NFS: Add sequence_priviliged_ops for nfs4_proc_sequence()
authorBryan Schumaker <bjschuma@netapp.com>
Mon, 12 Nov 2012 21:55:38 +0000 (16:55 -0500)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Wed, 21 Nov 2012 04:34:54 +0000 (23:34 -0500)
If I mount an NFS v4.1 server to a single client multiple times and then
run xfstests over each mountpoint I usually get the client into a state
where recovery deadlocks.  The server informs the client of a
cb_path_down sequence error, the client then does a
bind_connection_to_session and checks the status of the lease.

I found that bind_connection_to_session sets the NFS4_SESSION_DRAINING
flag on the client, but this flag is never unset before
nfs4_check_lease() reaches nfs4_proc_sequence().  This causes the client
to deadlock, halting all NFS activity to the server.  nfs4_proc_sequence()
is only called by the state manager, so I can change it to run in privileged
mode to bypass the NFS4_SESSION_DRAINING check and avoid the deadlock.

Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@vger.kernel.org
fs/nfs/nfs4proc.c

index 6300cdd..a32d953 100644 (file)
@@ -6136,13 +6136,26 @@ static void nfs41_sequence_prepare(struct rpc_task *task, void *data)
        rpc_call_start(task);
 }
 
+static void nfs41_sequence_prepare_privileged(struct rpc_task *task, void *data)
+{
+       rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
+       nfs41_sequence_prepare(task, data);
+}
+
 static const struct rpc_call_ops nfs41_sequence_ops = {
        .rpc_call_done = nfs41_sequence_call_done,
        .rpc_call_prepare = nfs41_sequence_prepare,
        .rpc_release = nfs41_sequence_release,
 };
 
-static struct rpc_task *_nfs41_proc_sequence(struct nfs_client *clp, struct rpc_cred *cred)
+static const struct rpc_call_ops nfs41_sequence_privileged_ops = {
+       .rpc_call_done = nfs41_sequence_call_done,
+       .rpc_call_prepare = nfs41_sequence_prepare_privileged,
+       .rpc_release = nfs41_sequence_release,
+};
+
+static struct rpc_task *_nfs41_proc_sequence(struct nfs_client *clp, struct rpc_cred *cred,
+                                            const struct rpc_call_ops *seq_ops)
 {
        struct nfs4_sequence_data *calldata;
        struct rpc_message msg = {
@@ -6152,7 +6165,7 @@ static struct rpc_task *_nfs41_proc_sequence(struct nfs_client *clp, struct rpc_
        struct rpc_task_setup task_setup_data = {
                .rpc_client = clp->cl_rpcclient,
                .rpc_message = &msg,
-               .callback_ops = &nfs41_sequence_ops,
+               .callback_ops = seq_ops,
                .flags = RPC_TASK_ASYNC | RPC_TASK_SOFT,
        };
 
@@ -6179,7 +6192,7 @@ static int nfs41_proc_async_sequence(struct nfs_client *clp, struct rpc_cred *cr
 
        if ((renew_flags & NFS4_RENEW_TIMEOUT) == 0)
                return 0;
-       task = _nfs41_proc_sequence(clp, cred);
+       task = _nfs41_proc_sequence(clp, cred, &nfs41_sequence_ops);
        if (IS_ERR(task))
                ret = PTR_ERR(task);
        else
@@ -6193,7 +6206,7 @@ static int nfs4_proc_sequence(struct nfs_client *clp, struct rpc_cred *cred)
        struct rpc_task *task;
        int ret;
 
-       task = _nfs41_proc_sequence(clp, cred);
+       task = _nfs41_proc_sequence(clp, cred, &nfs41_sequence_privileged_ops);
        if (IS_ERR(task)) {
                ret = PTR_ERR(task);
                goto out;