Steve Lord wrote:
>
> >
> > Are there significant changes to xfs_repair later than those of
> > 2.4.6-pre1-xfs?
> >
> > D. Stimits, stimits@xxxxxxxxxx
>
> No, repair itself has not changed, the messages you got about all the devices
> on your system having gone away make me nervous. I have not had a chance to
> read this thread yet.
>
> Steve
I will look at some things later tonight, but it may be of interest that
Redhat's kudzu does not distinguish between devices "missing" and having
changed. As an example, depending on whether I run with APIC or not, the
irq of the modem changes. Each time it changes, it says it has detected
new or missing hardware, and asks to configure it. In this case, it
wanted to reconfigure all pci slot cards, video, sound, modem, and even
the integrated scsi controller and psaux mouse. It is a fact that I had
previously removed and placed a drive back in its bay (it is a hot swap
bay, it never gets removed/added with power on though), to test another
setup that is completely independent (pull out one drive, put in the
other, test, swap back). I have never seen it do this with all things,
and in theory the configuration was 100% unchanged, since the drive was
pulled and added back without power. But mechanical device connections
are always suspect to me, I tend doubt them if an error occurs after
pulling and adding a drive back ends up with such an error. On the other
hand, I am absolutely positive that it was correctly/fully seated, it
uses a positive-lock mechanism on a high-end hot swap backplane, and
scsi was in fact detected correctly...the strangeness is that it said
all configurations changed. On another note, it was shut down correctly
with the normal shutdown sequence before this. One really big concern I
have is that had I not created an IDE hard drive install (I also have a
removable IDE tray...I can boot to that by setting the bios to boot IDE
first instead of integrated SCSI), I would not have been able to run
xfs_repair (so far all boot floppy systems have failed for me, I'm still
working on it). Although I cannot guarantee that anything was actually
wrong with the filesystem (and I doubt there was a filesystem error), it
did find one node that ended up in lost+found, but it was size 0...I
have no idea if normal journaling would have had any problems coping
with it. Is there no way to run xfs_repair with "no actual modify by -n"
on a mounted xfs partition, or read-only mounted xfs partition?
D. Stimits, stimits@xxxxxxxxxx
|