xfs_repair segfaults

[Eric Sandeen]

On 3/8/13 4:21 AM, Ole Tange wrote:
> On Mon, Mar 4, 2013 at 4:20 PM, Eric Sandeen <sandeen at sandeen.net> wrote:
> 
>> 2) so you could run a "real" non-"n" xfs_repair on a metadata image as a more realistic dry run
> 
> I have now done a 'xfs_repair' using the code in GIT. It failed, and I
> then did 'xfs_repair -L' which succeeded.
> 
> Am I correct that I should now be able to mount the sparse disk-image
> file and see all the filenames? In that case I am quite worried. I get
> filenames like:
> 
> /mnt/disk/??5?z+hEOgl/?7?Psr1?aIH<?ip:??/>S??+??z=ozK/8_0/???d)
> 5JCG?eiBd?EVsNF'A?v?m?f;Fi6v)d>/?M%?A??J?)B<soGlc??QuY!e-<,6G?
> X[Df?Wm^[?f 4|

By default, xfs_metadump scrambles filenames, so nothing to worry
about (it's for privacy reasons).  If you use the "-o" option it'll keep
it in the clear.

-Eric

> My guess is some superblock is corrupt and that it should instead try
> a backup superblock. It might be useful if xfs_repair could do this
> automatically based on the rule of thumb that more than 90% of
> filenames/dirnames match:
> 
> [- _.,=A-Za-z0-9':]* [([{]* [- _.,=A-Za-z0-9':]* []})]* [- _.,=A-Za-z0-9':]*
> 
> If it finds a superblock resulting in more then 10% not matching the
> above it should probably ignore that superblock (unless the file names
> are using non-latin characters - such as Japanese).
> 
> 
> /Ole
>