[bug-inetutils] Re: [PATCH] Make `logger' IPv6-capable.

[Mats Erik Andersson]

onsdag den 19 januari 2011 klockan 11:16 skrev Simon Josefsson detta:
> Mats Erik Andersson <address@hidden> writes:
> 
> > onsdag den 19 januari 2011 klockan 08:24 skrev Simon Josefsson detta:
> >> Mats Erik Andersson <address@hidden> writes:
> >> 
 [...]
> >
> > a) Do you like the call "logger --ipv6 ..." to print the warning to stderr,
> >    attempt the requested logging message via IPv4, and also exit execution
> >    with a non-zero exit code? Presently the exit code is zero, but the
> >    other two parts are included in my suggestion.
> >
> > b) Do you desire the IPv6 failure to log a local message about the
> >    inability to use that transport, thus producing two log messages:
> >    the intended text which is possibly non-local and the additional
> >    failure report which remains on the local workstation/server?
> 
> I haven't made up my mind, but wanted to bring this aspect up for
> discussion.
> 
> Other InetUtils tools fails when passed a -6 and IPv6 is not working,
> for example:
> 
> address@hidden:~$ telnet gnu.org
> Trying 140.186.70.148...
> ^C
> address@hidden:~$ telnet -6 gnu.org
> gnu.org/telnet: lookup failure: No address associated with hostname
> address@hidden:~$ 
> 
> Generally, it feels more natural for a tool to fail when it cannot do
> what it is asked to, rather than give a warning and do something else.

This behaviour is present if GNU Inetutils is compiled with IPv6 support:

   $ ./src/logger -6 -h gnu.org -p user.debug Testing IU-logger
   ./src/logger: no address associated with name, gnu.org:syslog.

   $ echo $?
   1

It is in a situation where compilation was done without IPv6 support,
that the logger client takes the liberty of ignoring `-6', uses IPv4,
sends the message, and emits a warning for the user. Thus it is really
single stacked IPv4-machines that are expected to rescue the user.
On all up to date machines the options `-4' and `-6' will abort with
errors if the host, port, and source are not able to fit the chosen
domain.

> 
> Typically users won't use --ipv4 or --ipv6 but rather use a hostname,
> and let the getaddrinfo() do its magic of selecting ipv4 vs ipv6, right?

This is an important point. Presently I have set AF_UNSPEC when both
flags are missing (and the compilation had "#define HAVE_IPV6 1").
Since UDP is carrying the message, this means that any dual stacked
host with existing A and AAAA DNS-entries might be contacted via a
domain where the syslog service does not listen, if it was set up
in an asymmetric manner.

Can it be argued that AF_INET still should be the default?
Bearing in mind that the migration to IPv6 is still in its
beginning. I do not know for certain yet.

Regards,
  Mats