]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
futex: Handle transient "ownerless" rtmutex state correctly
authorMike Galbraith <efault@gmx.de>
Wed, 4 Nov 2020 15:12:44 +0000 (16:12 +0100)
committerThomas Gleixner <tglx@linutronix.de>
Sat, 7 Nov 2020 21:07:04 +0000 (22:07 +0100)
Gratian managed to trigger the BUG_ON(!newowner) in fixup_pi_state_owner().
This is one possible chain of events leading to this:

Task Prio       Operation
T1   120 lock(F)
T2   120 lock(F)   -> blocks (top waiter)
T3   50 (RT) lock(F)   -> boosts T1 and blocks (new top waiter)
XX    timeout/  -> wakes T2
signal
T1   50 unlock(F) -> wakes T3 (rtmutex->owner == NULL, waiter bit is set)
T2   120 cleanup   -> try_to_take_mutex() fails because T3 is the top waiter
           and the lower priority T2 cannot steal the lock.
        -> fixup_pi_state_owner() sees newowner == NULL -> BUG_ON()

The comment states that this is invalid and rt_mutex_real_owner() must
return a non NULL owner when the trylock failed, but in case of a queued
and woken up waiter rt_mutex_real_owner() == NULL is a valid transient
state. The higher priority waiter has simply not yet managed to take over
the rtmutex.

The BUG_ON() is therefore wrong and this is just another retry condition in
fixup_pi_state_owner().

Drop the locks, so that T3 can make progress, and then try the fixup again.

Gratian provided a great analysis, traces and a reproducer. The analysis is
to the point, but it confused the hell out of that tglx dude who had to
page in all the futex horrors again. Condensed version is above.

[ tglx: Wrote comment and changelog ]

Fixes: c1e2f0eaf015 ("futex: Avoid violating the 10th rule of futex")
Reported-by: Gratian Crisan <gratian.crisan@ni.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/87a6w6x7bb.fsf@ni.com
Link: https://lore.kernel.org/r/87sg9pkvf7.fsf@nanos.tec.linutronix.de
kernel/futex.c

index f8614ef4ff311b759d37530930ba05052363ff15..ac328874f6e583c29c652f2980202f56877cfa60 100644 (file)
@@ -2380,10 +2380,22 @@ retry:
                }
 
                /*
-                * Since we just failed the trylock; there must be an owner.
+                * The trylock just failed, so either there is an owner or
+                * there is a higher priority waiter than this one.
                 */
                newowner = rt_mutex_owner(&pi_state->pi_mutex);
-               BUG_ON(!newowner);
+               /*
+                * If the higher priority waiter has not yet taken over the
+                * rtmutex then newowner is NULL. We can't return here with
+                * that state because it's inconsistent vs. the user space
+                * state. So drop the locks and try again. It's a valid
+                * situation and not any different from the other retry
+                * conditions.
+                */
+               if (unlikely(!newowner)) {
+                       err = -EAGAIN;
+                       goto handle_err;
+               }
        } else {
                WARN_ON_ONCE(argowner != current);
                if (oldowner == current) {