Date: Fri, 29 Mar 2013 12:10:02 GMT From: Gleb Smirnoff <glebius@FreeBSD.org> To: freebsd-net@FreeBSD.org Subject: Re: misc/177456: An error of calculating TCP sequence number will resault in the machine to restart Message-ID: <201303291210.r2TCA2d1038533@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR kern/177456; it has been noted by GNATS.
From: Gleb Smirnoff <glebius@FreeBSD.org>
To: HouYeFei&XiBoLiu <lglion718@163.com>
Cc: freebsd-gnats-submit@FreeBSD.org
Subject: Re: misc/177456: An error of calculating TCP sequence number will
resault in the machine to restart
Date: Fri, 29 Mar 2013 16:08:03 +0400
HouYeFei & XiBoLiu,
On Thu, Mar 28, 2013 at 11:55:04PM +0000, HouYeFei&XiBoLiu wrote:
H> >Number: 177456
H> >Category: misc
H> >Synopsis: An error of calculating TCP sequence number will resault in the machine to restart
H> >Confidential: no
H> >Severity: non-critical
H> >Priority: low
H> >Responsible: freebsd-bugs
H> >State: open
H> >Quarter:
H> >Keywords:
H> >Date-Required:
H> >Class: sw-bug
H> >Submitter-Id: current-users
H> >Arrival-Date: Fri Mar 29 00:00:00 UTC 2013
H> >Closed-Date:
H> >Last-Modified:
H> >Originator: HouYeFei&XiBoLiu
H> >Release: FreeBSD-9.0
H> >Organization:
H> H3C
H> >Environment:
H> FreeBSD www.unixnotes.net 9.0-RELEASE FreeBSD 9.0-RELEASE #0: Sun May 4 12:36:15 HKT 2012 root@www.unixnotes.net:/usr/src/sys/i386/compile/unixnotes i386
H> >Description:
H> There is a large number of TCP links between Client and Server, each link can transmit large amounts of data. When the Client is low on memory, at the same time it wants to establish a new TCP connection to the server. The Client sends SYN message and startups retransmission timer, but retransmission of the first time
H>
H> sends failed because there is not enough mbuf.At this time, a sequence number is transmitted messages on the tcpcb (tp->snd_nxt) regression. Then
H>
H> a syn+ack message is received and processing the tp->snd_una sequence number is increased by 1, resault in tp->snd_nxt < th->snd_una. It is likely that
H>
H> the sending buffer has data to send, but actually is empty, call
H>
H> Tcp_output to send ack to the Server. But Tcp_output enter to the mbuf replication process, leading to access a null pointer.
I am trying to reproduce the problem, with no success yet.
Can you please clarify the sequence of failures that is required? I understand
your submission in the following way:
Client performs connect(2).
Client TCP stack generates SYN packet, and this packet is lost in network.
Client TCP stack tries to retransmit SYN packet, buf mbuf allocation fails.
Client TCP stack retransmits SYN packet.
Server replies with SYN+ACK.
... and according to you smth should go wrong ...
But in my tests nothing goes wrong. Client successfully retransmits SYN and
connection is established.
This is how I instrument this. I have added special TCP option and set it
before doing connect. The tcp_output() emulates failures that you
described:
Index: tcp_output.c
===================================================================
--- tcp_output.c (revision 248873)
+++ tcp_output.c (working copy)
@@ -898,6 +898,13 @@ send:
else
TCPSTAT_INC(tcps_sndwinup);
+ /* Fail allocating second packet. */
+ if (tp->t_flags & TF_ZHOPA && tp->t_zhopa == 1) {
+ tp->t_zhopa = 2;
+ m = NULL;
+ error = ENOBUFS;
+ goto out;
+ } else
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
error = ENOBUFS;
@@ -1273,6 +1280,13 @@ timer:
if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
ip->ip_off |= htons(IP_DF);
+ /* Lose first packet. */
+ if (tp->t_flags & TF_ZHOPA && tp->t_zhopa == 0) {
+ tp->t_zhopa = 1;
+ m_freem(m);
+ error = 0;
+ } else
+
error = ip_output(m, tp->t_inpcb->inp_options, &ro,
((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
tp->t_inpcb);
Am I doing something wrong? Can you provide your way to reproduce this?
Do you have backtrace of the panic?
--
Totus tuus, Glebius.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201303291210.r2TCA2d1038533>