On 9 Dec 2008, at 17:49, Paul Hartman wrote:
...
xfs_repair exists, but the man page says it can only be used against
clealy unmounted drives... which seems kind of funny. I ran it anyway,
in read-only mode, and it was unable to find the primary superblock
and started scanning the disk for the secondary. I aborted it because
I don't know how much torture this drive can take.
When I run "less -f /dev/sdf1" i can see amongst the bits all of the
names of my directories and files etc... so I am pretty sure the data
is there and readable, but I can't figure out how to get XFS to go
ahead and mount it so I can copy all of that stuff off and swear off
USB hard drive adapters forever. :)
I would probably `dd` the whole drive to a new disk.
You want to copy everything off one way or the other, so the minimum
amount of torture you need to inflict upon the drive is reading every
bit from it once. Ok, you probably only want to read *some* bits of
the disk, so you only really need to read x% of that, but with an
uncertain recovery strategy you may find that you need to read some
bits of the fisk more than once.
If the drive can survive being completely read once, then `dd` allows
you to do what the heck you like with the image whilst you make
different attempts. I would personally try to find two drives - one at
least as large as the failing one and one a bit larger. Format the
larger disk ext3 and `dd if=/dev/failing-disk of=/mnt/largest-disk/
image.img` then `dd if=/mnt/largest-disk/image.img of=/dev/at-least-as-
large-disk`.
You can then work on the disk that is "at least as large", and even if
xfs_repair (or whatever) fails & b0rks the filesystem up even worse,
you can always recreate the last known "good" state.
Stroller.
