717758 - abort() isn't caught by Breakpad (e.g. pure virtual calls, stack buffer overflow checks)

Status: RESOLVED FIXED

(Toolkit :: Crash Reporting, defect)

Description

[Joe Drew (not getting mail)]

Pure virtual calls explicitly call abort(), killing Firefox, but they aren't caught by Breakpad, so Firefox just disappears on people when they happen.

Comment 1

[Benjamin Smedberg]

Yes, abort() on Mac doesn't trigger breakpad. Is there a way to alter the default behavior, perhaps by implementing or injecting an alternative _purecall?

Comment 2

[(not currently active) Ted Mielczarek]

We also don't handle stack overflow on Mac (bug 507876), or unhandled C++ exceptions (http://code.google.com/p/google-breakpad/issues/detail?id=318).

There's an upstream report about potentially handling SIGABRT, but it hasn't gone anywhere:
http://code.google.com/p/google-breakpad/issues/detail?id=343

Comment 4

[Jesse Ruderman]

I hit the same problem today with a stack buffer overflow (bug 866544).

Comment 5

[Jesse Ruderman]

Attachment: __stack_chk_fail crash, as seen by Mac OS X crash reporter

Comment 6

[Jesse Ruderman]

Also happens with release-mode assertion failures in system libraries (e.g. bug 903722).

Comment 7

[(not currently active) Ted Mielczarek]

This might be fixable with the WIP patch I mentioned in bug 525296, since that catches SIGABRT.

Comment 8

[(not currently active) Ted Mielczarek]

Patch up for upstream review:
https://breakpad.appspot.com/618002/

Turns out it was not hard to finish up, I should have done that ages ago.

Comment 9

[(not currently active) Ted Mielczarek]

Landed this upstream, I'll get our in-tree copy updated (or cherrypick this):
http://code.google.com/p/google-breakpad/source/detail?r=1205

Comment 10

[(not currently active) Ted Mielczarek]

Try build available with this patch if you want to sanity check some cases:
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-443970023f3a/try-macosx64/firefox-26.0a1.en-US.mac.dmg

Comment 11

[(not currently active) Ted Mielczarek]

Attachment: Breakpad test for abort() crashes

I'm just going to cherry-pick the patch from upstream and land it. I added a unit test here for sanity's sake. Try push here also for sanity:
https://tbpl.mozilla.org/?tree=Try&rev=24c8d5b62900

Comment 12

[(not currently active) Ted Mielczarek]

Attachment: kill hung processes with SIGABRT on OS X in automation.py

Being lazy, just going to put this patch here too. Since SIGABRT didn't work on Mac historically we didn't try to use it in automation.py. This patch changes that. It's also going to bitrot jhammel's big Mochitest-rework patch, but what can you do?

Comment 13

[(not currently active) Ted Mielczarek]

Attachment: handle abort() in Breakpad on OS X.

This is the upstream patch with r=mento.

Comment 14

[Joel Maher ( :jmaher ) (UTC -8)]

Comment on attachment 805459 [details] [diff] [review]
kill hung processes with SIGABRT on OS X in automation.py

Review of attachment 805459 [details] [diff] [review]:
-----------------------------------------------------------------

can't argue with this.

Comment 15

[(not currently active) Ted Mielczarek]

remote:   https://hg.mozilla.org/integration/mozilla-inbound/rev/62cba8b2fd7d
remote:   https://hg.mozilla.org/integration/mozilla-inbound/rev/742c8c54c8fc
remote:   https://hg.mozilla.org/integration/mozilla-inbound/rev/751208f98deb

Comment 16

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/62cba8b2fd7d
https://hg.mozilla.org/mozilla-central/rev/742c8c54c8fc
https://hg.mozilla.org/mozilla-central/rev/751208f98deb