NFS bug with 2.6.25 ?

Paul Lussier p.lussier at comcast.net
Tue Aug 5 08:54:41 EDT 2008


Hi all,

We've encountered a bug in 2.6.25 with sunrpc and the
nfs-kernel-server package.  In short, stopping/starting the
nfs-kernel-server daemon results in an over-decrement of the module
use count for sunrpc.  Here's an example:

   $ lsmod|grep sunrpc
   sunrpc                172672  35 nfsd,auth_rpcgss,nfs,lockd,nfs_acl
   $ sudo /etc/init.d/nfs-kernel-server stop

   $ lsmod|grep sunrpc
   sunrpc                172672  30 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

   $ sudo /etc/init.d/nfs-kernel-server stop
   $ lsmod|grep sunrpc
   sunrpc                172672  30 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

   $ sudo /etc/init.d/nfs-kernel-server start
   $ lsmod|grep sunrpc
   sunrpc                172672  33 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

   $ sudo /etc/init.d/nfs-kernel-server stop
   $ lsmod|grep sunrpc
   sunrpc                172672  28 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

   $ sudo /etc/init.d/nfs-kernel-server start
   $ lsmod|grep sunrpc
   sunrpc                172672  31 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

   $ sudo /etc/init.d/nfs-kernel-server stop
   $ lsmod|grep sunrpc
   sunrpc                172672  26 nfsd,auth_rpcgss,nfs,lockd,nfs_acl

This is on Debian/Etch for whatever that's worth.

This ends up eventually causing a negative use count.  We encountered
it when the use count hit -15:

  sunrpc 172928 4294967281 nfsd,auth_rpcgss,nfs,lockd,nfs_acl, Live 0xf8a94000

One of our programs tries to parse /proc/modules, and the field
containing 4294967281 was expected to be a 32bit int.  Oops.

Has anyone seen this before?  Googling seems to reveal nothing.
-- 
Seeya,
Paul


More information about the gnhlug-discuss mailing list