nfs: enforce FIFO ordering of operations trying to acquire slot

Signed-off-by: Alexandros Batsakis <batsakis@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index acd698b..02513da 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -346,8 +346,12 @@
  */
 static void nfs41_check_drain_session_complete(struct nfs4_session *ses)
 {
+	struct rpc_task *task;
+
 	if (!test_bit(NFS4CLNT_SESSION_DRAINING, &ses->clp->cl_state)) {
-		rpc_wake_up_next(&ses->fc_slot_table.slot_tbl_waitq);
+		task = rpc_wake_up_next(&ses->fc_slot_table.slot_tbl_waitq);
+		if (task)
+			rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 		return;
 	}
 
@@ -483,6 +487,14 @@
 		return -EAGAIN;
 	}
 
+	if (!rpc_queue_empty(&tbl->slot_tbl_waitq) &&
+	    !rpc_task_has_priority(task, RPC_PRIORITY_PRIVILEGED)) {
+		rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
+		spin_unlock(&tbl->slot_tbl_lock);
+		dprintk("%s enforce FIFO order\n", __func__);
+		return -EAGAIN;
+	}
+
 	slotid = nfs4_find_slot(tbl);
 	if (slotid == NFS4_MAX_SLOT_TABLE) {
 		rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
@@ -492,6 +504,7 @@
 	}
 	spin_unlock(&tbl->slot_tbl_lock);
 
+	rpc_task_set_priority(task, RPC_PRIORITY_NORMAL);
 	slot = tbl->slots + slotid;
 	args->sa_session = session;
 	args->sa_slotid = slotid;
@@ -4401,11 +4414,12 @@
 			(struct nfs4_get_lease_time_data *)calldata;
 
 	dprintk("--> %s\n", __func__);
+	rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 	/* just setup sequence, do not trigger session recovery
 	   since we're invoked within one */
 	ret = nfs41_setup_sequence(data->clp->cl_session,
-					&data->args->la_seq_args,
-					&data->res->lr_seq_res, 0, task);
+				   &data->args->la_seq_args,
+				   &data->res->lr_seq_res, 0, task);
 
 	BUG_ON(ret == -EAGAIN);
 	rpc_call_start(task);
@@ -4625,7 +4639,7 @@
 	tbl = &session->fc_slot_table;
 	tbl->highest_used_slotid = -1;
 	spin_lock_init(&tbl->slot_tbl_lock);
-	rpc_init_wait_queue(&tbl->slot_tbl_waitq, "ForeChannel Slot table");
+	rpc_init_priority_wait_queue(&tbl->slot_tbl_waitq, "ForeChannel Slot table");
 
 	tbl = &session->bc_slot_table;
 	tbl->highest_used_slotid = -1;
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index e76427e..0e45075 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -138,8 +138,22 @@
 static void nfs41_end_drain_session(struct nfs_client *clp,
 		struct nfs4_session *ses)
 {
-	if (test_and_clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state))
-		rpc_wake_up(&ses->fc_slot_table.slot_tbl_waitq);
+	int max_slots;
+
+	if (test_and_clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state)) {
+		spin_lock(&ses->fc_slot_table.slot_tbl_lock);
+		max_slots = ses->fc_slot_table.max_slots;
+		while (max_slots--) {
+			struct rpc_task *task;
+
+			task = rpc_wake_up_next(&ses->fc_slot_table.
+						slot_tbl_waitq);
+			if (!task)
+				break;
+			rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
+		}
+		spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
+	}
 }
 
 static int nfs41_begin_drain_session(struct nfs_client *clp,
diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 89ea8e6..aae6907 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -210,6 +210,7 @@
 {
 	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 }
+EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 
 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 {