netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
To: "David S. Miller" <davem@davemloft.net>, netdev@vger.kernel.org
Cc: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Subject: [PATCH 0/2] tcp: fix problems when tcp_fin_timeout is greater than 60
Date: Tue, 12 Feb 2013 21:49:44 +0900	[thread overview]
Message-ID: <1360673384.10638.10.camel@ubuntu-vm-makita> (raw)

Hi all,

I found a strange behavior of FIN_WAIT2 timer when tcp_fin_timeout is greater
than 60. 

When it is between 60 and 120, if the socket is closed, FIN_WAIT2 keepalive
timer of (tcp_fin_timeout - 60) seconds long starts. After it expires, timewait
timer of (tcp_fin_timeout - 60) seconds long starts again. This takes total time
of (tcp_fin_timeout - 60) * 2 seconds to disappear the FIN_WAIT2 socket, which
is shorter than tcp_fin_timeout.

# sysctl -w net.ipv4.tcp_fin_timeout=63
net.ipv4.tcp_fin_timeout = 63
# while :; do netstat -anot | grep 54321 | grep FIN_WAIT2; sleep 1; done
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   keepalive (2.62/0/0)
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   keepalive (1.59/0/0)
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   keepalive (0.56/0/0)
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   timewait (2.61/0/0)
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   timewait (1.59/0/0)
tcp        0      0 127.0.0.1:43034         127.0.0.1:54321         FIN_WAIT2   timewait (0.56/0/0)

When it is greater than 120, although timewait timer appears to start from
(tcp_fin_timeout - 60), it expires after 60 seconds elapse in practice.

# sysctl -w net.ipv4.tcp_fin_timeout=150
net.ipv4.tcp_fin_timeout = 150
# while :; do netstat -anot | grep 54321 | grep FIN_WAIT2; sleep 1; done
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (90.00/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (88.97/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (87.95/0/0)
...
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (2.76/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (1.73/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   keepalive (0.70/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (89.68/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (88.66/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (87.63/0/0)
...
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (32.21/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (31.18/0/0)
tcp        0      0 127.0.0.1:43036         127.0.0.1:54321         FIN_WAIT2   timewait (30.16/0/0)
(no more messages)

This seems to have been so for many years, but I think this behavior is not
desirable because it is confusing and does not match the documents.

Besides, it is also confusing that netstat shows keepalive timer first, and then
shows timewait timer, even though it is necessary to limit resources for the
orphaned socket.

I made patches that fix these problems.

Toshiaki Makita

             reply	other threads:[~2013-02-12 12:49 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-02-12 12:49 Toshiaki Makita [this message]
2013-02-12 12:52 ` [PATCH 1/2] tcp: fix too short FIN_WAIT2 time out Toshiaki Makita
2013-02-12 12:54   ` [PATCH 2/2] tcp: fix FIN_WAIT2 timer expression in /proc/net/tcp Toshiaki Makita
2013-02-12 14:57     ` Eric Dumazet
2013-02-12 15:13       ` Eric Dumazet
2013-02-15  8:26         ` Toshiaki Makita

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1360673384.10638.10.camel@ubuntu-vm-makita \
    --to=makita.toshiaki@lab.ntt.co.jp \
    --cc=davem@davemloft.net \
    --cc=netdev@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).