commit
566c09c53455d7c4f1 raid5: relieve lock contention in get_active_stripe()
modified the locking in get_active_stripe() reducing the range
protected by the (highly contended) device_lock.
Unfortunately it reduced the range too much opening up some races.
One race can occur if get_priority_stripe runs between the
test on sh->count and device_lock being taken.
This will mean that sh->lru is not empty while get_active_stripe
thinks ->count is zero resulting in a 'BUG' firing.
Another race happens if __release_stripe is called immediately
after sh->count is tested and found to be non-zero. If STRIPE_HANDLE
is not set, get_active_stripe should increment ->active_stripes
when it increments ->count from 0, but as it didn't think it was 0,
it doesn't.
Extending device_lock to cover the test on sh->count close these
races.
While we are here, fix the two BUG tests:
-If count is zero, then lru really must not be empty, or we've
lock the stripe_head somehow - no other tests are relevant.
-STRIPE_ON_RELEASE_LIST is completely independent of ->lru so
testing it is pointless.
Reported-and-tested-by: Brassow Jonathan <jbrassow@redhat.com>
Reviewed-by: Shaohua Li <shli@kernel.org>
Fixes: 566c09c53455d7c4f1
Signed-off-by: NeilBrown <neilb@suse.de>
} else
init_stripe(sh, sector, previous);
} else {
+ spin_lock(&conf->device_lock);
if (atomic_read(&sh->count)) {
BUG_ON(!list_empty(&sh->lru)
&& !test_bit(STRIPE_EXPANDING, &sh->state)
&& !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)
- && !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state));
+ );
} else {
- spin_lock(&conf->device_lock);
if (!test_bit(STRIPE_HANDLE, &sh->state))
atomic_inc(&conf->active_stripes);
- if (list_empty(&sh->lru) &&
- !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state) &&
- !test_bit(STRIPE_EXPANDING, &sh->state))
- BUG();
+ BUG_ON(list_empty(&sh->lru));
list_del_init(&sh->lru);
if (sh->group) {
sh->group->stripes_cnt--;
sh->group = NULL;
}
- spin_unlock(&conf->device_lock);
}
+ spin_unlock(&conf->device_lock);
}
} while (sh == NULL);