Re: [PATCH 1/3] futex: do not pagefault_disable in futex_atomic_cmpxchg_inatomic()

[Thomas Gleixner <tglx@linutronix.de>]

On Mon, 14 Mar 2011, Thomas Gleixner wrote:
> On Sun, 13 Mar 2011, Linus Torvalds wrote:
> > On Thu, Mar 10, 2011 at 6:47 PM, Michel Lespinasse <walken@google.com> wrote:
> That's my fault. 
> 
> I really checked the call sites of futex_atomic_cmpxchg_inatomic() and
> totally failed to see the one in handle_futex_death() which does not
> use the helper function cmpxchg_futex_value_locked(). That helper
> function is safe and does the right thing:
> 
> 	pagefault_disable();
> 	curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
> 	pagefault_enable();
> 
> So, that means we have all call sites covered except one, which needs
> to be fixed _AND_ also pushed into stable as all arch implementations
> except ARM rely on the caller doing the pagefault_disable().

After applying some coffee to my brain, I noticed that the ability to
fault in handle_futex_death() is desired. The get_user() before that
call covers the case where the futex is paged out, but it does not
handle the case where the futex is in a non writeable mapping. That
lacks a big fat comment at least.

So the removal of the pagefault_disable() in ARM is correct, just the
changelog and the comment there sucks. Sorry for not catching it.

Thinking more about it. Adding a comment is to handle_futex_death() is
good, but changing the code to make it entirely clear what is going on
is even better.

-------->
Subject: futex: Deobfuscate handle_futex_death()
From: Thomas Gleixner <tglx@linutronix.de>
Date: Mon, 14 Mar 2011 10:34:35 +0100

handle_futex_death() uses futex_atomic_cmpxchg_inatomic() without
disabling page faults. That's ok, but totally non obvious.

We don't hold locks so we actually can and want to fault here, because
the get_user() before futex_atomic_cmpxchg_inatomic() does not
guarantee a R/W mapping.

We could just add a big fat comment to explain this, but actually
changing the code so that the functionality is entirely clear is
better.

Use the helper function which disables page faults around the
futex_atomic_cmpxchg_inatomic() and handle a fault with a call to
fault_in_user_writeable() as all other places in the futex code do as
well.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/futex.c |   17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

Index: linux-2.6-tip/kernel/futex.c
===================================================================
--- linux-2.6-tip.orig/kernel/futex.c
+++ linux-2.6-tip/kernel/futex.c
@@ -2458,9 +2458,20 @@ retry:
 		 * userspace.
 		 */
 		mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
-		if (futex_atomic_cmpxchg_inatomic(&nval, uaddr, uval, mval))
-			return -1;
-
+		/*
+		 * We are not holding a lock here, but we want to have
+		 * the pagefault_disable/enable() protection because
+		 * we want to handle the fault gracefully. If the
+		 * access fails we try to fault in the futex with R/W
+		 * verification via get_user_pages. get_user() above
+		 * does not guarantee R/W access. If that fails we
+		 * give up and leave the futex locked.
+		 */
+		if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
+			if (fault_in_user_writeable(uaddr))
+				return -1;
+			goto retry;
+		}
 		if (nval != uval)
 			goto retry;