[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: don't-fragment-flag on ftp & icmp




Brian St. Denis <bstdenis@paranet.com> wrote:

>      Sun Solaris systems (at least 2.5 and up) do Path MTU discovery.  This 
>      is done by sending maximum sized ICMP packets with the "don't 
>      fragment" flag set.

Not quite.  

In IPv4, (which I'll assume is the topic of discussion) ,Path MTU
discovery (RFC1981) uses the DF bit in the IP header.  As such, it may
be present in any IP packet, TCP, UDP, ICMP, etc. If a router along
the path has a MTU that would require fragmentation, then it sends a
"fragmentation needed and DF set" ICMP message back (and drops the
original packet on the floor.)

Packets are never bigger than the 'first hop MTU'.  A maximally-sized
IP packet would be 65536 bytes long.  The only interface I know of
with a MTU this large is Hyperchannel (RFC1044).

> Once the Path MTU for the destination has been identified, all packets 
> sent to that destination will have the "don't fragment" flag set.  One 
> of the purposes of this is to detect if a network route has changed to 
> a network with a smaller MTU.

Though this how Solaris handles things, there is no requirement that
this happen.  (In fact, the RFC explicitly allows a host to stop
sending packets with DF set.)

> I know SunOS 4.x systems do not do Path MTU discovery and, therefore, 
> do not set the "don't fragment" flag.

Path MTU discovery is not the only only use for the DF bit.

-- 
Jim Thompson / Smallworks, Inc. / jim@smallworks.com  
      512 338 0619 phone / 512 338 0625 fax
"Faster, faster, until the thrill of speed overcomes the fear of death."
	 --Hunter S. Thompson




References: