Description of problem: In BZ 204638, dchapman, esandeen, agriffis and others have been chasing severe memory corruption on all platforms when using squashfs. We have chased this down to an issue with the filesystem using fragments. Avoiding fragments until squashfs is patched and fixed is a good idea and will increase stability in the long run. Version-Release number of selected component (if applicable): FC6-test3/20060913 How reproducible: 100% Steps to Reproduce: 1. Attempt to boot any arch FC6-test3 or rawhide 20060913. 2. Occasionally corruption will be seen. It varies from arch-to-arch, with ia64 being the most severely hit. Actual results: Various panics can be seen in 204638 Expected results: No memory corruption should occur. Additional info: Patches are currently being generated. We are continuing to discuss with Phillip, but at least the fix for anaconda will allow us to continue installing. The size difference between the two images is relatively minor (stage2.img was grabbed from a recent rawhide build. nofrag2.img was built with the -no-frag option using the files from stage2.img): ls -l nofrag2.img stage2.img -rwx------ 1 root root 95625216 Sep 14 12:09 nofrag2.img -rw-r--r-- 1 root root 94273536 Sep 13 14:32 stage2.img
Created attachment 136282 [details] Patch to add -no-fragments to mksquashfs calls
Comment #1: "but at least the fix for anaconda" should read "but at least the WORKAROUND for anaconda" P.
Should point out that this was tested on ppc, ia64, and x86_64. P.
Applied
*** Bug 206530 has been marked as a duplicate of this bug. ***