| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| Netty's HTTP/3 codec (io.netty:netty-codec-http3) versions 4.2.0.Final through 4.2.17.Final contain a denial-of-service vulnerability in the QPACK prefixed-integer decoder (QpackUtil.decodePrefixedInteger), which does not bound the number of continuation bytes it will process. A remote, unauthenticated peer can open a QPACK unidirectional stream (type 0x02 encoder or 0x03 decoder) and send a first byte with all prefix bits set (e.g. 0xFF for a 7-bit prefix or 0x3F for a 5-bit prefix) followed by an endless run of 0x80 continuation bytes. The decoder returns -1 ('need more bytes'), so callers never consume the input, the ByteToMessageDecoder cumulator grows without bound, and each decode() invocation re-scans the whole accumulated buffer, yielding O(N^2) CPU cost. The result is unbounded per-connection heap growth (OutOfMemoryError) and event-loop CPU starvation, reachable in every configuration. Fixed in 4.2.18.Final. |
| stoatchat before 0.15.5 fails to enforce account-level attempt limits on MFA login challenges, allowing attackers who know a password to guess TOTP codes with only IP-based rate limiting. Attackers can reuse MFA challenge tickets across multiple failed attempts and distribute guesses across IP addresses to bypass rate limiting and gain account access. |
| GitPython before 3.1.62 does not validate the `path` field read from an untrusted .gitmodules file when updating submodules. While a prior fix (GHSA-hmq2-w58f-27jc) added Submodule._validated_name() to constrain the `name` field, and GitPython's own containment guard Submodule._to_relative_path() is applied in add() and move(), Submodule.update() derives the absolute checkout location from the raw `path` value without that guard. A .gitmodules entry containing directory traversal components (e.g., path = ../../../tmp/escaped) can therefore cause directories to be created via os.makedirs() outside the repository working tree, populated from the submodule URL on the clone path, and removed via shutil.rmtree() when force_remove is used. Exploitation requires an application flow that updates submodules at a non-HEAD commit (such as a historical-commit API); the common clone-then-update flow re-derives the path from a canonical tree lookup and is not affected. The issue is fixed in GitPython 3.1.62. |
| vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. Attackers can submit requests with min_tokens greater than zero and out-of-vocabulary stop_token_ids to trigger CUDA tensor indexing failures that leave EngineCore in a fatal state requiring service restart. |
| vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaudiochat.py and the Qwen2VLConfig.from_pretrained call used by Tarsier2ProcessingInfo in vllm/model_executor/models/qwen2_vl.py. As a result, deployments pinned to a reviewed revision still resolve these behavior-affecting processor, tokenizer, and config artifacts from the repository's default revision, so a later change to the upstream default branch can alter audio preprocessing, speech tokenizer behavior, or Tarsier2 configuration without any change to the operator's configured pin. This is a supply-chain integrity and reproducibility failure for pinned deployments; it is residual to the earlier fix tracked as GHSA-3ww4-5jv9-j5gm / CVE-2026-47155 and does not constitute remote code execution or a trust_remote_code=False bypass. The issue is fixed in version 0.28.0. |
| stoatchat before 0.15.5 contains an account enumeration vulnerability in the login endpoint that exposes source file locations in error responses. Unauthenticated attackers can distinguish between registered and unregistered email addresses by comparing error location fields returned from POST /api/auth/session/login requests. |
| Budibase (@budibase/server) before 3.45.0 builds MySQL and MSSQL column-rename DDL in packages/backend-core/src/sql/sqlTable.ts by interpolating identifiers directly into a raw query string (backtick-quoted for MySQL, a single-quoted sp_rename literal for MSSQL) without applying the project's quoteMySqlIdentifier / quoteSqlServerIdentifier helpers. An attacker with DDL rights on a connected MySQL/MSSQL datasource can create a column whose name contains a backtick (MySQL) or single quote (MSSQL) plus additional SQL; Budibase's schema introspection stores the name verbatim, and when a Budibase builder later renames that column through the UI (POST /api/tables with _rename.old), the embedded quote character terminates the identifier and the injected SQL is executed. Because the MySQL connection is opened with multipleStatements: true, stacked statements run as Budibase's datasource user, allowing arbitrary reads, writes, or destructive operations on the connected database outside Budibase's row/table permission model. Fixed in 3.45.0. |
| Budibase versions before 3.45.0 fail to validate per-app authorization in the POST /api/global/groups/:groupId/apps endpoint, allowing builders to assign application roles across workspace boundaries. A builder of a single workspace can exploit missing per-app authorization checks to grant themselves admin roles in other workspaces by modifying user group role mappings. |
| Budibase Server before 3.45.0 fails to redact plaintext datasource credentials before broadcasting external table updates to the Builder collaboration websocket room. Attackers with Builder access can intercept unredacted datasource objects containing database passwords and API keys by observing table save or delete operations. |
| vllm before 0.29.0 fails to enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB limit in multimodal chat audio decoding, allowing unauthenticated clients to bypass file size restrictions. Attackers can submit oversized audio files through chat endpoints to consume excessive memory and CPU resources during decoding. |
| vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact. |
| vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' (multimodal) payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied token_ids, and GenerateRequest.token_ids (vllm/entrypoints/serve/disagg/protocol.py) is not checked against model_config.max_model_len. For multimodal processors that report skip_prompt_length_check=True (for example Nemotron Parse, Whisper, and FireRedLID), InputProcessor._validate_prompt_len() returns immediately for both encoder and decoder prompts, so an overlong prompt becomes an EngineCoreRequest and reaches the worker input-batch copy into a fixed max_model_len-wide NumPy row. A client able to reach the endpoint on an affected model configuration can therefore submit an overlong token_ids list to trigger a worker failure and denial of service. Fixed in 0.29.0. |
| stoatchat before 0.15.5 fails to revalidate usernames after Unicode sanitization, allowing attackers to create usernames with forbidden characters by submitting Unicode letters that transform into rejected characters. Attackers can bypass character allowlists and length limits to create reserved-name lookalikes, embed special characters, and exceed the 32-character storage limit. |
| stoatchat versions before 0.15.5 contain a denial of service vulnerability in the acknowledgement worker that processes mass mention messages. Authenticated users can send five crafted role-mention messages to terminate all acknowledgement workers, disabling push notifications and mention badges deployment-wide until the API process restarts. |
| January, the media proxy/embed service of stoatchat (stoatchat/stoatchat), before version 0.15.5 improperly resolves SVG <image href> values as local filesystem paths when a fetched resource is served as image/svg+xml. An unauthenticated remote attacker who causes the service to proxy an attacker-hosted SVG (e.g. via the /proxy endpoint) can determine whether local files exist through observable response-time differences, and can cause supported local image files to be disclosed after re-encoding. Because each referenced file is read in full with no effective limit on the number or total volume of reads, a single request can also generate an unbounded amount of local filesystem I/O and memory pressure (the published proof of concept drives about 4.34 GB of reads), leading to denial of service. The issue is fixed in 0.15.5. |
| stoatchat before 0.15.5 fails to validate that MFA tickets belong to the authenticated user, allowing attackers to bypass MFA by using their own valid ticket with another user's session token. Attackers can obtain a ticket from their own account and use it with a victim's session token to disable TOTP, view recovery codes, or perform other sensitive operations without providing the victim's credentials. |
| Budibase versions before 3.45.0 fail to disable external JSON reference resolution in the OpenAPI/Swagger import validator, allowing authenticated builders to read arbitrary local files. Attackers with builder access can embed file:// references in OpenAPI specifications submitted to the import endpoint to exfiltrate sensitive files including environment variables containing JWT secrets, API keys, and database credentials. |
| Budibase before 3.45.0 contains an unauthenticated server-side request forgery and credential exfiltration vulnerability in the Microsoft Teams webhook endpoint that accepts forged Bot Framework activities with arbitrary serviceUrl values. Attackers can submit a crafted POST request to inject an attacker-controlled serviceUrl that is persisted and used for all subsequent bot replies, causing the server to send live Microsoft OAuth access tokens in Authorization headers to the attacker's host and enabling blind internal network access. |
| Budibase Server before 3.45.0 contains an arbitrary file write vulnerability in the PWA icon upload endpoint that extracts user-supplied ZIP archives without proper symlink validation. Attackers with BUILDER role can craft a malicious ZIP with leaf symlink entries followed by duplicate file entries to write arbitrary files as root, enabling remote code execution. |
| Budibase versions 3.41.0 before 3.45.0 contain an authentication bypass in the OIDC/SSO login path of @budibase/server. In sso.authenticate, when no existing user matches the incoming SSO subject, the server looks up pending user invites by the IdP-asserted email address alone — without validating an invite code and without an email_verified check (the email_verified gate protects only the existing-account lookup). An attacker who can register at an IdP that the tenant trusts for OIDC and assert a victim's invited email address (even with email_verified=false) claims the pending invite and inherits all of its granted privileges, including builder and admin.global, with no admin exclusion. This results in takeover of the invited principal and, for admin invites, full tenant compromise (access to all apps, datasources including production credentials, and automations); the invite is consumed, denying onboarding to the legitimate invitee. |
