Short hostnames in URLs

[Thiago Macieira]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Cristian Tibirna wrote:
>Bind offers the cool feature of being able to list all records on a name
>server.
>
>$ nslookup
>
>> server <nameserver>.<domain>
>> ls -d <domain>
>
>I wonder if the bind libs/tools would allow to do this from code.

Yes, they would. What you're referring to is a simple Authority Transfer query 
(AXFR). If you want to test it, try this:

$ dig <domain> AXFR @<nameserver>.<domain>

It should list all records in that domain, starting and ending with the SOA 
(Start Of Authority).

>Problem is, AFAICT, this feature isn't well supported by all servers, and
>especially not by the wonderful windows2000 ones.

That's not the point. All DNS servers must support this call, since it's the 
standard way for seconday nameservers to obtain their records from the 
primary nameserver. (i.e., you set up your secondary and run named, it'll 
dispatch an AXFR to the primary and obtain the listing).

However, the behaviour you're referring to -- not all nameservers honour this 
request -- is because the DNS Admin took preventive measures. Normally, no 
one else but the secondary nameservers needs to do AXFR. So you can just 
block and ignore such requests from anyone else. In BIND, you'd add to your 
zone (in named.conf):
	allow-transfer { secondary-ip; secondary-ip; };

>With this information, we would be able to automatically build a cache, and
>probably update it once in a while (in the background).

You're asking us to maintain a cache of all domains searched. It wouldn't make 
sense to maintain such a cache at application-level. So, what you're really 
asking is to maintain a cache of the Internet. From my point of view, it's 
not feasible.

Now, AXFR queries are more sensitive than normal ones. First of all, as I said 
before, they could be simply ignored, which would mean timeout. And secondly, 
AXFR responses are large (more than 512 bytes), so they are usually 
transferred via TCP instead of UDP.

And mind you, what would happen if the code decided to transfer the whole of 
.com?

The solution to this is asynchronous DNS lookup, if at all. QDns honours 
/etc/hosts, but it doesn't honour /etc/nsswitch.conf, which means it won't 
try nis and nisplus lookups before trying DNS -- or anything else configured 
in nsswitch.conf, for that matter. I've been investigating how to do that for 
the past few days, and the only asynchronous lookup method in glibc is 
getaddrinfo_a. Short of that, only threading would work.

- -- 
  Thiago Macieira - UFOT Registry number: 1001
 thiagom at mail.com
   ICQ UIN: 1967141  PGP/GPG: 0x6EF45358
     Registered Linux user #65028
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.7 (GNU/Linux)

iD8DBQE9UlEGM/XwBW70U1gRAohLAJ4xOcrXXx3smzyvlha2tTfYqfD2PwCgsQpF
rVooDrDsOuAzS+jXAj7/whQ=
=02i/
-----END PGP SIGNATURE-----