {
@@ -7308,9 +7309,8 @@ mod tests {
         endpoints::{Endpoints, PoolEndpoints, SetupType},
     };
     use crate::services::tier::tier_mutation_intent::TIER_MUTATION_INTENT_RECORD_PREFIX;
-    use s3s::dto::{
-        BucketLifecycleConfiguration, ExpirationStatus, LifecycleRule, NoncurrentVersionTransition, Transition,
-        TransitionStorageClass,
+    use rustfs_gateway_types::persistence::{
+        PersistedLifecycleConfiguration, PersistedLifecycleRule, PersistedNoncurrentVersionTransition, PersistedTransition,
     };
 
     #[test]
@@ -8137,7 +8137,7 @@ mod tests {
 
     #[async_trait::async_trait]
     impl TierReferenceProofStore for LockingTierConfigStore {
-        async fn lifecycle_config_for_reference_proof(&self, _bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
+        async fn lifecycle_config_for_reference_proof(&self, _bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
             Ok(None)
         }
     }
@@ -10870,6 +10870,42 @@ 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());
@@ -15370,6 +15406,7 @@ 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,
@@ -15388,7 +15425,7 @@ mod tests {
         lock_manager: Arc<rustfs_lock::GlobalLockManager>,
         lock_requests: Mutex<Vec<(String, String)>>,
         listed_versions: Mutex<Vec<ObjectInfo>>,
-        lifecycle_configs: Mutex<HashMap<String, BucketLifecycleConfiguration>>,
+        lifecycle_configs: Mutex<HashMap<String, PersistedLifecycleConfiguration>>,
     }
 
     impl Default for CasConfigStore {
@@ -15399,6 +15436,7 @@ 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),
@@ -15451,7 +15489,7 @@ mod tests {
                 .retain(|version| version.bucket != bucket || version.name != object);
         }
 
-        fn add_lifecycle_config(&self, bucket: &str, config: BucketLifecycleConfiguration) {
+        fn add_lifecycle_config(&self, bucket: &str, config: PersistedLifecycleConfiguration) {
             self.lifecycle_configs
                 .lock()
                 .expect("tier reference fixture should not poison")
@@ -15560,6 +15598,19 @@ 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;
@@ -16188,7 +16239,7 @@ mod tests {
             .await
         }
 
-        async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<BucketLifecycleConfiguration>> {
+        async fn lifecycle_config_for_reference_proof(&self, bucket: &str) -> Result<Option<PersistedLifecycleConfiguration>> {
             if self.reference_lifecycle_error.load(Ordering::SeqCst) {
                 return Err(Error::CorruptedFormat);
             }
@@ -16323,10 +16374,10 @@ mod tests {
         });
         store.add_lifecycle_config(
             "photos",
-            BucketLifecycleConfiguration {
+            PersistedLifecycleConfiguration {
                 expiry_updated_at: None,
-                rules: vec![LifecycleRule {
-                    status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                rules: vec![PersistedLifecycleRule {
+                    status: "Enabled".to_owned(),
                     expiration: None,
                     abort_incomplete_multipart_upload: None,
                     del_marker_expiration: None,
@@ -16335,10 +16386,10 @@ mod tests {
                     noncurrent_version_expiration: None,
                     noncurrent_version_transitions: None,
                     prefix: None,
-                    transitions: Some(vec![Transition {
+                    transitions: Some(vec![PersistedTransition {
                         days: Some(1),
                         date: None,
-                        storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                        storage_class: Some("COLD-A".to_owned()),
                     }]),
                 }],
             },
@@ -16991,10 +17042,10 @@ mod tests {
             .save_tiering_config_if_current(store.clone(), None)
             .await
             .expect("reference proof fixture should persist");
-        let config = BucketLifecycleConfiguration {
+        let config = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -17003,10 +17054,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![Transition {
+                transitions: Some(vec![PersistedTransition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                    storage_class: Some("COLD-A".to_owned()),
                 }]),
             }],
         };
@@ -17436,23 +17487,23 @@ mod tests {
             });
             store.add_lifecycle_config(
                 "private-bucket",
-                BucketLifecycleConfiguration {
+                rustfs_gateway_types::persistence::PersistedLifecycleConfiguration {
                     expiry_updated_at: None,
-                    rules: vec![LifecycleRule {
-                        status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                    rules: vec![rustfs_gateway_types::persistence::PersistedLifecycleRule {
+                        status: "Enabled".to_string(),
                         id: Some("private-rule".to_string()),
                         noncurrent_version_transitions: noncurrent.then(|| {
-                            vec![NoncurrentVersionTransition {
+                            vec![rustfs_gateway_types::persistence::PersistedNoncurrentVersionTransition {
                                 noncurrent_days: Some(1),
                                 newer_noncurrent_versions: None,
-                                storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                                storage_class: Some("COLD-A".to_string()),
                             }]
                         }),
                         transitions: (!noncurrent).then(|| {
-                            vec![Transition {
+                            vec![rustfs_gateway_types::persistence::PersistedTransition {
                                 days: Some(1),
                                 date: None,
-                                storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                                storage_class: Some("COLD-A".to_string()),
                             }]
                         }),
                         expiration: None,
@@ -17855,10 +17906,10 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = BucketLifecycleConfiguration {
+        let config = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -17867,10 +17918,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![Transition {
+                transitions: Some(vec![PersistedTransition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                    storage_class: Some("COLD-A".to_owned()),
                 }]),
             }],
         };
@@ -17900,20 +17951,20 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = BucketLifecycleConfiguration {
+        let config = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 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![NoncurrentVersionTransition {
+                noncurrent_version_transitions: Some(vec![PersistedNoncurrentVersionTransition {
                     noncurrent_days: Some(1),
                     newer_noncurrent_versions: None,
-                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                    storage_class: Some("COLD-A".to_owned()),
                 }]),
                 prefix: None,
                 transitions: None,
@@ -17949,10 +18000,10 @@ mod tests {
             old_backend_identity: Some(current_identity),
             new_backend_identity: None,
         };
-        let config = BucketLifecycleConfiguration {
+        let config = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -17961,10 +18012,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![Transition {
+                transitions: Some(vec![PersistedTransition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some(TransitionStorageClass::from_static("COLD-A")),
+                    storage_class: Some("COLD-A".to_owned()),
                 }]),
             }],
         };

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

@@ -705,9 +705,25 @@ 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) =
-            load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await?;
+            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());
+        }
         let advanced = intent
             .advance_idempotent(next, committed_config_etag.clone())
             .map_err(tier_mutation_intent_store_error)?;
@@ -718,7 +734,15 @@ where
             Ok(()) => return Ok((intent, true)),
             Err(Error::PreconditionFailed) => {
                 let (mut current, _) =
-                    load_tier_mutation_intent_record_with_etag_at_prefix(api.clone(), prefix, mutation_id).await?;
+                    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 !current.same_identity_as(&intent) {
                     return Err(Error::PreconditionFailed);
                 }
@@ -733,6 +757,11 @@ 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

@@ -37,7 +37,9 @@ 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,
@@ -50,10 +52,6 @@ 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;
@@ -194,12 +192,15 @@ pub trait WarmBackend {
             .await
         {
             Ok(_) => Ok(TransitionCandidateProbe::VersionedPresent(remote_version_id.to_string())),
-            Err(err) if matches!(to_error_response(&err).code, S3ErrorCode::InvalidRange) => {
+            Err(err) if matches!(to_error_response(&err).code, RemoteErrorCode::INVALID_RANGE) => {
                 Ok(TransitionCandidateProbe::VersionedPresent(remote_version_id.to_string()))
             }
             Err(err)
                 if err.kind() == std::io::ErrorKind::NotFound
-                    || matches!(to_error_response(&err).code, S3ErrorCode::NoSuchKey | S3ErrorCode::NoSuchVersion) =>
+                    || matches!(
+                        to_error_response(&err).code,
+                        RemoteErrorCode::NO_SUCH_KEY | RemoteErrorCode::NO_SUCH_VERSION
+                    ) =>
             {
                 Ok(TransitionCandidateProbe::Missing)
             }
@@ -219,9 +220,9 @@ pub fn build_transition_put_options(storage_class: String, mut metadata: HashMap
     let mut opts = PutObjectOptions {
         storage_class,
         send_content_md5: true,
-        legalhold: ObjectLockLegalHoldStatus::from_static(""),
+        legalhold: ObjectLockLegalHoldStatus::custom(""),
         internal: AdvancedPutOptions {
-            replication_status: ReplicationStatus::from_static(""),
+            replication_status: ReplicationStatus::custom(""),
             ..Default::default()
         },
         ..Default::default()
@@ -252,7 +253,7 @@ 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 = ObjectLockRetentionMode::from(mode.to_ascii_uppercase());
+        opts.mode = ObjectLockMode::custom(mode.to_ascii_uppercase());
     }
 
     if let Some(retain_until_date) = metadata
@@ -263,7 +264,7 @@ 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::from(legalhold.to_ascii_uppercase());
+        opts.legalhold = ObjectLockLegalHoldStatus::custom(legalhold.to_ascii_uppercase());
     }
 
     // Promoted keys are read above through `lookup`, which accepts more than
@@ -683,7 +684,7 @@ fn is_remote_tier_auth_rejection(err: &std::io::Error) -> bool {
     matches!(response.status_code, StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN)
         || matches!(
             response.code,
-            S3ErrorCode::AccessDenied | S3ErrorCode::InvalidAccessKeyId | S3ErrorCode::SignatureDoesNotMatch
+            RemoteErrorCode::ACCESS_DENIED | RemoteErrorCode::INVALID_ACCESS_KEY_ID | RemoteErrorCode::SIGNATURE_DOES_NOT_MATCH
         )
 }
 
@@ -1331,7 +1332,7 @@ mod tests {
 
     struct AuthRejectingProbePutBackend {
         status_code: StatusCode,
-        code: S3ErrorCode,
+        code: RemoteErrorCode,
         probes: Arc<AtomicUsize>,
         removes: Arc<AtomicUsize>,
     }
@@ -1794,10 +1795,10 @@ mod tests {
     #[tokio::test]
     async fn check_warm_backend_reports_auth_rejected_probe_put_as_invalid_credentials() {
         for (status_code, code) in [
-            (StatusCode::FORBIDDEN, S3ErrorCode::AccessDenied),
-            (StatusCode::UNAUTHORIZED, S3ErrorCode::Custom("".into())),
-            (StatusCode::OK, S3ErrorCode::InvalidAccessKeyId),
-            (StatusCode::OK, S3ErrorCode::SignatureDoesNotMatch),
+            (StatusCode::FORBIDDEN, RemoteErrorCode::ACCESS_DENIED),
+            (StatusCode::UNAUTHORIZED, RemoteErrorCode::custom("")),
+            (StatusCode::OK, RemoteErrorCode::INVALID_ACCESS_KEY_ID),
+            (StatusCode::OK, RemoteErrorCode::SIGNATURE_DOES_NOT_MATCH),
         ] {
             let probes = Arc::new(AtomicUsize::new(0));
             let removes = Arc::new(AtomicUsize::new(0));
@@ -2192,18 +2193,31 @@ mod tests {
             metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(),
             ObjectLockLegalHoldStatus::ON.to_string(),
         );
-        metadata.insert(
-            metadata_keys::OBJECT_LOCK_MODE.to_string(),
-            ObjectLockRetentionMode::GOVERNANCE.to_string(),
-        );
+        metadata.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), ObjectLockMode::GOVERNANCE.to_string());
 
         let opts = build_transition_put_options("COLD".to_string(), metadata);
 
-        assert_eq!(opts.mode.as_str(), ObjectLockRetentionMode::GOVERNANCE);
-        assert_eq!(opts.legalhold.as_str(), ObjectLockLegalHoldStatus::ON);
+        assert_eq!(opts.mode, ObjectLockMode::GOVERNANCE);
+        assert_eq!(opts.legalhold, 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

@@ -28,6 +28,8 @@ 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,
@@ -42,7 +44,6 @@ 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>,
@@ -61,8 +62,8 @@ enum RemoteBucketVersioning {
 
 fn remote_bucket_versioning_from_status(status: Option<&str>) -> Result<RemoteBucketVersioning, std::io::Error> {
     Ok(match status {
-        Some(BucketVersioningStatus::ENABLED) => RemoteBucketVersioning::Enabled,
-        Some(BucketVersioningStatus::SUSPENDED) => RemoteBucketVersioning::Suspended,
+        Some(status) if status == BucketVersioningStatus::ENABLED.as_str() => RemoteBucketVersioning::Enabled,
+        Some(status) if status == BucketVersioningStatus::SUSPENDED.as_str() => RemoteBucketVersioning::Suspended,
         Some(status) => {
             return Err(std::io::Error::new(
                 std::io::ErrorKind::InvalidData,
@@ -237,7 +238,7 @@ impl WarmBackendS3 {
             }
             Err(err) => {
                 let response = to_error_response(&err);
-                if response.code == S3ErrorCode::NoSuchKey {
+                if response.code == RemoteErrorCode::NO_SUCH_KEY {
                     Ok(TransitionCandidateProbe::Missing)
                 } else {
                     Err(err)
@@ -324,7 +325,7 @@ impl WarmBackendS3 {
             }
             Err(err) => {
                 let response = to_error_response(&err);
-                if response.code == S3ErrorCode::NoSuchKey {
+                if response.code == RemoteErrorCode::NO_SUCH_KEY {
                     Ok(TransitionCandidateProbe::Missing)
                 } else {
                     Err(err)
@@ -688,7 +689,7 @@ 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, S3ErrorCode::AccessDenied);
+        assert_eq!(to_error_response(&err).code, RemoteErrorCode::ACCESS_DENIED);
         assert_eq!(
             backend
                 .probe_transition_candidate("delete-marker-hidden")
@@ -1066,11 +1067,13 @@ mod tests {
             RemoteBucketVersioning::Disabled
         );
         assert_eq!(
-            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::ENABLED)).expect("enabled status should parse"),
+            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::ENABLED.as_str()))
+                .expect("enabled status should parse"),
             RemoteBucketVersioning::Enabled
         );
         assert_eq!(
-            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::SUSPENDED)).expect("suspended status should parse"),
+            remote_bucket_versioning_from_status(Some(BucketVersioningStatus::SUSPENDED.as_str()))
+                .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

@@ -18,7 +18,6 @@ use std::{
     sync::atomic::{AtomicBool, Ordering},
 };
 
-use s3s::header::{X_AMZ_DELETE_MARKER, X_AMZ_VERSION_ID};
 use uuid::Uuid;
 
 use crate::services::tier::{
@@ -128,7 +127,7 @@ 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/mod.rs

@@ -153,8 +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_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_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,
 };
 use rustfs_utils::{
     HashAlgorithm,
@@ -14047,10 +14047,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14062,7 +14059,7 @@ mod tests {
         };
         let opts = ObjectOptions {
             object_lock_retention: Some(crate::storage_api_contracts::object::ObjectLockRetentionOptions {
-                mode: Some(s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string()),
+                mode: Some("COMPLIANCE".to_string()),
                 retain_until: Some(requested_until),
                 bypass_governance: true,
             }),
@@ -14082,10 +14079,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14097,7 +14091,7 @@ mod tests {
         };
         let opts = ObjectOptions {
             object_lock_retention: Some(crate::storage_api_contracts::object::ObjectLockRetentionOptions {
-                mode: Some(s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string()),
+                mode: Some("GOVERNANCE".to_string()),
                 retain_until: Some(requested_until),
                 bypass_governance: true,
             }),
@@ -14112,10 +14106,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14157,10 +14148,7 @@ mod tests {
             },
             restore_expires: Some(restore_expiry),
             user_defined: Arc::new(HashMap::from([
-                (
-                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                    s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                ),
+                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                 (
                     metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
                     retain_until.format(&time::format_description::well_known::Rfc3339).unwrap(),
@@ -14231,10 +14219,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14259,10 +14244,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14312,10 +14294,7 @@ 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(),
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "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(),
@@ -14370,11 +14349,11 @@ mod tests {
 
     #[test]
     fn test_object_lock_delete_check_required_keeps_object_lock_config_buckets() {
-        use s3s::dto::{ObjectLockConfiguration, ObjectLockEnabled};
+        use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
 
         let mut bm = crate::bucket::metadata::BucketMetadata::new("lock-config-bucket");
         bm.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             ..Default::default()
         });
         assert!(object_lock_delete_check_required(Some(&bm)));
@@ -17271,8 +17250,8 @@ 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(
-            s3s::dto::VersioningConfiguration {
-                status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::SUSPENDED)),
+            rustfs_gateway_types::persistence::PersistedVersioningConfiguration {
+                status: Some("Suspended".to_owned()),
                 ..Default::default()
             },
             None,

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

@@ -3039,6 +3039,86 @@ 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";
@@ -3405,6 +3485,140 @@ 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/multipart.rs

@@ -38,16 +38,16 @@ 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_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, restore_operation_id_from_metadata, should_persist_encryption_original_size,
-    strip_internal_multipart_metadata, to_object_err, warn,
+    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_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, restore_operation_id_from_metadata,
+    should_persist_encryption_original_size, strip_internal_multipart_metadata, to_object_err, warn,
 };
 use super::bitrot_self_verify::{BitrotSelfVerifyTarget, drop_failed_writer_disks, verify_written_bitrot_shards};
 #[cfg(test)]
@@ -3724,7 +3724,14 @@ 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);
-        if expected_restore_operation_id.is_some() {
+        // 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))
+        {
             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);
         }
@@ -4109,6 +4116,13 @@ 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

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

@@ -41,23 +41,24 @@ use super::super::{
     ObjectKey, ObjectLockConfigSnapshot, ObjectLockConfigState, ObjectOptions, ObjectReader, ObjectToDelete, OffsetDateTime,
     Ordering, Pin, PutObjReader, RUSTFS_META_BUCKET, RUSTFS_META_TMP_BUCKET, ReadPathPlan, ReaderImpl, ReplicateDecision,
     ReplicationObjectBridge, Result, SET_DISK_COMMIT_TAIL_WARN_THRESHOLD_MS, SLASH_SEPARATOR, SUFFIX_ACTUAL_SIZE,
-    SUFFIX_COMPRESSION, SUFFIX_COMPRESSION_SIZE, SUFFIX_RESTORE_OPERATION_ID, SUFFIX_RESTORE_WORKER_LOCK, SetDisks,
-    SmallWritePath, StorageError, TRANSITION_COMPLETE, UpdateMetadataOpts, Uuid, WriteLayout, adaptive_duplex_buffer_size,
-    build_get_object_info, build_inline_bitrot_readers, build_inline_bitrot_readers_from_refs, can_try_inline_data_shards_direct,
-    check_object_lock_delete, check_object_lock_for_deletion_with_state, check_object_lock_retention_update,
-    classify_get_codec_streaming_object_class, classify_put_write_path, classify_storage_error,
-    collect_inline_data_shard_fileinfos_by_index, contains_key_str, create_bitrot_writer_with_orig_volume, debug,
-    delete_file_info_version_id, disk, ensure_delete_commit_locks_held, error, explicit_delete_removed_marker,
-    finish_set_disk_read_lock, get_codec_streaming_reader_gate_with_plan, get_object_body_cache_hook, get_raw_etag,
-    get_small_object_direct_memory_decision_with_threshold_and_plan, get_small_object_direct_memory_threshold,
-    get_stage_timer_if_enabled, get_str, get_transitioned_object_reader_with_tier_manager, inline_admission_shard_size,
-    inline_erasure_shard_file_offset, inline_erasure_shard_size, insert_str, is_deadlock_detection_enabled,
-    is_err_object_not_found, is_err_version_not_found, is_explicit_null_version, is_get_codec_streaming_base_enabled,
-    is_get_small_object_direct_memory_enabled, is_lock_optimization_enabled, issue3031_diag_enabled, join_all,
-    known_put_object_storage_size, path_join_buf, put_restore_opts, record_compression_total_memory,
-    record_get_codec_streaming_gate_decision, record_get_direct_memory_decision, record_get_object_pipeline_failure,
-    record_get_object_pipeline_failure_for_path, record_get_object_reader_path_observation, record_get_stage_duration_if_enabled,
-    record_lock_acquire, reduce_write_quorum_errs, release_materialized_read_lock, replication_write_may_pass_worm_gate,
+    SUFFIX_COMPRESSION, SUFFIX_COMPRESSION_SIZE, SUFFIX_RESTORE_COPYBACK, SUFFIX_RESTORE_OPERATION_ID,
+    SUFFIX_RESTORE_WORKER_LOCK, SetDisks, SmallWritePath, StorageError, TRANSITION_COMPLETE, UpdateMetadataOpts, Uuid,
+    WriteLayout, adaptive_duplex_buffer_size, build_get_object_info, build_inline_bitrot_readers,
+    build_inline_bitrot_readers_from_refs, can_try_inline_data_shards_direct, check_object_lock_delete,
+    check_object_lock_for_deletion_with_state, check_object_lock_retention_update, classify_get_codec_streaming_object_class,
+    classify_put_write_path, classify_storage_error, collect_inline_data_shard_fileinfos_by_index, contains_key_str,
+    create_bitrot_writer_with_orig_volume, debug, delete_file_info_version_id, disk, ensure_delete_commit_locks_held, error,
+    explicit_delete_removed_marker, finish_set_disk_read_lock, get_codec_streaming_reader_gate_with_plan,
+    get_object_body_cache_hook, get_raw_etag, get_small_object_direct_memory_decision_with_threshold_and_plan,
+    get_small_object_direct_memory_threshold, get_stage_timer_if_enabled, get_str,
+    get_transitioned_object_reader_with_tier_manager, inline_admission_shard_size, inline_erasure_shard_file_offset,
+    inline_erasure_shard_size, insert_str, is_deadlock_detection_enabled, is_err_object_not_found, is_err_version_not_found,
+    is_explicit_null_version, is_get_codec_streaming_base_enabled, is_get_small_object_direct_memory_enabled,
+    is_lock_optimization_enabled, issue3031_diag_enabled, join_all, known_put_object_storage_size, path_join_buf,
+    put_restore_opts, record_compression_total_memory, record_get_codec_streaming_gate_decision,
+    record_get_direct_memory_decision, record_get_object_pipeline_failure, record_get_object_pipeline_failure_for_path,
+    record_get_object_reader_path_observation, record_get_stage_duration_if_enabled, record_lock_acquire,
+    reduce_write_quorum_errs, release_materialized_read_lock, replication_write_may_pass_worm_gate,
     republishes_data_movement_delete_marker, require_restore_operation_id, resolve_delete_version_state,
     resolve_tiered_decommission_write_quorum_result, resolve_write_layout, restore_commit_operation_id_from_metadata,
     restore_operation_id_from_metadata, send_event, set_disk_delete_creates_delete_marker,
@@ -2360,6 +2361,8 @@ fn is_restore_control_metadata(key: &str) -> bool {
             .is_some_and(|remainder| remainder.is_empty())
         || rustfs_utils::http::internal_key_strip_suffix_prefix(key, SUFFIX_RESTORE_WORKER_LOCK)
             .is_some_and(|remainder| remainder.is_empty())
+        || rustfs_utils::http::internal_key_strip_suffix_prefix(key, SUFFIX_RESTORE_COPYBACK)
+            .is_some_and(|remainder| remainder.is_empty())
 }
 
 fn restore_metadata_update_preserves_protected_metadata(
@@ -4310,7 +4313,13 @@ impl SetDisks {
             // state.
             merge_replication_metadata_lww(&mut user_defined, &HashMap::new(), opts);
         }
-        if expected_restore_operation_id.is_some() {
+        // Pending copy-back keeps its recovery identity until finalization.
+        if expected_restore_operation_id.is_some()
+            && !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))
+        {
             rustfs_utils::http::metadata_compat::remove_str(&mut user_defined, SUFFIX_RESTORE_OPERATION_ID);
             rustfs_utils::http::metadata_compat::remove_str(&mut user_defined, SUFFIX_RESTORE_WORKER_LOCK);
         }
@@ -4711,19 +4720,29 @@ impl SetDisks {
                     inline_proof.to_owned(),
                 );
             }
+            if opts.restore_copyback {
+                insert_str(
+                    &mut user_defined,
+                    SUFFIX_RESTORE_COPYBACK,
+                    rustfs_utils::http::metadata_compat::RESTORE_COPYBACK_PROTOCOL_V1.to_string(),
+                );
+            }
             fi.metadata = user_defined;
-            let put_tier_free_version_id =
-                if fi.version_id.is_none_or(|id| id.is_nil()) && !opts.data_movement && expected_restore_operation_id.is_none() {
-                    // Current disks publish the same cleanup owner alongside a
-                    // replaced null version. Keep the intent outside metadata:
-                    // older disks persist unknown metadata keys, which splits
-                    // the replacement's read-quorum identity during an upgrade.
-                    let free_version_id = Uuid::new_v4();
-                    fi.overwrite_tier_free_version_id = Some(free_version_id);
-                    Some(free_version_id)
-                } else {
-                    None
-                };
+            let put_tier_free_version_id = if fi.version_id.is_none_or(|id| id.is_nil())
+                && !opts.data_movement
+                && expected_restore_operation_id.is_none()
+                && !opts.restore_copyback
+            {
+                // Current disks publish the same cleanup owner alongside a
+                // replaced null version. Keep the intent outside metadata:
+                // older disks persist unknown metadata keys, which splits
+                // the replacement's read-quorum identity during an upgrade.
+                let free_version_id = Uuid::new_v4();
+                fi.overwrite_tier_free_version_id = Some(free_version_id);
+                Some(free_version_id)
+            } else {
+                None
+            };
             fi.mod_time = mod_time;
             fi.size = w_size as i64;
             fi.versioned = opts.versioned || opts.version_suspended;
@@ -11893,7 +11912,14 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         if fi.transition_status == TRANSITION_COMPLETE {
             return Ok(());
         }
-
+        if 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))
+        {
+            return Ok(());
+        }
         /*if fi.xlv1 {
             if let Err(err) = self.heal_object(bucket, object, "", &HealOpts {no_lock: true, ..Default::default()}) {
                 return err.expect("err");
@@ -12400,7 +12426,9 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         let actual_fi = actual.fi();
 
         let oi = ObjectInfo::from_file_info(actual_fi, bucket, object, opts.versioned || opts.version_suspended);
+        let mut restore_cleanup_info = oi.clone();
         let restore_result: Result<()> = async {
+            crate::services::notification_sys::require_restore_copyback_fleet_capability().await?;
             // The tier reader releases its own lease at EOF, before PUT or
             // CompleteMultipartUpload necessarily reaches metadata quorum. Keep a
             // second exact-generation lease for the whole restore so the final
@@ -12479,6 +12507,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     .put_object(bucket, object, &mut p_reader, &ropts)
                     .await
                     .map_err(|err| to_object_err(err, vec![bucket, object]))?;
+                restore_cleanup_info = restored_info.clone();
                 let restored_info = self_.finalize_restore_metadata(bucket, object, &restored_info, &opts).await?;
                 send_event(EventArgs {
                     event_name: EventName::ObjectRestoreCompleted.as_str().to_string(),
@@ -12598,6 +12627,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     bucket_lifecycle_lock_fence: opts.bucket_lifecycle_lock_fence.clone(),
                     user_defined: restore_commit_metadata,
                     no_lock: false,
+                    restore_copyback: true,
                     decommission_capacity_admission: opts.decommission_capacity_admission.clone(),
                     // The composite ETag would otherwise be recomputed from the
                     // parts as they were written back, which for an encrypted or
@@ -12622,6 +12652,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     return Err(err);
                 }
             };
+            restore_cleanup_info = restored_info.clone();
             let restored_info = self_.finalize_restore_metadata(bucket, object, &restored_info, opts).await?;
             send_event(EventArgs {
                 event_name: EventName::ObjectRestoreCompleted.as_str().to_string(),
@@ -12635,7 +12666,10 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         }
         .await;
         if let Err(primary_error) = restore_result {
-            if let Err(cleanup_error) = self_.update_restore_metadata(bucket, object, &oi, opts).await {
+            if let Err(cleanup_error) = self_
+                .update_restore_metadata(bucket, object, &restore_cleanup_info, opts)
+                .await
+            {
                 warn!(
                     event = EVENT_LIFECYCLE_RESTORE_CLEANUP,
                     component = LOG_COMPONENT_ECSTORE,
@@ -13608,8 +13642,8 @@ mod replication_quota_safety_tests {
         let suspended_opts = ObjectOptions {
             version_suspended: true,
             delete_replication_config_snapshot: Some(Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
-                s3s::dto::VersioningConfiguration {
-                    status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::SUSPENDED)),
+                rustfs_gateway_types::persistence::PersistedVersioningConfiguration {
+                    status: Some("Suspended".to_owned()),
                     ..Default::default()
                 },
                 None,
@@ -14242,11 +14276,11 @@ mod replication_lww_tests {
 
     fn default_retention_snapshot(mode: &'static str) -> Arc<ObjectLockConfigSnapshot> {
         Arc::new(ObjectLockConfigSnapshot::new(ObjectLockConfigState::Configured {
-            config: s3s::dto::ObjectLockConfiguration {
-                object_lock_enabled: Some(s3s::dto::ObjectLockEnabled::from_static(s3s::dto::ObjectLockEnabled::ENABLED)),
-                rule: Some(s3s::dto::ObjectLockRule {
-                    default_retention: Some(s3s::dto::DefaultRetention {
-                        mode: Some(s3s::dto::ObjectLockRetentionMode::from_static(mode)),
+            config: rustfs_gateway_types::persistence::PersistedObjectLockConfiguration {
+                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.to_owned()),
                         days: Some(1),
                         years: None,
                     }),
@@ -14263,10 +14297,7 @@ mod replication_lww_tests {
     /// timestamp the write passes and LWW judges it.
     #[tokio::test]
     async fn inbound_without_retention_timestamp_stays_rejected_under_bucket_default_retention() {
-        for mode in [
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE,
-            s3s::dto::ObjectLockRetentionMode::GOVERNANCE,
-        ] {
+        for mode in ["COMPLIANCE", "GOVERNANCE"] {
             let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
             let bucket = "lww-locked-default-retention";
             let object = "object";
@@ -16419,8 +16450,8 @@ mod metadata_mutation_generation_tests {
 
     #[tokio::test]
     async fn restore_metadata_copy_preserves_tags_expiry_and_version_order() {
+        use rustfs_lifecycle::RestoreRequest;
         use rustfs_utils::http::headers::AMZ_OBJECT_TAGGING;
-        use s3s::dto::RestoreRequest;
         use time::format_description::well_known::Rfc3339;
 
         let (_temp_dirs, disks, set) = hermetic_set_disks(4).await;
@@ -16697,7 +16728,7 @@ mod transition_commit_failure_tests {
     use crate::services::tier::tier::TierConfigMgr;
     use http::HeaderMap;
     use rustfs_filemeta::{RestoreStatusOps as _, parse_restore_obj_status};
-    use s3s::dto::RestoreRequest;
+    use rustfs_lifecycle::RestoreRequest;
     use tokio::io::AsyncReadExt;
 
     pub(super) fn restore_operation_id_metadata(operation_id: Uuid) -> HashMap<String, String> {
@@ -19276,7 +19307,7 @@ mod transition_upload_integrity_tests {
         for disk in disk_stores {
             disk.make_volume(bucket).await.expect("bucket volume should be created");
         }
-        let mut source = PutObjReader::from_vec(b"restore source body".repeat(1024));
+        let mut source = PutObjReader::from_vec(b"restore source body".repeat(64 * 1024));
         set_disks
             .put_object(bucket, object, &mut source, &ObjectOptions::default())
             .await
@@ -19293,7 +19324,7 @@ mod transition_upload_integrity_tests {
             .await
             .expect("pending restore metadata should be installed");
 
-        let mut restored_reader = PutObjReader::from_vec(b"restored body".repeat(1024));
+        let mut restored_reader = PutObjReader::from_vec(b"restored body".repeat(64 * 1024));
         set_disks
             .put_object(
                 bucket,
@@ -19301,18 +19332,19 @@ mod transition_upload_integrity_tests {
                 &mut restored_reader,
                 &ObjectOptions {
                     user_defined: restore_metadata(operation_id, true),
+                    restore_copyback: true,
                     ..Default::default()
                 },
             )
             .await
-            .expect("matching restore commit should consume its operation id")
+            .expect("matching restore copy-back should retain its pending operation id")
     }
 
     fn restore_finalize_options(operation_id: Uuid) -> ObjectOptions {
         ObjectOptions {
             user_defined: restore_operation_id_metadata(operation_id),
             transition: TransitionOptions {
-                restore_request: s3s::dto::RestoreRequest {
+                restore_request: rustfs_lifecycle::RestoreRequest {
                     days: Some(1),
                     ..Default::default()
                 },
@@ -19322,17 +19354,87 @@ mod transition_upload_integrity_tests {
         }
     }
 
-    async fn assert_committed_restore_remains_pending(set_disks: &Arc<SetDisks>, bucket: &str, object: &str) {
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn restore_cleanup_reclaims_committed_copyback_data_dir() {
+        let (temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "restore-cleanup-copyback-data-dir-bucket";
+        let object = "object.bin";
+        let operation_id = Uuid::new_v4();
+        let restored = write_committed_restore(&set_disks, &disk_stores, bucket, object, operation_id).await;
+        let data_dir = restored
+            .data_dir
+            .expect("committed copy-back must have a local data directory");
+        for temp_dir in &temp_dirs {
+            assert!(
+                temp_dir.path().join(bucket).join(object).join(data_dir.to_string()).exists(),
+                "committed copy-back must retain its local data directory before cleanup"
+            );
+        }
+
+        set_disks
+            .update_restore_metadata(
+                bucket,
+                object,
+                &restored,
+                &ObjectOptions {
+                    user_defined: restore_operation_id_metadata(operation_id),
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("matching restore cleanup should remove the failed copy-back");
+
+        for temp_dir in &temp_dirs {
+            assert!(
+                !temp_dir.path().join(bucket).join(object).join(data_dir.to_string()).exists(),
+                "failed copy-back cleanup must reclaim its committed local data directory"
+            );
+        }
+        let cleaned = set_disks
+            .get_object_info(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("cleaned object metadata should remain readable");
+        assert!(!cleaned.user_defined.contains_key(rustfs_filemeta::metadata_keys::RESTORE));
+        assert!(
+            rustfs_utils::http::metadata_compat::get_consistent_str(
+                cleaned.user_defined.as_ref(),
+                rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
+            )
+            .is_none(),
+            "failed copy-back cleanup must remove the local routing marker"
+        );
+    }
+
+    async fn assert_committed_restore_remains_pending(set_disks: &Arc<SetDisks>, bucket: &str, object: &str, operation_id: Uuid) {
         let current = set_disks
             .get_object_info(bucket, object, &ObjectOptions::default())
             .await
             .expect("pending restore metadata should remain readable");
-        assert!(
-            restore_operation_id_from_metadata(current.user_defined.as_ref())
-                .expect("operation id metadata should parse")
-                .is_none(),
-            "successful restore commit must have consumed the operation id"
+        assert_eq!(
+            restore_operation_id_from_metadata(current.user_defined.as_ref()).expect("operation id metadata should parse"),
+            Some(operation_id),
+            "unfinished restore copy-back must retain its exact operation id"
         );
+        let operation_id = operation_id.to_string();
+        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!(
+                    current.user_defined.get(&format!("{prefix}{suffix}")).map(String::as_str),
+                    Some(expected),
+                    "unfinished restore copy-back must retain each recovery marker alias"
+                );
+            }
+        }
         assert!(
             rustfs_filemeta::parse_restore_obj_status(
                 current
@@ -19390,7 +19492,7 @@ mod transition_upload_integrity_tests {
                 if error.kind() == std::io::ErrorKind::Other
                     && error.to_string() == "restore finalization lock lost before metadata update"
         ));
-        assert_committed_restore_remains_pending(&set_disks, bucket, object).await;
+        assert_committed_restore_remains_pending(&set_disks, bucket, object, operation_id).await;
     }
 
     #[tokio::test]
@@ -19427,7 +19529,7 @@ mod transition_upload_integrity_tests {
                 ..
             }
         ));
-        assert_committed_restore_remains_pending(&set_disks, bucket, object).await;
+        assert_committed_restore_remains_pending(&set_disks, bucket, object, operation_id).await;
     }
 
     async fn assert_local_source_intact(set_disks: &Arc<SetDisks>, bucket: &str, object: &str, payload: &[u8]) {
@@ -19977,7 +20079,7 @@ mod transition_upload_integrity_tests {
             no_lock: true,
             part_number: Some(1),
             transition: TransitionOptions {
-                restore_request: s3s::dto::RestoreRequest {
+                restore_request: rustfs_lifecycle::RestoreRequest {
                     days: Some(1),
                     ..Default::default()
                 },
@@ -24654,29 +24756,20 @@ mod put_object_tmp_cleanup_tests {
             (
                 "compliance",
                 HashMap::from([
-                    (
-                        metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                        s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                    ),
+                    (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                     (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until.clone()),
                 ]),
             ),
             (
                 "governance",
                 HashMap::from([
-                    (
-                        metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                        s3s::dto::ObjectLockRetentionMode::GOVERNANCE.to_string(),
-                    ),
+                    (metadata_keys::OBJECT_LOCK_MODE.to_string(), "GOVERNANCE".to_string()),
                     (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
                 ]),
             ),
             (
                 "legal-hold",
-                HashMap::from([(
-                    metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(),
-                    s3s::dto::ObjectLockLegalHoldStatus::ON.to_string(),
-                )]),
+                HashMap::from([(metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(), "ON".to_string())]),
             ),
         ];
 
@@ -24877,13 +24970,11 @@ mod put_object_tmp_cleanup_tests {
             .expect("versioned PUT should return a version ID")
             .to_string();
         let default_compliance = ObjectLockConfigState::Configured {
-            config: s3s::dto::ObjectLockConfiguration {
-                object_lock_enabled: Some(s3s::dto::ObjectLockEnabled::from_static(s3s::dto::ObjectLockEnabled::ENABLED)),
-                rule: Some(s3s::dto::ObjectLockRule {
-                    default_retention: Some(s3s::dto::DefaultRetention {
-                        mode: Some(s3s::dto::ObjectLockRetentionMode::from_static(
-                            s3s::dto::ObjectLockRetentionMode::COMPLIANCE,
-                        )),
+            config: rustfs_gateway_types::persistence::PersistedObjectLockConfiguration {
+                object_lock_enabled: Some("Enabled".to_owned()),
+                rule: Some(rustfs_gateway_types::persistence::PersistedObjectLockRule {
+                    default_retention: Some(rustfs_gateway_types::persistence::PersistedDefaultRetention {
+                        mode: Some("COMPLIANCE".to_owned()),
                         days: Some(1),
                         years: None,
                     }),
@@ -25043,10 +25134,7 @@ mod put_object_tmp_cleanup_tests {
                 &ObjectOptions {
                     version_id: Some(destination_version.clone()),
                     versioned: true,
-                    eval_metadata: Some(HashMap::from([(
-                        metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(),
-                        s3s::dto::ObjectLockLegalHoldStatus::ON.to_string(),
-                    )])),
+                    eval_metadata: Some(HashMap::from([(metadata_keys::OBJECT_LOCK_LEGAL_HOLD.to_string(), "ON".to_string())])),
                     ..Default::default()
                 },
             )
@@ -26079,7 +26167,7 @@ mod delete_objects_lock_gating_tests {
 
     #[tokio::test]
     async fn delete_objects_derives_per_object_versioning_from_the_request_snapshot() {
-        use s3s::dto::{BucketVersioningStatus, ExcludedPrefix, VersioningConfiguration};
+        use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 
         let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
         let bucket = "batch-versioning-snapshot-bucket";
@@ -26091,10 +26179,8 @@ mod delete_objects_lock_gating_tests {
         put_plain_object(&set_disks, bucket, "archive/unversioned-object").await;
         let snapshot = Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
-                excluded_prefixes: Some(vec![ExcludedPrefix {
-                    prefix: Some("archive/".to_string()),
-                }]),
+                status: Some("Enabled".to_owned()),
+                excluded_prefixes: Some(vec![Some("archive/".to_string())]),
                 ..Default::default()
             },
             None,
@@ -26140,11 +26226,13 @@ mod delete_objects_lock_gating_tests {
 
     #[tokio::test]
     async fn batch_version_delete_uses_tags_read_under_the_delete_lock() {
-        use rustfs_utils::http::headers::AMZ_OBJECT_TAGGING;
-        use s3s::dto::{
-            BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration,
-            ReplicationRule, ReplicationRuleFilter, ReplicationRuleStatus, Tag, VersioningConfiguration,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+        use rustfs_gateway_types::persistence::{
+            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationFilter,
+            PersistedReplicationRule, PersistedReplicationStatus, PersistedReplicationTag,
         };
+        use rustfs_utils::http::headers::AMZ_OBJECT_TAGGING;
 
         let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
         let bucket = "batch-delete-locked-tags";
@@ -26180,23 +26268,23 @@ mod delete_objects_lock_gating_tests {
 
         let snapshot = Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::ENABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
                     existing_object_replication: None,
-                    filter: Some(ReplicationRuleFilter {
-                        tag: Some(Tag {
+                    filter: Some(PersistedReplicationFilter {
+                        tag: Some(PersistedReplicationTag {
                             key: Some("generation".to_string()),
                             value: Some("locked".to_string()),
                         }),
@@ -26206,7 +26294,7 @@ mod delete_objects_lock_gating_tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
         ));
@@ -26244,9 +26332,11 @@ mod delete_objects_lock_gating_tests {
     #[tokio::test]
     #[serial]
     async fn lifecycle_delete_all_history_records_exact_replication_purge() {
-        use s3s::dto::{
-            BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration,
-            ReplicationRule, ReplicationRuleStatus, VersioningConfiguration,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+        use rustfs_gateway_types::persistence::{
+            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationRule,
+            PersistedReplicationStatus,
         };
 
         let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
@@ -26288,17 +26378,17 @@ mod delete_objects_lock_gating_tests {
 
         let snapshot = Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::ENABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
@@ -26308,7 +26398,7 @@ mod delete_objects_lock_gating_tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
         ));
@@ -26350,10 +26440,11 @@ mod delete_objects_lock_gating_tests {
 
     #[tokio::test]
     async fn synthetic_directory_delete_uses_decoded_prefix_and_marker_switch() {
-        use s3s::dto::{
-            BucketVersioningStatus, DeleteMarkerReplication, DeleteMarkerReplicationStatus, DeleteReplication,
-            DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule, ReplicationRuleFilter,
-            ReplicationRuleStatus, VersioningConfiguration,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+        use rustfs_gateway_types::persistence::{
+            PersistedOptionalReplicationStatus, PersistedReplicationConfiguration, PersistedReplicationDestination,
+            PersistedReplicationFilter, PersistedReplicationRule, PersistedReplicationStatus,
         };
 
         let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
@@ -26367,24 +26458,24 @@ mod delete_objects_lock_gating_tests {
 
         let snapshot = Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
-                    delete_marker_replication: Some(DeleteMarkerReplication {
-                        status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
+                rules: vec![PersistedReplicationRule {
+                    delete_marker_replication: Some(PersistedOptionalReplicationStatus {
+                        status: Some(Status::ENABLED.to_string()),
                     }),
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::DISABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
                     existing_object_replication: None,
-                    filter: Some(ReplicationRuleFilter {
+                    filter: Some(PersistedReplicationFilter {
                         prefix: Some("photos/".to_string()),
                         ..Default::default()
                     }),
@@ -26392,7 +26483,7 @@ mod delete_objects_lock_gating_tests {
                     prefix: None,
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
         ));
@@ -26524,10 +26615,7 @@ mod delete_objects_lock_gating_tests {
         // stat and the #4297 rejection.
         let retain_until = OffsetDateTime::now_utc() + Duration::from_secs(60 * 60 * 24 * 30);
         let mut user_defined = HashMap::new();
-        user_defined.insert(
-            metadata_keys::OBJECT_LOCK_MODE.to_string(),
-            s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-        );
+        user_defined.insert(metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string());
         user_defined.insert(
             metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(),
             retain_until
@@ -28157,9 +28245,10 @@ mod batch_delete_marker_tests {
 
     #[tokio::test]
     async fn batch_delete_explicit_marker_does_not_wait_for_pending_replication() {
-        use s3s::dto::{
-            BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration,
-            ReplicationRule, ReplicationRuleStatus, VersioningConfiguration,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::{
+            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationRule,
+            PersistedReplicationStatus, PersistedVersioningConfiguration,
         };
 
         let (_dirs, store) = isolated_store_over_temp_disks().await;
@@ -28167,18 +28256,18 @@ mod batch_delete_marker_tests {
         let (data_versions, markers) = marker_history(&store, bucket).await;
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let snapshot = Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
-            VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+            PersistedVersioningConfiguration {
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::ENABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
@@ -28188,7 +28277,7 @@ mod batch_delete_marker_tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
         ));

crates/ecstore/src/set_disk/read.rs

@@ -2979,10 +2979,11 @@ fn build_legacy_per_part_fallback_reader(
     part_length: usize,
     part_size: usize,
     read_repair: Option<(DecodeReadRepairContext, bool)>,
-) -> Box<dyn AsyncRead + Unpin + Send + Sync> {
+) -> Box<LegacyPerPartDecodeReader> {
     let buffer = adaptive_duplex_buffer_size(part_size as i64);
-    let (read_half, mut write_half) = tokio::io::duplex(buffer);
+    let (read_half, write_half) = tokio::io::duplex(buffer);
     let decode = tokio::spawn(async move {
+        let mut write_half = GetObjectDownstreamWriter::new(write_half);
         let outcome = erasure
             .decode_with_stripe_handles_and_reopeners_with_diagnostics(
                 &mut write_half,
@@ -8908,6 +8909,42 @@ mod tests {
         assert!(err.to_string().contains("inconsistent read source shards"));
     }
 
+    #[tokio::test]
+    async fn legacy_per_part_output_close_retains_downstream_classification() {
+        let erasure = coding::Erasure::new(2, 2, 64);
+        let data = (0..64u8).collect::<Vec<_>>();
+        let setup =
+            setup_codec_data_blocks_first_encoded_bitrot_readers(&erasure, &data, &[0], &[], &[], HashAlgorithm::HighwayHash256)
+                .await;
+        let mut reader = build_legacy_per_part_fallback_reader(
+            erasure,
+            setup.readers,
+            setup.deferred_stripe_handles,
+            Vec::new(),
+            None,
+            0,
+            data.len(),
+            data.len(),
+            None,
+        );
+        // Closing the actual output pipe models the body consumer disappearing
+        // while retaining the executor long enough to inspect its typed result.
+        assert!(!reader.finished);
+        let (replacement, _unused) = tokio::io::duplex(1);
+        drop(std::mem::replace(&mut reader.inner, replacement));
+        let handle = reader.decode.take().expect("fallback decode task");
+        let error = handle
+            .await
+            .expect("join the fallback task")
+            .expect("closed pipe must interrupt output");
+        assert_eq!(error.kind(), std::io::ErrorKind::BrokenPipe);
+        assert_eq!(
+            crate::diagnostics::get::classify_io_error(&error),
+            GetObjectFailureReason::DownstreamClosed,
+            "fallback output must carry the same marker as the normal GET path"
+        );
+    }
+
     // backlog#879: a later multipart part that loses a shard (read quorum still
     // holds) must be reconstructed in place through the legacy per-part decode
     // bridge instead of failing the stream. This exercises the exact reader the

crates/ecstore/src/set_disk/replication.rs

@@ -27,6 +27,7 @@ use std::sync::Arc;
 struct RestoreFinalizeBarrierState {
     bucket: String,
     object: String,
+    before_lock: bool,
     arrived: tokio::sync::Notify,
     release: tokio::sync::Notify,
 }
@@ -43,9 +44,18 @@ 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(),
         });
@@ -81,13 +91,13 @@ impl Drop for RestoreFinalizeBarrier {
 }
 
 #[cfg(all(test, feature = "test-util"))]
-async fn maybe_pause_restore_finalize(bucket: &str, object: &str) {
+async fn maybe_pause_restore_finalize(bucket: &str, object: &str, before_lock: bool) {
     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)
+        .filter(|barrier| barrier.bucket == bucket && barrier.object == object && barrier.before_lock == before_lock)
         .cloned();
     if let Some(barrier) = barrier {
         barrier.arrived.notify_one();
@@ -150,6 +160,8 @@ 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
@@ -208,7 +220,7 @@ 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).await;
+        maybe_pause_restore_finalize(bucket, object, false).await;
         let restore_expiry =
             lifecycle::expected_expiry_time(OffsetDateTime::now_utc(), opts.transition.restore_request.days.unwrap_or(1));
         fi.metadata.insert(
@@ -219,7 +231,14 @@ 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,
         ] {
@@ -315,10 +334,20 @@ 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,
@@ -348,6 +377,20 @@ 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

@@ -274,3 +274,452 @@ 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/heal.rs

@@ -527,7 +527,7 @@ impl ECStore {
     async fn acquire_heal_format_fence(
         &self,
     ) -> Result<(
-        tokio::sync::MutexGuard<'_, PoolMetaWriteState>,
+        crate::core::pools::PoolMetaSaveGateWriteGuard<'_>,
         NamespaceLockGuard,
         NamespaceLockGuard,
         PoolMeta,
@@ -538,7 +538,7 @@ impl ECStore {
             .first()
             .cloned()
             .ok_or_else(|| Error::other("heal format requires at least one storage pool"))?;
-        let mut write_state = self.pool_meta_save_gate.lock().await;
+        let mut write_state = self.pool_meta_save_gate.write().await;
         write_state.ensure_write_safe("heal format fence failed")?;
 
         // Metadata fence order is part of the decommission/rebalance protocol:
@@ -1079,7 +1079,7 @@ impl ECStore {
     pub(super) async fn handle_heal_bucket(&self, bucket: &str, opts: &HealOpts) -> Result<HealResultItem> {
         let movement_gate = self.ctx.data_movement_operation_gate();
         let _movement_guard = movement_gate.read().await;
-        let save_guard = self.pool_meta_save_gate.lock().await;
+        let save_guard = self.pool_meta_save_gate.write().await;
         save_guard.ensure_write_safe("bucket heal cannot run while pool metadata requires recovery")?;
         let mut fenced_pools = BTreeSet::new();
         {
@@ -2527,7 +2527,7 @@ mod tests {
             rebalance_meta: RwLock::new(None),
             decommission_cancelers: RwLock::new(Vec::new()),
             start_gate: Mutex::new(()),
-            pool_meta_save_gate: Mutex::default(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
             ctx: crate::runtime::instance::bootstrap_ctx(),
             bucket_fence_registry: std::sync::Arc::default(),
         }
@@ -3550,10 +3550,14 @@ mod tests {
     async fn full_format_heal_preblocked_pool_metadata_never_writes_format() {
         let (temp_dir, store, shutdown) = multi_pool_heal_store().await;
         remove_heal_test_format(&temp_dir, &store, 0, 3).await;
-        store.pool_meta_save_gate.lock().await.observe_replicas(PoolMetaReplicaState {
-            needs_repair: true,
-            repair_write_safe: false,
-        });
+        store
+            .pool_meta_save_gate
+            .write()
+            .await
+            .observe_replicas(PoolMetaReplicaState {
+                needs_repair: true,
+                repair_write_safe: false,
+            });
 
         let err = store
             .handle_heal_format(false)
@@ -3688,7 +3692,7 @@ mod tests {
         let target = remove_heal_test_format(&temp_dir, &store, 0, 3).await;
         store.pools[0].disk_set[0].disks.write().await[3] = None;
         let barrier = HealFormatAfterSaveBarrier::install(&store.pools[0], 3);
-        let recovery_latch = store.pool_meta_save_gate.lock().await.aborted_transaction_latch_for_test();
+        let recovery_latch = store.pool_meta_save_gate.write().await.aborted_transaction_latch_for_test();
         let mut heal = tokio::spawn({
             let store = Arc::clone(&store);
             async move { store.handle_heal_format(false).await }
@@ -3848,7 +3852,7 @@ mod tests {
             }
         }
         let endpoints: EndpointServerPools = pools.iter().map(|pool| pool.endpoints.clone()).collect::<Vec<_>>().into();
-        let state = source.pool_meta_save_gate.lock().await.independent_clone_for_test();
+        let state = source.pool_meta_save_gate.write().await.independent_clone_for_test();
         Arc::new(ECStore {
             id: Uuid::new_v4(),
             disk_map: Default::default(),
@@ -3858,7 +3862,7 @@ mod tests {
             rebalance_meta: tokio::sync::RwLock::new(None),
             decommission_cancelers: tokio::sync::RwLock::new(vec![None; 2]),
             start_gate: tokio::sync::Mutex::new(()),
-            pool_meta_save_gate: tokio::sync::Mutex::new(state),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::new(state),
             ctx: Arc::clone(&source.ctx),
             bucket_fence_registry: Arc::default(),
         })
@@ -4040,7 +4044,7 @@ mod tests {
         let missing = remove_pool_meta_shard(&store, 0).await;
         let (guard, admissions, mut state) = store.acquire_pool_meta_object_heal_fence(&[0]).await.expect("heal authority");
         assert!(admissions.into_iter().all(|admission| admission.is_ok()));
-        assert!(store.pool_meta_save_gate.try_lock().is_err(), "heal must retain the earlier save gate");
+        assert!(store.pool_meta_save_gate.try_write().is_err(), "heal must retain the earlier save gate");
         let writer_store = Arc::clone(&store);
         let writer = tokio::spawn(async move { writer_store.save_current_pool_meta_for_test(&[0]).await });
         tokio::task::yield_now().await;
@@ -4687,10 +4691,14 @@ mod tests {
     #[tokio::test]
     async fn bucket_heal_blocks_before_dispatch_after_unreadable_pool_meta_replica() {
         let store = minimal_heal_store().await;
-        store.pool_meta_save_gate.lock().await.observe_replicas(PoolMetaReplicaState {
-            needs_repair: true,
-            repair_write_safe: false,
-        });
+        store
+            .pool_meta_save_gate
+            .write()
+            .await
+            .observe_replicas(PoolMetaReplicaState {
+                needs_repair: true,
+                repair_write_safe: false,
+            });
 
         let err = store
             .handle_heal_bucket("bucket", &HealOpts::default())
@@ -4724,7 +4732,7 @@ mod tests {
             .await
             .expect("peer dispatch should start");
         assert!(
-            store.pool_meta_save_gate.try_lock().is_ok(),
+            store.pool_meta_save_gate.try_write().is_ok(),
             "coordinator must release its local save gate before waiting for peers"
         );
         assert!(
@@ -5116,7 +5124,7 @@ mod tests {
             rebalance_meta: RwLock::new(None),
             decommission_cancelers: RwLock::new(Vec::new()),
             start_gate: Mutex::new(()),
-            pool_meta_save_gate: Mutex::default(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
             ctx: crate::runtime::instance::bootstrap_ctx(),
             bucket_fence_registry: std::sync::Arc::default(),
         };
@@ -5136,7 +5144,7 @@ mod tests {
             .save_for_startup(store.pools.clone())
             .await
             .expect("pool metadata should be persisted before format heal");
-        *store.pool_meta_save_gate.lock().await = PoolMetaWriteState::default();
+        *store.pool_meta_save_gate.write().await = crate::core::pools::PoolMetaWriteState::default();
 
         let replacement_endpoint = store.pools[1].endpoints.endpoints.as_ref()[2].clone();
         tokio::fs::remove_file(

crates/ecstore/src/store/list_objects.rs

@@ -127,28 +127,30 @@ impl AuthoritativeListOperations for ECStore {
         marker: Option<String>,
         max_keys: i32,
     ) -> Result<ListObjectVersionsInfo> {
-        if max_keys <= 0 || max_keys > MAX_OBJECT_LIST {
+        self.list_unversioned_authoritative_page(bucket, prefix, marker, max_keys, None)
+            .await
+    }
+
+    async fn list_identifier_tree_leaves_authoritative(
+        self: Arc<Self>,
+        bucket: &str,
+        prefix: &str,
+        leaf: &str,
+        marker: Option<String>,
+        max_keys: i32,
+    ) -> Result<ListObjectVersionsInfo> {
+        let valid_prefix = prefix
+            .strip_suffix('/')
+            .is_some_and(|root| !root.is_empty() && root.split('/').all(|part| !part.is_empty() && !matches!(part, "." | "..")));
+        if !valid_prefix || !crate::disk::is_identifier_tree_leaf(leaf) {
             return Err(StorageError::InvalidArgument(
                 bucket.to_string(),
                 prefix.to_string(),
-                "authoritative page size is out of range".to_string(),
+                "identifier-tree listing requires a directory prefix and a dotted leaf name".to_string(),
             ));
         }
-        let opts = ListPathOptions {
-            bucket: bucket.to_string(),
-            prefix: prefix.to_string(),
-            marker: marker.clone(),
-            limit: max_keys,
-            recursive: true,
-            incl_deleted: true,
-            versioned: true,
-            authoritative: true,
-            caller: ListObjectsGatherCaller::S3ListVersions,
-            ..Default::default()
-        };
-        let mut page = self.list_path(&opts).await?;
-        page.entries.forward_past(marker);
-        unversioned_authoritative_page(bucket, page)
+        self.list_unversioned_authoritative_page(bucket, prefix, marker, max_keys, Some(leaf.to_string()))
+            .await
     }
 
     async fn walk_authoritative(
@@ -368,6 +370,9 @@ fn log_list_path_finished(
 #[derive(Debug, Default, Clone)]
 pub struct ListPathOptions {
     pub caller: ListObjectsGatherCaller,
+    /// Internal native-walk admission, shared with every disk producer so
+    /// aborting an HTTP request cannot release its budget ahead of cleanup.
+    pub read_budget: Option<Arc<tokio::sync::OwnedSemaphorePermit>>,
     pub id: Option<String>,
 
     // Bucket of the listing.
@@ -433,6 +438,8 @@ pub struct ListPathOptions {
     // Internal bounded pages require GET-equivalent per-set resolution.
     pub authoritative: bool,
 
+    pub identifier_tree_leaf: Option<String>,
+
     pub stop_disk_at_limit: bool,
 
     pub pool_idx: Option<usize>,
@@ -1098,6 +1105,39 @@ impl ListPathOptions {
 }
 
 impl ECStore {
+    async fn list_unversioned_authoritative_page(
+        self: Arc<Self>,
+        bucket: &str,
+        prefix: &str,
+        marker: Option<String>,
+        max_keys: i32,
+        identifier_tree_leaf: Option<String>,
+    ) -> Result<ListObjectVersionsInfo> {
+        if max_keys <= 0 || max_keys > MAX_OBJECT_LIST {
+            return Err(StorageError::InvalidArgument(
+                bucket.to_string(),
+                prefix.to_string(),
+                "authoritative page size is out of range".to_string(),
+            ));
+        }
+        let opts = ListPathOptions {
+            bucket: bucket.to_string(),
+            prefix: prefix.to_string(),
+            marker: marker.clone(),
+            limit: max_keys,
+            recursive: true,
+            incl_deleted: true,
+            versioned: true,
+            authoritative: true,
+            identifier_tree_leaf,
+            caller: ListObjectsGatherCaller::S3ListVersions,
+            ..Default::default()
+        };
+        let mut page = self.list_path(&opts).await?;
+        page.entries.forward_past(marker);
+        unversioned_authoritative_page(bucket, page)
+    }
+
     #[allow(clippy::too_many_arguments)]
     // @continuation_token marker
     // @start_after as marker when continuation_token empty
@@ -1478,6 +1518,11 @@ impl ECStore {
         if o.transient {
             o.create = false;
         }
+        let fanout = self.pools.iter().fold(0usize, |total, pool| {
+            total.saturating_add(pool.disk_set.len().saturating_mul(pool.set_drive_count))
+        });
+        o.read_budget = self.ctx.acquire_list_read_budget(fanout).await.map_err(Error::other)?;
+
         let log_context = ListPathLogContext::from_options(&o);
 
         // cancel channel
@@ -2878,6 +2923,12 @@ impl SetDisks {
         if o.transient {
             o.create = false;
         }
+        o.read_budget = self
+            .instance_ctx()
+            .acquire_list_read_budget(self.set_drive_count)
+            .await
+            .map_err(Error::other)?;
+
         let log_context = ListPathLogContext::from_options(&o);
 
         let cancel = CancellationToken::new();
@@ -2922,6 +2973,33 @@ impl SetDisks {
         Ok(page)
     }
 
+    fn resolve_authoritative_listed_entry(
+        bucket: &str,
+        selection: Option<(&str, &str)>,
+        entries: MetaCacheEntries,
+        errors: &[Option<DiskError>],
+        disk_count: usize,
+        parity: usize,
+    ) -> crate::disk::error::Result<Option<MetaCacheEntry>> {
+        if disk_count == 0 || entries.0.len() > disk_count || entries.0.len() != errors.len() {
+            return Err(DiskError::ErasureReadQuorum);
+        }
+        // An old peer may ignore the optional disk selector. Its noncandidate
+        // rows must be ignored before metadata parsing. Selected keys retain
+        // the original resolver and all of its quorum/corruption checks.
+        if selection.is_some_and(|(prefix, leaf)| {
+            entries
+                .0
+                .iter()
+                .flatten()
+                .next()
+                .is_some_and(|entry| !crate::disk::identifier_tree_leaf_matches(prefix, leaf, &entry.name))
+        }) {
+            return Ok(None);
+        }
+        Self::resolve_listed_versions(bucket, entries, errors, disk_count, parity)
+    }
+
     /// Version scan for set-scoped callers that act on absence: set walks
     /// (tier reference proofs, tier-delete journals, pool maintenance) and the
     /// recursive prefix-delete scan. Every online drive is read and each
@@ -2945,10 +3023,13 @@ impl SetDisks {
         let first_error = Arc::new(tokio::sync::Mutex::new(None));
         let error_sink = Arc::clone(&first_error);
         let bucket = opts.bucket.clone();
+        let prefix = opts.prefix.clone();
+        let identifier_tree_leaf = opts.identifier_tree_leaf.clone();
         let cancel = rx.clone();
         let result = list_path_raw_with_claim_tracker(
             rx,
             ListPathRawOptions {
+                read_budget: opts.read_budget.clone(),
                 disks: disks.into_iter().map(Some).collect(),
                 bucket: opts.bucket,
                 path: opts.base_dir,
@@ -2961,6 +3042,7 @@ impl SetDisks {
                 forward_to: opts.marker,
                 min_disks: read_quorum,
                 preserve_replica_metadata: true,
+                identifier_tree_leaf: opts.identifier_tree_leaf,
                 // A reader's local page may be consumed by stale entries. Only
                 // the resolved, merged page may stop the authoritative walk.
                 per_disk_limit: 0,
@@ -2968,7 +3050,14 @@ impl SetDisks {
                 walkdir_timeout: opts.walkdir_timeout,
                 walkdir_stall_timeout: opts.walkdir_stall_timeout,
                 partial: Some(Box::new(move |entries, errors| {
-                    let resolved = Self::resolve_listed_versions(&bucket, entries, errors, disk_count, parity);
+                    let resolved = Self::resolve_authoritative_listed_entry(
+                        &bucket,
+                        identifier_tree_leaf.as_deref().map(|leaf| (prefix.as_str(), leaf)),
+                        entries,
+                        errors,
+                        disk_count,
+                        parity,
+                    );
                     let sender = sender.clone();
                     let cancel = cancel.clone();
                     let first_error = Arc::clone(&error_sink);
@@ -3107,6 +3196,7 @@ impl SetDisks {
         let result = list_path_raw_with_outcome(
             rx,
             ListPathRawOptions {
+                read_budget: opts.read_budget.clone(),
                 disks: disks.into_iter().map(Some).collect(),
                 fallback_disks: fallback_disks.into_iter().map(Some).collect(),
                 bucket: opts.bucket,
@@ -6788,6 +6878,58 @@ mod test {
         (dirs, store)
     }
 
+    #[tokio::test]
+    async fn list_read_budget_bounds_store_and_set_entrypoints_and_recovers_after_errors() {
+        let (_dirs, store) = single_set_store(4).await;
+        let held = store
+            .ctx
+            .acquire_list_read_budget(usize::MAX)
+            .await
+            .expect("occupy native budget");
+        let semaphore = held.as_ref().expect("bounded instance budget").semaphore();
+        let owners = Arc::strong_count(semaphore);
+        let opts = ListPathOptions {
+            bucket: "list-budget-absent-bucket".into(),
+            limit: 1,
+            authoritative: true,
+            ..Default::default()
+        };
+        let set = store.pools[0].disk_set[0].clone();
+        set.disks.write().await.fill(None);
+        {
+            let mut store_walk = std::pin::pin!(store.clone().list_path(&opts));
+            let mut set_walk = std::pin::pin!(set.clone().list_path_result(&opts));
+            assert!(futures::poll!(&mut store_walk).is_pending(), "store walk must await its budget");
+            assert_eq!(Arc::strong_count(semaphore), owners + 1, "store must queue on the shared budget");
+            assert!(futures::poll!(&mut set_walk).is_pending(), "set walk shares the same instance budget");
+            assert_eq!(Arc::strong_count(semaphore), owners + 2, "set must queue on the same budget");
+            let empty = store
+                .clone()
+                .list_path(&ListPathOptions {
+                    limit: 0,
+                    ..opts.clone()
+                })
+                .await
+                .expect("empty page needs no disk admission");
+            assert!(empty.exhausted);
+        }
+        assert_eq!(Arc::strong_count(semaphore), owners, "cancelled entry points release the queue ownership");
+        drop(held);
+        let result = timeout(Duration::from_secs(5), store.clone().list_path(&opts))
+            .await
+            .expect("failed walk terminates");
+        assert_eq!(
+            result.err(),
+            Some(DiskError::ErasureReadQuorum.into()),
+            "offline drives retain their quorum error"
+        );
+        let next = timeout(Duration::from_secs(5), store.ctx.acquire_list_read_budget(usize::MAX))
+            .await
+            .expect("failed and cancelled walks release their grants")
+            .expect("budget remains open");
+        assert!(next.is_some());
+    }
+
     fn versioned_object_metadata(bucket: &str, name: &str, versions: &[(u128, i64)], slot: usize) -> Bytes {
         let mut meta = FileMeta::new();
         for (id, modified) in versions {
@@ -6929,6 +7071,155 @@ mod test {
         );
     }
 
+    #[test]
+    fn authoritative_identifier_tree_filters_old_peer_rows_before_metadata_resolution() {
+        let prefix = "table-catalog/namespaces/";
+        let leaf = "table-entry.json";
+        let corrupt = |name: &str| {
+            MetaCacheEntries(vec![
+                Some(MetaCacheEntry {
+                    name: format!("{prefix}{name}"),
+                    metadata: b"corrupt metadata".to_vec(),
+                    ..Default::default()
+                });
+                4
+            ])
+        };
+        let report = "sales/tables/orders/maintenance/hash/jobs/job.json";
+        let errors = vec![None; 4];
+        assert!(
+            SetDisks::resolve_authoritative_listed_entry(RUSTFS_META_BUCKET, None, corrupt(report), &errors, 4, 2,).is_err(),
+            "the control must actually parse and reject the corrupt metadata"
+        );
+        assert!(
+            SetDisks::resolve_authoritative_listed_entry(
+                RUSTFS_META_BUCKET,
+                Some((prefix, leaf)),
+                corrupt(report),
+                &errors,
+                4,
+                2,
+            )
+            .expect("an old peer's unrelated report is outside the selected scan")
+            .is_none()
+        );
+        assert!(
+            SetDisks::resolve_authoritative_listed_entry(
+                RUSTFS_META_BUCKET,
+                Some((prefix, leaf)),
+                corrupt("sales/tables/orders/maintenance/tables/owner/table-entry.json"),
+                &errors,
+                4,
+                2,
+            )
+            .is_err(),
+            "a matching nested owner must still fail closed on corruption"
+        );
+    }
+
+    #[tokio::test]
+    async fn authoritative_identifier_tree_pages_preserve_nested_owners_and_read_quorum() {
+        let (dirs, store) = single_set_store(4).await;
+        let prefix = "table-catalog/identifier-tree/";
+        let leaf = "table-entry.json";
+        for (root, selected) in [
+            ("table-catalog/identifier-tree", leaf),
+            (prefix, "../entry.json"),
+            (prefix, "table-entry"),
+        ] {
+            assert!(matches!(
+                store
+                    .clone()
+                    .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, root, selected, None, 2)
+                    .await,
+                Err(StorageError::InvalidArgument(..))
+            ));
+        }
+        let names = [
+            format!("{prefix}a/tables/first/{leaf}"),
+            format!("{prefix}sales/tables/orders/maintenance/tables/owner/{leaf}"),
+            format!("{prefix}z/tables/last/{leaf}"),
+        ];
+        for name in &names {
+            write_version_replicas(&dirs, RUSTFS_META_BUCKET, name, |slot| Some(unversioned_catalog_metadata(name, slot))).await;
+        }
+        for name in [
+            format!("{prefix}sales"),
+            format!("{prefix}sales/tables/orders/maintenance/hash/jobs/job.json"),
+        ] {
+            write_version_replicas(&dirs, RUSTFS_META_BUCKET, &name, |_| Some(b"corrupt noncandidate metadata".to_vec())).await;
+        }
+        assert!(
+            store
+                .clone()
+                .list_unversioned_objects_authoritative(RUSTFS_META_BUCKET, prefix, None, 100)
+                .await
+                .is_err(),
+            "the ordinary authoritative scan must expose the corrupt control objects"
+        );
+        let first = store
+            .clone()
+            .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, prefix, leaf, None, 2)
+            .await
+            .expect("noncandidate metadata must not be read");
+        assert!(first.is_truncated);
+        assert_eq!(
+            first.objects.iter().map(|object| &object.name).collect::<Vec<_>>(),
+            names[..2].iter().collect::<Vec<_>>()
+        );
+        let second = store
+            .clone()
+            .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, prefix, leaf, first.next_marker, 2)
+            .await
+            .expect("the next page must retain the final owner");
+        assert!(!second.is_truncated);
+        assert_eq!(second.objects.iter().map(|object| &object.name).collect::<Vec<_>>(), vec![&names[2]]);
+
+        write_version_replicas(&dirs, RUSTFS_META_BUCKET, &names[0], |_| Some(b"corrupt selected metadata".to_vec())).await;
+        assert!(
+            store
+                .clone()
+                .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, prefix, leaf, None, 2)
+                .await
+                .is_err(),
+            "corruption at a selected key cannot be skipped to fill the page"
+        );
+        write_version_replicas(&dirs, RUSTFS_META_BUCKET, &names[0], |slot| {
+            Some(unversioned_catalog_metadata(&names[0], slot))
+        })
+        .await;
+        write_version_replicas(&dirs, RUSTFS_META_BUCKET, &names[0], |slot| {
+            Some(versioned_object_metadata(RUSTFS_META_BUCKET, &names[0], &[(2, 1_700_000_002)], slot).to_vec())
+        })
+        .await;
+        assert!(
+            store
+                .clone()
+                .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, prefix, leaf, None, 2)
+                .await
+                .is_err(),
+            "a versioned ownership entry cannot be accepted as an unversioned proof"
+        );
+        write_version_replicas(&dirs, RUSTFS_META_BUCKET, &names[0], |slot| {
+            Some(unversioned_catalog_metadata(&names[0], slot))
+        })
+        .await;
+        let set = &store.pools[0].disk_set[0];
+        for disk in set.disks.write().await.iter_mut().skip(1) {
+            *disk = None;
+        }
+        assert!(
+            matches!(
+                store
+                    .clone()
+                    .list_identifier_tree_leaves_authoritative(RUSTFS_META_BUCKET, prefix, leaf, None, 2)
+                    .await,
+                Err(StorageError::ErasureReadQuorum)
+            ),
+            "unavailable replicas cannot prove owner absence"
+        );
+    }
+
     #[tokio::test]
     async fn authoritative_store_walk_includes_reserved_prefix_versions() {
         let bucket = "catalog-cleanup-versions";

crates/ecstore/src/store/mod.rs

@@ -34,7 +34,7 @@ use crate::bucket::utils::check_put_object_part_args;
 use crate::bucket::utils::{check_valid_bucket_name, check_valid_bucket_name_strict, is_meta_bucketname};
 use crate::cluster::rpc::{RemoteClient, S3PeerSys};
 use crate::config::storageclass;
-use crate::core::pools::{DecommissionCanceler, PoolMeta, PoolMetaWriteState};
+use crate::core::pools::{DecommissionCanceler, PoolMeta, PoolMetaSaveGate};
 use crate::disk::endpoint::{Endpoint, EndpointType};
 use crate::disk::{DiskAPI, DiskInfo, DiskInfoOptions};
 use crate::error::{Error, Result};
@@ -67,11 +67,13 @@ use lazy_static::lazy_static;
 use rand::RngExt as _;
 use rustfs_config::server_config::Config;
 use rustfs_filemeta::{FileInfo, FileMeta};
+use rustfs_gateway_types::persistence::{
+    PersistedObjectLockConfiguration as ObjectLockConfiguration, PersistedVersioningConfiguration as VersioningConfiguration,
+};
 use rustfs_heal_contracts::heal_channel::{HealItemType, HealOpts};
 use rustfs_lock::{LocalClient, LockClient, NamespaceLockWrapper};
 use rustfs_madmin::heal_commands::HealResultItem;
 use rustfs_utils::path::{decode_dir_object, encode_dir_object, path_join_buf};
-use s3s::dto::{BucketVersioningStatus, ObjectLockConfiguration, ObjectLockEnabled, VersioningConfiguration};
 use std::net::SocketAddr;
 use std::process::exit;
 use std::{collections::HashMap, sync::Arc, time::Duration};
@@ -480,9 +482,13 @@ pub struct ECStore {
     /// Serializes full-document pool metadata saves and retains a fail-closed
     /// write block after startup observes an unreadable replica.
     ///
+    /// Immutable admission proofs share authority; publication, recovery and
+    /// authority changes retain exclusive access through their existing I/O.
+    /// Each writer publishes its final state on release, and health
+    /// inspection reads that committed state without acquiring the gate.
     /// Lock order: acquire `pool_meta_save_gate`, then the distributed
     /// `pool.bin` fence, then clone `pool_meta` under a short read lock.
-    pub(crate) pool_meta_save_gate: Mutex<PoolMetaWriteState>,
+    pub(crate) pool_meta_save_gate: PoolMetaSaveGate,
     /// Per-instance runtime state (Phase 5, backlog#939).
     ///
     /// Carries this instance's identity/runtime out of the process globals so
@@ -555,7 +561,6 @@ impl Default for ScannerDataMovementPauseStatus {
 #[derive(Clone, Debug, PartialEq, Eq, serde::Serialize)]
 pub struct PoolMetaWriteGateStatus {
     pub writes_ready: bool,
-    pub check_timed_out: bool,
     pub write_blocked: bool,
     pub transaction_aborted: bool,
     pub pool_meta_absent: bool,
@@ -571,7 +576,6 @@ impl Default for PoolMetaWriteGateStatus {
     fn default() -> Self {
         Self {
             writes_ready: true,
-            check_timed_out: false,
             write_blocked: false,
             transaction_aborted: false,
             pool_meta_absent: false,
@@ -838,7 +842,7 @@ impl ECStore {
     }
 
     /// Get this store instance's configured region.
-    pub fn region(&self) -> Option<s3s::region::Region> {
+    pub fn region(&self) -> Option<rustfs_s3_types::Region> {
         self.ctx.region()
     }
 
@@ -1455,11 +1459,11 @@ impl ECStore {
 
 lazy_static! {
     static ref ENABLED_OBJECT_LOCK_CONFIG: ObjectLockConfiguration = ObjectLockConfiguration {
-        object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+        object_lock_enabled: Some("Enabled".to_owned()),
         ..Default::default()
     };
     static ref ENABLED_VERSIONING_CONFIG: VersioningConfiguration = VersioningConfiguration {
-        status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+        status: Some("Enabled".to_owned()),
         ..Default::default()
     };
 }
@@ -1936,7 +1940,7 @@ mod tests {
             rebalance_meta: RwLock::new(None),
             decommission_cancelers: RwLock::new(Vec::new()),
             start_gate: Mutex::new(()),
-            pool_meta_save_gate: Mutex::default(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
             ctx,
             bucket_fence_registry: Arc::default(),
         };
@@ -2011,7 +2015,7 @@ mod tests {
             rebalance_meta: RwLock::new(None),
             decommission_cancelers: RwLock::new(Vec::new()),
             start_gate: Mutex::new(()),
-            pool_meta_save_gate: Mutex::default(),
+            pool_meta_save_gate: crate::core::pools::PoolMetaSaveGate::default(),
             ctx,
             bucket_fence_registry: Arc::default(),
         })