Re: [Tsvwg] minimum TCP MSS

On Nov 3, 2003, at 12:11 PM, Rick Jones wrote:

> It is my understanding that there is no minimum MSS specified for TCP 
> (well, one
> byte is perhaps implicit in TCP being a byte stream rather than a bit
> stream...), although some TCP _implementations_ may have settings to 
> reject
> connections advertising an MSS less than some value.

> Again, unless I am getting beyond my knowledge, 536 is the _default_ 
> MSS - the
> MSS that TCP should assume when it does not send or see an MSS option 
> in the SYN
> or SYN|ACK segment.

Yes, see RFC 1122, section 4.2.2.6

>
> I _suspect_ but cannot quote chapter and verse to confirm, that 536 
> stems from
> the requirement that a conforming IP stack must be able to reassemble 
> IP
> datagram fragments into IP datagrams of at least 576 bytes. That is the
> "minimum, maximum IP datagram size" an IP stack must support, not a 
> minimum MTU.
> (IIRC, the minimum IPv4 MTU is 68 bytes, and 12XX for v6)

For IPv4, RFC 791 defines the minimum IP packet size as 68 bytes.  On 
page 25:

     Every internet module must be able to forward a datagram of 68
     octets without further fragmentation.  This is because an internet
     header may be up to 60 octets, and the minimum fragment is 8 octets.

     Every internet destination must be able to receive a datagram of 576
     octets either in one piece or in fragments to be reassembled.

So, avoiding any IP fragmentation, that puts the minimum TCP MSS at 28 
bytes.  But since every IP must be able to receive a 576 byte packet, 
either as a single piece or in fragments, that's how the TCP default 
MSS of 536 is derived.

> If TCP "knows" that the "minimum, maximum IP datagram size" is 576, 
> then it
> "knows" that using an MSS of 536 (576 less 20 bytes of TCP header and 
> 20 bytes
> of IP header, options may vary :) means that its traffic has at least a
> snowball's chance of getting through.

TCP MSS doesn't include IP or TCP options, it's the underlying MTU 
minus the fixed IP and TCP headers.  Senders need to adjust down the 
amount of TCP data based on any IP or TCP options that it is inserting.

> rick jones

			David Borman, dab@cray.com

_______________________________________________
tsvwg mailing list
tsvwg@ietf.org
https://www1.ietf.org/mailman/listinfo/tsvwg