Bugzilla@Mozilla – Bug 623791
[ANGLE] WebGLES shader assertion failed: (oldhashloc >= 0), function IncreaseHashTableSize
Last modified: 2011-05-18 22:01:17 PDT
Summon comment box
Created attachment 501885 [details] callstack
Created attachment 501886 [details] testcase
Can you please file this in the ANGLE issue tracker as well? Thanks
We have fixed this issue in ANGLE revision 605. Mozilla still crashes (under windows), but no longer in ANGLE code, so we expect that the validator must be statically compiled in and needs to be updated to verify this.
The upstream bug says A critical parser bug was fixed in r605. When loading the shader from a file, Chrome previously crashed but now properly reports the shader compilation failures. http://code.google.com/p/angleproject/issues/detail?id=117#c2 The patch fixes a buffer overflow http://code.google.com/p/angleproject/source/detail?r=605 This was also fixed in Chrome so we need to get it into Macaw asap
Chrome fixed an additional "off by three" overwrite in libGLESv2 http://code.google.com/p/angleproject/source/detail?r=611
This will be fixed with the patches in bug 649233.
This landed on mozilla-central. Going to land these patches on mozilla-2.0 once they are approved in bug 649233.
Credit for the additional security fix the Chrome team reported to us (comment 6) should go to "yuri.ko616". It wasn't actually reported as a security issue, just a crash.
The additional fix for yuri.ko616's bug is tracked by Google at http://code.google.com/p/chromium/issues/detail?id=70070 (hidden security bug) Christophe's original bug is tracked by them at http://code.google.com/p/chromium/issues/detail?id=78524 Announced (without detail) on 4/14 at http://googlechromereleases.blogspot.com/2011/04/stable-channel-update.html
Actually, the guy wanted to be credited as "Yuri Ko" I messed up in the Chrome release notes.
I was pinged for some additional information on why we landed the ANGLE updated the way we did. The Chrome maintainers created a chrome_m10 branch in upstream ANGLE, and cherrypicked the changes they wanted. They branched off ANGLE r551, but the last time we updated was ANGLE r550. So, in order to stay 100% current with the chrome_m10 branch, we applied both r551 and r611, the two changesets that branch had that mozilla-central didn't have.
Updating the CVE number: the Chrome security team already assigned CVE-2011-1302 to Christophe's bug and MITRE has duped -0068 to that one. Yuri's bug was CVE-2011-1300
I apologize for repeatedly misspelling Christoph's name.
No problem. :)