1342980 - ASan Firefox processes hanging around

Status: RESOLVED WORKSFORME

(Core :: XPCOM, defect)

Description

[Tyson Smith [:tsmith]]

Attachment: hang_bt.txt

When launching Firefox via Python using subprocess.Popen() we see Firefox process hanging around. We call terminate() (and also tried kill()) and when calling wait() on the process it returns a return code as if the process has terminated. Checking the running processes there is still a process hanging around. This happens intermittently and we cannot consistently reproduce. It seems to happen often on our fuzzing machines.

truber was able to get a back trace from one of the hanging processes. So far we have seen this on ASan opts builds with e10s disabled. It seem similar to bug 1341621

Comment 1

[Andrew McCreight [:mccr8]]

Just based on the stack, it looks more like an ASan bug, but I figured we should have a bugzilla bug for this issue.

Comment 2

[Eric Rahm [:erahm]]

(In reply to Andrew McCreight [:mccr8] from comment #1)
> Just based on the stack, it looks more like an ASan bug, but I figured we
> should have a bugzilla bug for this issue.

Arguably this is a bug in both ASAN and in glib when fork/execing processes. They could just change their file descriptor closing code to brute force rather than enumerating the proc directory (and allocating).

Comment 3

[Eric Rahm [:erahm]]

(In reply to Eric Rahm [:erahm] from comment #2)
> (In reply to Andrew McCreight [:mccr8] from comment #1)
> > Just based on the stack, it looks more like an ASan bug, but I figured we
> > should have a bugzilla bug for this issue.
> 
> Arguably this is a bug in both ASAN and in glib when fork/execing processes.
> They could just change their file descriptor closing code to brute force
> rather than enumerating the proc directory (and allocating).

glib already has a bug for this [1], I'll add links to our Fx bugs.

[1] https://bugzilla.gnome.org/show_bug.cgi?id=738620

Comment 4

[Eric Rahm [:erahm]]

Bug filed for ASAN [1] as well.

[1] https://github.com/google/sanitizers/issues/774

Comment 5

[Nathan Froyd [:froydnj]]

The upstream asan bug is fixed, I'm going to call this fixed as well.