SUNRPC: Handle ECONNREFUSED correctly in xprt_transmit()

If we get an ECONNREFUSED error, we currently go to sleep on the
'xprt->sending' wait queue. The problem is that no timeout is set there,
and there is nothing else that will wake the task up later.

We should deal with ECONNREFUSED in call_status, given that is where we
also deal with -EHOSTDOWN, and friends.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

3 files changed, 34 insertions, 37 deletions

diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 07e9b05321e6..145715b53115 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1117,10 +1117,12 @@ call_transmit_status(struct rpc_task *task)
                  * then hold onto the transport lock.
                  */
         case -ECONNREFUSED:
+        case -ECONNRESET:
         case -ENOTCONN:
         case -EHOSTDOWN:
         case -EHOSTUNREACH:
         case -ENETUNREACH:
+        case -EPIPE:
                 rpc_task_force_reencode(task);
         }
 }
@@ -1162,9 +1164,12 @@ call_status(struct rpc_task *task)
                         xprt_conditional_disconnect(task->tk_xprt,
                                         req->rq_connect_cookie);
                 break;
+        case -ECONNRESET:
         case -ECONNREFUSED:
-        case -ENOTCONN:
                 rpc_force_rebind(clnt);
+                rpc_delay(task, 3*HZ);
+        case -EPIPE:
+        case -ENOTCONN:
                 task->tk_action = call_bind;
                 break;
         case -EAGAIN:
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index d1afec640394..d588e755e107 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -901,32 +901,26 @@ void xprt_transmit(struct rpc_task *task)
         req->rq_connect_cookie = xprt->connect_cookie;
         req->rq_xtime = jiffies;
         status = xprt->ops->send_request(task);
-        if (status == 0) {
-                dprintk("RPC: %5u xmit complete\n", task->tk_pid);
-                spin_lock_bh(&xprt->transport_lock);
+        if (status != 0) {
+                task->tk_status = status;
+                return;
+        }
 
-                xprt->ops->set_retrans_timeout(task);
+        dprintk("RPC: %5u xmit complete\n", task->tk_pid);
+        spin_lock_bh(&xprt->transport_lock);
 
-                xprt->stat.sends++;
-                xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
-                xprt->stat.bklog_u += xprt->backlog.qlen;
+        xprt->ops->set_retrans_timeout(task);
 
-                /* Don't race with disconnect */
-                if (!xprt_connected(xprt))
-                        task->tk_status = -ENOTCONN;
-                else if (!req->rq_received)
-                        rpc_sleep_on(&xprt->pending, task, xprt_timer);
-                spin_unlock_bh(&xprt->transport_lock);
-                return;
-        }
+        xprt->stat.sends++;
+        xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
+        xprt->stat.bklog_u += xprt->backlog.qlen;
 
-        /* Note: at this point, task->tk_sleeping has not yet been set,
-         *       hence there is no danger of the waking up task being put on
-         *       schedq, and being picked up by a parallel run of rpciod().
-         */
-        task->tk_status = status;
-        if (status == -ECONNREFUSED)
-                rpc_sleep_on(&xprt->sending, task, NULL);
+        /* Don't race with disconnect */
+        if (!xprt_connected(xprt))
+                task->tk_status = -ENOTCONN;
+        else if (!req->rq_received)
+                rpc_sleep_on(&xprt->pending, task, xprt_timer);
+        spin_unlock_bh(&xprt->transport_lock);
 }
 
 static inline void do_xprt_reserve(struct rpc_task *task)
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 9d1898f6ee87..5e8198bede81 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -594,6 +594,8 @@ static int xs_udp_send_request(struct rpc_task *task)
                 /* Still some bytes left; set up for a retry later. */
                 status = -EAGAIN;
         }
+        if (!transport->sock)
+                goto out;
 
         switch (status) {
         case -ENOTSOCK:
@@ -603,19 +605,17 @@ static int xs_udp_send_request(struct rpc_task *task)
         case -EAGAIN:
                 xs_nospace(task);
                 break;
+        default:
+                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
+                        -status);
         case -ENETUNREACH:
         case -EPIPE:
         case -ECONNREFUSED:
                 /* When the server has died, an ICMP port unreachable message
                  * prompts ECONNREFUSED. */
                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
-                break;
-        default:
-                clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
-                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
-                        -status);
         }
-
+out:
         return status;
 }
 
@@ -697,6 +697,8 @@ static int xs_tcp_send_request(struct rpc_task *task)
                 status = -EAGAIN;
                 break;
         }
+        if (!transport->sock)
+                goto out;
 
         switch (status) {
         case -ENOTSOCK:
@@ -706,21 +708,17 @@ static int xs_tcp_send_request(struct rpc_task *task)
         case -EAGAIN:
                 xs_nospace(task);
                 break;
+        default:
+                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
+                        -status);
         case -ECONNRESET:
                 xs_tcp_shutdown(xprt);
         case -ECONNREFUSED:
         case -ENOTCONN:
         case -EPIPE:
-                status = -ENOTCONN;
-                clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
-                break;
-        default:
-                dprintk("RPC:       sendmsg returned unrecognized error %d\n",
-                        -status);
                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
-                xs_tcp_shutdown(xprt);
         }
-
+out:
         return status;
 }