On Wed, 2002-07-31 at 17:23, Christian Thalinger wrote:
> I'm just evaluating the best fs for my purposes and tried xfs under
> power failure conditions. Booted into linux with xfs as root partition
> and copied a linux kernel source tree into another directory on the same
> partition. While it was copying i just switched off the power and tried
> to get it up again. This is what i did get:
> Jul 31 23:33:30 (none) kernel: XFS mounting filesystem ide2(33,4)
> Jul 31 23:33:30 (none) kernel: Starting XFS recovery on filesystem:
> ide2(33,4) (dev: 33/4)
> Jul 31 23:33:31 (none) kernel: cmn_err level 1 Filesystem "ide2(33,4)":
> xfs_inode_recover: Bad inode magic number, dino ptr = 0xddc01100, dino
> bp = 0xddc33c80, ino = 25766753
> Jul 31 23:33:31 (none) kernel: XFS: log mount/recovery failed
> Jul 31 23:33:31 (none) kernel: XFS: log mount failed
> I could only mount this partition again after doing `xfs_repair -L'.
> Is there a way to have a xfs root partition and to get the machine up
> again, in most cases, without user interaction?
So are you running IDE with write caching turned on? Any journaled
filesystem is going to have issues with a write cache enabled
drive if you drop the power on it. In order to work a journaled
file system has ordering constraints between log writes and
metadata writes. A drive write cache can cause these constraints
to be broken - a metadata write was probably still in cache and
the relevant log space was overwritten on disk.