[e2e] Re: Question on "identification" field of IP header

[der Mouse]

> If the "Don't fragment bit" is set in the IP header, what purpose
> does the "identification" field serve?

It says which datagram is affected if an ICMP is returned in resposne
to it, if nothing else.

> I remember coming across some e-mail chain in one of the mailing
> lists (TCP-IMPL, e2e, TSVWG) about this issue and the interaction
> with NAT.

Heh.  My usual response to "NAT interacts badly with <foo>" is "don't
use NAT, then".  It just breaks too many of the assumptions underlying
IP-based networking.  But it wouldn't surprise me a bit if there were
NAT implementations broken enough to assume that packets with identical
IDs arriving within a small enough temporal window of one another were
actually the same packet.

Note that the portion of a packet returned in an ICMP does not include
even the IP source and destination addresses; the identification value
is almost the only value that can be relied upon to identify the
original packet upon getting an ICMP....

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse at rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B