[PATCH] libceph: queue_con only if the sock is connected

[Chen Yuanrun]

tcp_sendmsg_locked will trigger sk->sk_write_space while the sock
is still connecting, if con try to write to a connecting sock,
sk->sk_write_space will be called again, which lead to a endless loop.

This will happen if the public network of the cluster is down, and
the cpu will be in high usage.

Signed-off-by: Chen Yuanrun <chen-yuanrun@xxxxxxxxxxx>
---
 net/ceph/messenger.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
index d3bb656308b4..eef5d4dfd1f1 100644
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -367,7 +367,8 @@ static void ceph_sock_write_space(struct sock *sk)
      * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
      * and net/core/stream.c:sk_stream_write_space().
      */
-    if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
+    if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING) &&
+        atomic_read(&con->sock_state) == CON_SOCK_STATE_CONNECTED) {
         if (sk_stream_is_writeable(sk)) {
             dout("%s %p queueing write work\n", __func__, con);
             clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);

--
2.31.1