Re: [4.7-rc6 ext3 BUG] kernel BUG at fs/ext4/xattr.c:1331

[Dave Chinner <david@fromorbit.com>]

On Thu, Jul 28, 2016 at 03:54:32PM +0200, Jan Kara wrote:
> On Mon 18-07-16 15:24:47, Dave Chinner wrote:
> > On Sun, Jul 17, 2016 at 09:07:16PM -0700, Eric Sandeen wrote:
> > > On 7/17/16 8:02 PM, Dave Chinner wrote:
> > > > # rm !$
> > > > rm /mnt/scratch/fsr_test_file.27768.14.6
> > > > #
> > > > 
> > > > And, by removing an attribute, I can successfully remove the file.
> > > > So this definitely looks like a corner case xattr handling issue in
> > > > ext3/4.
> > > 
> > > I told xfs/227 that it could run on ext3 and ran it, but this
> > > didn't reproduce for me.
> > > 
> > > Can you provide a dumpe2fs -h for the root fs, this might depend on
> > > inode size etc.
> > 
> > # dumpe2fs -h /dev/sda1
> > dumpe2fs 1.43-WIP (18-May-2015)
> > Filesystem volume name:   <none>
> > Last mounted on:          /
> > Filesystem UUID:          b21615e5-fe8a-4ffc-ab80-c24cdc8b740a
> > Filesystem magic number:  0xEF53
> > Filesystem revision #:    1 (dynamic)
> > Filesystem features:      has_journal ext_attr resize_inode dir_index filetype needs_recovery sparse_super large_file
> > Filesystem flags:         signed_directory_hash 
> > Default mount options:    (none)
> > Filesystem state:         clean
> > Errors behavior:          Continue
> > Filesystem OS type:       Linux
> > Inode count:              624624
> > Block count:              2496091
> > Reserved block count:     124804
> > Free blocks:              567319
> > Free inodes:              352653
> > First block:              0
> > Block size:               4096
> > Fragment size:            4096
> > Reserved GDT blocks:      609
> > Blocks per group:         32768
> > Fragments per group:      32768
> > Inodes per group:         8112
> > Inode blocks per group:   507
> > Filesystem created:       Thu Mar 25 18:10:55 2010
> > Last mount time:          Tue Jul 19 01:21:57 2016
> > Last write time:          Tue Jul 19 01:21:57 2016
> > Mount count:              10
> > Maximum mount count:      27
> > Last checked:             Mon Jul 18 21:59:01 2016
> > Check interval:           15552000 (6 months)
> > Next check after:         Sat Jan 14 22:59:01 2017
> > Lifetime writes:          13 GB
> > Reserved blocks uid:      0 (user root)
> > Reserved blocks gid:      0 (group root)
> > First inode:              11
> > Inode size:               256
> > Required extra isize:     28
> > Desired extra isize:      28
> > Journal inode:            8
> > First orphan inode:       219355
> > Default directory hash:   half_md4
> > Directory Hash Seed:      740ffa95-af8d-4e89-b68c-5e768a27ece3
> > Journal backup:           inode blocks
> > Journal features:         journal_incompat_revoke
> > Journal size:             128M
> > Journal length:           32768
> > Journal sequence:         0x01c975b5
> > Journal start:            12
> 
> Thanks for report! So I see at least part of what happened: Your filesystem
> was created with 'extra inode size' 28 and likely your inodes were created
> with this amount of space reserved in the extended attribute area of the
> inode because you still created them with some older kernel (but that means
> that it had to be a kernel prior to commit 8b4953e13f4c which landed in
> 4.4-rc5 because newer kernels would automatically reserve 32-bytes in the
> inode, not 28 as specified by the superblock).

Well, yes, the filesystems were made prior to 4.4.-rc5. Only by a
little - it was made back in January 2010 and has been in use ever
since. :P

> The above mentioned commit has added project ID to the inode so new kernels
> now ask for 32 bytes in the extended attribute area. So when you tried to
> modify the inode with newer kernel, we were trying to shift extended
> attributes around to make space for those additional 4 bytes. So that makes
> it clear why Eric was not able to reproduce the issue.

Gotcha.

> I've tried creating file with an old kernel and deleting it with a new one
> and the bugon indeed triggers. Going through ext4_expand_extra_isize_ea() I
> see so many bugs that it's not nice. I guess we should add some inode size
> expansion tests...

Ouch. At least the problem is understood now - any idea on how long
it might take to fix?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com