[dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Ilya Maximets]

Application should be able to obtain information like 'ifname' from
the 'vid' passed to 'destroy_connection' callback. Currently, all the
API calls with passed 'vid' fails with 'device not found'.

Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
Cc: stable@dpdk.org

Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
---
 lib/librte_vhost/socket.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
index 3da9de62c..43f091d10 100644
--- a/lib/librte_vhost/socket.c
+++ b/lib/librte_vhost/socket.c
@@ -297,11 +297,12 @@ vhost_user_read_cb(int connfd, void *dat, int *remove)
 	if (ret < 0) {
 		close(connfd);
 		*remove = 1;
-		vhost_destroy_device(conn->vid);
 
 		if (vsocket->notify_ops->destroy_connection)
 			vsocket->notify_ops->destroy_connection(conn->vid);
 
+		vhost_destroy_device(conn->vid);
+
 		pthread_mutex_lock(&vsocket->conn_mutex);
 		TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 		pthread_mutex_unlock(&vsocket->conn_mutex);
-- 
2.17.1

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Jens Freimann]

On Tue, Apr 09, 2019 at 04:36:22PM +0300, Ilya Maximets wrote:
>Application should be able to obtain information like 'ifname' from
>the 'vid' passed to 'destroy_connection' callback. Currently, all the
>API calls with passed 'vid' fails with 'device not found'.
>
>Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
>Cc: stable@dpdk.org
>
>Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
>---
> lib/librte_vhost/socket.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>

Because the device is destroyed before the callback is done, vid is
not a valid index to look up device information any more. So, yes we
should destroy the device after the callback.


Reviewed-by: Jens Freimann <jfreimann@redhat.com>

regards,
Jens 

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[David Marchand]

On Tue, Apr 9, 2019 at 3:36 PM Ilya Maximets <i.maximets@samsung.com> wrote:

> Application should be able to obtain information like 'ifname' from
> the 'vid' passed to 'destroy_connection' callback. Currently, all the
> API calls with passed 'vid' fails with 'device not found'.
>
> Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
> Cc: stable@dpdk.org
>
> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
> ---
>  lib/librte_vhost/socket.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
> index 3da9de62c..43f091d10 100644
> --- a/lib/librte_vhost/socket.c
> +++ b/lib/librte_vhost/socket.c
> @@ -297,11 +297,12 @@ vhost_user_read_cb(int connfd, void *dat, int
> *remove)
>         if (ret < 0) {
>                 close(connfd);
>                 *remove = 1;
> -               vhost_destroy_device(conn->vid);
>
>                 if (vsocket->notify_ops->destroy_connection)
>                         vsocket->notify_ops->destroy_connection(conn->vid);
>
> +               vhost_destroy_device(conn->vid);
> +
>                 pthread_mutex_lock(&vsocket->conn_mutex);
>                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
>                 pthread_mutex_unlock(&vsocket->conn_mutex);
> --
> 2.17.1
>
>
Reviewed-by: David Marchand <david.marchand@redhat.com>

For vhost maintainers, looking at vhost_user_add_connection, aren't we
leaking a vid on errors ? either when new_connection notifier returns an
error, or after calling destroy_connection.


-- 
David Marchand

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Ilya Maximets]

On 10.04.2019 10:24, David Marchand wrote:
> 
> 
> On Tue, Apr 9, 2019 at 3:36 PM Ilya Maximets <i.maximets@samsung.com <mailto:i.maximets@samsung.com>> wrote:
> 
>     Application should be able to obtain information like 'ifname' from
>     the 'vid' passed to 'destroy_connection' callback. Currently, all the
>     API calls with passed 'vid' fails with 'device not found'.
> 
>     Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
>     Cc: stable@dpdk.org <mailto:stable@dpdk.org>
> 
>     Signed-off-by: Ilya Maximets <i.maximets@samsung.com <mailto:i.maximets@samsung.com>>
>     ---
>      lib/librte_vhost/socket.c | 3 ++-
>      1 file changed, 2 insertions(+), 1 deletion(-)
> 
>     diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
>     index 3da9de62c..43f091d10 100644
>     --- a/lib/librte_vhost/socket.c
>     +++ b/lib/librte_vhost/socket.c
>     @@ -297,11 +297,12 @@ vhost_user_read_cb(int connfd, void *dat, int *remove)
>             if (ret < 0) {
>                     close(connfd);
>                     *remove = 1;
>     -               vhost_destroy_device(conn->vid);
> 
>                     if (vsocket->notify_ops->destroy_connection)
>                             vsocket->notify_ops->destroy_connection(conn->vid);
> 
>     +               vhost_destroy_device(conn->vid);
>     +
>                     pthread_mutex_lock(&vsocket->conn_mutex);
>                     TAILQ_REMOVE(&vsocket->conn_list, conn, next);
>                     pthread_mutex_unlock(&vsocket->conn_mutex);
>     -- 
>     2.17.1
> 
> 
> Reviewed-by: David Marchand <david.marchand@redhat.com <mailto:david.marchand@redhat.com>>
> 
> For vhost maintainers, looking at vhost_user_add_connection, aren't we leaking a vid on errors ? either when new_connection notifier returns an error, or after calling destroy_connection.

I think that you're right.
I spotted that too yesterday while preparing this patch, just had no time to
check deeper. It should be safe to call 'vhost_destroy_device' in these cases.

Best regards, Ilya Maximets.

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[David Marchand]

On Wed, Apr 10, 2019 at 9:53 AM Ilya Maximets <i.maximets@samsung.com>
wrote:

> On 10.04.2019 10:24, David Marchand wrote:
> > For vhost maintainers, looking at vhost_user_add_connection, aren't we
> leaking a vid on errors ? either when new_connection notifier returns an
> error, or after calling destroy_connection.
>
> I think that you're right.
> I spotted that too yesterday while preparing this patch, just had no time
> to
> check deeper. It should be safe to call 'vhost_destroy_device' in these
> cases.
>

Yes, that's my understanding too.


-- 
David Marchand

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Ilya Maximets]

On 10.04.2019 11:07, David Marchand wrote:
> On Wed, Apr 10, 2019 at 9:53 AM Ilya Maximets <i.maximets@samsung.com <mailto:i.maximets@samsung.com>> wrote:
> 
>     On 10.04.2019 10:24, David Marchand wrote:
>     > For vhost maintainers, looking at vhost_user_add_connection, aren't we leaking a vid on errors ? either when new_connection notifier returns an error, or after calling destroy_connection.
> 
>     I think that you're right.
>     I spotted that too yesterday while preparing this patch, just had no time to
>     check deeper. It should be safe to call 'vhost_destroy_device' in these cases.
> 
> 
> Yes, that's my understanding too.

I'll prepare the patch for this.

> 
> 
> -- 
> David Marchand

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Tiwei Bie]

On Tue, Apr 09, 2019 at 04:36:22PM +0300, Ilya Maximets wrote:
> Application should be able to obtain information like 'ifname' from
> the 'vid' passed to 'destroy_connection' callback. Currently, all the
> API calls with passed 'vid' fails with 'device not found'.
> 
> Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
> ---
>  lib/librte_vhost/socket.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
> index 3da9de62c..43f091d10 100644
> --- a/lib/librte_vhost/socket.c
> +++ b/lib/librte_vhost/socket.c
> @@ -297,11 +297,12 @@ vhost_user_read_cb(int connfd, void *dat, int *remove)
>  	if (ret < 0) {
>  		close(connfd);
>  		*remove = 1;
> -		vhost_destroy_device(conn->vid);
>  
>  		if (vsocket->notify_ops->destroy_connection)
>  			vsocket->notify_ops->destroy_connection(conn->vid);
>  
> +		vhost_destroy_device(conn->vid);

By design, notify_ops->destroy_device() should be called before
notify_ops->destroy_connection(). I.e. something like:

	vhost_destroy_device_notify(...); // device stop

	if (vsocket->notify_ops->destroy_connection) // socket close
		vsocket->notify_ops->destroy_connection(conn->vid);

	vhost_destroy_device(conn->vid); // free

As destroy_device() callback means device stop (which also
happens in e.g. GET_VRING_BASE).


> +
>  		pthread_mutex_lock(&vsocket->conn_mutex);
>  		TAILQ_REMOVE(&vsocket->conn_list, conn, next);
>  		pthread_mutex_unlock(&vsocket->conn_mutex);
> -- 
> 2.17.1
> 

Re: [dpdk-stable] [PATCH] vhost: fix passing destroyed device to destroy callback

[Ilya Maximets]

On 12.04.2019 7:39, Tiwei Bie wrote:
> On Tue, Apr 09, 2019 at 04:36:22PM +0300, Ilya Maximets wrote:
>> Application should be able to obtain information like 'ifname' from
>> the 'vid' passed to 'destroy_connection' callback. Currently, all the
>> API calls with passed 'vid' fails with 'device not found'.
>>
>> Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
>> Cc: stable@dpdk.org
>>
>> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
>> ---
>>  lib/librte_vhost/socket.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
>> index 3da9de62c..43f091d10 100644
>> --- a/lib/librte_vhost/socket.c
>> +++ b/lib/librte_vhost/socket.c
>> @@ -297,11 +297,12 @@ vhost_user_read_cb(int connfd, void *dat, int *remove)
>>  	if (ret < 0) {
>>  		close(connfd);
>>  		*remove = 1;
>> -		vhost_destroy_device(conn->vid);
>>  
>>  		if (vsocket->notify_ops->destroy_connection)
>>  			vsocket->notify_ops->destroy_connection(conn->vid);
>>  
>> +		vhost_destroy_device(conn->vid);
> 
> By design, notify_ops->destroy_device() should be called before
> notify_ops->destroy_connection(). I.e. something like:
> 
> 	vhost_destroy_device_notify(...); // device stop
> 
> 	if (vsocket->notify_ops->destroy_connection) // socket close
> 		vsocket->notify_ops->destroy_connection(conn->vid);
> 
> 	vhost_destroy_device(conn->vid); // free
> 
> As destroy_device() callback means device stop (which also
> happens in e.g. GET_VRING_BASE).

Agree. Thanks. Will send v2.

> 
> 
>> +
>>  		pthread_mutex_lock(&vsocket->conn_mutex);
>>  		TAILQ_REMOVE(&vsocket->conn_list, conn, next);
>>  		pthread_mutex_unlock(&vsocket->conn_mutex);
>> -- 
>> 2.17.1
>>
> 
> 

[dpdk-stable] [PATCH v2] vhost: fix passing destroyed device to destroy callback

[Ilya Maximets]

Application should be able to obtain information like 'ifname' from
the 'vid' passed to 'destroy_connection' callback. Currently, all the
API calls with passed 'vid' fails with 'device not found'.

Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
Cc: stable@dpdk.org

Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
---

Version 2:
	* Fixed wrong order of 'destroy_device' and 'destroy_connection'.

 lib/librte_vhost/socket.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
index 3da9de62c..af09dfa17 100644
--- a/lib/librte_vhost/socket.c
+++ b/lib/librte_vhost/socket.c
@@ -295,13 +295,19 @@ vhost_user_read_cb(int connfd, void *dat, int *remove)
 
 	ret = vhost_user_msg_handler(conn->vid, connfd);
 	if (ret < 0) {
+		struct virtio_net *dev = get_device(conn->vid);
+
 		close(connfd);
 		*remove = 1;
-		vhost_destroy_device(conn->vid);
+
+		if (dev)
+			vhost_destroy_device_notify(dev);
 
 		if (vsocket->notify_ops->destroy_connection)
 			vsocket->notify_ops->destroy_connection(conn->vid);
 
+		vhost_destroy_device(conn->vid);
+
 		pthread_mutex_lock(&vsocket->conn_mutex);
 		TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 		pthread_mutex_unlock(&vsocket->conn_mutex);
-- 
2.17.1

Re: [dpdk-stable] [PATCH v2] vhost: fix passing destroyed device to destroy callback

[Tiwei Bie]

On Fri, Apr 12, 2019 at 11:32:28AM +0300, Ilya Maximets wrote:
> Application should be able to obtain information like 'ifname' from
> the 'vid' passed to 'destroy_connection' callback. Currently, all the
> API calls with passed 'vid' fails with 'device not found'.
> 
> Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
> ---
> 
> Version 2:
> 	* Fixed wrong order of 'destroy_device' and 'destroy_connection'.
> 
>  lib/librte_vhost/socket.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)

Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>

Re: [dpdk-stable] [PATCH v2] vhost: fix passing destroyed device to destroy callback

[Maxime Coquelin]

On 4/12/19 10:32 AM, Ilya Maximets wrote:
> Application should be able to obtain information like 'ifname' from
> the 'vid' passed to 'destroy_connection' callback. Currently, all the
> API calls with passed 'vid' fails with 'device not found'.
> 
> Fixes: efba12a78ddf ("vhost: add user callbacks for socket open/close")
> Cc:stable@dpdk.org
> 
> Signed-off-by: Ilya Maximets<i.maximets@samsung.com>
> ---
> 
> Version 2:
> 	* Fixed wrong order of 'destroy_device' and 'destroy_connection'.
> 
>   lib/librte_vhost/socket.c | 8 +++++++-
>   1 file changed, 7 insertions(+), 1 deletion(-)


Applied to dpdk-next-virtio/master.

Thanks,
Maxime