diff mbox series

virtio-blk: don't start dataplane during the stop of dataplane

Message ID 202310111414266586398@zte.com.cn
State New
Headers show
Series virtio-blk: don't start dataplane during the stop of dataplane | expand

Commit Message

lv.mengzhao@zte.com.cn Oct. 11, 2023, 6:14 a.m. UTC
From: hujian <hu.jian@zte.com.cn>

During the stop of dataplane for virtio-blk, virtio_bus_cleanup_host_notifier() is be
called to clean up notifier at the end, if polled ioeventfd, virtio_blk_handle_output()
is used to handle io request. But due to s->dataplane_disabled is false, it will be
returned directly, which drops io request.
Backtrace:
->virtio_blk_data_plane_stop
  ->virtio_bus_cleanup_host_notifier
    ->virtio_queue_host_notifier_read
      ->virtio_queue_notify_vq
        ->vq->handle_output
          ->virtio_blk_handle_output
            ->if (s->dataplane  && !s->dataplane_stoped)
              ->if (!s->dataplane_disabled)
                ->return *
            ->virtio_blk_handle_output_do
The above problem can occur when using "virsh reset" cmdline to reset guest, while
guest does io.
To fix this problem, don't try to start dataplane if s->stopping is true, and io would
be handled by virtio_blk_handle_vq().

Signed-off-by: hujian <hu.jian@zte.com.cn>
---
 hw/block/virtio-blk.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--
1.8.3.1

Comments

Stefan Hajnoczi Oct. 16, 2023, 7:39 p.m. UTC | #1
On Wed, Oct 11, 2023 at 02:14:26PM +0800, lv.mengzhao@zte.com.cn wrote:
> From: hujian <hu.jian@zte.com.cn>
> 
> During the stop of dataplane for virtio-blk, virtio_bus_cleanup_host_notifier() is be
> called to clean up notifier at the end, if polled ioeventfd, virtio_blk_handle_output()
> is used to handle io request. But due to s->dataplane_disabled is false, it will be
> returned directly, which drops io request.
> Backtrace:
> ->virtio_blk_data_plane_stop
>   ->virtio_bus_cleanup_host_notifier
>     ->virtio_queue_host_notifier_read
>       ->virtio_queue_notify_vq
>         ->vq->handle_output
>           ->virtio_blk_handle_output
>             ->if (s->dataplane  && !s->dataplane_stoped)
>               ->if (!s->dataplane_disabled)
>                 ->return *
>             ->virtio_blk_handle_output_do
> The above problem can occur when using "virsh reset" cmdline to reset guest, while
> guest does io.
> To fix this problem, don't try to start dataplane if s->stopping is true, and io would
> be handled by virtio_blk_handle_vq().
> 
> Signed-off-by: hujian <hu.jian@zte.com.cn>
> ---
>  hw/block/virtio-blk.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Thanks, applied to my block tree:
https://gitlab.com/stefanha/qemu/commits/block

Stefan
Stefan Hajnoczi Oct. 17, 2023, 2:03 p.m. UTC | #2
On Wed, Oct 11, 2023 at 02:14:26PM +0800, lv.mengzhao@zte.com.cn wrote:
> From: hujian <hu.jian@zte.com.cn>
> 
> During the stop of dataplane for virtio-blk, virtio_bus_cleanup_host_notifier() is be
> called to clean up notifier at the end, if polled ioeventfd, virtio_blk_handle_output()
> is used to handle io request. But due to s->dataplane_disabled is false, it will be
> returned directly, which drops io request.
> Backtrace:
> ->virtio_blk_data_plane_stop
>   ->virtio_bus_cleanup_host_notifier
>     ->virtio_queue_host_notifier_read
>       ->virtio_queue_notify_vq
>         ->vq->handle_output
>           ->virtio_blk_handle_output
>             ->if (s->dataplane  && !s->dataplane_stoped)
>               ->if (!s->dataplane_disabled)
>                 ->return *
>             ->virtio_blk_handle_output_do
> The above problem can occur when using "virsh reset" cmdline to reset guest, while
> guest does io.
> To fix this problem, don't try to start dataplane if s->stopping is true, and io would
> be handled by virtio_blk_handle_vq().
> 
> Signed-off-by: hujian <hu.jian@zte.com.cn>
> ---
>  hw/block/virtio-blk.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

I have dropped this patch again after Fiona pointed out it does not
compile and Kevin noticed that handling requests from the main loop
thread while the I/O is still being processed in the IOThread is going
to cause thread-safety issues.

Can you explain the problem you are seeing in more detail? You run
"virsh reset" while the guest is doing I/O. Then what happens?

Stefan

> 
> diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
> index 39e7f23..c2d5938 100644
> --- a/hw/block/virtio-blk.c
> +++ b/hw/block/virtio-blk.c
> @@ -1166,7 +1166,7 @@ static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
>  {
>      VirtIOBlock *s = (VirtIOBlock *)vdev;
> 
> -    if (s->dataplane && !s->dataplane_started) {
> +    if (s->dataplane && !s->dataplane_started && !s->stopping) {
>          /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
>           * dataplane here instead of waiting for .set_status().
>           */
> --
> 1.8.3.1
>
diff mbox series

Patch

diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 39e7f23..c2d5938 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -1166,7 +1166,7 @@  static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 {
     VirtIOBlock *s = (VirtIOBlock *)vdev;

-    if (s->dataplane && !s->dataplane_started) {
+    if (s->dataplane && !s->dataplane_started && !s->stopping) {
         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
          * dataplane here instead of waiting for .set_status().
          */