411874 - Crash [@ ffff068b nsGIFDecoder2::EndImageFrame] while loading MySpace page (hard to repro)

Status: VERIFIED FIXED

(Core :: Graphics: ImageLib, defect)

Description

[Stephen Donner [:stephend] Not actively reading bugmail]

Summary: Crash [@ ffff068b nsGIFDecoder2::EndImageFrame] while loading MySpace page (hard to repro)

Vague Steps to Reproduce:

1. Load http://profile.myspace.com/index.cfm?fuseaction=user.viewprofile&friendid=32991097 while Pandora.com is also loading
2. Crash, sometimes (once?) here:

0  	@0xffff068b  	
1 	nsGIFDecoder2::EndImageFrame() 	mozilla/modules/libpr0n/decoders/gif/nsGIFDecoder2.cpp:399
2 	nsGIFDecoder2::Close() 	mozilla/modules/libpr0n/decoders/gif/nsGIFDecoder2.cpp:165
3 	imgRequest::OnStopRequest(nsIRequest*, nsISupports*, unsigned int) 	mozilla/modules/libpr0n/src/imgRequest.cpp:723
4 	nsHttpChannel::OnStopRequest(nsIRequest*, nsISupports*, unsigned int) 	mozilla/netwerk/protocol/http/src/nsHttpChannel.cpp:4412
5 	nsInputStreamPump::OnStateStop() 	mozilla/netwerk/base/src/nsInputStreamPump.cpp:576
6 	nsInputStreamPump::OnInputStreamReady(nsIAsyncInputStream*) 	mozilla/netwerk/base/src/nsInputStreamPump.cpp:401
7 	nsInputStreamReadyEvent::Run() 	mozilla/gfx/cairo/cairo/src/cairo-hull.c:107
8 	nsThread::ProcessNextEvent(int, int*) 	mozilla/xpcom/threads/nsThread.cpp:510
9 	NS_ProcessPendingEvents_P(nsIThread*, unsigned int) 	nsThreadUtils.cpp:180
10 	nsBaseAppShell::NativeEventCallback() 	mozilla/widget/src/xpwidgets/nsBaseAppShell.cpp:112
11 	nsAppShell::ProcessGeckoEvents(void*) 	mozilla/widget/src/cocoa/nsAppShell.mm:294

Comment 1

[Alfred Kayser]

The problem seems to be here:
 394 reed          1.91   // Only add frame if we have any rows at all
 395                      if (mGIFStruct.rows_remaining != mGIFStruct.height) {
 396                        if (mGIFStruct.rows_remaining && mGIFStruct.images_decoded) {
 397                          // Clear the remaining rows (only needed for the animation frames)
 398                          PRUint8 *rowp = mImageData + ((mGIFStruct.rows_remaining - mGIFStruct.height) * mGIFStruct.width);
 399                          memset(rowp, 0, mGIFStruct.rows_remaining * mGIFStruct.width);
 400                        }

In the calculation for the offset into the part that needs to be cleared, I did: mGIFStruct.rows_remaining - mGIFStruct.height
But actually rows_remaining is number of rows not filled by the frame (e.g. 2 rows not filled), and height is total number rows is 15.
This means mGIFStruct.rows_remaining - mGIFStruct.height results in a negative number (-13) which is clearly the wrong offset...

Patch coming up... (I don't have cvs access during the day)

Comment 2

[Alfred Kayser]

But it would help greatly if one could determine which .gif file is causing this problem. Both sites have a number of GIF's, and they are also different depending current advertising and/or location (For example, I got Dutch advertising...)

Comment 3

[Stephen Donner [:stephend] Not actively reading bugmail]

(In reply to comment #2)
> But it would help greatly if one could determine which .gif file is causing
> this problem. Both sites have a number of GIF's, and they are also different
> depending current advertising and/or location (For example, I got Dutch
> advertising...)

Alfred, I believe a somewhat-more-reliable testcase than I've given is the following:

1. Load http://profile.myspace.com/index.cfm?fuseaction=user.viewprofile&friendid=132672179 (different profile, crasher for me more often)
2. Click on the profile picture (http://viewmorepics.myspace.com/index.cfm?fuseaction=user.viewAlbums&friendID=132672179) before the background (http://img.photobucket.com/albums/v221/ashbernie311/welife/welife3/vintageglam_floral.gif), i.e., when the page has a white background.

Keep playing with that, and hopefully you'll crash :-)

Comment 4

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Sadly you need an account on myspace to find a reliable testcase. I don't have one so I cannot help ATM. :/

Comment 5

[Alfred Kayser]

Attachment: Patch to correct the offset calculation

With patch I cannot reproduce this crash anymore. (the current FF3 build does seem to crash to the testcase above).

Comment 6

[Mike Schroepfer]

Can we get this image into our testsuite Dolske?

Comment 7

[Justin Dolske [:Dolske]]

Was anyone able to narrow this down to a specific image? Maybe enable network logging to figure out what image url was loading right before the crash?

Comment 8

[Reed Loden [:reed]]

Checking in modules/libpr0n/decoders/gif/nsGIFDecoder2.cpp;
/cvsroot/mozilla/modules/libpr0n/decoders/gif/nsGIFDecoder2.cpp,v  <--  nsGIFDecoder2.cpp
new revision: 1.92; previous revision: 1.91
done

Comment 9

[Stephen Donner [:stephend] Not actively reading bugmail]

Just an update: dolske and I spent at least an hour each trying to isolate the crashing GIF, and weren't able to.

Comment 10

[Sander]

Suspicion in bug 412294 is that it was a duplicate of this bug. If so, I have isolated the responsible GIF there.

Comment 12

[Justin Dolske [:Dolske]]

Attachment: GIF from bug 412294

This is the GIF causing bug 412294 (duped to this bug).

It's a looping animated GIF, so it can probably only be used as a "doesn't crash when loaded" test.

Comment 13

[Justin Dolske [:Dolske]]

Hmm. I checked with a nightly (OS X, Gecko/2008011304) from before the patch landed, and that GIF doesn't cause a crash for me. Seems to load and display the same as in FF2.

Comment 14

[Henrik Skupin [:whimboo][⌚️UTC+2]]

I can reproduce a crash with the given steps in bug 412294 comment 3 with Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.4; en-US; rv:1.9b3pre) Gecko/2008011504 Minefield/3.0b3pre ID:2008011504 and it doesn't occur anymore in current nightly builds.

The stack looks like following: bp-3cb9a9d4-ccd6-11dc-a87f-001a4bd43ed6

But it doesn't look like that it has anything to do with this bug. I have isolated the gif which is another one as the latest attachment on this bug. Hitting shift-reload as fast as possible lets an nightly build crash.

Comment 15

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Attachment: Gif which crashes the browser when shift-reload

Comment 16

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Perhaps you have to save the image locally before you notice the crash! I'm not able to crash the browser while viewing the attachment?

Comment 17

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Alfred, this kind of crash still happens on Windows. You duped bug 412294 which is about RtlpCoalesceFreeBlocks against this bug. Reopening?

http://crash-stats.mozilla.com/report/list?range_unit=weeks&version=Firefox%3A3.0b3pre&range_value=2&signature=RtlpCoalesceFreeBlocks

Comment 18

[Alfred Kayser]

If you look at:
http://crash-stats.mozilla.com/report/list?range_unit=weeks&version=Firefox%3A3.0b3pre&range_value=2&signature=RtlpCoalesceFreeBlocks
and sort on 'build', you can see that the latest build from which these crash reports came, you can see: 2008020104
That is precisely the date when this patch was committed to the trunk.

So, the are no crash reports after that, so I still consider this bug and its dup(s) has fixed.

Comment 19

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Do you mean with "this patch" the patch on this bug? As I can see in comment 8 it was committed to trunk on 2008-01-18 and not 2008-02-01.

Comment 21

[Henrik Skupin [:whimboo][⌚️UTC+2]]

No crashes anymore since 2008-02-05. Verified with Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9b4pre) Gecko/2008021004 Minefield/3.0b4pre ID:2008021004