Veille CVE

Vendredi 28 août 2026

3 CVE

Critique 0 Élevée 2 Moyenne 1 Faible 0 Non évaluée 0 Exploitées (KEV) Tri : plus récentes « Gitlab gitlab » ✕
  1. CVE-2026-77680 5.3 Moyenne

    Red Hat Enterprise Linux 10

    An algorithmic complexity flaw exists in libsoup's HTTP Range header processing that persists after the CVE-2025-32907 fix. CVE-2025-32907 addressed memory amplification when a client repeated the same range many times in a single Range header. Commit 9bb92f7a corrected merge correctness in soup_message_headers_get_ranges_internal() in libsoup/soup-message-headers.c, but the coalescing loop still removes merged ranges using g_array_remove_index() for each coalesced element. Because GArray is contiguous, each mid-array removal performs an O(N) memmove. When many identical satisfiable ranges are supplied (for example bytes=0-0 repeated thousands of times), the loop performs O(N²) work coalescing them into a single range. The vulnerable path is reachable server-side from handle_partial_get() in libsoup/server/http1/soup-server-message-io-http1.c when a SoupServer handler returns HTTP 200 with a non-empty body. No authentication is required. The number of ranges is bounded only by the maximum request header size (~100 KiB), allowing roughly 25,000 ranges per request. Reporter measurements on libsoup HEAD containing the CVE-2025-32907 fix show ~90 ms single-core CPU per such request at the wire maximum, blocking the server's event loop for that duration. This is a CPU exhaustion / availability issue only. No memory corruption or information disclosure occurs. Affected: libsoup versions containing the CVE-2025-32907 fix but not merge request !550. Fixed upstream: MR !550 merged 2026-08-20, replacing per-element removal with O(N) in-place compaction and rejecting Range headers requesting more than 200 ranges. Upstream report: https://gitlab.gnome.org/GNOME/libsoup/-/issues/538 Related: CVE-2025-32907

  2. CVE-2026-19889 8.2 Élevée

    GitLab AI Gateway

    GitLab has remediated a vulnerability in the GitLab AI Gateway component affecting all versions of the AI Gateway from 18.9.0 to 19.0.12, 19.1 to 19.1.7, and 19.2 to 19.2.2 that could have allowed an authenticated user with Duo Agent Platform access to redirect model requests to an externally-controlled endpoint via crafted model metadata, resulting in the disclosure of Google Vertex AI or AWS Bedrock cloud service credentials.

  3. CVE-2026-77658 7.8 Élevée

    GNOME Dia

    A stack-based buffer overflow vulnerability exists in the Dia diagram editor when processing Network Bus objects from Dia XML project files. In objects/network/bus.c, bus_load() reads the number of bus handles from the file attribute "bus_handles" using attribute_num_data() without validating an upper bound: bus->num_handles = attribute_num_data(attr); When a bus handle is subsequently moved, bus_handle_moved() allocates two temporary arrays on the stack: parallel = (real *)g_alloca(num_handles * sizeof(real)); perp = (real *)g_alloca(num_handles * sizeof(real)); Because num_handles is fully attacker-controlled via the project file, sufficiently large values (for example 262144 or higher) cause g_alloca() to consume more stack space than the default thread stack limit (typically 8 MB on Linux), resulting in stack overflow, SIGSEGV, and potential stack frame / return-address corruption. An attacker can embed a Bus object with an excessive bus_handles count in a malicious .dia file. Exploitation requires the victim to open the file in Dia (file dialog, command line, or file association) and trigger handle manipulation (moving a bus handle), which exercises the vulnerable code path. The identical g_alloca pattern is present in objects/Misc/tree.c (copied from bus.c) and is likely vulnerable to the same class of attack via Tree objects. Affected versions: Dia 0.98.0 and earlier versions containing this code; issue confirmed on upstream master as of 2026-08-21. Upstream report: https://gitlab.gnome.org/GNOME/dia/-/issues/581