Subject: Re: direct copy() between two proc address spaces
To: der Mouse <mouse@Rodents.Montreal.QC.CA>
From: Jonathan Stone <>
List: tech-kern
Date: 06/21/2004 13:55:51
In message <200406211913.PAA12969@Sparkle.Rodents.Montreal.QC.CA>,
der Mouse writes:

I must be missing something.  Oh. Its that if you mmap with MAP_ANON,
you're not allowed to specify the fd, right?

If we made /dev/zero cloning, shouldn't you be able to mmap()
/dev/zero as MAP_SHARED, pass that fd to another (possibly unrelate)
process via PF_LOCAL, then mmap() that fd into the second process, and
(assuming cloning devices and map-to-anon works as expected) get the same
shared region?

Hmm. Catching up with email, I guess that's what Thor is asking, too.

Would it help if we redefine MAP_ANON|MAP_SHARED mmap()s to allow a
valid fd, and use the supplied fd to find the (single, by-new-definition,
to-be-shared) anon object? Or would that break standards conformance?