"Patrick J. LoPresti" wrote:
> > Alternatively there may be Yet Another Bug in the nfsd file handle
> > processing... the sad truth is that the file handle processing code in
> > ALL existing 2.2.X kernels is a Festering Heap of Wombat Turds.
>
> This is my guess, actually. knfsd seems somewhat fragile, especially
> on SMP systems (like our file server). Here are some later log
> messages from the same problem. We received several hundred thousand
> of these:
>
> Jan 12 16:31:36 kernel: lookup_by_inode: ino 1080899 not found in elaborate
> Jan 12 16:31:36 kernel: find_fh_dentry: 30:12/1080899 dir/1080856 not found!
Hmm, slightly different syndrome: we didn't find a dentry in the cache,
couldn't build a name tree up to the root: but in this case we appeared to find
a parent directory dentry in one of the (numerous, hideously overcomplex) local
caches in the existing nfsfh.c code... but then failed to find our inode's
number in that directory. Does smell like a rename race or something similar.
> > Neil Brown has done a heroic job of cleaning this up, and we hope to
> > get his fixes into a stable release as soon as possible: but his
> > fixes are tied in with other NFS3 stuff & needs a bit more airtime
> > before it's ready for a stable kernel....
>
> It would be nice if several people could audit whatever knfsd stuff
> shows up in the stable kernels, particularly with an eye towards
> concurrency issues. (It seems that all of our problem NFS boxes are
> SMP systems.)
Actually knfsd hasn't been changing all that much over the last several 2.2.X
releases. But a lot of work has been going on behind the scenes by Neil &
Trond. It's just a question of getting it merged, stabilized & sufficiently
tested.
> It would also be nice to see a working knfsd in kernel 2.2.x...
It's my top priority at this point. Alan Cox has agreed to accept a patch of
the merged latest "best of breed" once it's been reasonably checked out.
Dave Higgen
-
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/
This archive was generated by hypermail 2b29 : Sat Jan 15 2000 - 21:00:25 EST