Subject: Re: ICMP handling for a fragmented packet.
To: Rohit Dhawan <rohit@samsung.com>
From: Charles M. Hannum <root@ihack.net>
List: tech-net
Date: 03/18/2001 22:33:46
On Mon, Mar 19, 2001 at 11:51:44AM +0530, Rohit Dhawan wrote:
> 
> > You're thinking too hard.  ICMP TIME EXCEEDED is basically ignored by
> > a host, so the presence or absence of connection information in the
> > packet is irrelevant.
> >
> TLL was just an example, my real concern is regarding the handling of ICMP
> messages associated with fragments other than the first fragment, as they
> donot have any information of the higher layer protocol.

And the answer is the same.  The ones we respond to are all either MTU
related (and are used for host-based MTU discovery) or generate a
host-specific response rather than a connection-specific response.  The
TCP/UDP/RDP header is not needed.  I believe other systems follow the
same rules.