moving audit_free() up into do_exit()

[Alexander Viro]

On Thu, Mar 30, 2006 at 01:03:55AM -0800, Andrew Morton wrote:
> > +static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk, gfp_t gfp_mask)
> >  {
> >  -	char name[sizeof(current->comm)];
> >  -	struct mm_struct *mm = current->mm;
> >  +	char name[sizeof(tsk->comm)];
> >  +	struct mm_struct *mm = tsk->mm;
> >   	struct vm_area_struct *vma;
> >   
> >  -	get_task_comm(name, current);
> >  +	get_task_comm(name, tsk);
> >   	audit_log_format(ab, " comm=");
> >   	audit_log_untrustedstring(ab, name);
> > 
> 
> Am too lazy to apply the thing and check, but does `tsk' always equal
> `current' here?   If not, what pins the mm_struct?

It's either current or something being freed.  In the latter case we'd
already got ->mm set to NULL by that point.  Note that with the second
patch we'll _always_ have current here and it will have intact ->mm;
current code is called too late to do anything useful and the best we
can do is to avoid using ->mm of unrelated process (deadlock country
if we do that and it ends up non-NULL by accident; that's exactly what
the guy had hit).