privsep patch, Please test (take 3)

[Ben Lindstrom]

On Thu, 6 Jun 2002, Tim Rice wrote:

> On Thu, 6 Jun 2002, Kevin Steves wrote:
>
> > On Wed, Jun 05, 2002 at 06:22:39PM -0700, Tim Rice wrote:
> [snip]
> > > +#elif defined(HAVE_MMAP_DEV_ZERO_PRIVATE)
> > > +	address = mmap(NULL, size, PROT_WRITE|PROT_READ, MAP_PRIVATE,
> > > +	    open("/dev/zero", O_RDWR), 0);
> > > +#endif
> > >  	if (address == MAP_FAILED)
> > >  		fatal("mmap(%lu)", (u_long)size);
> > >  #else
> >
> > hmm, more ifdefs.  can there be xmmap() so there's one line of diff
> > between openbsd?  and again, i think we should use mm if possible.
>
> Privsep may be important enough to introduce another library dependency.
> It seems like it may address the platforms that have no mmap.
> Ben said he didn't like the idea of requiring another library.
> We haven't heard from Damien yet.
>

Not extactly what I said.  I said I would perfer to use the native mmap
*IF* it will work.  I have no problems falling back to mm if there is no
usable mmap().

Personally I could care less if we do xmmap()  that defaults to mmap() if
it works with anonymous.. Otherwise fail over to mm library.

I just think for the 80% of the population that will be using OpenSSH on
reasonable OSes to be punished into installing yet another library.

I've spent way too many hours tracking down software dependancies lately
while building a video editing platform that drive this point home.


BTW, I still don't think we can use MAP_PRIVATE.  Do we have a platform
where we can prove that MAP_PRIVATE w/ /dev/zero and compression all work
correctly together?

 - Ben