CVE-2024-5225Discovered Published
LiteLLM SQLi via spend logs
View patch bundle on GitHubFull published report
The reviewed summary below describes the scope of the fixture testing.
LiteLLM SQLi via spend logs (CVE-2024-5225)
cve.now entry. Product: LiteLLM
(lab: gabriel-litellm-app).
Eden terminates the HTTP connection to the affected product, evaluates a declarative policy and — in tier 2 — a separately compiled native Rust module under signed module authority, before any byte reaches the application. It does not patch the affected component; the vendor's fixed release remains the durable repair.
Status: Remediated (tier 1 + tier 2), live-proven 2026-11-11T16:48:23Z on the
installed eden-product-release:1.103.0 lab container.
The vulnerability
The LiteLLM spend-logs endpoint (/global/spend/logs) interpolates a
caller-supplied query value into SQL. The fixture (make_litellm) reproduces
the shape: a body carrying a quote-or tautology (id' or 1=1) records a
SQLi mark.
The Eden patch
Tier 1 — declarative policy (blunt, fail-closed): deny every request under
/global/spend/logs, ordinary log queries included — false positives measured
below.
Tier 2 — native Rust module gabriel_litellm_v1 (surgical): refuses
bodies carrying SQL quote-or tautology fragments (' or, 1=1, case
insensitive); ordinary log queries pass unchanged.
Measured comparison (fixtures9, run 2026-11-11)
| Probe | Baseline | Tier 1 | Tier 2 (module) | Backend effect |
|---|---|---|---|---|
benign log query {"key":"val"} | 200 | 403 (false positive) | 200 | none |
attack: SQLi {"q":"id' or 1=1"} | 200 (sqli3 recorded) | 403 | 403 | blocked at tier 2 |
Performance trace (Eden telemetry, live 2026-11-11)
Measured with Eden's built-in request telemetry (gateway.request.baseline
spans, read back through /api/v1/analytics/telemetry?signal=traces), not
client wall-clock. Latency Eden introduces per remediation arm for this case:
| Arm | n | mean | p50 | p95 | min–max |
|---|---|---|---|---|---|
| Tier 1 — declarative policy, blocked request (403) | 15 | 27 µs | 22 µs | 56 µs | 14 µs–56 µs |
| Tier 2 — native module, benign passes through (200) | 21 | 402 µs | 372 µs | 523 µs | 324 µs–619 µs |
| Tier 2 — native module, attack refused (403) | 14 | 131 µs | 128 µs | 181 µs | 107 µs–181 µs |
Eden-added latency for the tier-2 remediation: benign requests complete in 402 µs mean (p50 372 µs, p95 523 µs) in-product, module evaluation included. Honest scope: lab fixtures on one host, fresh connection per measured request, warm requests only; module reactivated shards 2/2 with a fresh signed statement per measurement; spans carry the full authority lineage (module SHA-256, operation, publication). Not the EDEN-3640 performance qualification contract.
Tested against: eden-product-release:1.103.0-single (image sha256:809293e1f0b525ac…, build target
eden_service_single, tooling revision 81a0724a…). Raw data:
module2/telemetry-ab/litellm-telemetry.json and module2/tier1-ab/litellm-tier1.json.
Authority lifecycle and negatives
Import 201, activation: explicit_required; publisher enroll_key, host
inventory + approve, reader inspect refused 403; tampered-statement admit
400; invalid-signature admit 403; admit prepared (not serving), activate
→ receipt serving_now, applied shards 2/2. Revoke → the identical attack
gets 503 (fail-closed); re-approve + re-activate → 403 again. All recorded
in the evidence JSON.
Honest scope
The fixture reproduces the interpolation shape, not the full LiteLLM API
surface. Bypass variants (comment tricks, encoding, alternate parameters) are
not bypass-tested; the rule is a lab-grade partial. Module binary SHA-256
02ea5e3d…, source modules/gabriel_litellm_v1.rs.
Cost record
Run window 16:48:23Z → 16:48:33Z (per-CVE ledger, marginal session cost).
Evidence appendix
- Fixture:
gabriel-2026-11-11/fixtures9.py(make_litellm, port 18308) - Evidence:
gabriel-2026-11-11/module2/evidence/litellm-module.json - Module source:
docs/case-studies/modules/gabriel_litellm_v1.rs
The vulnerability
The LiteLLM spend-logs endpoint (/global/spend/logs) interpolates a
caller-supplied query value into SQL. The fixture (make_litellm) reproduces
the shape: a body carrying a quote-or tautology (id' or 1=1) records a
SQLi mark.
The Eden patch
Tier 1 — declarative policy (blunt, fail-closed): deny every request under
/global/spend/logs, ordinary log queries included — false positives measured
below.
Tier 2 — native Rust module gabriel_litellm_v1 (surgical): refuses
bodies carrying SQL quote-or tautology fragments (' or, 1=1, case
insensitive); ordinary log queries pass unchanged.
Measured comparison (fixtures9, run 2026-11-11)
| Probe | Baseline | Tier 1 | Tier 2 (module) | Backend effect |
|---|---|---|---|---|
benign log query {"key":"val"} | 200 | 403 (false positive) | 200 | none |
attack: SQLi {"q":"id' or 1=1"} | 200 (sqli3 recorded) | 403 | 403 | blocked at tier 2 |
Performance trace (Eden telemetry, live 2026-11-11)
Measured with Eden's built-in request telemetry (gateway.request.baseline
spans, read back through /api/v1/analytics/telemetry?signal=traces), not
client wall-clock. Latency Eden introduces per remediation arm for this case:
| Arm | n | mean | p50 | p95 | min–max |
|---|---|---|---|---|---|
| Tier 1 — declarative policy, blocked request (403) | 15 | 27 µs | 22 µs | 56 µs | 14 µs–56 µs |
| Tier 2 — native module, benign passes through (200) | 21 | 402 µs | 372 µs | 523 µs | 324 µs–619 µs |
| Tier 2 — native module, attack refused (403) | 14 | 131 µs | 128 µs | 181 µs | 107 µs–181 µs |
Authority lifecycle and negatives
Import 201, activation: explicit_required; publisher enroll_key, host
inventory + approve, reader inspect refused 403; tampered-statement admit
400; invalid-signature admit 403; admit prepared (not serving), activate
→ receipt serving_now, applied shards 2/2. Revoke → the identical attack
gets 503 (fail-closed); re-approve + re-activate → 403 again. All recorded
in the evidence JSON.
Honest scope
The fixture reproduces the interpolation shape, not the full LiteLLM API
surface. Bypass variants (comment tricks, encoding, alternate parameters) are
not bypass-tested; the rule is a lab-grade partial. Module binary SHA-256
02ea5e3d…, source modules/gabriel_litellm_v1.rs.
Cost record
Run window 16:48:23Z → 16:48:33Z (per-CVE ledger, marginal session cost).