Created attachment 577852 [details] crash report 0. enable Java(TM) Platform SE 6 U29 1. enable popups 2. disable malware detector 3. http://www.sheru100.in/view/index585.html 4. shutdown http://cpm1.jmxy.com:899/cpm/tg_href.aspx?tg_d=www.ads80.com will be loaded which is reported as a malware attack site. Nightly/11 Windows 7 bp-f1059123-a28f-499c-9003-5ed772111129 Crash Report [@ mozalloc_abort(char const* const) | NS_DebugBreak_P | mozilla::plugins::PluginModuleChild::ShouldContinueFromReplyTimeout() ] See also bug 681385 Automation hit this with ecx pointing to deleted memory. I wasn't able to reproduce locally though. Operating system: Windows NT 6.1.7601 Service Pack 1 CPU: x86 GenuineIntel family 6 model 44 stepping 2 2 CPUs Crash reason: EXCEPTION_ACCESS_VIOLATION_WRITE Crash address: 0xffffffffddddde1d Thread 0 (crashed) 0 xul.dll!nsNPAPIPluginInstance::InvalidateOwner() [nsNPAPIPluginInstance.cpp : 1431 + 0x3] eip = 0x6f9ed2ea esp = 0x0031b128 ebp = 0x0031b12c ebx = 0x00000001 esi = 0x04ae00e8 edi = 0xffffff87 eax = 0xdddddddd ecx = 0xdddddddd edx = 0x00000001 efl = 0x00210246 Found by: given as instruction pointer in context 1 xul.dll!nsDummyJavaPluginOwner::Destroy() [nsGlobalWindow.cpp : 483 + 0x11] eip = 0x6f28450a esp = 0x0031b134 ebp = 0x0031b138 Found by: call frame info 2 xul.dll!nsGlobalWindow::FreeInnerObjects(int) [nsGlobalWindow.cpp : 1311 + 0x14] eip = 0x6f2874ff esp = 0x0031b140 ebp = 0x0031b168 Found by: call frame info 3 xul.dll!nsGlobalWindow::SetDocShell(nsIDocShell *) [nsGlobalWindow.cpp : 2397 + 0x10] eip = 0x6f28bde4 esp = 0x0031b170 ebp = 0x0031b1c8 Found by: call frame info 4 xul.dll!nsDocShell::Destroy() [nsDocShell.cpp : 4611 + 0x1c] eip = 0x6f82852a esp = 0x0031b1d0 ebp = 0x0031b22c Found by: call frame info 5 xul.dll!nsFrameLoader::Finalize() [nsFrameLoader.cpp : 573 + 0x18] eip = 0x6efbc3a1 esp = 0x0031b234 ebp = 0x0031b248 Found by: call frame info
Can we capture the testcase? If the site is already flagged as a malware site the contents might go away at any time.
I'll try again. It was late and the attempts I made using wget didn't seem to catch anything useful.
Created attachment 578507 [details] [diff] [review] Possible fix. It's unclear exactly how this crash happens, but I did find this inconsistency where we can create a dummy java plugin (which we create when accessing window.java n friends) and then navigate switch documents in the window w/o creating a new inner. In that case we end up with a dummy java plugin whose owner has a reference to the old document rather than the new one. This fixes that case, but I have not had a chance to test whether this fixes the problem here. Bob, any chance you can test this out? I can spin a try build if that helps.
dveditz was right that this might be fleeting. I couldn't get anything to reproduce except for some really high cpu usage. I can retry again. A try build would be cool, though I can build my own with the patch if needed. I'll get to this after I take care of some other things this morning.
Have we seen evidence of this on the 1.9.2 branch, or is it a relatively-new-build crash?
I have seen only once instance of a related crash with signature mozilla::plugins::PluginModuleChild::ShouldContinueFromReplyTimeout. Socorro shows 300 in the last week though: https://crash-stats.mozilla.com/query/query?do_query=1&query_type=contains&query=mozilla%3A%3Aplugins%3A%3APluginModuleChild%3A%3AShouldContinueFromReplyTimeout I have seen only one instance of a related crash with signature nsNPAPIPluginInstance::InvalidateOwner. There are no instances of this in Socorro. I haven't been able to reproduce locally. The original SocorroRecord where the crash automation obtained the url to test had signature mozilla::widget::WindowHook::Lookup(unsigned int) which has ~3000 crashes in the last week. https://crash-stats.mozilla.com/query/query?do_query=1&query_type=contains&query=mozilla%3A%3Awidget%3A%3AWindowHook%3A%3ALookup See 556524. There were two mozilla::widget::WindowHook::Lookup at sheru100 in Nov/Dec. I'll try to collect a set of urls to retest to see if I can get more information. Johnny, without more information want to go ahead and see if this helps in other ways? Perhaps the fix will affect other "unrelated" signatures.
Comment on attachment 578507 [details] [diff] [review] Possible fix. Yeah, I think we should just land this patch, requesting review.
Looks like this applies to 1.9.2 as well
Comment on attachment 578507 [details] [diff] [review] Possible fix. Review of attachment 578507 [details] [diff] [review]: ----------------------------------------------------------------- ::: dom/base/nsGlobalWindow.cpp @@ +2177,5 @@ > + // this window. > + mDummyJavaPluginOwner->Destroy(); > + mDummyJavaPluginOwner = nsnull; > + > + mDidInitJavaProperties = PR_FALSE; How come we don't set mDidInitJavaProperties to false in the other places where we destroy the dummy instance, like in "nsGlobalWindow::FreeInnerObjects"?
All other places that tear down the dummy java plugin we're at the very end of the lifetime of the window, so it doesn't matter. But for consistency's sake it wouldn't hurt to do the same there as well.
https://hg.mozilla.org/mozilla-central/rev/025fa101a51b Leaving this bug open though, as it's not clear whether this patch actually changes things here, so we should track this in crash-stats etc for a while before we can tell.
Probably not much chance of getting this into Fx10 at this point, but it is a safe contained fix. Are there enough crashes that we could decide in a couple of days?
If this patch did help then we should get it into the ESR at some point.
Setting the ESR tracking flag to 12+ since this hasn't yet been uplifted to 11.
Bob, any signs of this left? Or can we even tell?
I haven't seen any signs of this but it is hard to tell. I wonder if this is related at all to bug 724781
Ok, calling this fixed then. We can always reopen this bug if additional information emerges.
[Triage Comment] Can this patch be landed as is to ESR branch? Please nominate for approval-mozilla-esr10 if so. See https://wiki.mozilla.org/Release_Management/ESR_Landing_Process for details.
qa- until testcase-wanted is satisfied
Comment on attachment 578507 [details] [diff] [review] Possible fix. This applies to ESR10
Nothing much to verify here since we've never been able to reproduce the problem again.