NFS: flush signals before taking down callback thread
authorJeff Layton <jlayton@redhat.com>
Tue, 19 Feb 2008 17:51:35 +0000 (12:51 -0500)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Wed, 20 Feb 2008 18:32:43 +0000 (13:32 -0500)
commit1227a74e2e0217a4ca155d1677bdbf5f69e32bed
tree5be4e69ea8f008ec488ac0980470f6acc1da9029
parent5d9c4a7de64d398604a978d267a6987f1f4025b7
NFS: flush signals before taking down callback thread

Now that the reference counting on the callback thread is working as
expected, it uncovers another problem.  Peter Staubach noticed while
testing that patch on an older kernel that he would occasionally see
this printk in rpc_register fire:

    "RPC: failed to contact portmap (errno -512).

The NFSv4 callback thread is signaled by nfs_callback_down(), but never
flushes that signal. All of the shutdown processing is done with that
signal pending. This makes it fail the call to unregister the port with
the portmapper.

In actuality, this rpc_register call isn't necessary at all since the
port isn't actually registered with the portmapper anymore. Regardless,
there doesn't seem to be any reason to leave the signal pending while
the thread is being shut down and flushing it should generally silence
that printk.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/callback.c