Skip to content

Update five dependencies with OSV advisories - #3035

Merged
MarcusSorealheis merged 1 commit into
mainfrom
fix/osv-advisories-oct-2026
Oct 24, 2026
Merged

MarcusSorealheis merged 1 commit into
mainfrom
fix/osv-advisories-oct-2026

Conversation

@MarcusSorealheis

@MarcusSorealheis MarcusSorealheis commented Oct 23, 2026 •

Copy link
Copy Markdown
Member

What and why

Five OSV advisories are open against this repository. Three of them reach the server binary, one reaches only the docs site, and one only the Azure store's test runner.

Advisory Package Moves Reaches
RUSTSEC-2026-0386 tokio 1.52.3 → 1.53.3 the server binary (direct dependency)
RUSTSEC-2026-0385 ctutils 0.4.2 → 0.4.4 the server binary, through digest
RUSTSEC-2026-0383 webpki-root-certs 1.0.3 → 1.0.10 the server binary, through rustls-platform-verifier
GHSA-mrhx-c7xc-m3w5 zod 4.4.3 → 4.6.7 web/ only, through fumadocs-mdx
GHSA-27r3-jxmh-p828 body-parser 2.3.0 → 2.4.0 the Azure store's test runner only, through azurite and express

tokio moves in every crate manifest and not only in the lock, so a later resolution cannot fall back onto an affected 1.52.x. Nothing in this repository constructs an AsyncFd, which is the API RUSTSEC-2026-0386 is about, so that one is runtime exposure rather than a call we make.

zod needed no manifest change: fumadocs-mdx already asks for ^4.4.3, which admits the fixed 4.6.6, so only the lock moved. body-parser is pulled in by Azurite's Express and needs an override, which the runner's package.json already uses for three other advisories.

MODULE.bazel.lock records crate metadata per version and several workflows resolve it with --lockfile_mode=error, so it is regenerated here for the three new crate versions.

How was this verified?

  • cargo check --workspace --all-targets: clean, no warnings.
  • cargo test --workspace --tests --no-fail-fast: 123 test binaries, 1374 passed, 0 failed, 5 ignored.
  • cargo-audit 0.22.2 against the updated Cargo.lock: no vulnerabilities. It still reports two yanked crates, chacha20 0.10.0 and spin 0.10.0, which are pre-existing and unrelated to these advisories.
  • node --test dependency-security.test.cjs in the Azurite runner: 8 of 8 pass, including the two added here.
  • bazelisk mod show_extension '@@rules_rs+//rs:extensions.bzl%crate' regenerated the module lock; the diff is the three crate facts and nothing else.

The new coverage is a pair: a floor on the body-parser version Express resolves, which is what the override has to keep holding, and a request whose trailing bytes arrive after the body has been read, which the patched unpipe-and-drain path has to answer and finish.

An honest gap on that second test: it does not fail on body-parser 2.3.0 on this machine. I tried to reproduce the crash first, including the raw-socket shape from the upstream fix's own test, and on Node v26.8.1 the finished inflate stream simply never errors on the trailing bytes, so the uncaught exception the advisory describes does not happen. The version floor is therefore the part of the pair that actually gates the regression, and the behavioural test guards the patched code path rather than reproducing the bug. I removed an earlier test that appeared to cover it but passed on the vulnerable version too.

Neither the Bazel build nor the docs site was built locally; CI covers both.

Risk

Low to moderate, and concentrated in tokio. It is a minor version bump under the whole server, so anything subtly wrong shows up as runtime behaviour rather than a compile error — the full cargo test sweep above is the main evidence against that. The other four are either transitive patch bumps or confined to test and docs tooling.

body-parser carries one behaviour change worth knowing: after the body is read the patched version unpipes and destroys the decompression stream and drains the remaining request bytes. That only affects the Azurite test runner, and its suite passes.

No configuration default, wire format, public API or stored data changes.

AI assistance

An agent made this change end to end: it looked up each advisory, traced where each package enters the dependency graph, moved the versions, regenerated the locks, wrote the two tests, ran the verification above, and drafted this description. The reproduction attempt and its failure are described as they happened rather than smoothed over. Maintainer review is still required; this does not claim human review of every line.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XntM7FzDafzhyVxr7mvqd2


This change is Reviewable

Three of the five reach the server binary, one reaches only the docs site and
one only the Azure store's test runner.

- `tokio` 1.52.3 -> 1.53.3 (RUSTSEC-2026-0386): `AsyncFd::new` and the
  related constructors violate I/O safety; 1.53.3 is the fix for the 1.52
  line. Nothing here constructs an `AsyncFd` itself. The floor moves in every
  crate manifest and not only in the lock, so a later resolution cannot fall
  back onto 1.52.x.
- `ctutils` 0.4.2 -> 0.4.4 (RUSTSEC-2026-0385): non-constant-time mask
  generation and `u128` comparisons. Transitive, through `digest`.
- `webpki-root-certs` 1.0.3 -> 1.0.10 (RUSTSEC-2026-0383): the bundled root
  store trusts distrusted intermediates. Transitive, through
  `rustls-platform-verifier`.
- `zod` 4.4.3 -> 4.6.7 (GHSA-mrhx-c7xc-m3w5): memory exhaustion while
  validating oversized constrained collections. It reaches `web/` through
  `fumadocs-mdx`, whose `^4.4.3` range already allowed the fixed 4.6.6.
- `body-parser` 2.3.0 -> 2.4.0 (GHSA-27r3-jxmh-p828): trailing bytes after a
  compressed body reach a decompression stream that was left piped with no
  error listener, and that uncaught error ends the process. It arrives
  through `azurite` and `express` in the Azure store's test runner alone, so
  it takes an override beside the three that runner already carries.

The runner's dependency-security suite gains the matching pair: a floor on
the resolved `body-parser`, which is what the override has to hold, and a
request whose trailing bytes land after the body has been read, which the
patched unpipe-and-drain path has to answer and finish. That second test
does not fail on 2.3.0 here: on Node 26 the finished inflate stream never
errors on the trailing bytes, so the crash the advisory describes needs a
runtime this one is not.

`MODULE.bazel.lock` records crate metadata per version and CI resolves it
with `--lockfile_mode=error`, so it is regenerated for the three new crate
versions.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XntM7FzDafzhyVxr7mvqd2
@vercel

vercel Bot commented Oct 23, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
nativelink Ready Ready Preview Oct 23, 2026 11:53pm UTC
nativelink-aidm Ready Ready Preview Oct 23, 2026 11:53pm UTC

Request Review

@MarcusSorealheis
MarcusSorealheis merged commit 42b4b4d into main Oct 24, 2026
30 checks passed
@MarcusSorealheis
MarcusSorealheis deleted the fix/osv-advisories-oct-2026 branch October 24, 2026 00:17

This branch was successfully deployed

2 active and 1 inactive deployments
Preview – nativelink — 8145c8c5 Deployed Oct 23, 2026 by vercel[bot]
Preview – nativelink-aidm — 8145c8c5 Deployed Oct 23, 2026 by vercel[bot]
cluster-ci-benchmark — 8145c8c5 Deployed Oct 23, 2026 by MarcusSorealheis via NativeLink / Linux #39
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.

2 participants