Re: [PATCH 1/9] exec_path 1/9: introduce ->exec_path and switch /proc/*/exe

[Alexey Dobriyan <adobriyan@gmail.com>]

On Fri, Jun 05, 2009 at 08:10:50AM -0700, Linus Torvalds wrote:
> On Fri, 5 Jun 2009, Christoph Hellwig wrote:
> >
> > On Thu, Jun 04, 2009 at 03:04:22AM +0400, Alexey Dobriyan wrote:
> > > ->exec_path is copied on clone(2) and put at do_exit() time.
> > > 
> > > ->exec_path is going to replace struct mm_struct::exe_file et al
> > > and allows to remove VM_EXECUTABLE flag while keeping readlink("/proc/*/exe")
> > > without loop over all VMAs.
> > 
> > Why don't you leave it in mm_struct?  That'll avoid having to mess with
> > it in clone,

There is no mess in clone:

	+       get_task_exec_path(current, &p->exec_path);

and there is no mess in exit:

	+       set_task_exec_path(tsk, &(struct path){ .mnt = NULL, .dentry = NULL });

> > and given that exec always replaces the whole VM it's a more natural fit anyway.
> 
> Oh, I didn't even notice that, because I just assumed it was in mm_struct 
> already due to the earlier discussion.
> 
> So I concur with Christoph - that field should be in the mm_struct. The 
> executable is a "mapping" issue,

This mapping issue is what created VM_EXECUTABLE/MAP_EXECUTABLE in
the first place, I assume? Never mind it's cheap hack.

> and goes along with task->mm, and should be there, not in task_struct.

Because ->mm can be borrowed to unrelated task and user has to check
for it.

> So it should be copied at fork() time when you do the dup_mm(), not 
> anywhere else.