PT-2026-50491 · Pypi · Vllm

Publicado 2026-06-17 · Atualizado 2026-06-17 · CVE-2026-54236

OSV

vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via the Anthropic API router

Researcher: Kai Aizen — SnailSploit (@SnailSploit), Adversarial & Offensive Security Research Severity: CVSS 3.1 5.3 (Medium) AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N Target: https://github.com/vllm-project/vllm

---

Summary

The fix for CVE-2026-22778 / GHSA-4r2x-xpjr-7cvv (PRs #31987 and #32319) introduced sanitize message and applied it at four FastAPI exception-handling sites in the OpenAI router. The sanitizer strips object-repr memory addresses (< io.BytesIO object at 0x7a95e299e750> → < io.BytesIO object>) before error messages reach the client, defeating the ASLR-bypass primitive that CVE-2026-22778 chained with a libopenjp2 heap overflow for RCE.

The fix is incomplete: response paths added to vLLM at or after the same time as the fix continue to echo str(exc) directly to clients without sanitize message. The original Stage 1 primitive — sending malformed image bytes so PIL raises UnidentifiedImageError whose message contains the BytesIO object repr — reaches all of them unmodified and leaks the heap address verbatim in the response body.

All five lines below are present in main HEAD (771e1e48b, 2026-05-26).

Affected sites

Current main HEAD (771e1e48b, 2026-05-26):

#	File	Line	Code
1	vllm/entrypoints/anthropic/api router.py	78	message=str(e), (inside POST /v1/messages exception handler)
2	vllm/entrypoints/anthropic/api router.py	124	message=str(e), (inside POST /v1/messages/count tokens)
3	vllm/entrypoints/anthropic/serving.py	808	error=AnthropicError(type="internal error", message=str(e)), (SSE streaming converter)
4	vllm/entrypoints/speech to text/realtime/connection.py	75	await self.send error(str(e), "processing error") (WebSocket event loop)
5	vllm/entrypoints/speech to text/realtime/connection.py	265	await self.send error(str(e), "processing error") (WebSocket generation loop)

Why the global exception handler does not save these paths

api server.py registers a catch-all app.exception handler(Exception)(exception handler) at line 262, and that handler calls create error response(exc) which DOES apply sanitize message. However, FastAPI exception handlers fire only on unhandled exceptions that propagate out of a route function.

All affected HTTP paths catch Exception inside the route coroutine and construct the response themselves:

# vllm/entrypoints/anthropic/api router.py:71-81 (POST /v1/messages)
try:
  generator = await handler.create messages(request, raw request)
except Exception as e:
  logger.exception("Error in create messages: %s", e)
  return JSONResponse(
    status code=HTTPStatus.INTERNAL SERVER ERROR.value,
    content=AnthropicErrorResponse(
      error=AnthropicError(
        type="internal error",
        message=str(e),    # <-- unsanitized
      )
    ).model dump(),
  )

Because the exception is caught and a JSONResponse is returned in-route, every registered FastAPI exception handler — including the sanitizing global one — is bypassed. The WebSocket path bypasses it for a different reason: WebSocket frames don't traverse FastAPI's HTTP exception handler chain at all.

Reachability — the same primitive as the parent CVE

The Anthropic Messages API accepts image content parts in the request body (type: "image" with base64 source.data or type: "image url"). Image bytes are passed to the same multimodal loader used by the OpenAI router. Malformed bytes cause PIL.Image.open to raise:

UnidentifiedImageError: cannot identify image file < io.BytesIO object at 0x7a95e299e750>

The exception propagates up through handler.create messages into the except Exception as e: at api router.py:75. str(e) returns the exception message verbatim, including the address. The address ends up in the error.message field of the JSON response body returned to the attacker. ASLR entropy on the affected process drops from ~4 billion to ~8 candidates, identically to CVE-2026-22778 Stage 1.

The same primitive is reachable on POST /v1/messages/count tokens (route #2), inside the SSE streaming converter when an exception is raised mid-stream (route #3), and over the realtime speech-to-text WebSocket when audio decoder or generation paths raise an exception containing any object repr (routes #4, #5).

Chronology — these are scope misses, not legacy code

• 2026-01-09: PR #31987 (aa125ecf0) introduces sanitize message and applies it to OpenAI router HTTP exception handlers.
• 2026-01-15 (six days later): PR #32369 (4c1c501a7) adds vllm/entrypoints/anthropic/api router.py containing line 78's message=str(e). The fix was not applied to the new router.
• 2026-03-02 (~two months later): PR #35588 (9a87b0578) adds the Anthropic count tokens endpoint, replicating the same message=str(e) pattern at line 124.
• 2026-05-12 (~four months later): PR #42370 (d37e25ffb) consolidates speech-to-text entrypoints and the realtime WebSocket uses send error(str(e), ...) for both error paths.
• 2026-05-26: current main HEAD, all five lines still present.

Remediation

1. Apply sanitize message symmetrically to the five sites

# vllm/entrypoints/anthropic/api router.py — add at top:
from vllm.entrypoints.utils import sanitize message

# Line 78 (POST /v1/messages) and Line 124 (count tokens):
message=sanitize message(str(e)),

# vllm/entrypoints/anthropic/serving.py — add at top:
from vllm.entrypoints.utils import sanitize message

# Line 808:
error=AnthropicError(type="internal error", message=sanitize message(str(e))),

# vllm/entrypoints/speech to text/realtime/connection.py — add at top:
from vllm.entrypoints.utils import sanitize message

# Lines 75 and 265:
await self.send error(sanitize message(str(e)), "processing error")

2. Tighten the regex (defense in depth)

The current regex r" at 0x[0-9a-f]+>" is narrow — it only matches the exact CPython builtin object-repr suffix in lowercase hex with a trailing >. Future Python versions, C extensions, or custom repr methods could produce non-matching formats that re-enable the leak:

# vllm/entrypoints/utils.py
def sanitize message(message: str) -> str:
  # Strip any standalone hex address; downstream observers don't need them.
  return re.sub(r"b0x[0-9a-fA-F]{6,}b", "0x?", message)

3. Future-proofing: consider a response middleware

Both the route-local exception handling pattern (Anthropic router) and the WebSocket path bypass FastAPI's exception handler chain. A response-level middleware that always invokes sanitize message on outgoing error bodies would prevent this class of regression entirely.

Affected versions

• All vLLM versions containing vllm/entrypoints/anthropic/api router.py (introduced 2026-01-15 in PR #32369).
• All vLLM versions containing vllm/entrypoints/speech to text/realtime/connection.py (introduced 2026-05-12 in PR #42370).
• Confirmed present in main HEAD 771e1e48b (2026-05-26).

Steps to reproduce

1. Clone the target: git clone --depth 1 https://github.com/vllm-project/vllm
2. Run the proof of concept (PoC.py) against the cloned source.
3. Observe the result shown under Verified result below.

Credit

Kai Aizen — SnailSploit (@SnailSploit). Adversarial & Offensive Security Research.

Fix

A fix for this vulnerability was added here: https://github.com/vllm-project/vllm/pull/45119

Insertion into Log File

Enumeração de Fraquezas

CWE-532

Identificadores relacionados

CVE-2026-54236

GHSA-HGG8-FQQC-VFMW

Produtos afetados

Vllm

Referências · 5

• https://github.com/vllm-project/vllm/pull/45119 · Nota
• https://github.com/vllm-project/vllm/commit/94923629729381d7f7c9efde72071a2441f7fd82 · Nota
• https://github.com/vllm-project/vllm/security/advisories/GHSA-hgg8-fqqc-vfmw · Nota
• https://github.com/vllm-project/vllm · Nota
• https://osv.dev/vulnerability/GHSA-hgg8-fqqc-vfmw · Aviso do Fabricante