Thread (5 messages) 5 messages, 2 authors, 2020-10-22

Re: WARNING in md_ioctl

From: Song Liu <song@kernel.org>
Date: 2020-10-19 06:18:29
Also in: lkml

On Sat, Oct 17, 2020 at 4:13 AM Dae R. Jeong [off-list ref] wrote:
Hi,

I looked into the warning "WARNING in md_ioctl" found by Syzkaller.
(https://syzkaller.appspot.com/bug?id=fbf9eaea2e65bfcabb4e2750c3ab0892867edea1)
I suspect that it is caused by a race between two concurrenct ioctl()s as belows.

CPU1 (md_ioctl())                          CPU2 (md_ioctl())
------                                     ------
set_bit(MD_CLOSING, &mddev->flags);
did_set_md_closing = true;
                                           WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));

if(did_set_md_closing)
    clear_bit(MD_CLOSING, &mddev->flags);

If the above is correct, this warning is introduced
in the commit 065e519e("md: MD_CLOSING needs to be cleared after called md_set_readonly or do_md_stop").
Could you please take a look into this?
This is an interesting case. We try to protect against concurrent
ioctl via mddev->openers:

                if (mddev->pers && atomic_read(&mddev->openers) > 1) {
                        mutex_unlock(&mddev->open_mutex);
                        err = -EBUSY;
                        goto out;
                }

But in this case, we are sending multiple ioctl from the same fd, so
openers == 1.

We can probably do something like:
diff --git i/drivers/md/md.c w/drivers/md/md.c
index 6072782070230..49442a3f4605b 100644
--- i/drivers/md/md.c
+++ w/drivers/md/md.c
@@ -7591,8 +7591,10 @@ static int md_ioctl(struct block_device *bdev,
fmode_t mode,
                        err = -EBUSY;
                        goto out;
                }
-               WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
-               set_bit(MD_CLOSING, &mddev->flags);
+               if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
+                       err = -EBUSY;
+                       goto out;
+               }
                did_set_md_closing = true;
                mutex_unlock(&mddev->open_mutex);
                sync_blockdev(bdev);

Could you please test whether this fixes the issue?

Thanks,
Song
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help