1808236 - Intermittent TSAN socketprocess netwerk/test/unit/test_ech_grease.js | data race in memcpy

Status: NEW

(Core :: Networking, defect, P2)

Description

[Treeherder Bug Filer]

Filed by: rvandermeulen [at] mozilla.com
Parsed log: https://treeherder.mozilla.org/logviewer?job_id=400695175&repo=try
Full log: https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/F7r4GxYBQYGZ1BtMzgRFVQ/runs/0/artifacts/public/logs/live_backing.log

---

Only happens in socketprocess mode.

Comment 1

[Ryan VanderMeulen [:RyanVM]]

Context - I'm looking at scheduling socketprocess xpcshell tests on more Linux configurations and this data race is happening in test_ech_grease.js pretty frequently (but not as often as the race in bug 1808235).

Comment 2

[Ryan VanderMeulen [:RyanVM]]

Attachment: tsan log

Comment 3

[Andrew McCreight [:mccr8]]

The "previous write" stack involves setting some SSL integer pref, which the test does in a few places. Users hopefully are not changing their crypto prefs all of the time so it doesn't really seem exploitable. I'll mark it moderate but maybe it could be unhidden.

Dana, it looks like you touched this test last, so this might be of interest. Thanks.

Comment 4

[Dana Keeler (she/her) (use needinfo) [:keeler] (on leave)]

Yeah, this doesn't seem exploitable to me. The relevant NSS API isn't thread-safe, but it's just enabling or disabling some ciphersuites. We could either make this an NSS problem and add some locks, or we could only touch NSS configuration stuff on the socket thread (which wouldn't be too hard, and is arguably the correct approach anyway).

Comment 5

[Ryan VanderMeulen [:RyanVM]]

FYI, test_ech_grease.js was skipped for the time-being when bug 1808234 landed, so it'll need re-enabling as part of the patch for this bug.