tcp: flush DMA queue before sk_wait_data if rcv_wnd is zero
authorMichal Kubeček <mkubecek@suse.cz>
Fri, 14 Sep 2012 04:59:52 +0000 (04:59 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 12 Oct 2012 20:38:43 +0000 (05:38 +0900)
commit2033554a2fe3c5e54764f2f1dba0baff3261f8b5
treeac3133307617931783ead5088f05e6de216a89c4
parent410eafac650a906e990351a01ec70451064df83d
tcp: flush DMA queue before sk_wait_data if rcv_wnd is zero

[ Upstream commit 15c041759bfcd9ab0a4e43f1c16e2644977d0467 ]

If recv() syscall is called for a TCP socket so that
  - IOAT DMA is used
  - MSG_WAITALL flag is used
  - requested length is bigger than sk_rcvbuf
  - enough data has already arrived to bring rcv_wnd to zero
then when tcp_recvmsg() gets to calling sk_wait_data(), receive
window can be still zero while sk_async_wait_queue exhausts
enough space to keep it zero. As this queue isn't cleaned until
the tcp_service_net_dma() call, sk_wait_data() cannot receive
any data and blocks forever.

If zero receive window and non-empty sk_async_wait_queue is
detected before calling sk_wait_data(), process the queue first.

Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/tcp.c