Skip to content

openssl, openssl-3.5: carry OpenSSL's November 2026 security fixes (8 Low CVEs) as patches - #1571

Draft
bryan-minimal wants to merge 1 commit into
mainfrom
openssl-nov-2026-backports
Draft

bryan-minimal wants to merge 1 commit into
mainfrom
openssl-nov-2026-backports

Conversation

@bryan-minimal

Copy link
Copy Markdown
Member

OpenSSL published eight advisories on 2026-11-10 and 2026-11-17 (vulnerabilities list). All are rated Low. The fixes are in 3.5.10, 4.0.4 and 4.1.1, and none of those versions has been released yet. This PR carries upstream's own fix commits as patches on both of our OpenSSL lines:

  • packages/openssl 4.1.0 takes the commits from the openssl-4.1 branch.
  • packages/openssl-3.5 3.5.9 takes the commits from the openssl-3.5 branch.
CVE Issue openssl (4.1.0) openssl-3.5 (3.5.9)
CVE-2026-88823 AES-GCM-SIV AAD forgery after a zero-length incremental AAD update openssl/openssl@6747b1a openssl/openssl@159e1de
CVE-2026-107496 Infinite loop in CMS_encrypt() with an EC group order of 1 openssl/openssl@36e15b6 openssl/openssl@a0051cf
CVE-2026-88822 QUIC uses uninitialised stack bytes during automated peer detection openssl/openssl@7b4f534 openssl/openssl@1a73dc5
CVE-2026-95692 AES-GCM-SIV OOB read and incorrect IV length handling openssl/openssl@70a47ac openssl/openssl@257a342
CVE-2026-95691 Retransmitted QUIC CRYPTO frames can cause a DoS or a one-byte disclosure openssl/openssl@dd90ceb openssl/openssl@f624467
CVE-2026-107493 Use-after-free in PEM_X509_INFO_read_bio_ex() openssl/openssl@c57a5cc openssl/openssl@c8f6d04
CVE-2026-35190 TLS 1.3 key-share OOB read with the Suite B flag openssl/openssl@ade6461 openssl/openssl@9491b0d
CVE-2026-124709 QUIC server keeps unbounded memory for rejected streams openssl/openssl@ead311d (test hunks dropped, see below) openssl/openssl@f3acb0a

Every commit is the one the advisory links for that branch. I also checked that each is an ancestor of the branch head.

How they're applied

  • The patches follow the llama.cpp stb style. build.sh applies them in the order they landed on the branch, using patch -Np1 -F0. Every file a patch touches is sha256-checked before and after it is applied. A different tarball or a partial apply therefore fails the build.
  • The patches are unchanged git format-patch output from upstream, with one exception. In the 4.1 copy of CVE-2026-124709, I removed the test/quicapitest.c hunks because one of them does not apply to the 4.1.0 test file. The code hunk (ssl/quic/quic_stream_map.c) is verbatim and byte-identical to the 3.5 backport, which upstream itself shipped without tests. A note at the top of the patch file records the change.
  • I checked locally against our gs:// mirrored tarballs, whose sha256 matches build.ncl, extracted into fresh dirs:
    • Every patch applies with zero fuzz and zero offset.
    • I ran the exact build.sh patch block, and only that block, against a fresh tree for each package, and both pass.
    • Running the block on an already-patched tree fails at the first hash, as it should.
    • build.ncl evaluates on amd64 and arm64 with the local Nickel shim.
    • Nothing has been compiled. The upstream test-file hunks that do apply (test/ectest.c, test/x509_check_cert_pkey_test.c) are compiled by make, so the build-bot is the first real compile check.
  • Both packages gain a patch build dep. patch reaches openssl only through glibc → python, which is the same route toolchain (already a build dep) takes, so the dependency graph keeps its existing shape. The planner is the real judge of this, though.

Not carried

None. All eight CVEs are carried on both packages. (We ship no 4.0.x, 3.0.x or 1.1.1 OpenSSL.)

⚠ Rebuild impact

Merging this rebuilds essentially the whole world:

  • openssl is a build dep of python, which is a build dep of glibc. By import closure, 569 of 570 packages sit downstream.
  • openssl-3.5 feeds rustc-1.98.1 / rustc-1.99.1, the Rust ladder, plus actions-runner and foundationdb.

It needs a full queue window: the build-bot is serial and the queue has a 6h timeout.

When 4.1.1 / 3.5.10 ship

Drop the patch block in build.sh and the cve-2026-*.patch files on the version bump. The before-hashes make a bump that keeps the block fail closed, so a missed drop cannot go unnoticed.

🤖 Generated with Claude Code

… patches

OpenSSL published eight Low-severity advisories on 2026-11-10 and
2026-11-17, fixed in 3.5.10 / 4.0.4 / 4.1.1, none of which is released
yet. Carry upstream's own fix commits, from the openssl-4.1 branch for
packages/openssl (4.1.0) and the openssl-3.5 branch for
packages/openssl-3.5 (3.5.9):

  CVE-2026-88823   AES-GCM-SIV AAD forgery after zero-length AAD update
  CVE-2026-107496  CMS_encrypt() infinite loop with EC group order 1
  CVE-2026-88822   QUIC uninitialised stack bytes in peer detection
  CVE-2026-95692   AES-GCM-SIV OOB read / IV length handling
  CVE-2026-95691   QUIC CRYPTO frame retransmit DoS / 1-byte disclosure
  CVE-2026-107493  PEM_X509_INFO_read_bio_ex() use-after-free
  CVE-2026-35190   TLS 1.3 key-share OOB read with the Suite B flag
  CVE-2026-124709  QUIC server unbounded memory for rejected streams

Patches are applied in branch order with -F0, every touched file
sha256-checked before and after (llama.cpp's style), and all apply with
no fuzz or offset to the mirrored release tarballs. One change from
upstream: the 4.1 CVE-2026-124709 patch drops its test/quicapitest.c
hunks (one does not apply to 4.1.0); its code hunk is verbatim and
identical to the code-only 3.5 backport.

Both packages gain a `patch` build dep. Drop the blocks when
version >= 4.1.1 / 3.5.10.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@coderabbitai

coderabbitai Bot commented Nov 18, 2026

Copy link
Copy Markdown
Contributor

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant