I don't see an xs_error_report() call here, which would confirm that the socket took a trip through tcp_disconnect().

From my reading of tcp_disconnect(), it calls sk->sk_error_report(sk) unconditionally so as there's no xs_error_report(), that surely means the exact opposite: tcp_disconnect() wasn't called. If it's not called, sk_shutdown is not cleared. And my revised tracing confirmed that it was set to 
SEND_SHUTDOWN.

-- 
Andy, BlueArc Engineering


-----Original Message-----
From: Chuck Lever [mailto:chuck.lever@oracle.com] 
Sent: 27 July 2010 18:29
To: Eric Dumazet
Cc: Andy Chittenden; Linux Kernel Mailing List (linux-kernel@vger.kernel.org); Trond Myklebust; netdev; Linux NFS Mailing List
Subject: Re: nfs client hang

Add CC: linux-nfs@vger.kernel.org

On 07/27/10 08:21 AM, Eric Dumazet wrote:

Le mardi 27 juillet 2010 à 11:53 +0100, Andy Chittenden a écrit :

quoted

quoted

quoted

quoted

IE the client starts a connection and then closes it again without sending data.

Once this happens, here's some rpcdebug info for the rpc module using 2.6.34.1 kernel:

... lots of the following nfsv3 WRITE requests:
[ 7670.026741] 57793 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026759] 57794 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026778] 57795 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026797] 57796 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026815] 57797 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026834] 57798 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026853] 57799 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026871] 57800 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026890] 57801 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7670.026909] 57802 0001    -11 ffff88012e32b000   (null)        0 ffffffffa03beb10 nfsv3 WRITE a:call_reserveresult q:xprt_backlog
[ 7680.520042] RPC:       worker connecting xprt ffff88013e62d800 via tcp to 10.1.6.102 (port 2049)
[ 7680.520066] RPC:       ffff88013e62d800 connect status 99 connected 0 sock state 7
[ 7680.520074] RPC: 33550 __rpc_wake_up_task (now 4296812426)
[ 7680.520079] RPC: 33550 disabling timer
[ 7680.520084] RPC: 33550 removed from queue ffff88013e62db20 "xprt_pending"
[ 7680.520089] RPC:       __rpc_wake_up_task done
[ 7680.520094] RPC: 33550 __rpc_execute flags=0x1
[ 7680.520098] RPC: 33550 xprt_connect_status: retrying
[ 7680.520103] RPC: 33550 call_connect_status (status -11)
[ 7680.520108] RPC: 33550 call_transmit (status 0)
[ 7680.520112] RPC: 33550 xprt_prepare_transmit
[ 7680.520118] RPC: 33550 rpc_xdr_encode (status 0)
[ 7680.520123] RPC: 33550 marshaling UNIX cred ffff88012e002300
[ 7680.520130] RPC: 33550 using AUTH_UNIX cred ffff88012e002300 to wrap rpc data
[ 7680.520136] RPC: 33550 xprt_transmit(32920)
[ 7680.520145] RPC:       xs_tcp_send_request(32920) = -32
[ 7680.520151] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7680.520156] RPC:       state 7 conn 0 dead 0 zapped 1

quoted

I changed that debug to output sk_shutdown too. That has a value of 2
(IE SEND_SHUTDOWN). Looking at tcp_sendmsg(), I see this:

quoted

          err = -EPIPE;
          if (sk->sk_err || (sk->sk_shutdown&  SEND_SHUTDOWN))
                  goto out_err;

quoted

which correlates with the trace "xs_tcp_send_request(32920) = -32". So,
this looks like a problem in the sockets/tcp layer. The rpc layer issues
a shutdown and then reconnects using the same socket. So either
sk_shutdown needs zeroing once the shutdown completes or should be
zeroed on subsequent connect. The latter sounds safer.

quoted

This patch for 2.6.34.1 fixes the issue:

--- /home/company/software/src/linux-2.6.34.1/net/ipv4/tcp_output.c     2010-07-27 08:46:46.917000000 +0100
+++ net/ipv4/tcp_output.c       2010-07-27 09:19:16.000000000 +0100

@@ -2522,6 +2522,13 @@
         struct tcp_sock *tp = tcp_sk(sk);
         __u8 rcv_wscale;

+       /* clear down any previous shutdown attempts so that
+        * reconnects on a socket that's been shutdown leave the
+        * socket in a usable state (otherwise tcp_sendmsg() returns
+        * -EPIPE).
+        */
+       sk->sk_shutdown = 0;
+
         /* We'll fix this up when we get a response from the other end.
          * See tcp_input.c:tcp_rcv_state_process case TCP_SYN_SENT.
          */

As I mentioned in my first message, we first saw this issue in 2.6.32 as supplied by debian (linux-image-2.6.32-5-amd64 Version: 2.6.32-17). It looks like the same patch would fix the problem there too.

CC netdev

This reminds me a similar problem we had in the past, fixed with commit
1fdf475a (tcp: tcp_disconnect() should clear window_clamp)

But tcp_disconnect() already clears sk->sk_shutdown

If NFS calls tcp_disconnect(), then shutdown(), there is a problem.

If tcp_disconnect() was called at some point, I would expect to see a 
message from xs_error_report() in the debugging output.  Perhaps 
tcp_disconnect() is not being invoked at all?

Maybe xs_tcp_shutdown() should make some sanity tests ?

Following sequence is legal, and your patch might break it.

fd = socket(...);
shutdown(fd, SHUT_WR);
...
connect(fd, ...);

I looked closely at some of Andy's debugging output from the 
linux-kernel mailing list archive.  I basically agree that the network 
layer is returning -EPIPE from tcp_sendmsg(), which the RPC client logic 
does not expect. But it's not clear to me how it gets into this state.

[ 7728.520042] RPC:       worker connecting xprt ffff88013e62d800 via tcp to 10.1.6.102 (port 2049)

 > [ 7728.520093] RPC:       ffff88013e62d800 connect status 115 
connected 0 sock state 2

"sock state 2" => sk->sk_state == TCP_SYN_SENT

 > [ 7728.520884] RPC:       xs_tcp_state_change client ffff88013e62d800...
 > [ 7728.520889] RPC:       state 1 conn 0 dead 0 zapped 1

"state 1" => sk->sk_state == TCP_ESTABLISHED

RPC client wakes up this RPC task now that the connection is established.

[ 7728.520896] RPC: 33550 __rpc_wake_up_task (now 4296824426)
[ 7728.520900] RPC: 33550 disabling timer
[ 7728.520906] RPC: 33550 removed from queue ffff88013e62db20 "xprt_pending"
[ 7728.520912] RPC:       __rpc_wake_up_task done
[ 7728.520932] RPC: 33550 __rpc_execute flags=0x1
[ 7728.520937] RPC: 33550 xprt_connect_status: retrying
[ 7728.520942] RPC: 33550 call_connect_status (status -11)

The awoken RPC task's status is -EAGAIN, which prevents a reconnection 
attempt.

[ 7728.520947] RPC: 33550 call_transmit (status 0)
[ 7728.520951] RPC: 33550 xprt_prepare_transmit
[ 7728.520957] RPC: 33550 rpc_xdr_encode (status 0)
[ 7728.520962] RPC: 33550 marshaling UNIX cred ffff88012e002300
[ 7728.520969] RPC: 33550 using AUTH_UNIX cred ffff88012e002300 to wrap rpc data
[ 7728.520976] RPC: 33550 xprt_transmit(32920)

RPC client encodes the request and attempts to send it.

[ 7728.520984] RPC:       xs_tcp_send_request(32920) = -32

Network layer says -EPIPE, for some reason.  RPC client calls 
kernel_sock_shutdown(SHUT_WR).

[ 7728.520997] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7728.521007] RPC:       state 4 conn 1 dead 0 zapped 1

"state 4" => sk->sk_state == TCP_FIN_WAIT1

The RPC client sets up a linger timeout.

[ 7728.521013] RPC: 33550 call_status (status -32)
[ 7728.521018] RPC: 33550 call_bind (status 0)
[ 7728.521023] RPC: 33550 call_connect xprt ffff88013e62d800 is not connected
[ 7728.521028] RPC: 33550 xprt_connect xprt ffff88013e62d800 is not connected
[ 7728.521035] RPC: 33550 sleep_on(queue "xprt_pending" time 4296824426)
[ 7728.521040] RPC: 33550 added to queue ffff88013e62db20 "xprt_pending"
[ 7728.521045] RPC: 33550 setting alarm for 60000 ms

RPC client puts this RPC task to sleep.

[ 7728.521439] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7728.521444] RPC:       state 5 conn 0 dead 0 zapped 1

"state 5" => sk->sk_state == TCP_FIN_WAIT2

[ 7728.521602] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7728.521608] RPC:       state 7 conn 0 dead 0 zapped 1
[ 7728.521612] RPC:       disconnected transport ffff88013e62d800

"state 7" => sk->sk_state == TCP_CLOSE

The RPC client marks the socket closed. and the linger timeout is 
cancelled.  At this point, sk_shutdown should be set to zero, correct? 
I don't see an xs_error_report() call here, which would confirm that the 
socket took a trip through tcp_disconnect().

[ 7728.521617] RPC: 33550 __rpc_wake_up_task (now 4296824426)
[ 7728.521621] RPC: 33550 disabling timer
[ 7728.521626] RPC: 33550 removed from queue ffff88013e62db20 "xprt_pending"
[ 7728.521631] RPC:       __rpc_wake_up_task done

RPC client wakes up the RPC task.  Meanwhile...

[ 7728.521636] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7728.521641] RPC:       state 7 conn 0 dead 0 zapped 1
[ 7728.521645] RPC:       disconnected transport ffff88013e62d800
[ 7728.521649] RPC:       xs_tcp_data_ready...

... network layer calls closed socket's data_ready method... while the 
awoken RPC task gets underway.

[ 7728.521662] RPC: 33550 __rpc_execute flags=0x1
[ 7728.521666] RPC: 33550 xprt_connect_status: retrying
[ 7728.521671] RPC: 33550 call_connect_status (status -11)

The awoken RPC task's status is -EAGAIN, which prevents a reconnection 
attempt, even though there is no established connection.

RPC client barrels on to send the request again.

[ 7728.521675] RPC: 33550 call_transmit (status 0)
[ 7728.521679] RPC: 33550 xprt_prepare_transmit
[ 7728.521683] RPC: 33550 rpc_xdr_encode (status 0)
[ 7728.521688] RPC: 33550 marshaling UNIX cred ffff88012e002300
[ 7728.521694] RPC: 33550 using AUTH_UNIX cred ffff88012e002300 to wrap rpc data
[ 7728.521699] RPC: 33550 xprt_transmit(32920)
[ 7728.521704] RPC:       xs_tcp_send_request(32920) = -32

RPC client attempts to send again, gets -EPIPE, and calls 
kernel_sock_shutdown(SHUT_WR).  If there is no connection established, 
the RPC client expects -ENOTCONN, in which case it will attempt to 
reconnect here.

[ 7728.521709] RPC:       xs_tcp_state_change client ffff88013e62d800...
[ 7728.521714] RPC:       state 7 conn 0 dead 0 zapped 1
[ 7728.521718] RPC:       disconnected transport ffff88013e62d800

"state 7" => sk->sk_state == TCP_CLOSE

Following this, the RPC client attempts to retransmit the request 
repeatedly, but the socket remains in the TCP_CLOSE state.