The Ceph messenger would sometimes queue multiple work items to write
data to a socket when the socket buffer was full.

Fix this problem by making ceph_write_space() use SOCK_NOSPACE in the
same way that net/core/stream.c:sk_stream_write_space() does, i.e.,
clearing it only when sufficient space is available in the socket buffer.

Signed-off-by: Jim Schutt <[email protected]>
---
 net/ceph/messenger.c |   18 ++++++++++++------
 1 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
index 67973f0..a4df90b 100644
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -175,16 +175,22 @@ static void ceph_write_space(struct sock *sk)
        struct ceph_connection *con =
                (struct ceph_connection *)sk->sk_user_data;
 
-       /* only queue to workqueue if there is data we want to write. */
+       /* only queue to workqueue if there is data we want to write,
+        * and there is sufficient space in the socket buffer to accept
+        * more data.  clear SOCK_NOSPACE so that ceph_write_space()
+        * doesn't get called again until try_write() fills the socket
+        * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
+        * and net/core/stream.c:sk_stream_write_space().
+        */
        if (test_bit(WRITE_PENDING, &con->state)) {
-               dout("ceph_write_space %p queueing write work\n", con);
-               queue_con(con);
+               if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
+                       dout("ceph_write_space %p queueing write work\n", con);
+                       clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
+                       queue_con(con);
+               }
        } else {
                dout("ceph_write_space %p nothing to write\n", con);
        }
-
-       /* since we have our own write_space, clear the SOCK_NOSPACE flag */
-       clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 }
 
 /* socket's state has changed */
-- 
1.7.1


--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to