> 1.2.2-4?? That is ancient, and there was a bug in knfsd's rmdir
> operation which could cause exactly these symptoms. Please upgrade to
> knfsd-1.4.7 and see if you can reprodce the problem.
Well I did this yesterday night an upgraded to knfsd-1.4.7-7 and
kernel 2.2.13 as Alan Cox suggested. But there again the very first
backup run yielded:
Oct 29 01:59:37 iwr01 kernel: attempt to access beyond end of device
Oct 29 01:59:37 iwr01 kernel: 08:11: rw=0, want=2045247456, limit=8924076
Oct 29 01:59:37 iwr01 kernel: dev 08:11 blksize=1024 blocknr=-102236193 sector=-204472386 size=1024 count=1
Oct 29 01:59:37 iwr01 kernel: attempt to access beyond end of device
Oct 29 01:59:37 iwr01 kernel: 08:11: rw=0, want=1879039744, limit=8924076
Oct 29 01:59:37 iwr01 kernel: dev 08:11 blksize=1024 blocknr=-268443905 sector=-536887810 size=1024 count=1
(Btw, what do the negative values mean, or it is just an integer overflow?)
> We know that (a) dodgy hardware
I don't think that my problem is in hardware. I suspected my hardware
for quite a while. Other boxes with nearly identical hardware run rock
solid for some months now.
> and (b) the knfsd problem can explain
> huge numbers of these reports,
Well. I through out knsfd and installed the nfs-server-2.2beta40-1
package and at *very* first glance things seem to be better now. At
least the first two backup runs were fine. I keep my fingers crossed,
though, since the above errors occurred only from time to time.
> but it's hard to completely exclude the
> possibility of other faults in the current kernels. It would really
> help to eliminate variables if you could go to a more recent knfsd.
Well, I did that. That box is not my playground, but the central
server of our (small, but working) group that I cannot shutdown at
will.
For this machine, stable operation is of much more importance
than the last 20% performance, so I won't tweek it if not really
necessary.
I tried to reproduce the effects on a (client) machine without nfs
server and tar'ed both to /dev/null and to real tape, but a handfull
of runs could not reproduce it. Dammned thing.
Would it help to have two playground boxes set aside? Is there such a
thing like a nfs-client-simulator? I reaaly do not want to abuse my
nonvirtual users for that purpose.
regards,
-- Martin Schulz schulz@iwrmm.math.uni-karlsruhe.de Uni Karlsruhe, Institut f. wissenschaftliches Rechnen u. math. Modellbildung Engesser Str. 6, 76128 Karlsruhe- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/