Re: [Freestandards-ldps] Re: LDPS 0.91 - even closer to release

["Julie" <jockgrrl@ix.netcom.com>]

From: Andi Kleen <ak@suse.de>
> Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
>
> > > events (signals etc.) you don't need any additional locking neither,
> > > because the normal locking of the task (grabbing exit_sem, task list
lock etc.)
> > > is enough.
> >
> > no. People assume it cant change during a syscall
>
> There is nothing I could find in Single Unix nor in POSIX that requires
that.
>
> Who assumes that ? And if they do, do they assume the current per-thread
> cred model too?

I have to agree with Andi.  I've not encountered anything in any of the
specs which require that the effective UID cannot change during a
system call.  And the only way that assumption can be made is if
there are per-thread credentials, and that is definitely not required
by POSIX.

In AIX (and this is public knowledge -- it's published in the various
kernel programming texts as well as taught in our kernels classes)
we take a reference against the credentials structure for the
duration of their use.  When that reference is no longer needed,
it's freed.  Other threads may have other credentials during the
execution of that first thread's kernel code.

-- Julie.