CVE Vulnerability Catalog

Backstage before version 3.9.1 in the @backstage/plugin-catalog-backend package has inconsistent catalog property permission evaluation. In deployments that use affected value-based catalog permission conditions as a confidentiality boundary, an authenticated user could receive catalog entity data that policy authors intended to restrict.

The @backstage/plugin-catalog-backend package in Backstage before 3.9.1 inconsistently enforces allowed location types during catalog processing. Under certain configurations, the backend could process location types that were not intended to be allowed, potentially leading to unintended file access on the backend host.

The Gitea API routes for issue attachments also accepted attachments that belong to comments on the issue. Because the author of an issue may edit and delete the issue's attachments, a user who opened an issue could rename or delete attachments that other users had posted in comments on that issue. The contents of the attachments could not be changed.

The Gitea web route for deleting tags requires only write access to the Code unit, but shares its handler with release deletion and did not check that the target was a plain tag. A collaborator with Code write access and without Releases write access could permanently delete published releases of that repository, including their attachments. Protected tag rules covering the release tag still blocked the deletion.

When migrating a repository from another Gitea instance, Gitea used the page size reported in the source server's API settings to end its paginated downloads. A source that reported max_response_items as 0 made these loops run indefinitely and grow server memory until it was exhausted. Any user who can migrate repositories could point a migration at a server they control and cause a denial of service.

Gitea's OAuth2 token endpoint verified the signature and grant of a token submitted with the refresh_token grant type, but not that the token was a refresh token. An unexpired access token for the same OAuth2 application and grant could be exchanged for a new access token and refresh token. Whoever holds such an access token could keep access beyond the token's original lifetime.

Improper input validation in the AMD ROCm Communication Collectives Library (RCCL) could allow a compromised peer rank or network-adjacent attacker to dereference an attacker-controlled pointer, potentially resulting in remote code execution.

In HDF5 through version 2.2.0, there is a heap-based buffer over-read in H5Z__filter_scaleoffset() in src/H5Zscaleoffset.c. An attacker can cause a denial of service (application crash) with a crafted HDF5 file. When the stored minimum bits equal the full precision of the datatype, the decoder copies d_nelmts * size bytes from the compressed chunk without checking that the chunk holds that many bytes.

In sshd in OpenSSH before 10.6, the restrict keyword (in authorized_keys) was supposed to apply to tunnel forwarding but was not enforced. This is a different vulnerability than CVE-2026-73283.

In sshd and ssh in OpenSSH before version 10.6, there is no check for whether the maximum packet length is exceeded during decompression of highly compressed data.

In ssh-keygen in OpenSSH before 10.6, certificates could have incorrect expiration times due to Daylight Saving mishandling. There can be a slightly more severe effect on users in certain Antarctic locations.

In sshd and ssh in OpenSSH before 10.6, an LZ77 dictionary coder can be used even though this is contraindicated by the arXiv 2609.07709 "Crossing the Streams" findings.

In sshd in OpenSSH before 10.6, GSSAPIAuthentication authentication state can incorrectly be persisted across authentication attempts.

In sshd in OpenSSH before 10.6, credentials can incorrectly persist after failure of a GSSAPIAuthentication authentication attempt.

In sftp in OpenSSH before version 10.6, a server can trigger directory traversal (causing files to be written to unintended locations) during a recursive copy operation.

Mozilla's Node-convict (version 6.2.2 and later) is vulnerable to a Denial of Service vulnerability caused by incomplete prototype-pollution protections in config.set(). An attacker controlling the configuration key can write arbitrary properties to constructor.<key>, which walk() resolves to the global Object function. This allows overwriting core JavaScript methods such as Object.assign, leading to persistent process-wide failures and requiring a restart. The issue bypasses existing filters that only block constructor.prototype.* and __proto__.*. Exploitation requires an endpoint that forwards attacker-controlled keys into config.set().

A heap-based buffer overflow in H5VM_array_fill() in src/H5VM.c in HDF5 before 2.2.0. A crafted HDF5 file can cause an application crash and possibly execute arbitrary code when unallocated chunks of a dataset are read and metadata is inconsistent with the buffer's allocated size.

Backstage before version 0.17.8 in the @backstage/backend-defaults package has improper input validation in cloud storage url readers. An attacker with write access to a cloud storage bucket used by Backstage could craft object names that could collide with protected files in the output directory. In certain deployment configurations, this could lead to content injection.

Cloud storage catalog providers in Backstage before 1.54.6 did not sufficiently validate object paths. A principal able to create or rename objects in a configured Azure Blob Storage or AWS S3 catalog source could cause catalog descriptors to be read from outside the intended storage boundary, limited to locations reachable with the backend's configured credentials.

The @backstage/backend-defaults package before versions 0.16.1 and 0.17.8 improperly preserves access restrictions during service credential delegation. An external service credential configured with access restrictions (e.g., read-only) could bypass those restrictions by routing requests through plugin delegation paths.