Messages in this thread | | | From | "Myklebust, Trond" <> | Subject | Re: [PATCH v2] SUNRPC: skip dead but not buried clients on PipeFS events | Date | Wed, 25 Apr 2012 18:54:55 +0000 |
| |
On Wed, 2012-04-25 at 13:30 -0400, J. Bruce Fields wrote: > On Fri, Apr 20, 2012 at 06:11:02PM +0400, Stanislav Kinsbursky wrote: > > v2: atomic_inc_return() was replaced by atomic_inc_not_zero(). > > > > These clients can't be safely dereferenced if their counter in 0. > > I'm pretty confused by how these notifiers work.... > > rpc_release_client decrements cl_count to zero temporarily, to have it > immediately re-incremented by rpc_free_auth. > > So if we're called concurrently with rpc_release_client then it's sort > of random whether someone gets this callback. > > Is that a problem?
Not really. If we re-increment the client->cl_count in rpc_free_auth() then it would be so that we can send off a bunch of NULL rpc calls to destroy existing RPCSEC_GSS contexts. We shouldn't need to do any more upcalls in pipefs.
If we care, we could simply move the call to rpc_unregister_client() into rpc_free_auth() so that the pipefs notifier doesn't see us, or we could set a flag to have it ignore us.
> Also, is this an existing bug? (In which case Trond should take it > now.)
-- Trond Myklebust Linux NFS client maintainer
NetApp Trond.Myklebust@netapp.com www.netapp.com
| |