From: NeilBrown <neilb@suse.de>
Date: Wed, 30 Aug 2023 13:46:57 +1000
Subject: [PATCH] SUNRPC: always clear XPRT_SOCK_CONNECTING before
 xprt_clear_connecting on TCP xprt
Patch-mainline: Submitted, linux-nfs@vger.kernel.org - Wed, 30 Aug 2023 14:14:21 +1000
References: bsc#1214453

sunrpc/xprtsock.c has an XPRT_SOCK_CONNECTING flag which is used only on
TCP xprts while connecting.
The purpose appears to be to protect against delayed TCP_CLOSE
transitions from calling xprt_clear_connecting() asynchronously.

Normally it is set after calling xprt_set_connecting() and before
calling kernel_connect(), and cleared before calling
xprt_clear_connecting().  It is only tested on a state change to
TCP_CLOSE.

Unfortunately there is one time that it is *not* explicitly cleared
before calling xprt_clear_connecting().  I don't know what all to
consequences of this are.  It may well relate to the underlying problem
that resulted in
Commit 3be232f11a3c ("SUNRPC: Prevent immediate close+reconnect")
That close/reconnect pattern can happen if a TCP_CLOSE is seen
while XPRT_SOCK_CONNECTING is set but shouldn't be.

local and udp connections don't use XPRT_SOCK_CONNECTING.

Signed-off-by: NeilBrown <neilb@suse.de>
Acked-by: NeilBrown <neilb@suse.com>

---
 net/sunrpc/xprtsock.c |    2 ++
 1 file changed, 2 insertions(+)

--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -2472,8 +2472,10 @@ static void xs_tcp_setup_socket(struct w
 		 */
 		xprt_wake_pending_tasks(xprt, status);
 		xs_tcp_force_close(xprt);
+		clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
 		goto out;
 	}
+	clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
 	status = -EAGAIN;
 out:
 	xprt_clear_connecting(xprt);