Re: [RFC][PATCH] Replacing the /proc/<pid|self>/exe symlink code

[Matt Helsley]

On Wed, 2007-05-30 at 13:09 -0500, Serge E. Hallyn wrote:
> Quoting Matt Helsley (matthltc@xxxxxxxxxx):
> > This patch avoids holding the mmap semaphore while walking VMAs in response to
> > programs which read or follow the /proc/<pid|self>/exe symlink. This also allows us
> > to merge mmu and nommu proc_exe_link() functions. The costs are holding a separate
> > reference to the executable file stored in the task struct and increased code in
> > fork, exec, and exit paths.
> > 
> > Signed-off-by: Matt Helsley <matthltc@xxxxxxxxxx>
> > ---
> > 
> > Compiled and passed simple tests for regressions when patched against a 2.6.20
> > and 2.6.22-rc2-mm1 kernel.
> > 
> >  fs/exec.c             |    5 +++--
> >  fs/proc/base.c        |   20 ++++++++++++++++++++
> >  fs/proc/internal.h    |    1 -
> >  fs/proc/task_mmu.c    |   34 ----------------------------------
> >  fs/proc/task_nommu.c  |   34 ----------------------------------
> >  include/linux/sched.h |    1 +
> >  kernel/exit.c         |    2 ++
> >  kernel/fork.c         |   10 +++++++++-
> >  8 files changed, 35 insertions(+), 72 deletions(-)

<snip>

> > Index: linux-2.6.22-rc2-mm1/kernel/exit.c
> > ===================================================================
> > --- linux-2.6.22-rc2-mm1.orig/kernel/exit.c
> > +++ linux-2.6.22-rc2-mm1/kernel/exit.c
> > @@ -924,10 +924,12 @@ fastcall void do_exit(long code)
> >  	if (unlikely(tsk->audit_context))
> >  		audit_free(tsk);
> >  
> >  	taskstats_exit(tsk, group_dead);
> >  
> > +	if (tsk->exe_file)
> > +		fput(tsk->exe_file);
> 
> Hi,
> 
> just taking a cursory look so I may be missing something, but doesn't
> this leave the possibility that right here, with tsk->exe_file being
> put, another task would try to look at tsk's /proc/tsk->pid/exe?
> 
> thanks,
> -serge
>
>       exit_mm(tsk);
>
  
<snip>

Good question. To be precise, I think the problem doesn't exist here but
after the exit_mm() because there's a VMA that holds a reference to the
same file.

The existing code appears to solve the race between
reading/following /proc/tsk->pid/exe and exit_mm() in the exit path by
returning -ENOENT for the case where there is no executable VMA with a
reference to the file backing it.

So I need to put NULL in the exe_file field and adjust the return value
to be -ENOENT instead of -ENOSYS.

Thanks for the review!

Cheers,
	-Matt Helsley

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/