tipc: re-enable compensation for socket receive buffer double counting
authorJon Paul Maloy <jon.maloy@ericsson.com>
Mon, 2 May 2016 15:58:45 +0000 (11:58 -0400)
committerDavid S. Miller <davem@davemloft.net>
Tue, 3 May 2016 19:51:14 +0000 (15:51 -0400)
In the refactoring commit d570d86497ee ("tipc: enqueue arrived buffers
in socket in separate function") we did by accident replace the test

if (sk->sk_backlog.len == 0)
     atomic_set(&tsk->dupl_rcvcnt, 0);

with

if (sk->sk_backlog.len)
     atomic_set(&tsk->dupl_rcvcnt, 0);

This effectively disables the compensation we have for the double
receive buffer accounting that occurs temporarily when buffers are
moved from the backlog to the socket receive queue. Until now, this
has gone unnoticed because of the large receive buffer limits we are
applying, but becomes indispensable when we reduce this buffer limit
later in this series.

We now fix this by inverting the mentioned condition.

Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/socket.c

index 3eeb50a27b89b6d9607b51b80a5ed5ce715235e4..d37a9401e182f599532ee4d02b647e1ecaac4e20 100644 (file)
@@ -1748,7 +1748,7 @@ static void tipc_sk_enqueue(struct sk_buff_head *inputq, struct sock *sk,
 
                /* Try backlog, compensating for double-counted bytes */
                dcnt = &tipc_sk(sk)->dupl_rcvcnt;
-               if (sk->sk_backlog.len)
+               if (!sk->sk_backlog.len)
                        atomic_set(dcnt, 0);
                lim = rcvbuf_limit(sk, skb) + atomic_read(dcnt);
                if (likely(!sk_add_backlog(sk, skb, lim)))