feat(kms): observe unknown fields in the last three silent persisted formats (#6003)

This commit is contained in:
Zhengchao An
2026-08-13 03:08:49 +08:00
committed by GitHub
parent 73bd5d9d95
commit 3f9b84ec70
7 changed files with 542 additions and 3 deletions
+4
View File
@@ -35,6 +35,10 @@ tokio = { workspace = true, features = ["fs", "io-util", "macros", "rt-multi-thr
uuid = { workspace = true, features = ["serde", "v4", "fast-rng", "macro-diagnostics"] }
jiff = { workspace = true, features = ["serde"] }
serde = { workspace = true, features = ["derive"] }
# Observes fields a persisted-format deserialization ignored, per the
# repository rule that formats too compatibility-bound for
# deny_unknown_fields must at least warn (AGENTS.md).
serde_ignored = { workspace = true }
serde_json = { workspace = true, features = ["raw_value"] }
tracing = { workspace = true }
thiserror = { workspace = true }
+220 -1
View File
@@ -26,6 +26,7 @@ use crate::backends::{
use crate::config::{KmsConfig, VaultConfig};
use crate::encryption::{AesDekCrypto, DataKeyEnvelope, DekCrypto, generate_key_material};
use crate::error::{KmsError, Result};
use crate::persisted_observability::{BoundedUnknownFieldName, UnknownFieldSummary};
use crate::policy::{self, AttemptError, OpClass, RetryPolicy};
use crate::types::*;
use async_trait::async_trait;
@@ -60,7 +61,12 @@ pub struct VaultKmsClient {
}
/// Key data stored in Vault
#[derive(Debug, Clone, Serialize, Deserialize)]
///
/// `Deserialize` is hand-written so fields the current build does not know
/// are counted and warned about instead of vanishing silently — this record
/// is compatibility-bound in both directions (older and newer builds read
/// each other's writes), so `deny_unknown_fields` is not an option.
#[derive(Debug, Clone, Serialize)]
struct VaultKeyData {
/// Key algorithm
algorithm: String,
@@ -108,6 +114,187 @@ struct VaultKeyData {
baseline_version: Option<u32>,
}
impl UnknownFieldSummary {
fn record_for_vault_kv2_key(&self) {
let Some((field, field_name_truncated, field_count)) = self.record("vault-kv2-key") else {
return;
};
static RECORDS_WITH_UNKNOWN_FIELDS: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
let observed_records = RECORDS_WITH_UNKNOWN_FIELDS
.fetch_add(1, std::sync::atomic::Ordering::Relaxed)
.saturating_add(1);
if observed_records.is_power_of_two() {
tracing::warn!(
field = ?field,
field_name_truncated,
field_count,
observed_records,
"Vault KV2 key record contains unknown fields"
);
}
}
}
impl<'de> Deserialize<'de> for VaultKeyData {
fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
use serde::de::{self, IgnoredAny, MapAccess, Visitor};
use std::fmt;
enum Field {
Algorithm,
Usage,
CreatedAt,
Status,
Version,
Description,
Metadata,
Tags,
DeletionDate,
RotatedAt,
EncryptedKeyMaterial,
BaselineVersion,
Unknown(BoundedUnknownFieldName),
}
impl<'de> Deserialize<'de> for Field {
fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
struct FieldVisitor;
impl Visitor<'_> for FieldVisitor {
type Value = Field;
fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
formatter.write_str("a Vault KV2 key record field name")
}
fn visit_str<E>(self, value: &str) -> std::result::Result<Self::Value, E>
where
E: de::Error,
{
Ok(match value {
"algorithm" => Field::Algorithm,
"usage" => Field::Usage,
"created_at" => Field::CreatedAt,
"status" => Field::Status,
"version" => Field::Version,
"description" => Field::Description,
"metadata" => Field::Metadata,
"tags" => Field::Tags,
"deletion_date" => Field::DeletionDate,
"rotated_at" => Field::RotatedAt,
"encrypted_key_material" => Field::EncryptedKeyMaterial,
"baseline_version" => Field::BaselineVersion,
_ => Field::Unknown(BoundedUnknownFieldName::new(value)),
})
}
}
deserializer.deserialize_identifier(FieldVisitor)
}
}
struct VaultKeyDataVisitor;
impl<'de> Visitor<'de> for VaultKeyDataVisitor {
type Value = VaultKeyData;
fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
formatter.write_str("a Vault KV2 key record")
}
fn visit_map<A>(self, mut map: A) -> std::result::Result<Self::Value, A::Error>
where
A: MapAccess<'de>,
{
macro_rules! read_field {
($slot:ident, $name:literal) => {{
if $slot.is_some() {
return Err(de::Error::duplicate_field($name));
}
$slot = Some(map.next_value()?);
}};
}
let mut algorithm = None;
let mut usage = None;
let mut created_at = None;
let mut status = None;
let mut version = None;
let mut description = None;
let mut metadata = None;
let mut tags = None;
let mut deletion_date = None;
let mut rotated_at = None;
let mut encrypted_key_material = None;
let mut baseline_version = None;
let mut unknown_fields = UnknownFieldSummary::default();
while let Some(field) = map.next_key()? {
match field {
Field::Algorithm => read_field!(algorithm, "algorithm"),
Field::Usage => read_field!(usage, "usage"),
Field::CreatedAt => read_field!(created_at, "created_at"),
Field::Status => read_field!(status, "status"),
Field::Version => read_field!(version, "version"),
Field::Description => read_field!(description, "description"),
Field::Metadata => read_field!(metadata, "metadata"),
Field::Tags => read_field!(tags, "tags"),
Field::DeletionDate => read_field!(deletion_date, "deletion_date"),
Field::RotatedAt => read_field!(rotated_at, "rotated_at"),
Field::EncryptedKeyMaterial => read_field!(encrypted_key_material, "encrypted_key_material"),
Field::BaselineVersion => read_field!(baseline_version, "baseline_version"),
Field::Unknown(field) => {
let _: IgnoredAny = map.next_value()?;
unknown_fields.observe(field);
}
}
}
let key_data = VaultKeyData {
algorithm: algorithm.ok_or_else(|| de::Error::missing_field("algorithm"))?,
usage: usage.ok_or_else(|| de::Error::missing_field("usage"))?,
created_at: created_at.ok_or_else(|| de::Error::missing_field("created_at"))?,
status: status.ok_or_else(|| de::Error::missing_field("status"))?,
version: version.ok_or_else(|| de::Error::missing_field("version"))?,
description: description.unwrap_or(None),
metadata: metadata.ok_or_else(|| de::Error::missing_field("metadata"))?,
tags: tags.ok_or_else(|| de::Error::missing_field("tags"))?,
deletion_date: deletion_date.unwrap_or(None),
rotated_at: rotated_at.unwrap_or(None),
encrypted_key_material: encrypted_key_material
.ok_or_else(|| de::Error::missing_field("encrypted_key_material"))?,
baseline_version: baseline_version.unwrap_or(None),
};
unknown_fields.record_for_vault_kv2_key();
Ok(key_data)
}
}
const FIELDS: &[&str] = &[
"algorithm",
"usage",
"created_at",
"status",
"version",
"description",
"metadata",
"tags",
"deletion_date",
"rotated_at",
"encrypted_key_material",
"baseline_version",
];
deserializer.deserialize_struct("VaultKeyData", FIELDS, VaultKeyDataVisitor)
}
}
/// Immutable per-version master key material record stored under
/// `{prefix}/{key_id}/versions/{N}`.
///
@@ -2465,6 +2652,38 @@ mod tests {
assert_eq!(legacy.version, 1);
}
#[test]
fn vault_key_data_unknown_fields_remain_readable_and_are_observed() {
// A record written by a newer build carries fields this build does not
// know. It must stay readable — and the drop must be visible, not
// silent (rustfs/backlog#1641). Only the field name may be logged; the
// value can sit next to key material.
let mut value = serde_json::to_value(healthy_key_data()).expect("serialize key data");
let object = value.as_object_mut().expect("key data serializes to an object");
object.insert("field_from_the_future".to_string(), serde_json::json!("field value must not be logged"));
let logs = crate::test_support::CapturedLogs::default();
let subscriber = tracing_subscriber::fmt()
.with_ansi(false)
.with_max_level(tracing::Level::WARN)
.with_writer(logs.clone())
.finish();
let dispatch = tracing::Dispatch::new(subscriber);
let recorder = metrics_util::debugging::DebuggingRecorder::new();
let parsed: VaultKeyData = metrics::with_local_recorder(&recorder, || {
tracing::dispatcher::with_default(&dispatch, || {
serde_json::from_value(value).expect("unknown fields must remain readable")
})
});
assert_eq!(parsed.algorithm, healthy_key_data().algorithm);
assert_eq!(crate::test_support::unknown_field_metric(&recorder, "vault-kv2-key"), 1);
let output = logs.output();
assert!(output.contains("Vault KV2 key record contains unknown fields"), "got: {output}");
assert!(output.contains("field_from_the_future"));
assert!(!output.contains("field value must not be logged"));
}
#[test]
fn test_is_cas_conflict_only_matches_cas_failures() {
let cas = ClientError::APIError {
+204 -1
View File
@@ -27,6 +27,7 @@ use crate::backends::{
use crate::config::{KmsConfig, VaultTransitConfig};
use crate::encryption::{DataKeyEnvelope, generate_key_material};
use crate::error::{KmsError, Result};
use crate::persisted_observability::{BoundedUnknownFieldName, UnknownFieldSummary};
use crate::policy::{self, AttemptError, OpClass, RetryPolicy};
use crate::types::*;
use async_trait::async_trait;
@@ -114,7 +115,12 @@ struct TransitKeyMetadata {
}
/// Serializable version of TransitKeyMetadata for KV v2 persistence.
#[derive(Debug, Clone, Serialize, Deserialize)]
///
/// `Deserialize` is hand-written so fields the current build does not know
/// are counted and warned about instead of vanishing silently — this record
/// is compatibility-bound in both directions (older and newer builds read
/// each other's writes), so `deny_unknown_fields` is not an option.
#[derive(Debug, Clone, Serialize)]
struct TransitKeyMetadataPersisted {
key_usage: KeyUsage,
description: Option<String>,
@@ -127,6 +133,168 @@ struct TransitKeyMetadataPersisted {
current_version: u32,
}
impl UnknownFieldSummary {
fn record_for_transit_key_metadata(&self) {
let Some((field, field_name_truncated, field_count)) = self.record("vault-transit-key-metadata") else {
return;
};
static RECORDS_WITH_UNKNOWN_FIELDS: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
let observed_records = RECORDS_WITH_UNKNOWN_FIELDS
.fetch_add(1, std::sync::atomic::Ordering::Relaxed)
.saturating_add(1);
if observed_records.is_power_of_two() {
tracing::warn!(
field = ?field,
field_name_truncated,
field_count,
observed_records,
"Vault Transit key metadata record contains unknown fields"
);
}
}
}
impl<'de> Deserialize<'de> for TransitKeyMetadataPersisted {
fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
use serde::de::{self, IgnoredAny, MapAccess, Visitor};
use std::fmt;
enum Field {
KeyUsage,
Description,
Tags,
KeyState,
CreatedAt,
DeletionDate,
Origin,
CreatedBy,
CurrentVersion,
Unknown(BoundedUnknownFieldName),
}
impl<'de> Deserialize<'de> for Field {
fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
struct FieldVisitor;
impl Visitor<'_> for FieldVisitor {
type Value = Field;
fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
formatter.write_str("a Vault Transit key metadata field name")
}
fn visit_str<E>(self, value: &str) -> std::result::Result<Self::Value, E>
where
E: de::Error,
{
Ok(match value {
"key_usage" => Field::KeyUsage,
"description" => Field::Description,
"tags" => Field::Tags,
"key_state" => Field::KeyState,
"created_at" => Field::CreatedAt,
"deletion_date" => Field::DeletionDate,
"origin" => Field::Origin,
"created_by" => Field::CreatedBy,
"current_version" => Field::CurrentVersion,
_ => Field::Unknown(BoundedUnknownFieldName::new(value)),
})
}
}
deserializer.deserialize_identifier(FieldVisitor)
}
}
struct TransitKeyMetadataPersistedVisitor;
impl<'de> Visitor<'de> for TransitKeyMetadataPersistedVisitor {
type Value = TransitKeyMetadataPersisted;
fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
formatter.write_str("a Vault Transit key metadata record")
}
fn visit_map<A>(self, mut map: A) -> std::result::Result<Self::Value, A::Error>
where
A: MapAccess<'de>,
{
macro_rules! read_field {
($slot:ident, $name:literal) => {{
if $slot.is_some() {
return Err(de::Error::duplicate_field($name));
}
$slot = Some(map.next_value()?);
}};
}
let mut key_usage = None;
let mut description = None;
let mut tags = None;
let mut key_state = None;
let mut created_at = None;
let mut deletion_date = None;
let mut origin = None;
let mut created_by = None;
let mut current_version = None;
let mut unknown_fields = UnknownFieldSummary::default();
while let Some(field) = map.next_key()? {
match field {
Field::KeyUsage => read_field!(key_usage, "key_usage"),
Field::Description => read_field!(description, "description"),
Field::Tags => read_field!(tags, "tags"),
Field::KeyState => read_field!(key_state, "key_state"),
Field::CreatedAt => read_field!(created_at, "created_at"),
Field::DeletionDate => read_field!(deletion_date, "deletion_date"),
Field::Origin => read_field!(origin, "origin"),
Field::CreatedBy => read_field!(created_by, "created_by"),
Field::CurrentVersion => read_field!(current_version, "current_version"),
Field::Unknown(field) => {
let _: IgnoredAny = map.next_value()?;
unknown_fields.observe(field);
}
}
}
let metadata = TransitKeyMetadataPersisted {
key_usage: key_usage.ok_or_else(|| de::Error::missing_field("key_usage"))?,
description: description.unwrap_or(None),
tags: tags.ok_or_else(|| de::Error::missing_field("tags"))?,
key_state: key_state.ok_or_else(|| de::Error::missing_field("key_state"))?,
created_at: created_at.ok_or_else(|| de::Error::missing_field("created_at"))?,
deletion_date: deletion_date.unwrap_or(None),
origin: origin.ok_or_else(|| de::Error::missing_field("origin"))?,
created_by: created_by.unwrap_or(None),
current_version: current_version.ok_or_else(|| de::Error::missing_field("current_version"))?,
};
unknown_fields.record_for_transit_key_metadata();
Ok(metadata)
}
}
const FIELDS: &[&str] = &[
"key_usage",
"description",
"tags",
"key_state",
"created_at",
"deletion_date",
"origin",
"created_by",
"current_version",
];
deserializer.deserialize_struct("TransitKeyMetadataPersisted", FIELDS, TransitKeyMetadataPersistedVisitor)
}
}
impl TransitKeyMetadata {
fn from_create_request(request: &CreateKeyRequest) -> Self {
Self {
@@ -2133,6 +2301,41 @@ mod tests {
assert!(metadata.deletion_date.is_none());
}
#[test]
fn transit_key_metadata_unknown_fields_remain_readable_and_are_observed() {
// A record written by a newer build carries fields this build does not
// know. It must stay readable — and the drop must be visible, not
// silent (rustfs/backlog#1641). Only the field name may be logged.
let persisted: TransitKeyMetadataPersisted = TransitKeyMetadata::synthesized().into();
let mut value = serde_json::to_value(&persisted).expect("serialize metadata record");
let object = value.as_object_mut().expect("metadata record serializes to an object");
object.insert("field_from_the_future".to_string(), serde_json::json!("field value must not be logged"));
let logs = crate::test_support::CapturedLogs::default();
let subscriber = tracing_subscriber::fmt()
.with_ansi(false)
.with_max_level(tracing::Level::WARN)
.with_writer(logs.clone())
.finish();
let dispatch = tracing::Dispatch::new(subscriber);
let recorder = metrics_util::debugging::DebuggingRecorder::new();
let parsed: TransitKeyMetadataPersisted = metrics::with_local_recorder(&recorder, || {
tracing::dispatcher::with_default(&dispatch, || {
serde_json::from_value(value).expect("unknown fields must remain readable")
})
});
assert_eq!(parsed.key_state, KeyState::Enabled);
assert_eq!(crate::test_support::unknown_field_metric(&recorder, "vault-transit-key-metadata"), 1);
let output = logs.output();
assert!(
output.contains("Vault Transit key metadata record contains unknown fields"),
"got: {output}"
);
assert!(output.contains("field_from_the_future"));
assert!(!output.contains("field value must not be logged"));
}
/// KV2 write acknowledgement (`SecretVersionMetadata`) for `kv2::set`.
fn kv2_write_ack() -> serde_json::Value {
serde_json::json!({
+99
View File
@@ -1129,6 +1129,53 @@ pub fn allow_immediate_deletion_from_env() -> bool {
get_env_bool(ENV_KMS_ALLOW_IMMEDIATE_DELETION, false)
}
impl crate::persisted_observability::UnknownFieldSummary {
fn record_for_kms_config(&self) {
let Some((field, field_name_truncated, field_count)) = self.record("kms-config") else {
return;
};
static RECORDS_WITH_UNKNOWN_FIELDS: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
let observed_records = RECORDS_WITH_UNKNOWN_FIELDS
.fetch_add(1, std::sync::atomic::Ordering::Relaxed)
.saturating_add(1);
if observed_records.is_power_of_two() {
tracing::warn!(
field = ?field,
field_name_truncated,
field_count,
observed_records,
"persisted KMS configuration contains unknown fields"
);
}
}
}
/// Deserialize a persisted KMS configuration, observing ignored fields.
///
/// The persisted configuration deliberately tolerates unknown fields — a
/// rolling upgrade writes fields the previous build does not know, and
/// rejecting them would turn every upgrade into a hard stop (see the
/// regression test pinning that tolerance). Tolerated must not mean
/// invisible: this loader wraps the deserializer with `serde_ignored`, so
/// every field the configuration silently dropped is counted and sampled
/// into a warning, per the repository rule that formats too
/// compatibility-bound for `deny_unknown_fields` must at least log unknown
/// fields. Only field paths are recorded, never values — a mistyped field
/// name can sit next to a secret.
pub fn kms_config_from_persisted_json(data: &[u8]) -> serde_json::Result<KmsConfig> {
use crate::persisted_observability::{BoundedUnknownFieldName, UnknownFieldSummary};
let mut deserializer = serde_json::Deserializer::from_slice(data);
let mut unknown_fields = UnknownFieldSummary::default();
let config: KmsConfig = serde_ignored::deserialize(&mut deserializer, |path| {
unknown_fields.observe(BoundedUnknownFieldName::new(&path.to_string()));
})?;
deserializer.end()?;
unknown_fields.record_for_kms_config();
Ok(config)
}
fn vault_tls_config(skip_tls_verify: bool) -> Option<TlsConfig> {
skip_tls_verify.then_some(TlsConfig {
ca_cert_path: None,
@@ -1979,6 +2026,58 @@ mod tests {
});
}
#[test]
fn persisted_config_unknown_fields_remain_readable_and_are_observed() {
// Unknown fields in a persisted config are deliberately tolerated (a
// rolling upgrade writes fields the previous build does not know), but
// tolerated must not mean invisible (rustfs/backlog#1641): the
// observing loader counts and warns, naming only the field path —
// never the value, which can sit next to a secret. Coverage includes a
// field nested inside the backend variant, which the externally tagged
// enum exposes to the observer.
let mut value = serde_json::to_value(KmsConfig::default()).expect("serialize config");
value.as_object_mut().expect("config serializes to an object").insert(
"top_level_field_from_the_future".to_string(),
serde_json::json!("top-level value must not be logged"),
);
value
.pointer_mut("/backend_config/Local")
.expect("default config has a Local backend section")
.as_object_mut()
.expect("Local backend section is an object")
.insert(
"nested_field_from_the_future".to_string(),
serde_json::json!("nested value must not be logged"),
);
let data = serde_json::to_vec(&value).expect("encode config");
let logs = crate::test_support::CapturedLogs::default();
let subscriber = tracing_subscriber::fmt()
.with_ansi(false)
.with_max_level(tracing::Level::WARN)
.with_writer(logs.clone())
.finish();
let dispatch = tracing::Dispatch::new(subscriber);
let recorder = metrics_util::debugging::DebuggingRecorder::new();
let config = metrics::with_local_recorder(&recorder, || {
tracing::dispatcher::with_default(&dispatch, || {
kms_config_from_persisted_json(&data).expect("unknown fields must remain readable")
})
});
assert!(matches!(config.backend_config, BackendConfig::Local(_)));
assert_eq!(crate::test_support::unknown_field_metric(&recorder, "kms-config"), 2);
let output = logs.output();
assert!(output.contains("persisted KMS configuration contains unknown fields"), "got: {output}");
assert!(!output.contains("must not be logged"));
// A clean config observes nothing and logs nothing.
let clean = serde_json::to_vec(&KmsConfig::default()).expect("encode clean config");
let recorder = metrics_util::debugging::DebuggingRecorder::new();
metrics::with_local_recorder(&recorder, || kms_config_from_persisted_json(&clean).expect("clean config must parse"));
assert_eq!(crate::test_support::unknown_field_metric(&recorder, "kms-config"), 0);
}
#[test]
fn test_validate_rejects_incomplete_approle() {
let mut config = KmsConfig::vault_approle(