Date: Tue, 27 Jan 1998 21:15:45 +0000 From: njs3@doc.ic.ac.uk (Niall Smart) To: Stefan Bethke <stefan@promo.de>, Niall Smart <njs3@doc.ic.ac.uk> Cc: hackers@FreeBSD.ORG Subject: Re: TCP_MAXSEG and path MTU discovery Message-ID: <E0xxIMD-0001jw-00@oak67.doc.ic.ac.uk> In-Reply-To: Stefan Bethke <stefan@promo.de> "Re: TCP_MAXSEG and path MTU discovery" (Jan 27, 9:29pm)
next in thread | previous in thread | raw e-mail | index | archive | help
On Jan 27, 9:29pm, Stefan Bethke wrote:
} Subject: Re: TCP_MAXSEG and path MTU discovery
> --On Die, 27. Jan 1998 19:10 Uhr +0000 "Niall Smart" <njs3@doc.ic.ac.uk>
> wrote:
>
> > if (setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &opt, optlen) < 0)
> > perror("setsockopt"), exit(1);
> >
> > if (connect(fd, (struct sockaddr*) &sin, sizeof(sin)) < 0)
> > perror("connect"), exit(1);
>
> > With argv[3] == "300", I'm getting this, 146.169.50.56 being my host.
> >
> > 19:04:07.726242 146.169.50.56.1171 > 146.169.46.7.21: S
> 1185413053:1185413053(0) win 16384 <mss 1460,nop,wscale
> 0,nop,nop,timestamp[|tcp]> (DF)
> > 19:04:07.728505 146.169.46.7.21 > 146.169.50.56.1171: S
> 4095294624:4095294624(0) ack 1185413054 win 8760 <mss 1460> (DF)
> > 19:04:07.728603 146.169.50.56.1171 > 146.169.46.7.21: . ack 4095294625 win
> 17520 (DF)
> >
> > What happened to my maximum segment size?? (Yes, the program prints 1460)
>
> Quoting from Stevens, TCP/IP Illustrated, Vol. II, pp. 1022:
>
> TCP_MAXSEG option
> [...] When a SYN is received from the other end with a MSS option, tcp_input
> calls tcp_mss, and t_maxseg can be set as high as the outgoing interface
> MTU. [1460 for Ethernet ...] Therefore, after a call to socket but before a
> connection is established, a process can only decrease the MSS from its
> default of 512. After a connection is established, the process can decrease
> the MSS from whatever was selected by tcp_mss.
>
> The logic here seems to be that a SYN with MSS is the equivalent of a
> setsockopt(TCP_MAXSEG).
>
> I've haven't read enought to say *why* this is useful, and Net/3 even
> doesn't include path MTU discovery, so Stevens doesn't explain it's
> implementation.
Stefen,
As far as I can see, this processing is only done on incoming SYN segments
(nothing to do with incoming SYN|ACK segments, or outgoing SYN segments)
>From tcp_input.c:1947
*
* NOTE that this routine is only called when we process an incoming
* segment, for outgoing segments only tcp_mssopt is called.
*
The function tcp_mssopt is commented:
/*
* Determine the MSS option to send on an outgoing SYN.
*/
(Bingo?) It uses a simple algorithm:
rt = tcp_rtlookup(tp->t_inpcb);
if (rt == NULL)
return tcp_mssdflt;
return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
I'm not familiar with the sockets data structures, but shouldn't this
function lookup the current value for t_maxseg and if it is valid,
use it? I'll look a little more into this now and send another email
soon.
Regards
Niall
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E0xxIMD-0001jw-00>