Hi,
On 10/09/2015 10:24 AM, Felix Hübner wrote:
Hi all,
I have just reported a concurrency issue in the implementation of
sem_lock, see https://bugzilla.kernel.org/show_bug.cgi?id=105651
[...]
# P0 does spin_lock(&sem->lock); in line 336.
spin_lock(&sem->lock);
[...]
# P2 performs rest of semtimedop, increments complex_count and ends up
in line 1961 and starts to sleep.
return -1;
}
That is the problem: semtimedop() increments complex_count - thus
sem_wait_array() returns without a spin_unlock_wait() loop - but P0
already owns spin_lock(&sem->lock).
How do we want to fix it?
- revert my patch (simplify code, but slower for one corner case)
- add the missing sem_wait_array (more complex, but also better for
complex semops).
what do you think?
(patch untested)
--
Manfred
>From 0ce84d118e2ee7ebc98ad4a8cfd23f04ad45115c Mon Sep 17 00:00:00 2001
From: Manfred Spraul <[email protected]>
Date: Sat, 10 Oct 2015 08:37:22 +0200
Subject: [PATCH] ipc/sem.c: Alternative for fixing Concurrency bug
Two ideas for fixing the bug found by Felix:
- Revert my initial patch.
Problem: Significant slowdown for application that use large sem
arrays and complex operations: Every semop() does a loop
with spin_lock() on all semaphores.
- Add another sem_wait_array() that catches operations that are in
the middle of sem_lock().
What do you think? Is it worth to optimize for complex ops?
Reported-by: [email protected]
Signed-off-by: Manfred Spraul <[email protected]>
---
ipc/sem.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/ipc/sem.c b/ipc/sem.c
index b471e5a..9a55cfb 100644
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -1936,9 +1936,16 @@ SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
list_add_tail(&queue.list, &curr->pending_const);
}
} else {
- if (!sma->complex_count)
+ if (!sma->complex_count) {
merge_queues(sma);
+ /*
+ * squeeze out any simple operations that are in the middle
+ * of sem_lock()
+ */
+ sem_wait_array(sma);
+ }
+
if (alter)
list_add_tail(&queue.list, &sma->pending_alter);
else
--
2.4.3
