Closed Bug 588931 Opened 9 years ago Closed 9 years ago
Crash in [@ nss
Trust Domain _Lock Cert Cache ]
Seen while sifting through trunk crash stats. http://tinyurl.com/332aa82 links to all the crashes. No comments in any of the reports. Crash data indicates it started showing up in 2010081800 builds and there have been a few using today's nightly as well. Apologies in advance if I selected the incorrect location for this bug. Frame Module Signature [Expand] Source 0 libnss3.dylib nssTrustDomain_LockCertCache tdcache.c:405 1 libnss3.dylib nssCertificate_Destroy certificate.c:144 2 libssl3.dylib ssl3_CleanupPeerCerts ssl3con.c:7744 3 libssl3.dylib ssl3_DestroySSL3Info ssl3con.c:9452 4 libssl3.dylib ssl_DestroySocketContents sslsock.c:407 5 libssl3.dylib ssl_FreeSocket sslsock.c:478 6 libssl3.dylib ssl_DefClose ssldef.c:233 7 XUL nsNSSSocketInfo::CloseSocketAndDestroy security/manager/ssl/src/nsNSSIOLayer.cpp:1825 8 XUL nsSSLThread::requestClose security/manager/ssl/src/nsSSLThread.cpp:437 9 XUL nsSSLIOLayerClose security/manager/ssl/src/nsNSSIOLayer.cpp:1814 10 XUL nsSocketTransport::OnSocketDetached netwerk/base/src/nsSocketTransport2.cpp:1395 11 XUL nsSocketTransportService::DetachSocket netwerk/base/src/nsSocketTransportService2.cpp:187 12 XUL nsSocketTransportService::DoPollIteration netwerk/base/src/nsSocketTransportService2.cpp:658 13 XUL nsSocketTransportService::OnProcessNextEvent netwerk/base/src/nsSocketTransportService2.cpp:543 14 XUL nsThread::ProcessNextEvent xpcom/threads/nsThread.cpp:517 15 XUL NS_ProcessPendingEvents_P nsThreadUtils.cpp:200 16 XUL nsSocketTransportService::Run netwerk/base/src/nsSocketTransportService2.cpp:579 17 XUL nsThread::ProcessNextEvent xpcom/threads/nsThread.cpp:547 18 XUL NS_ProcessNextEvent_P nsThreadUtils.cpp:250 19 XUL nsThread::ThreadFunc xpcom/threads/nsThread.cpp:263 20 libnspr4.dylib _pt_root nsprpub/pr/src/pthreads/ptthread.c:228 21 libSystem.B.dylib _pthread_start 22 libSystem.B.dylib thread_start
I believe that this occurs after Firefox sync connects to server. Before this crash, firefox shows "The operation can not be completed because of an internal failure. A secure network communication has not been cleaned up correctly." dialog message.
Oh, so this is due to the browser shutting down NSS while some SSL socket is still open, then trying to shut down that SSL socket. I thought PSM was supposed to prevent that somehow. (?) Most likely I'll give this bug to PSM.
#2 top crash in early firefox 4.0b5 data
Assignee: nobody → nobody
Component: Libraries → Security: PSM
Product: NSS → Core
QA Contact: libraries → psm
Whiteboard: See comment 2
Version: trunk → Trunk
I've seen this twice today - both times it was after installing b5 after b4 had been open, I speculate based on tabs to yahoo and gmail - it looked like it tried to restore the tabs after install, but something went really awry. Opened a blank browser and couldn't do anything until program shutdown. After restart, seemed to be fine. Error box said "The operation can not be completed because of an internal failure. A secure network communication has not been cleaned up correctly."
Who can we talk to about this? It looks like someone is suddenly shutting down NSS. Both this one and bug 517615 . Maybe we can put an assert in shutdown and find out who is causing this? bob
(of course we should comment out the shutdown call in PSM). Do we have any indication if these are happening when we are trying to shutdown the browser. Is there some new profile switching feature in FF 4.0? bob
Here are some user comments from the crash reports: "It restarted after applying the update to b5 and it promptly crashed." "Running firefox from remote desktop" "Updated to the lastest version of 4b5. Crashes after trying to restore previous session." I have never hit this crash stack myself, but I have definitely seen the error message in Bug 588511 when testing on trunk Win XP and Vista, and this is after I had applied a software update.
Bob, As you know, FF is still using cert8.db, the Berkeley DB version. IIRC, that DB has the property that, whenever the DB is enlarged, it MUST be closed properly or it becomes corrupt. It's not enough to force the flush of the dirty data pages (as we do before each softoken operation completes) when the DB enlarges, because this does not write the DB header (page 0), IIRC. Only file close does that, IIRC. So, I'm worried that removing the NSS_Shutdown call will guarantee the creation of corrupt cert8.DBs in new profiles.
I could reproduce this reliably a few days ago with these steps: - Run Firefox 4 (with some add-ons) and close it - Run Firefox 3.6 with same profile (it will check for add-on compatibility) and close it - Run Firefox 4 again with same profile. After checking add-on compatibility again, bug 588511 is shown. After clicking Ok the browser opens (sometimes it can't load any data from the web) and after a few seconds it crashed. Today with the same steps I've been unable to reproduce the crash, although I still get webpages failing to load after bug 588511 happens
mconner/ed, can you have a look from the sync side? other comments talk about update invovled, but maybe that is just a by stander.
I'm betting this is tied to bug 588511. We're way way out of the startup path on trunk (no more binary component) so I'm betting it's tied to the addon manager invoking NSS really early.
Agreed -- haven't seen this since the fix for bug 588511 landed (was first in the sep 15 nightly). DUP!
Status: NEW → RESOLVED
Closed: 9 years ago
Resolution: --- → DUPLICATE
Duplicate of bug: 588511
Crash Signature: [@ nssTrustDomain_LockCertCache ]
You need to log in before you can comment on or make changes to this bug.