Squashed Inodes on 4.2
notes at isucs1.UUCP
notes at isucs1.UUCP
Tue Oct 30 06:09:58 AEST 1984
After having several crashes in the area of _iget() in the
kernel, I did a pstat on the inode list. We seem to accumulate inodes
with very large reference counts. Apparantly the inode count is being
decremented past zero. These inodes sometimes have their links messed
up and that is what causes the system to crash. Does anybody else have
this problem and is there a fix? We are running 4.2BSD on a VAX 11/730,
but I have also seen the bad inodes on a VAX 11/780 running 4.2BSD.
Pstat reports the link count to be 255, but it is actually 65535 as
pstat trucates to byte size for output.
% pstat -i
68/212 active inodes
LOC FLAGS CNT DEVICE RDC WRC INO MODE NLK UID SIZE/DEV
80103540 24 11, 0 0 0 2 41ed 15 0 1024
80103690 T 1 11, 0 0 0 41 81ed 1 0 13312
(Intervening entries deleted)
80105a00 255 11, 7 0 0 16379 41ed 2 6 512
(Intervening entries deleted)
After seeing this from pstat, I used adb to look at the inode. Adb
reports that the count is negative(rolled over).
% adb -k /vmunix /dev/mem
sbr 80022c64 slr 19dc
p0br 802a9200 p0lr c4 p1br 7faa9600 p1lr 1ffff2
80105a00$<inode
80105a00: forw back
80105418 80109d98
80105a08: flag count min maj shc exc ino
0 -1 07 013 0 0 16379
80105a18: fs dquot
800c1400 8011a4f4
80105a20: lastr freef freeb
0 0 0
80105a28: mode links uid gid size
040755 2 6 10 512
80105a38: atime 1984 Oct 11 12:24:23
mtime 1984 Oct 23 16:43:16
ctime 1984 Oct 23 16:43:16
80105a50: addresses
80105a50: 882a 0 0 0
0 0 0 0
0 0 0 0
0 0 0
80105a8c: flags
0
As a final note, the system was initially running near the max users limit.
After I recompilied the system for five times as many users the number of
these corrupted inodes went down, but now I waste memory that I can't afford.
Please help as we only get about 8 days to a crash.
Jon Allen
Iowa State Universiy
...umn-cs!isucs1!EXNET!s1allen
More information about the Comp.unix.wizards
mailing list