usage of the extended __res_state

[Petar Bogdanovic <petar at smokva dot net>]

Hi,

how exactly does the extended __res_state work?  I've spent several
hours with trial'n'error (resolver related documentation seems to be
scarce) and still wasn't able to get something like a general idea.

The basic stuff seems simple: _ext.nsaddr contains ipv6 addresses,
_ext.nscount6 contains the sum of all ipv6-servers.  What's left is
_ext.nsmap and according to Jakub Jelinek (in a post which is about
eight years old):

> ----- http://sourceware.org/ml/libc-hacker/2002-05/msg00035.html
>
> nsmap[X] = Y means that..
>
> 1. nsaddrs[X] is unused slot if Y == MAXNS
>
> 2. nsaddrs[X] is an IPv6 address if Y == MAXNS + 1
>
> 3. nsmap[X] = Y means that nsaddrs[X] corresponds to
>    nsaddr_list[Y] if Y < MAXNS
>
> -----

I can confirm 1. and 2. but while I think I also understood number 3.,
it certainly didn't work the way I expected it to work.  Here are three
things that I can remember:

   - When one address in nsaddr_list was `0.0.0.0', the resolver
     didn't just skip it, it tried 127.0.0.1.  Even if there were
     other ipv4 servers in the queue.  I thought 127.0.0.1 was the
     last resort when there is no resolv.conf an nothing was
     provided manually.

   - When having MAXNS nameservers in a ipv4/ipv6 resolv.conf, with
     ::1 being on the first place, the resolver tried ::1,
     127.0.0.1, 1.1.1.1 and skipped the last nameserver, 2.2.2.2.

   - When having MAXNS nameservers in a ipv4 resolv.conf (1.1.1.1,
     2.2.2.2, 3.3.3.3), manually setting nsaddr_list[0] to 0.0.0.0
     and nsaddrs[0] to ::1, the resolver tried 2.2.2.2 (twice) and
     then fell back to ::1.

I've read the related parts in resolv/res_init.c and it seemed mostly
consistent with the things I did in order to set/override an ipv6-
address.  So is there maybe something else that I have to keep in mind
when messing with a mixed (ipv4/ipv6) serverlist (or messing with
nsaddr_list/nsaddrs in general)?  A variation of res_setservers()
perhaps?

Also should one use __res_state at all?  The prefix somehow wants to
tell me that it's res_state I'm actually looking for.

Thanks,



   Petar Bogdanovic