CVE Database

IBM Langflow OSS 1.0.0 through 1.10.2 is vulnerable to server-side request forgery (SSRF). This may allow an unauthenticated attacker to send unauthorized requests from the system, potentially leading to network enumeration or facilitating other attacks.

IBM Langflow OSS 1.0.0 through 1.11.5 could allow an authenticated attacker to access another user's MCP server context due to improper cache key isolation in the MCP Tools component.

A vulnerability was found in FFmpeg 8.0.x. This affects the function parse_playlist of the file libavformat/hlsproto.c of the component Duration Parser. Performing a manipulation of the argument duration/target_duration results in denial of service. The attack is possible to be carried out remotely. Upgrading to version 8.1 and 9.0 is able to mitigate this issue. The patch is named 64fafd63f0b4. Upgrading the affected component is recommended.

A vulnerability has been found in FFmpeg up to 4.4.6/5.1.8/6.1.4/7.1.3/8.0.1. Affected by this issue is the function setup_3x3 of the file libavfilter/vf_convolution.c of the component Convolution Filter. Such manipulation leads to out-of-bounds read. The attack can be executed remotely. The exploit has been disclosed to the public and may be used. Upgrading to version 4.4.7, 5.1.9, 6.1.5, 7.1.4, 8.0.2, 8.1.1 and 9.0 can resolve this issue. The name of the patch is 8970658472/e24b9820b4. It is suggested to upgrade the affected component.

A flaw has been found in cosmicstack-labs mercury-agent up to 1.1.13. Affected by this vulnerability is the function githubRequest of the file src/utils/github.ts of the component GitHub API Handler. This manipulation of the argument path causes server-side request forgery. Remote exploitation of the attack is possible. The exploit has been published and may be used. The project was informed of the problem early through an issue report but has not responded yet.

A vulnerability was detected in cosmicstack-labs mercury-agent up to 1.1.13. Affected is the function checkShellCommand of the file src/capabilities/permissions.ts of the component Shell Command Execution. The manipulation results in incorrect behavior order: validate before canonicalize. The attack may be launched remotely. The exploit is now public and may be used. The project was informed of the problem early through an issue report but has not responded yet.

Magistrala before 1.0.0 contains a SQL injection vulnerability in the timescale-reader and postgres-reader HTTP API services that allows authenticated attackers to inject arbitrary SQL by supplying a malicious format query parameter that is interpolated directly into the FROM clause without parameterization or identifier quoting. Attackers with a self-registered account can substitute arbitrary subqueries to achieve cross-tenant database reads, extract pg_shadow password hashes, read and write arbitrary files, and execute arbitrary code as the postgres OS user by loading attacker-supplied shared objects, with all injected SQL executing at superuser privilege due to the default PostgreSQL role configuration.

MCP Atlassian is a Model Context Protocol (MCP) server for Atlassian products (Confluence and Jira). From 0.17.0 until 0.22.0, validate_url_for_ssrf resolves the attacker-controlled X-Atlassian-Jira-Url and X-Atlassian-Confluence-Url header host once at middleware time, but the outbound request is built with the raw hostname and resolves it again at connection time with no IP pinning. An attacker-controlled DNS-rebinding name can return a public IP during validation and 169.254.169.254 or another internal IP during connection, enabling unauthenticated server-side requests to cloud metadata or internal services. The flaw spans src/mcp_atlassian/utils/urls.py, src/mcp_atlassian/servers/main.py, and src/mcp_atlassian/servers/dependencies.py; validate_url_for_ssrf returns only a verdict rather than a pinned IP, UserTokenMiddleware processes the attacker-controlled headers before fetcher creation, and the Jira and Confluence fetchers use the raw hostname. This issue is fixed in version 0.22.0.

MCP Atlassian is a Model Context Protocol (MCP) server for Atlassian products (Confluence and Jira). Prior to 0.22.0, the confluence_upload_attachment and confluence_upload_attachments tools pass a client-controlled file_path through src/mcp_atlassian/confluence/attachments.py upload_attachment, and the jira_update_issue attachments parameter reaches src/mcp_atlassian/jira/attachments.py upload_attachment, without confining either path to an approved server workspace. In a remote HTTP, SSE, or multi-user deployment, absolute or traversing paths are resolved on the MCP server and uploaded to Atlassian, allowing a client with write-tool access to disclose server files, environment-held Atlassian credentials, or another tenant's data. A local single-user stdio deployment does not cross this trust boundary because the server runs in the caller's environment. This issue is fixed in version 0.22.0.

Skipper is an HTTP router and reverse proxy for service composition. Prior to 0.27.35, the opaAuthorizeRequestWithBody filter in filters/openpolicyagent/openpolicyagent.go can allow an oversized declared Content-Length request to bypass a deny-on-presence Rego policy because ExtractHttpBodyOptionally leaves OPA with an empty parsed_body while forwarding the complete request body upstream. This incomplete remediation of CVE-2026-50197 affects deployments that authorize request-body content and exceed -open-policy-agent-max-request-body-size, which defaults to 1 MB. Policy logic that does not reject input.attributes.request.http.truncated_body can therefore fail open and permit a forbidden payload to reach the protected service, while small bodies and the previously fixed chunked-body case are evaluated normally. This issue is fixed in version 0.27.35.

ASTEVAL is an evaluator of Python expressions and statements. Prior to 1.0.9, FROM_PY in asteval/astutils.py exposes BaseException, SystemExit, KeyboardInterrupt, and GeneratorExit to expressions evaluated by asteval.Interpreter.eval(), while run() and eval() in asteval/asteval.py catch Exception rather than these non-Exception BaseException subclasses. When an attacker-controlled expression raises one of these classes, on_raise() passes the class to raise_exception(), and the resulting exception bypasses the interpreter's safety handlers and propagates into the calling application. A consuming service that evaluates untrusted expressions can therefore be terminated or have signal and cleanup handling disrupted, causing denial of service. The separately documented read-only open() capability is not part of this vulnerability. This issue is fixed in version 1.0.9.

resdata is software for reading and writing result files from the Eclipse reservoir simulator. Prior to 6.2.9, resdata insufficiently validates numeric fields, grid dimensions, keyword sizes, and array indexes while parsing untrusted GRDECL files in lib/resdata/rd_kw_grdecl.cpp and lib/resdata/rd_grid.cpp. Malformed COORD, ZCORN, CORSNUM, ACTNUM, or MAPAXES data can reach rd_grid_alloc_GRDECL_kw__ with inconsistent lengths, while unbounded floating-point conversion can exceed the intended parser buffer. In a network service that accepts untrusted GRDECL files, these conditions can cause a classic buffer overflow, out-of-bounds reads, invalid array access, NULL pointer dereference, memory corruption, or service termination. This issue is fixed in version 6.2.9.

Tract is a tiny, no-nonsense, self-contained TensorFlow and ONNX inference toolkit. Prior to 0.21.16, 0.22.2, and 0.23.1, tract-nnef uses unchecked usize multiplication in nnef/src/tensors.rs read_tensor for attacker-controlled tensor dimensions, the allocation size, and the reported tensor length. Loading a crafted NNEF archive through model_for_path or model_for_read reaches the default DatLoader and can make the wrapped size check accept a small allocation while data/src/tensor.rs as_slice_unchecked creates a much larger logical slice. Model construction through as_uniform can then read beyond the heap allocation and disclose adjacent data, and later access can terminate the process with a segmentation fault. The affected dense numeric tensor path does not include the independently guarded bool, String, or block-quant paths, and no out-of-bounds write or code execution was demonstrated. This issue is fixed in versions 0.21.16, 0.22.2, and 0.23.1.

SIPSorcery is a WebRTC, SIP, and VoIP library for C# and .NET. Prior to 10.0.9, RTPChannel.OnRTPPacketReceived and the STUNAttribute.ParseMessageAttributes, STUNXORAddressAttribute, and STUNAddressAttribute parsing path index untrusted bytes without sufficient length checks, while UdpReceiver.EndReceiveFrom closes the channel when those operations raise a non-socket exception. A remote party can send a single short RTP packet or malformed zero-to-seven-byte STUN address attribute to the shared RTP/ICE socket, including during ICE connectivity checks before DTLS or STUN MESSAGE-INTEGRITY verification, and terminate the active RTP or WebRTC media session. The attacker must reach or learn the advertised ephemeral RTP/ICE port, but no authentication or user interaction is required, and the impact is limited to availability. This issue is fixed in version 10.0.9.

Anyquery is an SQL query engine built on top of SQLite. Prior to 0.4.5, anyquery server exposes file-backed SQLite virtual table modules such as csv_reader and log_reader through its MySQL-compatible server port without authentication, authorization, or directory restrictions. A remote attacker can use SQLite CREATE VIRTUAL TABLE statements to provide a local path to these modules, which use hashicorp/go-getter under the Anyquery server process and return the selected file contents as queryable table rows. The disclosure is limited only by the filesystem permissions of the server process and can expose system configuration, credentials, and private keys. This issue is fixed in version 0.4.5.

Anyquery is an SQL query engine built on top of SQLite. Prior to 0.4.5, anyquery server exposes URL-capable SQLite virtual table modules such as json_reader and log_reader through its unauthenticated MySQL-compatible server port without restricting outbound destinations. A remote attacker can provide a loopback, private-network, or link-local cloud metadata URL, causing go-getter in the Anyquery server process to fetch the selected resource and expose its response as queryable table data. This permits internal network probing, access to internal APIs, and disclosure of cloud credentials; low-integrity impact is possible when a reached internal API performs state-changing actions. This issue is fixed in version 0.4.5.

PocketSphinx is a small speech recognizer. Prior to 5.1.1, the trie language-model loaders in src/lm/ngram_model_trie.c do not adequately validate boundary conditions in ARPA, DMP, and binary format headers, and the acoustic-model loaders in src/mdef.c and src/util/bio.c use sscanf with unbounded string fields. Loading an invalid, corrupted, or malicious language or acoustic model can therefore cause stack or heap buffer overflows and memory corruption. An attacker who can write to a directory selected by POCKETSPHINX_PATH can replace or add a model file that PocketSphinx later loads; users of PocketSphinx 5prealpha have no backported patch and must migrate to the fixed release. This issue is fixed in version 5.1.1.

garminconnect is a Python 3 API wrapper for Garmin Connect that retrieves statistics and manages activities. Prior to 0.3.5, garminconnect/client.py Client.dump creates the OAuth token directory and garmin_tokens.json without explicit owner-only modes, so a permissive umask such as 022 can leave the directory mode at 0755 and the token file mode at 0644. garmin_tokens.json contains di_refresh_token, and another unprivileged user on a shared Linux or macOS host can read the token and obtain persistent access to the victim's Garmin Connect account, including health, fitness, activity, and device data. The Garmin.login tokenstore path is affected, and a pre-existing loosely permissioned token file remains exposed until rewritten or manually restricted. This issue is fixed in version 0.3.5.

UEFI Firmware Parser parses BIOS, Intel ME, and UEFI firmware structures including volumes, file systems, and files. Prior to 1.14, ReadCLen() in uefi_firmware/compression/Tiano/Decompress.c reads Number from GetBits(Sd, CBIT) with CBIT = 9 and can obtain 511 entries for the 510-element Sd->mCLen heap array because its loop does not enforce Index < NC. The CharC == 2 run-length path can additionally request up to 531 zero writes through Sd->mCLen[Index++] = 0. The normal CompressedSection.process() to efi_compressor.TianoDecompress() to TianoDecompress() to DecodeC() to ReadCLen() parsing path therefore permits crafted Tiano or EFI compressed firmware to corrupt heap memory, deterministically crash the parsing process, and potentially execute code depending on build and runtime details. This issue is fixed in version 1.14.

UEFI Firmware Parser parses BIOS, Intel ME, and UEFI firmware structures including volumes, file systems, and files. Prior to 1.14, MakeTable() in uefi_firmware/compression/Tiano/Decompress.c does not validate that bit-length values read from a crafted Tiano or EFI compressed firmware bitstream remain within the expected range from 0 through 16. The normal CompressedSection.process() to efi_compressor.TianoDecompress() to TianoDecompress() to ReadPTLen() to MakeTable() parsing path can consequently write beyond the stack-allocated Count[17] array and related decode tables. The resulting stack corruption deterministically crashes the parsing process and may permit code execution depending on build and runtime details. This issue is fixed in version 1.14.

Skipper is an HTTP router and reverse proxy for service composition. Prior to 0.26.22, Handler in dataclients/kubernetes/admission/admission.go passes the body of requests to the Kubernetes admission endpoint at :9443/admission directly to io.ReadAll(r.Body) without a size limit. An attacker with in-cluster network access and a valid Kubernetes client certificate can send a very large body that causes unbounded memory allocation and an out-of-memory termination of the Skipper process. The disruption is limited to Ingress and RouteGroup admission rather than pod creation or unrelated admission controllers, and Kubernetes normally restarts the process. This issue is fixed in version 0.26.22.

Skipper is an HTTP router and reverse proxy for service composition. Prior to 0.27.13, the routesrv component serves cluster-wide control-plane data without application-layer authentication through /routes, /routes/{zone}, /swarm/redis/shards, and /swarm/valkey/shards. The handlers registered in routesrv/routesrv.go, eskipBytes.ServeHTTP in routesrv/eskipbytes.go, RedisHandler.ServeHTTP in routesrv/redishandler.go, and ValkeyHandler.ServeHTTP in routesrv/valkeyhandler.go only restrict HTTP methods and do not authenticate callers. A compromised or otherwise attacker-controlled pod that can reach routesrv on the Kubernetes cluster network can bypass its namespace-scoped RBAC and read Ingress and RouteGroup backend URLs, filter-chain and OAuth or OIDC path details, and Redis or Valkey shard addresses across namespaces. NetworkPolicy can restrict reachability but does not remove the missing-authentication condition, and the established impact is confidentiality exposure rather than direct integrity or availability loss. This issue is fixed in version 0.27.13.

Envoy Gateway is an open source project for managing Envoy Proxy as a standalone or Kubernetes-based application gateway. Prior to 1.7.4 and 1.8.1, internal/wasm/imagefetcher.go follows tenant-controlled EnvoyExtensionPolicy spec.wasm[].code.image.url values to Docker or OCI Wasm layers, and extractWasmPluginBinary uses the untrusted tar-header h.Size value to allocate memory before validating the entry name or declared size. A small PAX or GNU tar header can therefore claim a multi-terabyte entry even though the surrounding LimitReader restricts only the bytes read from the stream, and no registry allowlist prevents a permitted tenant from selecting an attacker-controlled registry that the controller can reach. The allocation is attempted for every tar entry and can cause an unrecoverable Go runtime out-of-memory failure; because the custom resource persists, reconciliation repeatedly crash-loops the shared controller and causes a single-request, non-volumetric, cluster-wide control-plane denial of service. This issue is fixed in versions 1.7.4 and 1.8.1.

Anyquery is an SQL query engine built on top of SQLite. Prior to 0.4.5, anyquery server forwards unauthenticated SQL from its MySQL-compatible server port to SQLite without restricting ATTACH DATABASE filesystem targets. A remote attacker can select any path writable by the Anyquery server process, cause SQLite to create a database file there, and place attacker-controlled table content in that file. This permits arbitrary file creation or overwrite, causing filesystem integrity loss and denial of service; remote code execution is possible only when another service interprets the written file or the process has a suitably privileged writable target. This issue is fixed in version 0.4.5.

Anyquery is an SQL query engine built on top of SQLite. Prior to 0.4.5, the clear_plugin_cache(plugin) SQL scalar function in namespace/other_functions.go passes the caller-controlled plugin parameter through path.Join to os.RemoveAll without rejecting traversal segments. A low-privileged bearer-token holder can invoke the function through the /v1/query HTTP endpoint, causing path.Join to resolve .. segments outside $XDG_CACHE_HOME/anyquery/plugins/ and os.RemoveAll to recursively delete any reachable directory writable by the Anyquery server process. This causes permanent data loss and denial of service without disclosing file contents. This issue is fixed in version 0.4.5.

A flaw was found in PackageKit. PackageKit skips the polkit authorization check for transactions carrying the SIMULATE (dry-run) flag. In the dnf5 backend, the RepoRemove handler ignores that contract and always executes the real transaction because its guard is written as (role == REPO_REMOVE || !SIMULATE), which is always true for RepoRemove. An unprivileged local user can therefore perform a genuine package uninstall while claiming to simulate. This vulnerability only affects systems using PackageKit with the dnf5 backend.

A flaw was found in NetworkManager-l2tp. The plugin writes attacker-controlled VPN connection properties (vpn.data and vpn.secrets values) unescaped into a generated ipsec.conf file that pluto loads as root. A local unprivileged user can create and activate their own L2TP VPN profile containing a newline-injected leftupdown directive; pluto executes that command as root when the IKE security association is established, resulting in local privilege escalation. This is the same bug class as CVE-2018-10900 (NetworkManager-vpnc).

IBM i 7.6, 7.5, 7.4, and 7.3 could allow a remote attacker to obtain sensitive information due to improper validation of the WebSocket origin.

Concrete CMS below 9.5.3 did not sanitize custom style values in the Block Design dialog before writing them into page CSS via a DOM sink, permitting stored cross-site scripting. An editor-level user could execute script in an administrator's session and escalate privileges. The Concrete CMS security team gave this vulnerability a CVSS v4.0 score of 7.0 with vector CVSS:4.0/AV:N/AC:L/AT:N/PR:H/UI:P/VC:L/VI:H/VA:H/SC:N/SI:N/SA:N. Thanks Nguyen Manh Thuan for reporting.

IBM i 7.6, 7.5, 7.4, and 7.3 could allow a remote unauthenticated attacker to gain access to sensitive information through session IP binding bypass in Navigator for i.