🔒 Aisle Security Analysis

We found 1 potential security issue(s) in this PR:

1. 🟡 Potential DoS from unbounded recursive media cleanup on gateway maintenance timer

Description

A new hourly gateway maintenance interval runs a recursive filesystem cleanup over the entire media directory tree:

• startGatewayMaintenanceTimers() calls cleanOldMedia() with { recursive: true, pruneEmptyDirs: true }.
• cleanOldMedia() performs readdir/lstat over every entry and descends into subdirectories.
• Inbound media files can be created via normal remote interactions (attachments downloaded and stored under ~/.openclaw/media/...), so an attacker who can send many small attachments can inflate the number of files/directories.
• The cleanup has no cap/time budget/backpressure, so the periodic scan can become very expensive (CPU + disk IO) and degrade gateway responsiveness or cause an availability issue.

Vulnerable code (trigger):

mediaCleanupInFlight = cleanOldMedia(params.mediaCleanupTtlMs, {
  recursive: true,
  pruneEmptyDirs: true,
});
...
const mediaCleanup = setInterval(() => {
  void runMediaCleanup();
}, 60 * 60_000);

Note: media.ttlHours is loaded from local config (operator-controlled), not directly from a remote request, but once enabled, remote users can still amplify the cost by causing many inbound media files to exist within the retention window.

Recommendation

Mitigate worst-case traversal costs and attacker-amplification:

• Apply quotas to inbound media creation (per remote sender/session/day): max file count, max total bytes, and/or eviction on write.
• Make cleanup incremental with a time budget / max-entries-per-run (store a cursor or maintain an index), rather than scanning the full tree each time.
• Consider using a background worker/process (or worker_threads) so heavy IO doesn’t block the gateway event loop.
• Add jitter to the schedule to avoid synchronized cleanup across multiple instances.

Example (time/entry budget):

async function cleanOldMediaBudgeted(ttlMs: number, budget: { maxEntries: number; maxMs: number }) {
  const start = Date.now();
  let seen = 0;
  ​// walk directories but stop when (seen >= maxEntries || Date.now() - start >= maxMs)
  ​// persist cursor for next run
}

setInterval(() => void cleanOldMediaBudgeted(ttlMs, { maxEntries: 10_000, maxMs: 2_000 }), 60*60_000);

This prevents a single run from monopolizing IO/CPU when the media tree is large.

Analyzed PR: #38292 at commit d58111f

_{Last updated on: 2026-03-07T03:39:00Z}

Greptile Summary

This PR introduces a configurable inbound media retention TTL (media.ttlHours, default 24 h) and upgrades the media cleanup logic to recurse through nested subdirectories (e.g. remote cache trees), addressing the gap where buffer-backed writes via saveMediaBuffer had no automatic cleanup. A startup sweep and an hourly gateway-managed interval are added alongside the existing per-save cleanup.

Key observations:

• The recursive removeExpiredFilesRecursively function correctly uses fs.lstat to avoid following symlinks when traversing subdirectories, which is an improvement over the previous fs.stat-based approach.
• saveMediaSource (line 275 of store.ts) still calls cleanOldMedia() with no arguments, meaning it uses the hardcoded 2-minute DEFAULT_TTL_MS, not media.ttlHours. Users who configure a longer retention window may be surprised to see files written via that code-path evicted much sooner during active use.
• After recursive deletion, empty directory trees (e.g. remote-cache/session-abc/images/) are never pruned. For long-running servers with many sessions, empty directories will accumulate indefinitely in the media tree.
• The schema, types, labels, and help-text are all updated consistently.
• New tests cover both the recursive store cleanup and the maintenance timer wiring.

Confidence Score: 4/5

• Safe to merge; changes are well-tested and the core logic is correct, with two minor style concerns that don't affect correctness.
• The implementation is logically sound: the recursive cleanup correctly handles symlinks with lstat, the hourly interval follows established patterns, and the new tests provide adequate coverage. The two deductions are: (1) saveMediaSource still uses the 2-minute default TTL rather than the user-configured ttlHours, creating a subtle inconsistency with user expectations; and (2) empty directories from cleaned-up session trees are never removed, which could accumulate over time. Neither is a runtime error, but both are worth addressing before the feature is widely relied upon.
• src/media/store.ts — the saveMediaSource inline cleanOldMedia() call and the lack of empty-directory pruning after recursive deletion.

_{Last reviewed commit: a97cd56}