queue-4.9/tcp-refine-memory-limit-test-in-tcp_fragment.patch

   1 From b6653b3629e5b88202be3c9abc44713973f5c4b4 Mon Sep 17 00:00:00 2001
   2 From: Eric Dumazet <edumazet@google.com>
   3 Date: Fri, 21 Jun 2019 06:09:55 -0700
   4 Subject: tcp: refine memory limit test in tcp_fragment()
   5
   6 From: Eric Dumazet <edumazet@google.com>
   7
   8 commit b6653b3629e5b88202be3c9abc44713973f5c4b4 upstream.
   9
  10 tcp_fragment() might be called for skbs in the write queue.
  11
  12 Memory limits might have been exceeded because tcp_sendmsg() only
  13 checks limits at full skb (64KB) boundaries.
  14
  15 Therefore, we need to make sure tcp_fragment() wont punish applications
  16 that might have setup very low SO_SNDBUF values.
  17
  18 Fixes: f070ef2ac667 ("tcp: tcp_fragment() should apply sane memory limits")
  19 Signed-off-by: Eric Dumazet <edumazet@google.com>
  20 Reported-by: Christoph Paasch <cpaasch@apple.com>
  21 Tested-by: Christoph Paasch <cpaasch@apple.com>
  22 Signed-off-by: David S. Miller <davem@davemloft.net>
  23 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  24
  25 ---
  26  net/ipv4/tcp_output.c |    3 ++-
  27  1 file changed, 2 insertions(+), 1 deletion(-)
  28
  29 --- a/net/ipv4/tcp_output.c
  30 +++ b/net/ipv4/tcp_output.c
  31 @@ -1185,7 +1185,8 @@ int tcp_fragment(struct sock *sk, struct
  32         if (nsize < 0)
  33                 nsize = 0;
  34
  35 -       if (unlikely((sk->sk_wmem_queued >> 1) > sk->sk_sndbuf)) {
  36 +       if (unlikely((sk->sk_wmem_queued >> 1) > sk->sk_sndbuf &&
  37 +                    tcp_queue != TCP_FRAG_IN_WRITE_QUEUE)) {
  38                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPWQUEUETOOBIG);
  39                 return -ENOMEM;
  40         }