Repository navigation
SECURITY: chromium-bin: distrust the compromised eMudhra keys via CADistrustedCertificates policy (CVE-2026-111943) - #961
Conversation
…erprise policy (CVE-2026-111943) New package chromium-policies installs /etc/chromium/policies/managed/ca-distrusted-certificates.json, a CADistrustedCertificates policy (Chrome 132+, Linux) listing the 8 compromised eMudhra intermediate keys and emSign ECC Root CA - G3. Chromium matches by SPKI, so one certificate per key covers the emSign copy, the Sectigo AAA cross-sign and the ECDSA-malleated twins. build.sh fails unless the file holds exactly those 9 keys, and five negative controls prove the check can fail. chromium-bin takes it as a runtime dep; versions and sources unchanged. chromium-headless-shell-bin cannot read platform policy (no loader since Chromium ~150; the 155 binary has no policy path or policy name), so it only gets a comment documenting the gap. A CI job checks every policy key against the `spki` lines of packages/ca-certificates/distrust.txt; it warns until #948 lands that file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reached
This review includes 5 billable files and costs up to $1.25.
Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing. Or wait 7 minutes for your next included review. View limit details
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to
|
mitodrummer
left a comment
There was a problem hiding this comment.
Approving with nits. Verified by computation: all 9 entries decode to the 8 eMudhra intermediates (EM DV/OV/EV TLS CA and OV CS CA, G3B and G3C) plus emSign ECC Root CA - G3. The 8 intermediates match Chromium's da63b94c blocklist by certificate SHA-256 and SPKI SHA-256; the root is byte-identical to the one removed in 8793082f. The SPKI set in build.sh, the standalone test and the Python checker agree, and the checker reports OK against the now-merged distrust.txt from #948. By reading Chromium 155: the CADistrustedCertificates loop in ProfileNetworkContextService has no feature gate, policy_paths.cc reads /etc/chromium/policies for Chromium branding, and headless/lib/browser/policy is gone by 150, so the headless-shell claim holds. build.sh's five negative controls each exercise a distinct branch and none can pass vacuously.
Should fix
chromium_policy_distrust_check.py/chromium-policy-distrust.yml: the "distrust.txt absent, warn and exit 0" branch is now a silent-disable hole since #948 merged. If the file is renamed or moved the guard goes green with only a warning annotation. Make absence an::error::exit 1.- CI has only run the warn path (both runs predate #948). Update the branch and confirm the job logs the OK line.
- chromium-headless-shell-bin stays vulnerable and is what agent-browser drives against untrusted sites. Open a tracking issue whose trigger is the first CfT Stable carrying da63b94c and 8793082f (Chrome expects its root store update in Stable 2026-10-27), so the comment isn't the only record.
Note: the in-browser proof (emSign chain rejected, letsencrypt loads, negative control without the dep connects) is in #960's arm64 end-to-end run.
Nits
chromium-policies/build.nclstandalone test repeats build.sh's 9 hashes; drift fails loudly, but there are two places to edit and the test prints no diff on failure.build.sh: the summary line hard-codes "5 negative controls"; the SPKI hash comes from openssl's re-encoded key while Chromium and the Python script hash raw bytes (they agree for these 9).- Checker: a non-string list entry raises an uncaught TypeError (still non-zero, but no
::error::annotation). - Workflow checkout lacks
persist-credentials: false(matches sibling workflows).
Review follow-ups (mitodrummer): - chromium_policy_distrust_check.py: a missing distrust.txt is now an ::error:: and exit 1, not a warning and exit 0 (#948 has merged, so absence means the file moved and the guard must not go green). A non-string CADistrustedCertificates entry is an ::error:: annotation instead of an uncaught TypeError. - workflow: persist-credentials: false on the checkout; drop the "warns until #948" comment. - chromium-policies standalone test prints a diff on mismatch; comment that the 9 hashes are repeated from build.sh. - build.sh counts its negative controls instead of hard-coding 5, and notes the SPKI hash method difference (openssl re-encoded key vs the certificate's raw SPKI bytes; equal for all 9, checked). - chromium-headless-shell-bin: cite gominimal/pkgmgr-rs#1028, the tracker for the unmitigated headless shell. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Picks up #948 (distrust.txt) so the policy check runs its enforcing path. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks @mitodrummer. All points addressed in 226374a, plus a merge of main (439799d) that brings in #948. Should fix
Nits
🤖 Generated with Claude Code |
…ith #960) #960 merged first carrying an earlier copy of this branch (2b4583a), so the add/add conflicts are that copy against this branch's review-fix commit 226374a. Resolved to exactly main + 226374a (verified identical to a cherry-pick of 226374a onto origin/main): chromium-bin/build.ncl is main's (ungoogled, #960); the checker, workflow, chromium-policies and the headless-shell comment carry 226374a's changes. The checker passes against main's distrust.txt (9/9 policy keys are spki entries). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What's compromised
CVE-2026-111943. The private keys of eight eMudhra intermediate CAs (EM DV/OV/EV TLS CA and EM OV CS CA, generations G3B and G3C) are recoverable from their public certificates (Mozilla bug 2083449). The ca-certificates fix is #948, and sibling PRs cover the other TLS stacks.
Chromium does not use
/etc/ssl/certs. On Linux it verifies against its own compiled-in Chrome Root Store, so #948 does nothing for it. Our builds are also exposed:Chromium fixed this on main today (2026-10-20), and no release carries the fix yet:
da63b94c(main@{#1723080}) blocklists the 8 intermediates;8793082f(main@{#1723181}, Chrome Root Store v43) removes emSign ECC Root CA - G3;What changes
New package
chromium-policies. It installs/etc/chromium/policies/managed/ca-distrusted-certificates.json, which sets theCADistrustedCertificatesenterprise policy.supported_on: chrome.*:132-, so Linux is covered, and our builds are 153 and 155.ProfileNetworkContextService::GetCertificatePolicyextracts each listed certificate's SPKI intodistrusted_spkis, with no feature gate. Every certificate with that key is then distrusted for TLS server auth.da63b94c. Matching is by SPKI, so one certificate per key also covers the Sectigo AAA cross-sign and the ECDSA-malleated twins. Listing all 16 would add nothing.policy::kPolicyPathincomponents/policy/core/common/policy_paths.cc). All three Chromium-branded binaries contain the string/etc/chromium/policies: the snapshot x64 build, the Playwright arm64 build and ungoogled-chromium 155. Chrome and CfT branding would read/etc/opt/chrome{,_for_testing}/policies, but nothing in the tree uses those, so they are not shipped.*.jsonin the directory, but a key set in two files is not merged. Future distrust entries therefore go into this file, never into a second one.chromium-bin. It gets
chromium-policiesas a runtime dep, plus apolicy_installedstandalone test. Version and sources are unchanged./etc/chromium/policiesand containsCADistrustedCertificates.chromium-headless-shell-bin: comment only. ⚠ It stays vulnerable, and nothing at package level can fix it.
chrome-headless-shellcannot load platform policy. Chromium deletedheadless/lib/browser/policy/: it is present at the 145.0.7632.0 tag and gone by 150.0.7871.46. Even before that, it handled only HeadlessMode, URLBlocklist/Allowlist and RemoteDebuggingAllowed./etc/.../policiespath and noCADistrustedCertificatesstring.$HOMEand matches whole certificates, not SPKIs.chromium-policieshere would only look like a fix.da63b94cand8793082f;chromium --headlessfrom chromium-bin.Alignment with the canonical denylist (pkgmgr-rs#1020).
.github/scripts/chromium_policy_distrust_check.pyand thechromium-policy-distrustworkflow check that every certificate in the policy has its SPKI SHA-256 listed as anspkiline inpackages/ca-certificates/distrust.txt(#948). The policy needs whole certificates, so it can't be generated from those hashes. Until #948 merges, distrust.txt is absent and the job warns and passes. It enforces from then on.Tests
Build-time (
build.sh, gates the artifact):{"CADistrustedCertificates": [...]}and nothing else.Standalone:
chromium-policies.policy_installedre-derives the 9 SPKI hashes from the installed file with openssl.chromium-bin.policy_installedchecks that the file is where the browser looks.Verified locally (no package build, no container):
build.shon the real file (jq + LibreSSL 3.3.6): it passes. Three mutants all fail at the expected check: an expected key dropped, the set comparison disabled, and a negative control disabled.distrust.txt(18spkilines), it reports OK;spkiline removed, it fails;86a1ecba…be6b.mip dumpevaluates on amd64 and arm64.Online check: not run, needs network. The PR's chromium-bin should reject the emSign chain and still load a normal site:
/bin/chromium-headless-shellis expected to still load the first URL, for the reasons above.🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Bug Fixes
Documentation