Re: ifconfig and some minor fixes.

[Alain Magloire]

> 
> > >   Not sure I understand the #include <sys/sockio.h>
> > >   but I've use it in if_index.c too.
> > 
> > We should remove it if nobody knows what it is for.  Sergey?
> > (I don't want to start the first release of ifconfig with anonymous cruft ;)
> 
> sys/sockio.h is where Solaris defines SIOCSIF.* and SIOCGIF.* stuff.
> It can safely be omitted provided <sys/ioctl.h> is included instead
> (which is the case with new if_index.c).

When I had a Solaris box (2.5.1)  bringing <sys/ioctl.h>
was not enough, you had to define BSD_COMP

So
#if defined(__svr4__)
# define BSD_COMP 1
#endif
#include <sys/ioctl.h>

Whithout BSD_COMP the SIOC{G,S}* would not be in.

> 
> > I reverted the SYSTEM_{SHORT,LONG}_OPTIONS change in options.c.  I don't
> > know why everybody thinks they should be defined.  They are only to be
> > defined if the system provides additional options.  I plan to use them to
> > decide if such options are provided later on.  Defining them to nothing will
> > defeat this.  Every place where they are used (only one currently, this is
> > still not worked out) will be protected by #ifdef.
> 
> Actually, the reason for this change was that SYSTEM_.*_OPTIONS
> were not protected by ifdef in the version 1.1 of options.c :^)
> 
> 
> Another note about if_index.c: It works fine, but there seems to be
> an inconsistency in using SIOCGIFINDEX. if_nameindex() contains:
> 
> # if defined(SIOCGIFINDEX)
>       if (ioctl (fd, SIOCGIFINDEX, cur) < 0)
>       idx[i].if_index = cur->ifr_index;
>       else
> # endif
>       idx[i].if_index =  i + 1;
> 
> i.e. if SIOCGIFINDEX is defined but fails, the information about
> the interface is returned anyway. On the other hand,
> if_nametoindex() just returns error when it fails:
> 
>   if (ioctl (fd, SIOCGIFINDEX, &ifr) < 0)
>     {
>       int saved_errno = errno;
>       close (fd);
>       if (saved_errno == EINVAL)
>       errno = ENOSYS;
>       return 0;
>     }
> 
> As usual, the problem manifests itself on Solaris :^)
> On the Solaris box where I am testing the stuff, SIOCGIFINDEX
> call fails for all interfaces. if_nameindex() returns all the
> interfaces correctly, but if_nametoindex() bails out and makes
> the program quit. After changing if_nametoindex() to work as
> if SIOCGIFINDEX was not defined, everything worked smoothly.
> Apparently, SIOCGIFINDEX is broken on this version of Solaris
> (it was 5.7 Generic). I am going to check it on SPARCs also. 
> Nevertheless I would suggest that both if_nametoindex() and
> if_nameindex() behave in the same manner: if SIOCGIFINDEX fails
> they both should fall back to the similar method of assigning
> the index. 

Ok, far the fallback, I can submit a patch tonight
if noone beats me to it.

--
alain