Subject: Re: kernel ip_randomid() and libc randomid(3) still "broken"
To: Dennis Ferguson <dennis@juniper.net>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: tech-net
Date: 11/26/2003 14:11:23
Dennis,

Huh? Who mentioned theoretical arguments?  Not me, I'm talking engineering.

People do run NFS over UDP over gig-e. that's just a fact.

Given that fact, reducing the space of the (already worrisomely small)
ip-id space from 65,536 down to the (even-more-worrisome) 12,000 or
so, is, to my mind, not acceptable. I honeslty dont' understand why
the proposal continues to be made.

I think the phrase "breaking IP" is justified. (I think it was kre's,
but I'm not 100% certain).

especially when the *strongest* argument made here in favour of
ranodmized IDs is that they're listed on some ramdom Web page as a
"low threat" security issue.

Quite frankly, if I wanted to do DoS attacks or inject splicing errors
into IP reassembly, randomizing IP Ids doesn't even slow me down much.