NFS: fix disabling of swap

When swap is activated to a file on an NFSv4 mount we arrange that the
state manager thread is always present as starting a new thread requires
memory allocations that might block waiting for swap.

Unfortunately the code for allowing the state manager thread to exit when
swap is disabled was not tested properly and does not work.
This can be seen by examining /proc/fs/nfsfs/servers after disabling swap
and unmounting the filesystem.  The servers file will still list one
entry.  Also a "ps" listing will show the state manager thread is still
present.

There are two problems.
 1/ rpc_clnt_swap_deactivate() doesn't walk up the ->cl_parent list to
    find the primary client on which the state manager runs.

 2/ The thread is not woken up properly and it immediately goes back to
    sleep without checking whether it is really needed.  Using
    nfs4_schedule_state_manager() ensures a proper wake-up.

Reported-by: Olga Kornievskaia <aglo@umich.edu>
Fixes: 4dc73c679114 ("NFSv4: keep state manager thread active if swap is enabled")
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

1 files changed, 2 insertions, 0 deletions

diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 0b0b9f1eed46..fd7e1c630493 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -3350,6 +3350,8 @@ rpc_clnt_swap_deactivate_callback(struct rpc_clnt *clnt,
 void
 rpc_clnt_swap_deactivate(struct rpc_clnt *clnt)
 {
+	while (clnt != clnt->cl_parent)
+		clnt = clnt->cl_parent;
 	if (atomic_dec_if_positive(&clnt->cl_swapper) == 0)
 		rpc_clnt_iterate_for_each_xprt(clnt,
 				rpc_clnt_swap_deactivate_callback, NULL);