tcp: purge write queue upon RST
authorSoheil Hassas Yeganeh <soheil@google.com>
Tue, 27 Feb 2018 23:32:18 +0000 (18:32 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 9 Mar 2018 06:41:17 +0000 (22:41 -0800)
commitabb4a8b870b5c644c9bb6d9bd372298002f0a65f
tree3c05c6da9b91014dce55c58b3c361aa1cc92fe7a
parenteec434c573e78c1190a812cb48a8b16c254541ea
tcp: purge write queue upon RST

[ Upstream commit a27fd7a8ed3856faaf5a2ff1c8c5f00c0667aaa0 ]

When the connection is reset, there is no point in
keeping the packets on the write queue until the connection
is closed.

RFC 793 (page 70) and RFC 793-bis (page 64) both suggest
purging the write queue upon RST:
https://tools.ietf.org/html/draft-ietf-tcpm-rfc793bis-07

Moreover, this is essential for a correct MSG_ZEROCOPY
implementation, because userspace cannot call close(fd)
before receiving zerocopy signals even when the connection
is reset.

Fixes: f214f915e7db ("tcp: enable MSG_ZEROCOPY")
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/tcp_input.c