New: kernel BUG at fs/ext4/inode.c:1063 if attempted to use non-ext4 partition with ext4

Quoted 8 lines · Rafael J. Wysocki

This message has been generated automatically as a part of a report of recent regressions.

The following bug entry is on the current list of known regressions from 2.6.32. Please verify if it still should be listed and let me know (either way).

Quoted 7 lines

Bug-Entry : http://bugzilla.kernel.org/show_bug.cgi?id=14936
Subject : kernel BUG at fs/ext4/inode.c:1063 if attempted to use non-ext4
partition with ext4 Submitter : Gene Heskett <[email protected]>
Date : 2009-12-26 14:13 (4 days old)
References : http://marc.info/?l=linux-kernel&m=126183682330809&w=4

		  http://marc.info/?l=linux-kernel&m=126169372613898&w=4

I would assume its still valid. Even though I have rebuilt 2.6.33-rc2 3 times now, it is so unstable, and so much stuff doesn't "Just Work" that I haven't rebooted to it, or to 2.6.32.2 in about a week now. USB and bluetooth are somewhat busted in 2.6.32.2. 2.6.32 final seems bulletproof here. The first thing that pops its ugly head up during either a 2.6.32.2, or a 2.6.33-rc2 boot is that it cannot find the firmware patch that is sitting in /lib/firmware for my phenom X4 cpu. The load attempt, it looks like, has been moved forward in the boot cycle, before it has actually able to access a file by the /lib/firmware path known after the filesystems are mounted. And a make xconfig no longer has the ability to edit the firmware inclusions in the kernel, so that is 2 bugs even before we get to the numerous oopsen that will bring it down in 15 minutes or so.

Quoted 7 lines · Rafael J. Wysocki

Bug-Entry : http://bugzilla.kernel.org/show_bug.cgi?id=14936
Subject : kernel BUG at fs/ext4/inode.c:1063 if attempted to use non-ext4 partition with ext4
Submitter : Gene Heskett <[email protected]>
Date : 2009-12-26 14:13 (4 days old)
References : http://marc.info/?l=linux-kernel&m=126183682330809&w=4

		  http://marc.info/?l=linux-kernel&m=126169372613898&w=4

Bisected to commit d21cd8, and reverting this commit seems to fix the problem. The problem appeared between -rc1 and -rc2. It was a fix that was pushed as part of a quota/ext4 fixes. For more information:

http://marc.info/?l=linux-ext4&m=126197233225936&w=2

I will take a look at this to see if I can find an easy fix; otherwise, we'll probably want to revert commit d21cd8 before -rc3.

Quoted 9 lines · On Tue, Dec 29, 2009 at 11:57:12AM -0500, [email protected] wrote:

Quoted 7 lines · Rafael J. Wysocki

Bug-Entry : http://bugzilla.kernel.org/show_bug.cgi?id=14936
Subject : kernel BUG at fs/ext4/inode.c:1063 if attempted to use non-ext4 partition with ext4
Submitter : Gene Heskett <[email protected]>
Date : 2009-12-26 14:13 (4 days old)
References : http://marc.info/?l=linux-kernel&m=126183682330809&w=4

		  http://marc.info/?l=linux-kernel&m=126169372613898&w=4

Quoted 8 lines

Bisected to commit d21cd8, and reverting this commit seems to fix the problem. The problem appeared between -rc1 and -rc2. It was a fix that was pushed as part of a quota/ext4 fixes. For more information:

http://marc.info/?l=linux-ext4&m=126197233225936&w=2

I will take a look at this to see if I can find an easy fix; otherwise, we'll probably want to revert commit d21cd8 before -rc3.

I'm working on a fix, and have a proposed patch. See the thread on linux-ext4 here:

http://marc.info/?l=linux-kernel&m=126219515024315&w=2

This patch also fixes the dq_claim_space kerneloops problem as well....

Quoted 9 lines · Rafael J. Wysocki

Quoted 7 lines

I'm working on a fix, and have a proposed patch. See the thread on linux-ext4 here:

http://marc.info/?l=linux-kernel&m=126219515024315&w=2

This patch also fixes the dq_claim_space kerneloops problem as well....

Thanks, I've added your patch to the Bugzilla entry.

This patch has been merged to mainline. I'm working on improving things so that we don't lose performance by unnecessarily forcing delalloc blocks out to disk when the (wildly overestimated) number of required metadata blocks approaches the user's remaining quota or remaining free space on disk, but the fundamental fix is in mainline.

(The performance hit was something I knew about when I pushed the patch to Linus, but after discussing things with Eric Sandeen, we figured that for users who were converting from ext3, which didn't have delayed allocation support at all, even with the performance hit when the user approached the quota limit, ext4 with delalloc+quota was still better than ext3 --- and I wanted a fix in mainline ASAP. Fortunately, it's not going to be that hard make the required metadata much more accurate.)