| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a peer-to-peer secondary tier. Attackers can supply arbitrary remote host and port values in kv_transfer_params to create unreachable peer sessions that retain ZeroMQ sockets until the context quota is exhausted, causing an uncaught ZMQError that crashes EngineCore and stops all inference. |
| vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block counts across multi-prompt completion requests in prefill/decode disaggregated deployments. Attackers can trigger an assertion failure in NixlBaseConnectorWorker._apply_prefix_caching by submitting completion requests with multiple prompts of varying lengths, causing the decode worker to terminate and become unavailable until restarted. |
| vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments. Attackers can send requests with incomplete kv_transfer_params dictionary entries to trigger an uncaught KeyError in EngineCore scheduling, causing the decode engine to terminate and making all routed requests fail until manual restart. |
| vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of concurrent requests, causing different in-flight HTTP requests to return incorrect tokens. |
| vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow concurrent requests to sample tokens outside their allowlists. |
| CSSOM through 0.5.0 contains a denial of service vulnerability in CSSStyleDeclaration.setProperty() that fails to validate reserved property names. Attackers can supply a stylesheet with a declaration named length to replace the internal counter and trigger excessive memory allocation during cssText serialization, causing process termination. |
| uri-js through 4.4.1 contains an improper UTF-8 decoding vulnerability in pctDecChars() that decodes invalid and overlong percent-encoded sequences into ASCII metacharacters. Attackers can craft percent-encoded payloads to bypass platform decoder validation and inject path traversal or CRLF sequences that downstream consumers process without filtering. |
| source-map-js through 1.2.1 fails to validate the per-section offset line value in indexed source maps, allowing attackers to specify arbitrary numeric values. Attackers can supply extremely large offset line values that cause synchronous event loop blocking for extended periods, preventing the service from handling other requests. |
| uri-js through 4.4.1 contains a denial of service vulnerability in the removeDotSegments function that loops infinitely when a path segment begins with Unicode line or paragraph separators. Attackers can trigger this by calling removeDotSegments directly or through normalize/resolve functions with IRI handling enabled, causing the Node.js event loop to block indefinitely until heap exhaustion. |
| braces through 3.0.3 contains a stack overflow vulnerability in the recursive AST walkers that lack depth guards. Attackers can supply deeply nested brace patterns under the character limit to exhaust the call stack and terminate the Node.js process with an uncaught RangeError. |
| rustls-webpki versions before 0.103.10 and 0.104.0-alpha.5 contain faulty CRL authority-matching logic that compares only the first distributionPoint against each CRL's IssuingDistributionPoint, ignoring additional distributionPoints. Attackers with a compromised trusted issuing authority can present revoked certificates that pass revocation checks under UnknownStatusPolicy::Allow, or cause incorrect errors under the default deny policy. |
| rustls-webpki through 0.103.12 (and 0.104.0-alpha releases before 0.104.0-alpha.7) contains a reachable panic in bit_string_flags() in src/der.rs. The input guard fails to reject a named-bit BIT STRING whose content is exactly [0x00] (zero padding bits and no data bytes), so raw_bits.len() - 1 underflows on the empty slice and the subsequent index operation panics (subtract-with-overflow in debug, index-out-of-bounds in release). The condition is reachable through the public API BorrowedCertRevocationList::from_der() when a CRL contains an issuingDistributionPoint extension with such an onlySomeReasons value. Exploitation requires an application that explicitly opts in to CRL revocation checking by passing RevocationOptions to verify_for_usage() and that parses CRL bytes obtained from a source the attacker can influence; the default rustls configuration, which does not use RevocationOptions, is unaffected. A crafted CRL causes a denial of service via the panic. Fixed in 0.103.13 and 0.104.0-alpha.7. |
| vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with max_tokens=0 to exhaust decode-worker memory without bound until the worker restarts. |
| vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. Attackers can submit multimodal audio requests with tokens equal to vocabulary size, causing out-of-bounds writes that corrupt concurrent requests' sampler state and alter repetition penalty behavior. |
| vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. A single request with a negative token ID triggers a CUDA device-side assertion that poisons the GPU context, causing all subsequent requests to fail until the process restarts. |
| lwIP (Lightweight IP) has a double free vulnerability, which could crash the system, cause a DoS, memory corruption, or allow code execution on the victim system. |
| OpenClaw iOS before 2026.8.11 logs complete agent deep-link URLs including persistent bearer keys to unified logs as public diagnostic data. Attackers who obtain diagnostic archives can recover unrotated keys and replay them in forged deep links to submit agent requests without local confirmation prompts. |
| Vaultwarden through 1.37.3 omits organization membership status validation from three cipher access-restriction queries, allowing revoked and not-yet-confirmed members to retain read, write, delete, and attachment access to organization ciphers. Attackers with revoked or pending membership can exploit missing status filters in get_user_collections_access_flags, get_group_collections_access_flags, and is_in_full_access_group to access protected cipher data server-side. |
| e621ng versions before 26.09.16 pass untrusted request parameters directly to Rails url_for in PaginatorComponent and controller navigation links, allowing attackers to redirect pagination and navigation controls to attacker-controlled origins. Attackers can supply host, protocol, and port query parameters that are interpreted as URL generation options, causing pagination links to point to malicious domains while the initial page loads from the legitimate site. |
| The Okta Access Gateway backup restore function does not validate the filename embedded in an encrypted backup payload. This results in writing file contents to unintended locations on the appliance filesystem. |
