Subject: truncated inode, and painful death
To: None <firstname.lastname@example.org>
From: Wolfgang Rupprecht <email@example.com>
Date: 01/26/2001 09:37:31
Can softdeps on the source partition of a massive tar-gzip ever cause
corruption on the destination partition? I was trying to tar and
compress 30 gigs of ascii data files into an estimated 4-gig tar.gz
file into my home directory.
Without softdeps on the source side things were only going at a
megabyte or two per second. I figured the access time updates were
slowing me so I unmounted the source and remounted it with softdeps.
Things did pick up a bit. After 20 minutes or so I knew something
strange was happening. My xterms and emacs started closing (exiting)
on their own. Restarting them got me an "libXaw.so.xx not found".
Oh-oh. Quick reboot attempt. Hang. Try DDB. Nope. Wait a minute
just in case. Nope. Power button time.
Upon reboot fsck asked for assistance. ("I'd like to newfs your disk
one inode at a time. ok? [yn]") The initial gripe was a truncated
inode. After fixing that it moved onto a never ending stream of DUP
inodes, intermixed with the occasional digital joke. ("CANNOT SEEK:
BLK -2, THE FOLLOWING DISK SECTORS COULD NOT BE READ: -2, -1") I let
fsck run for half an hour watching the bad news scroll by screen after
screen. I finally bored of that game and went multiuser without my
So the question is, is this the sort of thing other folks are seeing
with softdeps under extreme load? I can't rule out hardware, but up
to this point things have been somewhat calm on my system ever since I
stopped using the Promise ultra-100 hardware on my motherboard and I
got a 400Watt power supply.
Wolfgang Rupprecht <firstname.lastname@example.org>
Coming soon: GPS mapping tools for Open Systems. http://www.gnomad-mapping.com/