[netperf-talk] MacOSX vs Linux

[Rick Jones]

Well, looks like Andrew was at the beginning of a trend.  Someone else 
has noticed a similar problem in 10.5.5 (and 10.5.6 IIRC) and posted 
about it in both comp.protocols.tcp-up and comp.sys.mac.system.  And it 
seems that someone may have an idea of what is wrong:


Path: 
usenet01.boi.hp.com!sdd.hp.com!ihnp4.UCSD.Edu!newsfeed.berkeley.edu!ucber
keley!newsfeed-00.mathworks.com!newsfeed2.dallas1.level3.net!news.level3.com!po
stnews.google.com!news2.google.com!Xl.tags.giganews.com!border1.nntp.dca.gigane
ws.com!nntp.giganews.com!local02.nntp.dca.giganews.com!nntp.speakeasy.net!news.
speakeasy.net.POSTED!not-for-mail
NNTP-Posting-Date: Wed, 17 Dec 2008 20:02:51 -0600
Newsgroups: comp.sys.mac.system
Subject: Re: Mac OSX tcp/ip problem
References: <gi8dgi$i1q$1 at newsreader2.netcologne.de> 
<jollyroger-D7EBD1.1023121
6122008 at news.individual.net> <giafnh$v5o$03$2 at news.t-online.com>
Organization: DiS(L)organized
From: russotto at grace.speakeasy.net (Matthew Russotto)
Message-ID: <07GdnZHmN-fWMtTUnZ2dnUVZ_q7inZ2d at speakeasy.net>
Date: Wed, 17 Dec 2008 20:02:51 -0600
Lines: 44
X-Usenet-Provider: http://www.giganews.com
NNTP-Posting-Host: 66.92.234.100
X-Trace: 
sv3-eN7cxmeZtq3w+cbrU6TJ6qd465PjZEzHQQSYtnx/Mi+eUIi6x/6DHNcbXBbpn7pTTh
8BgWO0Ex872by!nK7JDw6uobYeLf8xHGnpK/2U3CYuOCnSFMUgJdsgz+OUGdN28EjP1O31P2MJYmzF8
6tZuydP7/wJ!NJqhp6VpD8iZ+ZdCA3dHBZda2i0H5uekhg==
X-Abuse-and-DMCA-Info: Please be sure to forward a copy of ALL headers
X-Abuse-and-DMCA-Info: Otherwise we will be unable to process your 
complaint pr
operly
X-Postfilter: 1.3.39
Xref: usenet01.boi.hp.com comp.sys.mac.system:372389

In article <giafnh$v5o$03$2 at news.t-online.com>,
Peter Sprenger  <sprenger at moving-bytes.de> wrote:
 >Jolly Roger wrote:
 >> In article <gi8dgi$i1q$1 at newsreader2.netcologne.de>,
 >>  Peter Sprenger <sprenger at moving-bytes.de> wrote:
 >>
 >>> perhaps can somebody give me some info about a problem with the Mac
 >>> TCP/IP stack I encounter at the moment. The issue is that at the end of
 >>> a TCP/IP connection the Mac is not ACKing the last FIN of the opposite
 >>> connection. Example:
 >>>
 >>> Mac:  FIN/ACK
 >>> IPdevice: FIN/ACK
 >>> <here should come an ACK from Mac>
 >>> .
 >>> <timeout>
 >>> IPdevice: FIN/ACK
 >>> Mac:  ACK
 >>>
 >>> Any idea?
 >>
 >> How about providing some details, like which model Mac, what version
 >> operating system, how you noticed the problem to begin with, what tools
 >> you used to test the TCP/IP stack, and so on?
 >>
 >

 >It was a MacBook from a colleague, ver 10.5.6. I used Wireshark to
 > trace it, and you can see the problem with every HTTP connection to
 > the internet. Regardless if you use Safari, Firefox etc..

It's a bug in Apple's TCP implementation.  If you look in the XNU
project, file bsd/netinet/tcp_input.c, you'll see that the stack
processes the ACK flag before the FIN flag.  There's a special case in 
the ACK
processing for when the stack is in FIN_WAIT_1 state (sent FIN, not yet
ACKed), which switches the state to FIN_WAIT_2, then terminates
processing of the packet ("goto drop;") without processing the FIN
flag.  It's not clear why it does this; it seems to me at a quick
glance that simply removing the goto would result in correct
behavior.

It appears this bug has been here for a long time; xnu-792.13.8 (which
I happened to have lying around) has it.

rick jones