sunrpc: remove an offlined xprt using sysfs

Once a transport has been put offline, this transport can be also
removed from the list of transports. Any tasks that have been stuck
on this transport would find the next available active transport
and be re-tried. This transport would be removed from the xprt_switch
list and freed.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

include/linux/sunrpc/xprt.h

@@ -428,6 +428,7 @@ void			xprt_release_write(struct rpc_xprt *, struct rpc_task *);
 #define XPRT_BINDING		(5)
 #define XPRT_CLOSING		(6)
 #define XPRT_OFFLINE		(7)
+#define XPRT_REMOVE		(8)
 #define XPRT_CONGESTED		(9)
 #define XPRT_CWND_WAIT		(10)
 #define XPRT_WRITE_SPACE	(11)

net/sunrpc/clnt.c

@@ -2106,6 +2106,30 @@ call_connect_status(struct rpc_task *task)
 	case -ENOTCONN:
 	case -EAGAIN:
 	case -ETIMEDOUT:
+		if (!(task->tk_flags & RPC_TASK_NO_ROUND_ROBIN) &&
+		    (task->tk_flags & RPC_TASK_MOVEABLE) &&
+		    test_bit(XPRT_REMOVE, &xprt->state)) {
+			struct rpc_xprt *saved = task->tk_xprt;
+			struct rpc_xprt_switch *xps;
+
+			rcu_read_lock();
+			xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
+			rcu_read_unlock();
+			if (xps->xps_nxprts > 1) {
+				long value;
+
+				xprt_release(task);
+				value = atomic_long_dec_return(&xprt->queuelen);
+				if (value == 0)
+					rpc_xprt_switch_remove_xprt(xps, saved);
+				xprt_put(saved);
+				task->tk_xprt = NULL;
+				task->tk_action = call_start;
+			}
+			xprt_switch_put(xps);
+			if (!task->tk_xprt)
+				return;
+		}
 		goto out_retry;
 	case -ENOBUFS:
 		rpc_delay(task, HZ >> 2);

net/sunrpc/sysfs.c

@@ -133,7 +133,7 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
 	ssize_t ret;
 	int locked, connected, connecting, close_wait, bound, binding,
-	    closing, congested, cwnd_wait, write_space, offline;
+	    closing, congested, cwnd_wait, write_space, offline, remove;
 
 	if (!xprt)
 		return 0;
@@ -152,8 +152,9 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 		cwnd_wait = test_bit(XPRT_CWND_WAIT, &xprt->state);
 		write_space = test_bit(XPRT_WRITE_SPACE, &xprt->state);
 		offline = test_bit(XPRT_OFFLINE, &xprt->state);
+		remove = test_bit(XPRT_REMOVE, &xprt->state);
 
-		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s %s\n",
+		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s %s %s\n",
 			      locked ? "LOCKED" : "",
 			      connected ? "CONNECTED" : "",
 			      connecting ? "CONNECTING" : "",
@@ -164,7 +165,8 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 			      congested ? "CONGESTED" : "",
 			      cwnd_wait ? "CWND_WAIT" : "",
 			      write_space ? "WRITE_SPACE" : "",
-			      offline ? "OFFLINE" : "");
+			      offline ? "OFFLINE" : "",
+			      remove ? "REMOVE" : "");
 	}
 
 	xprt_put(xprt);
@@ -251,7 +253,7 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 					   const char *buf, size_t count)
 {
 	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
-	int offline = 0, online = 0;
+	int offline = 0, online = 0, remove = 0;
 	struct rpc_xprt_switch *xps = rpc_sysfs_xprt_kobj_get_xprt_switch(kobj);
 
 	if (!xprt)
@@ -261,6 +263,8 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 		offline = 1;
 	else if (!strncmp(buf, "online", 6))
 		online = 1;
+	else if (!strncmp(buf, "remove", 6))
+		remove = 1;
 	else
 		return -EINVAL;
 
@@ -282,6 +286,20 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 		spin_lock(&xps->xps_lock);
 		xps->xps_nactive++;
 		spin_unlock(&xps->xps_lock);
+	} else if (remove) {
+		if (test_bit(XPRT_OFFLINE, &xprt->state)) {
+			set_bit(XPRT_REMOVE, &xprt->state);
+			xprt_force_disconnect(xprt);
+			if (test_bit(XPRT_CONNECTED, &xprt->state)) {
+				if (!xprt->sending.qlen &&
+				    !xprt->pending.qlen &&
+				    !xprt->backlog.qlen &&
+				    !atomic_long_read(&xprt->queuelen))
+					rpc_xprt_switch_remove_xprt(xps, xprt);
+			}
+		} else {
+			count = -EINVAL;
+		}
 	}
 
 release_tasks: