]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
SUNRPC: Respond promptly to server TCP resets
authorTrond Myklebust <Trond.Myklebust@netapp.com>
Tue, 28 Oct 2008 19:21:39 +0000 (15:21 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Tue, 28 Oct 2008 19:21:39 +0000 (15:21 -0400)
If the server sends us an RST error while we're in the TCP_ESTABLISHED
state, then that will not result in a state change, and so the RPC client
ends up hanging forever (see
http://bugzilla.kernel.org/show_bug.cgi?id=11154)

We can intercept the reset by setting up an sk->sk_error_report callback,
which will then allow us to initiate a proper shutdown and retry...

We also make sure that if the send request receives an ECONNRESET, then we
shutdown too...

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
net/sunrpc/xprtsock.c

index 9a288d5eea646e65953bf92b015f2e557d15239e..0a50361e3d83535492160f7f08a36a5a5c34d4be 100644 (file)
@@ -249,6 +249,7 @@ struct sock_xprt {
        void                    (*old_data_ready)(struct sock *, int);
        void                    (*old_state_change)(struct sock *);
        void                    (*old_write_space)(struct sock *);
+       void                    (*old_error_report)(struct sock *);
 };
 
 /*
@@ -698,8 +699,9 @@ static int xs_tcp_send_request(struct rpc_task *task)
        case -EAGAIN:
                xs_nospace(task);
                break;
-       case -ECONNREFUSED:
        case -ECONNRESET:
+               xs_tcp_shutdown(xprt);
+       case -ECONNREFUSED:
        case -ENOTCONN:
        case -EPIPE:
                status = -ENOTCONN;
@@ -742,6 +744,22 @@ out_release:
        xprt_release_xprt(xprt, task);
 }
 
+static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
+{
+       transport->old_data_ready = sk->sk_data_ready;
+       transport->old_state_change = sk->sk_state_change;
+       transport->old_write_space = sk->sk_write_space;
+       transport->old_error_report = sk->sk_error_report;
+}
+
+static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
+{
+       sk->sk_data_ready = transport->old_data_ready;
+       sk->sk_state_change = transport->old_state_change;
+       sk->sk_write_space = transport->old_write_space;
+       sk->sk_error_report = transport->old_error_report;
+}
+
 /**
  * xs_close - close a socket
  * @xprt: transport
@@ -765,9 +783,8 @@ static void xs_close(struct rpc_xprt *xprt)
        transport->sock = NULL;
 
        sk->sk_user_data = NULL;
-       sk->sk_data_ready = transport->old_data_ready;
-       sk->sk_state_change = transport->old_state_change;
-       sk->sk_write_space = transport->old_write_space;
+
+       xs_restore_old_callbacks(transport, sk);
        write_unlock_bh(&sk->sk_callback_lock);
 
        sk->sk_no_check = 0;
@@ -1179,6 +1196,28 @@ static void xs_tcp_state_change(struct sock *sk)
        read_unlock(&sk->sk_callback_lock);
 }
 
+/**
+ * xs_tcp_error_report - callback mainly for catching RST events
+ * @sk: socket
+ */
+static void xs_tcp_error_report(struct sock *sk)
+{
+       struct rpc_xprt *xprt;
+
+       read_lock(&sk->sk_callback_lock);
+       if (sk->sk_err != ECONNRESET || sk->sk_state != TCP_ESTABLISHED)
+               goto out;
+       if (!(xprt = xprt_from_sock(sk)))
+               goto out;
+       dprintk("RPC:       %s client %p...\n"
+                       "RPC:       error %d\n",
+                       __func__, xprt, sk->sk_err);
+
+       xprt_force_disconnect(xprt);
+out:
+       read_unlock(&sk->sk_callback_lock);
+}
+
 /**
  * xs_udp_write_space - callback invoked when socket buffer space
  *                             becomes available
@@ -1454,10 +1493,9 @@ static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
 
                write_lock_bh(&sk->sk_callback_lock);
 
+               xs_save_old_callbacks(transport, sk);
+
                sk->sk_user_data = xprt;
-               transport->old_data_ready = sk->sk_data_ready;
-               transport->old_state_change = sk->sk_state_change;
-               transport->old_write_space = sk->sk_write_space;
                sk->sk_data_ready = xs_udp_data_ready;
                sk->sk_write_space = xs_udp_write_space;
                sk->sk_no_check = UDP_CSUM_NORCV;
@@ -1589,13 +1627,13 @@ static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
 
                write_lock_bh(&sk->sk_callback_lock);
 
+               xs_save_old_callbacks(transport, sk);
+
                sk->sk_user_data = xprt;
-               transport->old_data_ready = sk->sk_data_ready;
-               transport->old_state_change = sk->sk_state_change;
-               transport->old_write_space = sk->sk_write_space;
                sk->sk_data_ready = xs_tcp_data_ready;
                sk->sk_state_change = xs_tcp_state_change;
                sk->sk_write_space = xs_tcp_write_space;
+               sk->sk_error_report = xs_tcp_error_report;
                sk->sk_allocation = GFP_ATOMIC;
 
                /* socket options */