Re: [PATCH 1/1] vhost: Protect the virtqueue from being cleared whilst still in use

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Mar 02, 2022 at 04:36:43PM +0100, Stefano Garzarella wrote:
> On Wed, Mar 02, 2022 at 09:50:38AM -0500, Michael S. Tsirkin wrote:
> > On Wed, Mar 02, 2022 at 03:11:21PM +0100, Stefano Garzarella wrote:
> > > On Wed, Mar 02, 2022 at 08:35:08AM -0500, Michael S. Tsirkin wrote:
> > > > On Wed, Mar 02, 2022 at 10:34:46AM +0100, Stefano Garzarella wrote:
> > > > > On Wed, Mar 02, 2022 at 07:54:21AM +0000, Lee Jones wrote:
> > > > > > vhost_vsock_handle_tx_kick() already holds the mutex during its call
> > > > > > to vhost_get_vq_desc().  All we have to do is take the same lock
> > > > > > during virtqueue clean-up and we mitigate the reported issues.
> > > > > >
> > > > > > Link: https://syzkaller.appspot.com/bug?extid=279432d30d825e63ba00
> > > > >
> > > > > This issue is similar to [1] that should be already fixed upstream by [2].
> > > > >
> > > > > However I think this patch would have prevented some issues, because
> > > > > vhost_vq_reset() sets vq->private to NULL, preventing the worker from
> > > > > running.
> > > > >
> > > > > Anyway I think that when we enter in vhost_dev_cleanup() the worker should
> > > > > be already stopped, so it shouldn't be necessary to take the mutex. But in
> > > > > order to prevent future issues maybe it's better to take them, so:
> > > > >
> > > > > Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
> > > > >
> > > > > [1]
> > > > > https://syzkaller.appspot.com/bug?id=993d8b5e64393ed9e6a70f9ae4de0119c605a822
> > > > > [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=a58da53ffd70294ebea8ecd0eb45fd0d74add9f9
> > > >
> > > >
> > > > Right. I want to queue this but I would like to get a warning
> > > > so we can detect issues like [2] before they cause more issues.
> > > 
> > > I agree, what about moving the warning that we already have higher up, right
> > > at the beginning of the function?
> > > 
> > > I mean something like this:
> > > 
> > > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
> > > index 59edb5a1ffe2..1721ff3f18c0 100644
> > > --- a/drivers/vhost/vhost.c
> > > +++ b/drivers/vhost/vhost.c
> > > @@ -692,6 +692,8 @@ void vhost_dev_cleanup(struct vhost_dev *dev)
> > >  {
> > >         int i;
> > > +       WARN_ON(!llist_empty(&dev->work_list));
> > > +
> > >         for (i = 0; i < dev->nvqs; ++i) {
> > >                 if (dev->vqs[i]->error_ctx)
> > >                         eventfd_ctx_put(dev->vqs[i]->error_ctx);
> > > @@ -712,7 +714,6 @@ void vhost_dev_cleanup(struct vhost_dev *dev)
> > >         dev->iotlb = NULL;
> > >         vhost_clear_msg(dev);
> > >         wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
> > > -       WARN_ON(!llist_empty(&dev->work_list));
> > >         if (dev->worker) {
> > >                 kthread_stop(dev->worker);
> > >                 dev->worker = NULL;
> > > 
> > 
> > Hmm I'm not sure why it matters.
> 
> Because after this new patch, putting locks in the while loop, when we
> finish the loop the workers should be stopped, because vhost_vq_reset() sets
> vq->private to NULL.
> 
> But the best thing IMHO is to check that there is no backend set for each
> vq, so the workers have been stopped correctly at this point.
> 
> Thanks,
> Stefano

It's the list of workers waiting to run though. That is not affected by
vq lock at all.

-- 
MST