net: tcp: Scale the TCP backlog queue to absorb packet bursts
authorHarout Hedeshian <harouth@codeaurora.org>
Mon, 2 Feb 2015 20:30:42 +0000 (13:30 -0700)
committerDanny Wood <danwood76@gmail.com>
Tue, 28 Jul 2020 08:47:02 +0000 (09:47 +0100)
commit27adc682bf05227c57f2ce781e075fa701bfe24f
treefa280a8b7ae8863fd1a5ed88390d10da12d485a0
parentb3a20fc392f857f11bb622cf5f05e6badc86d8b8
net: tcp: Scale the TCP backlog queue to absorb packet bursts

A large momentary influx of packets flooding the TCP layer may cause
packets to get dropped at the socket's backlog queue. Bump this up to
prevent these drops. Note that this change may cause the socket memory
accounting to allow the total backlog queue length to exceed the user
space configured values, sometimes by a substantial amount, which can
lead to out of order packets to be dropped instead of being queued. To
avoid these ofo drops, the condition to drop an out of order packet is
modified to allow out of order queuing to continue as long as it falls
within the now increased backlog queue limit.

Change-Id: I447ffc8560cb149fe84193c72bf693862f7ec740
Signed-off-by: Harout Hedeshian <harouth@codeaurora.org>
include/net/sock.h
net/ipv4/tcp_input.c