Another 4.x ABI question; uidinfo

[Robert Watson]

On Sat, 2 Aug 2003, Mike Silbersack wrote:

> Ok, so I took another at the uidinfo ui_ref field being only a u_short
> in 4.x.  As I recall, the reason this was not changed to a u_int was
> binary compatibility... however, as I look around the source tree, it
> appears that the only places uidinfos are used are within kern/, and
> then generally only by things touching procs.  Thirdly, it appears that
> the refcount is only modified within three functions in kern_resource.c,
> and that ui_ref is the last member in that structure.
> 
> So, is there _really_ a problem in promoting ui_ref to an int from a
> short?  As far as I can tell, any network or sound driver should be
> completely insulated from such a change; do we have any other class of
> binary modules to worry about? 

I'm guessing it would be fine, then -- if nothing is accessing it using
libkvm, it shouldn't be a problem.  Changing the last field in the
structure won't change dereferences from other compiled kernel modules --
the only thing I think you'd need to worry about are reference count
changes in other modules (possibly, but unlikely), or statically allocated
uidinfo storage (unlikely).

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Network Associates Laboratories