On Saturday 24 June 2006 1:02 am, Johannes Berg wrote:
> I got a corrupted xfs device that printed out, when trying to recover:
> Bad inode magic number, ....
> Internal error xlog_recover_do_inode_trans(1) ...
Looks like the inode item in the transaction is bad in the ondisk log.
Did you mount and replay the log with a different word-size linux by any
chance? I.e. replay with 32 bit linux having crashed on a 64 bit linux or
vice versa? (Seen a few of those lately)
If it was that case, then this is a known bug and now handled in recovery with
a new kernel. Normally, though that bug is shown slightly ealier when it
tries to read from a bad inode offset IIRC.
It would be interesting to see the ondisk log which can be saved using
# xfs_logprint -C filename devicename
However, "xfs_repair -L" will have zeroed yours out now.