[59] in linux-net channel archive
nfs_refresh_inode: inode number mismatch
daemon@ATHENA.MIT.EDU (Werner Almesberger)
Fri Feb 3 17:19:22 1995
Date: Fri, 3 Feb 1995 21:41:39 +0100
From: almesber@di.epfl.ch (Werner Almesberger)
To: linux-net@vger.rutgers.edu
I've observed "nfs_refresh_inode: inode number mismatch" errors recently.
running 1.1.84 and 1.1.88. They frequently cause file corruption by
writing incomplete data to the file. Most of those errors don't cause
visible corruption, though.
The NFS client is a PC (i586/90, PCI, Neptune chipset, WD8013) running
Linux 1.1.84 or 1.1.88 with default rsize/wsize (I think it's 1024) and
8192. The NFS server is a SPARCserver 1000E running Solaris 2.4. Only
this client is accessing the files.
There are a few others PCs (same hardware, 1.1.68 kernel, rsize/wsize
always default) running off the same server which haven't had this error
yet, but they're also doing less IO.
All of the PCs are also getting
NFS: notice message: result=-11
errors, which seem to be harmless. (They occur whenever I SIGINT a
process writing to an NFS FS mounted with -o intr. Maybe they should
be suppressed in this case, because they might hide some real
problems.)
Before, when all the PCs were running 1.1.68 for about two months and
one of the PCs was acting as the file server for the others, no "inode
number mismatch" occurred. (Got a few "notice message: result=-11",
though.)
Unfortunately, it's hard to reproduce the error. Yesterday, I didn't
have a single "inode number mismatch". But today, it struck again
several times. I've also done a few unsystematic stress tests (disk IO,
CPU, net, NFS alone), which didn't yield anything interesting.
I suspect that these are either undetected network errors, some subtle
bugs in the Linux NFS code or a Solaris problem. Any ideas ?
- Werner
