Re: [MiNT] SYS_NMLN

[Guido Flohr <guido@imperia.net>]

Hi,

Frank Naumann wrote:
> Hello!
>
>
> > > > Brief check gave me that this constan represents length of the
> > > > entries of the utsname structure defined in sys/utsname.h.

Correct.

> > > > This constant should be also defined in sys/utsname.h and should
> > > > be at least 256, so adding

No, 256 is a quite large value.  It is not a required minimum.

> > > > #define SYS_NMLN       256
> > > >
> > > > seems resonable.
> > > >
> > > > Can the library mainainer(s) express their opinion about it?

I'm the ex-maintainer...

> After looking at it it's not POSIX (POSIX explicitly state that the length

The macro comes from SVID (System V Interface Description).

Since it is not ANSI C, it should not be defined unconditionally, but 
rather like this:

    #ifdef __USE_SVID
    # define SYS_NMLN 9
    #endif

But you should not define that macro at all for the MiNTLib because the 
definition of that constant suggests that all members of "struct 
utsname" have a constant length, which is _not_ the case with the 
MiNTLib.  Defining it to a wrong value, may lead to a buffer overflow. 
The only "safe" definition would be the minimum value 9.

So, what's the use of the macro? If it is used as the *constant* size of 
the structure members, you have to re-write the sources anyhow, since 
this assumption is wrong for MiNT.  If it is used to retrieve the size 
of a particular member, you can just as well use the sizeof operator.

As long as "struct utsname" uses these small sizes, I would rather not 
define SYS_NMLN at all.  This will intentionally provoke compile-time 
errors, that are easy to fix.

Ciao

Guido
--
Imperia AG, Development
Leyboldstr. 10 - D-50354 Hürth - http://www.imperia.net/