[PATCH -next] SUNRPC: Clean XPRT_CONGESTED of xprt->state when rpc task is killed

[Bixuan Cui <cuibixuan@linux.alibaba.com>]

There is a problem with nfs hang on my machine, Steps to reproduce:

1.
echo "options sunrpc tcp_slot_table_entries=2" >> /etc/modprobe.d/sunrpc.conf
echo "options sunrpc tcp_max_slot_table_entries=2" >>  /etc/modprobe.d/sunrpc.conf
and remount nfsv3: sudo mount -t nfs -o vers=3,nolock,proto=tcp 127.0.0.1:/root/nfs_server/ /root/nfs_client/

2.Run case
# cat case.sh 
while true
do
    ls &
done &

while true
do
    killall ls
done &

[root@cuibixuan nfs_client]# cd /root/nfs_client/; sh case.sh

3. Run 'killall sh' to kill case after waiting 10 seconds

4. The nfs hang and xprt->state is 0x212(XPRT_CONGESTED):
[root@cuibixuan sunrpc]# cat /sys/kernel/debug/sunrpc/rpc_xprt/4/info 
netid: tcp
addr:  127.0.0.1
port:  2049
state: 0x212

[root@cuibixuan sunrpc]# cat /sys/kernel/debug/sunrpc/rpc_clnt/4/tasks 
  155 0080    -11 0x4 0x0        0 rpc_default_ops [sunrpc] nfsv3 GETATTR a:call_reserveresult [sunrpc] q:xprt_backlog
15418 0080    -11 0x4 0x0        0 rpc_default_ops [sunrpc] nfsv3 GETATTR a:call_reserveresult [sunrpc] q:xprt_backlog

5. Enable ftrace
 # echo 1 > /sys/kernel/debug/tracing/events/sunrpc/enable
 # cat /sys/kernel/debug/tracing/trace
	      ...
              ls-18792 [001] ....   668.541062: rpc_task_run_action: task:28885@4 flags=0080 state=0005 status=0 action=call_start [sunrpc]
              ls-18792 [001] ....   668.541063: rpc_request: task:28885@4 nfsv3 GETATTR (sync)
              ls-18792 [001] ....   668.541063: rpc_task_run_action: task:28885@4 flags=0080 state=0005 status=0 action=call_reserve [sunrpc]
              ls-18792 [001] ....   668.541065: rpc_task_sleep: task:28885@4 flags=0080 state=0005 status=-11 timeout=0 queue=xprt_backlog
              ls-18792 [000] ....   670.815066: rpc_task_wakeup: task:28885@4 flags=0180 state=0006 status=-512 timeout=0 queue=xprt_backlog
              ls-18792 [000] ....   670.815069: rpc_task_run_action: task:28885@4 flags=0180 state=0005 status=-512 action=rpc_exit_task [sunrpc]
              ls-18846 [001] ....   682.302970: rpc_task_begin: task:28886@4 flags=0080 state=0004 status=0 action=          (null)
              ls-18846 [001] ....   682.302972: rpc_task_run_action: task:28886@4 flags=0080 state=0005 status=0 action=call_start [sunrpc]
              ls-18846 [001] ....   682.302973: rpc_request: task:28886@4 nfsv3 GETATTR (sync)
              ls-18846 [001] ....   682.302973: rpc_task_run_action: task:28886@4 flags=0080 state=0005 status=0 action=call_reserve [sunrpc]
              ls-18846 [001] ....   682.302974: rpc_task_sleep: task:28886@4 flags=0080 state=0005 status=-11 timeout=0 queue=xprt_backlog
              ls-18846 [000] ....   684.690924: rpc_task_wakeup: task:28886@4 flags=0180 state=0006 status=-512 timeout=0 queue=xprt_backlog
              ls-18846 [000] ....   684.690928: rpc_task_run_action: task:28886@4 flags=0180 state=0005 status=-512 action=rpc_exit_task [sunrpc]

The kernel version of my machine is 4.19.91, and it can be fixed with
the following patch (in v4.19):

diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 793f595..2656048 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -1358,6 +1358,7 @@ void xprt_release(struct rpc_task *task)
        if (req == NULL) {
                if (task->tk_client) {
                        xprt = task->tk_xprt;
+                       xprt_wake_up_backlog(xprt);
                        if (xprt->snd_task == task)
                                xprt_release_write(xprt, task);
                }

I checked the -next code, there should be the same problem.

Bixuan Cui (1):
  SUNRPC: Clean XPRT_CONGESTED of xprt->state when rpc task is killed

 net/sunrpc/xprt.c | 1 +
 1 file changed, 1 insertion(+)

-- 
1.8.3.1