Re: modutils-2.3.15 'insmod'

[Andy Isaacson <adi@hexapodia.org>]

On Wed, Jul 09, 2003 at 12:08:23PM -0400, Kurt Wall wrote:
> Quoth Richard B. Johnson:
> > modutils-2.3.15, and probably later, has a bug that can prevent
> > modules from being loaded from initrd, this results in not
> > being able to mount a root file-system. The bug assumes that
> > malloc() will return a valid pointer when given an allocation
> > size of zero.
> 
> This isn't a bug. The standard allow returning a non-null pointer
> for malloc(0).

It's not literally a bug in libc -- the C standard says it's
implementation-defined whether malloc(0) returns NULL or a cookie -- but
it is definitely a bug (in a portable program) to depend on either
behavior from libc.  See ISO/IEC 9899:1999 7.20.3 paragraph 1.

> > The most recent `man` pages that RH 9.0 distributes states that
> > malloc() can return either NULL of a pointer that is valid for
> > free(). This, of course, depends upon the 'C' runtime library's
> > malloc() implementation.
> 
> Perhaps, but IIRC, the rationale in the GNU C library was that
> existing programs assume malloc(0) != 0, which allows you to call
> realloc on the pointer. Returning NULL only makes sense if the 
> malloc() call fails.

This paragraph is nonsensical, because realloc(malloc(0), 10) is
allowed, regardless of whether malloc(0) returns NULL or a cookie.
realloc(NULL, n) is allowed, and defined to be identical to malloc(n).
7.20.3.4 paragraph 3.

Geez, why does a trivial post about a bug in some program have to turn
into a pile of misleading statements and citations to ISO documents?

-andy