From: Stuart Anderson <anderson@ligo.caltech.edu>
Subject: Re: kernel Oops in rpc.mountd
Date: Sat, 12 Feb 2005 14:23:01 -0800
Message-ID: <200502122223.j1CMN1sK003668@m27.ligo.caltech.edu>
Cc: nfs@lists.sourceforge.net
To: neilb@cse.unsw.edu.au
Sender: nfs-admin@lists.sourceforge.net
Errors-To: nfs-admin@lists.sourceforge.net

Neil,
	This patch did the trick! We have now run our 290 cluster nodes for
72 hours with this extra locking patch without any kernel crashes. This is
to be compared to 1-5 crashes per day before the patch.

	Many thanks!

	What is the next step in getting this patch integrated into FC3
and the mainline kernel branch?

According to Neil Brown:
> On Monday February 7, anderson@ligo.caltech.edu wrote:
> > We are going to rebuild 2.6.10-1.760_FC3smp with 8k stack (just my
> > paranoia about the 4k stacks), and remove absolutely everything we do
> > not need, however, I will add in the CONFIG_DEBUG_SLAB. Are there any
> > other kernel debug flags that might be helpful?
>
> No.  However the following patch might be worth a try.
> It adds some extra locking.  I'm pretty sure there is a race that this
> closes that could possibly cause your problem.  I also think the
> locking here is a bit heavy handed, but I am sure it is safe.
>
> >
> > Perhaps the bug is due to having a large list of static NFS mounts (290)?
>
> Having lots of mounts won't hurt.  But having lots of clients mount
> this server might have make it more likely to trigger the bug as there
> is more activity on the export cache.
>
> NeilBrown
>
>
> Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
>
> ### Diffstat output
>  ./include/linux/sunrpc/cache.h |   14 +++++++++-----
>  1 files changed, 9 insertions(+), 5 deletions(-)
>
> diff ./include/linux/sunrpc/cache.h~current~ ./include/linux/sunrpc/cache.h
> --- ./include/linux/sunrpc/cache.h~current~   2005-02-08 16:23:21.000000000 +1
100
> +++ ./include/linux/sunrpc/cache.h    2005-02-08 16:41:09.000000000 +1100
> @@ -268,15 +268,19 @@ static inline struct cache_head  *cache_
>
>  static inline int cache_put(struct cache_head *h, struct cache_detail *cd)
>  {
> -     atomic_dec(&h->refcnt);
> +     int rv = 0;
> +     read_lock(&cd->hash_lock);
> +     if (atomic_dec_and_test(&h->refcnt))
> +             rv = 1;
>       if (!atomic_read(&h->refcnt) &&
>           h->expiry_time < cd->nextcheck)
>               cd->nextcheck = h->expiry_time;
> -     if (!test_bit(CACHE_HASHED, &h->flags) &&
> -         !atomic_read(&h->refcnt))
> -             return 1;
> +     if (test_bit(CACHE_HASHED, &h->flags) ||
> +         atomic_read(&h->refcnt))
> +             rv = 0;
> +     read_unlock(&cd->hash_lock);
>
> -     return 0;
> +     return rv;
>  }
>
>  extern void cache_init(struct cache_head *h);
>


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs