[dpdk-dev] *rte_vhost_rx_queue_count* should be protected by vq->access_lock

[dpdk-dev] *rte_vhost_rx_queue_count* should be protected by vq->access_lock

[He Peng]

Hi，


We found that *rte_vhost_rx_queue_count* is not protected by vq->access_lock,
and the access to vq->avail->idx is not thread-safe, since at the same time, the vq->avail  might be 

set by *vring_invalidate* when some vhost-user messages arrived, such as VRING_SET_ADDRESS,
VRING_SET_MEM_TABLE, etc.


Thanks. 

Re: [dpdk-dev] *rte_vhost_rx_queue_count* should be protected by vq->access_lock

[Tiwei Bie]

On Wed, Aug 14, 2019 at 03:31:09AM +0000, He Peng wrote:
> Hi，
> 
> We found that *rte_vhost_rx_queue_count* is not protected by vq->access_lock,
> and the access to vq->avail->idx is not thread-safe, since at the same time, the vq->avail  might be 
> 
> set by *vring_invalidate* when some vhost-user messages arrived, such as VRING_SET_ADDRESS,
> VRING_SET_MEM_TABLE, etc.

You are right. And other similar APIs also need to be protected.
Thanks for reporting this!

Thanks,
Tiwei


> 
> 
> Thanks. 
> 
> 
> 
> 

Re: [dpdk-dev] *rte_vhost_rx_queue_count* should be protected by vq->access_lock

[He Peng]

Hi,


Would you please provide  a list of impacted APIs, thus we can check if our code has further issues. b.t.w, we are now using dpdk 18.11.


Thanks.

2019年8月14日 下午11:09，Tiwei Bie <tiwei.bie@intel.com> 写道：


On Wed, Aug 14, 2019 at 03:31:09AM +0000, He Peng wrote:

Hi，


We found that *rte_vhost_rx_queue_count* is not protected by vq->access_lock,
and the access to vq->avail->idx is not thread-safe, since at the same time, the vq->avail  might be 


set by *vring_invalidate* when some vhost-user messages arrived, such as VRING_SET_ADDRESS,
VRING_SET_MEM_TABLE, etc.

You are right. And other similar APIs also need to be protected.
Thanks for reporting this!

Thanks,
Tiwei







Thanks. 

Re: [dpdk-dev] *rte_vhost_rx_queue_count* should be protected by vq->access_lock

[Tiwei Bie]

On Thu, Aug 15, 2019 at 09:14:52AM +0000, He Peng wrote:
> Hi,
> 
> 
> Would you please provide  a list of impacted APIs, thus we can check if our code has further issues. b.t.w, we are now using dpdk 18.11.

I'm looking into this issue. The API of the builtin backend
is supposed to be protected like enqueue/dequeue. So e.g.
rte_vhost_enable_guest_notification should also be protected.
I'll keep you in the CC list of the fix patch.

Thanks,
Tiwei

> 
> 
> Thanks.
> 
> 2019年8月14日 下午11:09，Tiwei Bie <tiwei.bie@intel.com> 写道：
> 
> 
> On Wed, Aug 14, 2019 at 03:31:09AM +0000, He Peng wrote:
> 
> Hi，
> 
> 
> We found that *rte_vhost_rx_queue_count* is not protected by vq->access_lock,
> and the access to vq->avail->idx is not thread-safe, since at the same time, the vq->avail  might be 
> 
> 
> set by *vring_invalidate* when some vhost-user messages arrived, such as VRING_SET_ADDRESS,
> VRING_SET_MEM_TABLE, etc.
> 
> You are right. And other similar APIs also need to be protected.
> Thanks for reporting this!
> 
> Thanks,
> Tiwei
> 
> 
> 
> 
> 
> 
> 
> Thanks. 
> 
> 
> 
> 
> 
> 
>