Infinite loops in client processes --

David Blacka davidb at verisignlabs.com
Fri Jul 28 11:23:51 EDT 2006


Paul Townsend wrote:

> Grumble.  Maybe, if a sufficient number of client programs that
> encountered this problem sent a "-notify ...", the filesystem containing
> the log would fill and the operators would have to look at it to see
> what was wrong (nasty grin).

Part of the issue here is that I'm not sure that rwhois.covad.net is
actually wrong.  Issuing a "punt" referral is the defined behavior.

>> Basically, it is just important for the client to to handle the referral
>> loop.
> 
> FWIW - rwhois-client-1.6.1 does not - it keeps following the referrals
> until it's interrupted.

I thought it was supposed to detect referral loops, so I think that is a
bug.

>> You are right, though.  It is totally unclear to which server the
>> -notify directive should be sent.  Looking at your examples, the problem
>> is actually rwhois.covad.net.
> 
> I agree - rwhois.covad.net	in the 74.0.0.0/15 loop
> 	  rwhois.insightns.com	in the 74.128.0.0/12 loop
> 	  rwhois.allstream.com	in the 207.245.0.0/18 loop
> 
> All three servers have a punt(?) referral to root.rwhois.net which has
> a punt(?) referral to rwhois.arin.net which has a link(?) referral to
> the starting server which ....  Perhaps a note should be added to all
> rwhois server setup/maintenance documentation that a referral following
> client should be used to verify that each piece actually works.

I'm thinking that, when encountering a "punt" referral (i.e., a referral
upwards), the client needs to just stop right there, which is how the
equivalent concept works in DNS.

-- 
David Blacka                      <davidb at verisignlabs.com>
Sr. Engineer    VeriSign Infrastructure Product Engineering


More information about the Rwhois mailing list