locking issue in vhost_net_set_backend

locking issue in vhost_net_set_backend

[Jiri Slaby]

Hi,

Stanse found a locking problem in the following function:

static long vhost_net_set_backend(struct vhost_net *n, unsigned index, 
int fd)
{
         struct socket *sock, *oldsock;
         struct vhost_virtqueue *vq;
         int r;

         mutex_lock(&n->dev.mutex);
         r = vhost_dev_check_owner(&n->dev);
         if (r)
                 goto err;

         if (index >= VHOST_NET_VQ_MAX) {
                 r = -ENOBUFS;
                 goto err;
         }
         vq = n->vqs + index;
         mutex_lock(&vq->mutex);        <---  locked

         /* Verify that ring has been setup correctly. */
         if (!vhost_vq_access_ok(vq)) {
                 r = -EFAULT;
                 goto err;              <--- not unlocked
         }
         sock = get_socket(fd);
         if (IS_ERR(sock)) {
                 r = PTR_ERR(sock);
                 goto err;              <--- not unlocked
         }

         /* start polling new socket */
         oldsock = vq->private_data;
         if (sock == oldsock)
                 goto done;             <--- not unlocked

         vhost_net_disable_vq(n, vq);
         rcu_assign_pointer(vq->private_data, sock);
         vhost_net_enable_vq(n, vq);
         mutex_unlock(&vq->mutex);
done:
         if (oldsock) {
                 vhost_net_flush_vq(n, index);
                 fput(oldsock->file);
         }
err:
         mutex_unlock(&n->dev.mutex);
         return r;
}


I don't see how the lock is unlocked on the error paths and as it is not 
on none of the them maybe I'm missing something?

thanks,
-- 
js

Re: locking issue in vhost_net_set_backend

[Jiri Slaby]

On 03/16/2010 02:58 PM, Jiri Slaby wrote:
> I don't see how the lock is unlocked on the error paths and as it is not
> on none of the them maybe I'm missing something?

And there is one more issue in vhost_set_vring, several returns from 
VHOST_SET_VRING_KICK, VHOST_SET_VRING_CALL, VHOST_SET_VRING_ERR with the 
vq->mutex held.

-- 
js

Re: locking issue in vhost_net_set_backend

[Michael S. Tsirkin]

On Tue, Mar 16, 2010 at 03:01:51PM +0100, Jiri Slaby wrote:
> On 03/16/2010 02:58 PM, Jiri Slaby wrote:
>> I don't see how the lock is unlocked on the error paths and as it is not
>> on none of the them maybe I'm missing something?
>
> And there is one more issue in vhost_set_vring, several returns from  
> VHOST_SET_VRING_KICK, VHOST_SET_VRING_CALL, VHOST_SET_VRING_ERR with the  
> vq->mutex held.

Thanks!
I have patches for these that I'll push out shortly.

> -- 
> js

Re: locking issue in vhost_net_set_backend

[Jeff Dike]

On Tue, Mar 16, 2010 at 02:58:23PM +0100, Jiri Slaby wrote:
> I don't see how the lock is unlocked on the error paths and as it is not  
> on none of the them maybe I'm missing something?

I sent Michael a patch for this a couple of weeks ago.

				Jeff

-- 
Work email - jdike at linux dot intel dot com