380967 - "ASSERTION: nsNSSComponent relies on profile manager to wait for synchronous shutdown of all network activity" starting firefox on clean profile

Status: RESOLVED WORKSFORME

(Core :: Security: PSM, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I see an PSM assertion when starting Firefox on a new profile.

Steps to reproduce:
 cd /path/to/firefox
 mkdir profiledir
 ./firefox -no-remote -profile profiledir

I get the following assertion, during the shutdown before the extension-manager restart (so it's essentially an execution of Mozilla in which we did relatively little):

###!!! ASSERTION: nsNSSComponent relies on profile manager to wait for synchronous shutdown of all network activity: 'mIsNetworkDown', file /builds/trunk/mozilla/security/manager/ssl/src/
nsNSSComponent::DoProfileBeforeChange(nsISupports*) (/builds/trunk/mozilla/security/manager/ssl/src/nsNSSComponent.cpp:2236)
nsNSSComponent::Observe(nsISupports*, char const*, unsigned short const*) (/builds/trunk/mozilla/security/manager/ssl/src/nsNSSComponent.cpp:1878)
nsObserverList::NotifyObservers(nsISupports*, char const*, unsigned short const*) (/builds/trunk/mozilla/xpcom/ds/nsObserverList.cpp:127)
nsObserverService::NotifyObservers(nsISupports*, char const*, unsigned short const*) (/builds/trunk/mozilla/xpcom/ds/nsObserverService.cpp:184)
nsXREDirProvider::DoShutdown() (/builds/trunk/mozilla/toolkit/xre/nsXREDirProvider.cpp:765)
~ScopedXPCOMStartup (/builds/trunk/mozilla/toolkit/xre/nsAppRunner.cpp:793)
XRE_main (/builds/trunk/mozilla/toolkit/xre/nsAppRunner.cpp:2868)
main (/builds/trunk/mozilla/browser/app/nsBrowserApp.cpp:66)
__libc_start_main (/usr/src/debug/glibc-2.5-20061008T1257/csu/libc-start.c:262)

Comment 1

[Frank Wein [:mcsmurf]]

This assertion was added as part of Bug 379582.

Comment 2

[Kai Engert (:KaiE:)]

CC'ing Boris, as Frank said the assertion was caused by his patch from bug 379582.

Comment 3

[Boris Zbarsky [:bzbarsky]]

I just moved an existing assertion, sorry.  David, what topic is being passed to Observe() in this case?  That would at least tell us which call to DoProfileBeforeChange we're looking at.

That said, it sounds like we just never saw the PROFILE_CHANGE_NET_TEARDOWN_TOPIC and are now getting a PROFILE_BEFORE_CHANGE_TOPIC.  Or that we saw the former and got reinitialized before we saw the latter.

Now everyone in this bug knows at least as much as I do about this code.  ;)

Comment 4

[Mats Palmgren (inactive)]

Attachment: stack + data

I bumped into this when looking for something else in gdb...
The topic was "profile-before-change".

Comment 5

[Kai Engert (:KaiE:)]

I can't reproduce this any more (trunk, linux).