tcp: limit sk_rcvlowat by the maximum receive buffer
authorSoheil Hassas Yeganeh <soheil@google.com>
Sat, 9 Jun 2018 02:47:10 +0000 (22:47 -0400)
committerDavid S. Miller <davem@davemloft.net>
Sun, 10 Jun 2018 21:12:50 +0000 (14:12 -0700)
commit867f816badc01e6da655028810d468c9f935b37c
tree8638841a4e39647569a24e5021bfa45586c90a3b
parentb718e8c8f4f5920aaddc2e52d5e32f494c91129c
tcp: limit sk_rcvlowat by the maximum receive buffer

The user-provided value to setsockopt(SO_RCVLOWAT) can be
larger than the maximum possible receive buffer. Such values
mute POLLIN signals on the socket which can stall progress
on the socket.

Limit the user-provided value to half of the maximum receive
buffer, i.e., half of sk_rcvbuf when the receive buffer size
is set by the user, or otherwise half of sysctl_tcp_rmem[2].

Fixes: d1361840f8c5 ("tcp: fix SO_RCVLOWAT and RCVBUF autotuning")
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp.c