[pbs-devel] [PATCH proxmox-backup v4 4/4] pbs-config: add TTL window to token secret cache

[Samuel Rufinatscha]

Verify_secret() currently calls refresh_cache_if_file_changed() on every
request, which performs a metadata() call on token.shadow each time.
Under load this adds unnecessary overhead, considering also the file
usually should rarely change.

This patch introduces a TTL boundary, controlled by
TOKEN_SECRET_CACHE_TTL_SECS. File metadata is only re-loaded once the
TTL has expired; documents TTL effects.

Signed-off-by: Samuel Rufinatscha <s.rufinatscha at proxmox.com>
---
Changes from v3 to 4:
* Adjusted commit message

Changes from v2 to v3:
* Refactored refresh_cache_if_file_changed TTL logic.
* Remove had_prior_state check (replaced by last_checked logic).
* Improve TTL bound checks.
* Reword documentation warning for clarity.

Changes from v1 to v2:
* Add TOKEN_SECRET_CACHE_TTL_SECS and last_checked.
* Implement double-checked TTL: check with try_read first; only attempt
  refresh with try_write if expired/unknown.
* Fix TTL bookkeeping: update last_checked on the “file unchanged” path
  and after API mutations.
* Add documentation warning about TTL-delayed effect of manual
  token.shadow edits.

 docs/user-management.rst       |  4 ++++
 pbs-config/src/token_shadow.rs | 29 ++++++++++++++++++++++++++++-
 2 files changed, 32 insertions(+), 1 deletion(-)

diff --git a/docs/user-management.rst b/docs/user-management.rst
index 41b43d60..8dfae528 100644
--- a/docs/user-management.rst
+++ b/docs/user-management.rst
@@ -156,6 +156,10 @@ metadata:
 Similarly, the ``user delete-token`` subcommand can be used to delete a token
 again.
 
+.. WARNING:: Direct/manual edits to ``token.shadow`` may take up to 60 seconds (or
+   longer in edge cases) to take effect due to caching. Restart services for
+   immediate effect of manual edits.
+
 Newly generated API tokens don't have any permissions. Please read the next
 section to learn how to set access permissions.
 
diff --git a/pbs-config/src/token_shadow.rs b/pbs-config/src/token_shadow.rs
index a5bd1525..24633f6e 100644
--- a/pbs-config/src/token_shadow.rs
+++ b/pbs-config/src/token_shadow.rs
@@ -31,6 +31,8 @@ static TOKEN_SECRET_CACHE: LazyLock<RwLock<ApiTokenSecretCache>> = LazyLock::new
         shadow: None,
     })
 });
+/// Max age in seconds of the token secret cache before checking for file changes.
+const TOKEN_SECRET_CACHE_TTL_SECS: i64 = 60;
 
 #[derive(Serialize, Deserialize)]
 #[serde(rename_all = "kebab-case")]
@@ -72,11 +74,29 @@ fn write_file(data: HashMap<Authid, String>) -> Result<(), Error> {
 fn refresh_cache_if_file_changed() -> bool {
     let now = epoch_i64();
 
-    // Best-effort refresh under write lock.
+    // Fast path: cache is fresh if shared-gen matches and TTL not expired.
+    if let (Some(cache), Some(shared_gen_read)) =
+        (TOKEN_SECRET_CACHE.try_read(), token_shadow_shared_gen())
+    {
+        if cache.shared_gen == shared_gen_read
+            && cache.shadow.as_ref().is_some_and(|cached| {
+                now >= cached.last_checked
+                    && (now - cached.last_checked) < TOKEN_SECRET_CACHE_TTL_SECS
+            })
+        {
+            return true;
+        }
+        // read lock drops here
+    } else {
+        return false;
+    }
+
+    // Slow path: best-effort refresh under write lock.
     let Some(mut cache) = TOKEN_SECRET_CACHE.try_write() else {
         return false;
     };
 
+    // Re-read generation after acquiring the lock (may have changed meanwhile).
     let Some(shared_gen_now) = token_shadow_shared_gen() else {
         return false;
     };
@@ -86,6 +106,13 @@ fn refresh_cache_if_file_changed() -> bool {
         invalidate_cache_state_and_set_gen(&mut cache, shared_gen_now);
     }
 
+    // TTL check again after acquiring the lock
+    if cache.shadow.as_ref().is_some_and(|cached| {
+        now >= cached.last_checked && (now - cached.last_checked) < TOKEN_SECRET_CACHE_TTL_SECS
+    }) {
+        return true;
+    }
+
     // Stat the file to detect manual edits.
     let Ok((new_mtime, new_len)) = shadow_mtime_len() else {
         return false;
-- 
2.47.3