Re: [dpdk-dev] [PATCH] vhostuser: fix deadlock when vhost unregister

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Sunwenjie,

Thanks for your patch.

Please resend with following contribution guidelines, that can be
found in doc/guides/contributing/.

First, the commit message prefix should be vhost.
For other comments, please see inline:


On 1/8/19 12:45 PM, sunwenjie wrote:
>   When rte_vhost_driver_unregister  delete the connection fd, fdset_try_del will
>   always try and donot release the  vhostuser.mutex if the fd is busy, but the
>   fdset_event_dispatch will set the  fd to busy and call vhost_user_msg_handler
>   to get vhostuser.mutex, which will  cause deadlock.
> 

The commit message should not be indented, and wrapped at 72 chars.

> To fix it:
>      Unlock the vhost_user.mutex if fdset_try_del fail and relock it when
>      retry.

You need to add your Sined-off-by with your full name.

> ---
>   lib/librte_vhost/socket.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
> index 9cf34ad17..a9effa115 100644
> --- a/lib/librte_vhost/socket.c
> +++ b/lib/librte_vhost/socket.c
> @@ -961,13 +961,12 @@ rte_vhost_driver_unregister(const char *path)
>   	int count;
>   	struct vhost_user_connection *conn, *next;
>   
> +again:
>   	pthread_mutex_lock(&vhost_user.mutex);
>   
>   	for (i = 0; i < vhost_user.vsocket_cnt; i++) {
>   		struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
> -

Keep this new line, it is out of the scope of the change.

>   		if (!strcmp(vsocket->path, path)) {
> -again:
>   			pthread_mutex_lock(&vsocket->conn_mutex);
>   			for (conn = TAILQ_FIRST(&vsocket->conn_list);
>   			     conn != NULL;
> @@ -981,6 +980,7 @@ rte_vhost_driver_unregister(const char *path)
>   				 */
>   				if (fdset_try_del(&vhost_user.fdset,
>   						  conn->connfd) == -1) {
> +					pthread_mutex_unlock(&vhost_user.mutex);
>   					pthread_mutex_unlock(
>   							&vsocket->conn_mutex);
>   					goto again;
> 

Thanks,
Maxime