[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20180518190141.899-1-dbanerje@akamai.com>
Date: Fri, 18 May 2018 15:01:40 -0400
From: Debabrata Banerjee <dbanerje@...mai.com>
To: "David S . Miller" <davem@...emloft.net>, netdev@...r.kernel.org
Cc: Alexey Kuznetsov <kuznet@....inr.ac.ru>,
Hideaki YOSHIFUJI <yoshfuji@...ux-ipv6.org>,
dbanerje@...mai.com
Subject: [PATCH RFC net-next 0/1] tcp: close socket without reset on incoming data
There is a basic problem with TCP sockets, where sending and closing of
data is unreliable. One good example of this is a web server that wants
to send an error back on a HTTP POST and close the socket, however
assuming the POST was of any significant size what really happens is
that the browser gets a broken socket while it is trying to post, and
never reads the error, possible retrying the whole POST a number of
times. This has been well documented by other people, for example this
blog post:
https://blog.netherlabs.nl/articles/2009/01/18/the-ultimate-so_linger-page-or-why-is-my-tcp-not-reliable
Without this patch, our server application has to hang on to a socket
sink all of the POST data, eating up memory and cpu. With this patch
the task is offloaded to the kernel, which uses only a timewait socket
to efficiently ack and discard any incoming data. We've been using a
similar patch internally for years, I think it has applications for
everyone.
Debabrata Banerjee (1):
tcp: close socket without reset on incoming data
include/linux/tcp.h | 4 +++-
include/uapi/linux/tcp.h | 2 +-
net/ipv4/tcp.c | 23 +++++++++++++++++++++--
net/ipv4/tcp_input.c | 16 ++++++++++++----
net/ipv4/tcp_minisocks.c | 15 +++++++++++++++
5 files changed, 52 insertions(+), 8 deletions(-)
--
2.17.0
Powered by blists - more mailing lists