Bug 123583

Created attachment 100333 [details]
repro case for getnameinfo bug

This behavior is by design.  getnameinfo will return the IPv6 address
if /etc/resolv.conf has an "options" line containing "inet6".  I.e., add

options inet6

to that file and try again.

What we probably should do is to add this option to resolv.conf
whenever  /etc/sysconfig/network contains

NETWORKING_IPV6=yes

So this bug should probably be handled as a system-config-network bug.
 I'll reassign.

Well, that might work if every gethostbyname user would cope with
16-byte IPv4-mapped IPv6 addresses in the hostent, but they don't,
they expect 4-byte IPv4 addresses. So the gethostbyname world breaks.

Why can't getnameinfo() just DTRT here?

On FC2:

$ grep options /etc/resolv.conf
options inet6
$ ping www.redhat.com
PING www.redhat.com (127.0.0.1) 56(84) bytes of data.
64 bytes from localhost.localdomain (127.0.0.1): icmp_seq=0 ttl=64
time=0.038 ms64 bytes from localhost.localdomain (127.0.0.1):
icmp_seq=1 ttl=64 time=0.073 ms 
--- www.redhat.com ping statistics ---
2 packets transmitted, 2 received, 0% packet loss, time 999ms
rtt min/avg/max/mdev = 0.038/0.055/0.073/0.019 ms, pipe 2
$ grep options /etc/resolv.conf
$ ping www.redhat.com
PING www.redhat.com (209.132.177.50) 56(84) bytes of data.
 
--- www.redhat.com ping statistics ---
2 packets transmitted, 0 received, 100% packet loss, time 999ms

> Well, that might work if every gethostbyname user would cope with
> 16-byte IPv4-mapped IPv6 addresses in the hostent, but they don't,
> they expect 4-byte IPv4 addresses. So the gethostbyname world breaks.

No, why should it?  gethostbyname never returns IPv6 addresses. 
gethostbyname2 does it only if af == AF_INET6.  Likewise for
gethostbyaddr.  No program gets addresses in an unexpected format.


> Why can't getnameinfo() just DTRT here?

This is the intended behavior.

Empirically, that's exactly what it does, have you tried it? Also
matches what resolv.conf(5) says:

     inet6     sets RES_USE_INET6 in _res.options.  This has the
               effect of trying a AAAA query before an A query inside
               the gethostbyname function, and of mapping IPv4
               responses in IPv6 ââtunnelled formââ if no AAAA
               records are found but an A record set exists.

> > Why can't getnameinfo() just DTRT here?
> This is the intended behavior.

Why, what's the rationale for specifically ignoring /etc/hosts in this
case?

and what about bug #124376 in this context?

(in case it wasn't clear; getnameinfo does work correctly for
IPv4-mapped IPv6 addresses when it really uses DNS; this bug is just
that it fails to use /etc/hosts)

I have checked in a change which will, for this case, indeed cause the
correct result to be returned.  Still, adding the inet6 option is
still a good idea.


> and what about bug #124376 in this context?

Doesn't have much to do with it.  If the getaddrinfo() function is
used correctly, it will look up IPv6 addresses only if any interface
has an IPv6 address configured.  Since loading the ipv6 kernel modules
automatically assigns link-local addresses to all interface this means
IPv6 lookup happens every time when the ipv6 module is loaded.  To
prevent it the module must be stopped from being loaded.

The patch is in glibc-2.3.2-95.28 which ought to appear in U4 beta.

An errata has been issued which should help the problem 
described in this bug report. This report is therefore being 
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files, 
please follow the link below. You may reopen this bug report 
if the solution does not work for you.

http://rhn.redhat.com/errata/RHSA-2004-586.html