[PATCH 1/2] ublk: cancel device on START_DEV failure

[PATCH 1/2] ublk: cancel device on START_DEV failure

[Ming Lei]

When ublk_ctrl_start_dev() fails after waiting for completion, the
device needs to be properly cancelled to prevent leaving it in an
inconsistent state. Without this, pending I/O commands may remain
uncompleted and the device cannot be cleanly removed.

Add ublk_cancel_dev() call in the error path to ensure proper cleanup
when START_DEV fails.

Cc: <stable@vger.kernel.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/block/ublk_drv.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
index f6e5a0766721..2d6250d61a7b 100644
--- a/drivers/block/ublk_drv.c
+++ b/drivers/block/ublk_drv.c
@@ -2953,8 +2953,10 @@ static int ublk_ctrl_start_dev(struct ublk_device *ub,
 	if (wait_for_completion_interruptible(&ub->completion) != 0)
 		return -EINTR;
 
-	if (ub->ublksrv_tgid != ublksrv_pid)
-		return -EINVAL;
+	if (ub->ublksrv_tgid != ublksrv_pid) {
+		ret = -EINVAL;
+		goto out;
+	}
 
 	mutex_lock(&ub->mutex);
 	if (ub->dev_info.state == UBLK_S_DEV_LIVE ||
@@ -3017,6 +3019,9 @@ static int ublk_ctrl_start_dev(struct ublk_device *ub,
 		put_disk(disk);
 out_unlock:
 	mutex_unlock(&ub->mutex);
+out:
+	if (ret)
+		ublk_cancel_dev(ub);
 	return ret;
 }
 
-- 
2.47.0

Re: [PATCH 1/2] ublk: cancel device on START_DEV failure

[Caleb Sander Mateos]

On Sun, Jan 11, 2026 at 8:12 PM Ming Lei <ming.lei@redhat.com> wrote:
>
> When ublk_ctrl_start_dev() fails after waiting for completion, the
> device needs to be properly cancelled to prevent leaving it in an
> inconsistent state. Without this, pending I/O commands may remain
> uncompleted and the device cannot be cleanly removed.
>
> Add ublk_cancel_dev() call in the error path to ensure proper cleanup
> when START_DEV fails.

It's not clear to me why the UBLK_IO_FETCH_REQ commands must be
cancelled if UBLK_CMD_START_DEV fails. Wouldn't they get cancelled
whenever the ublk device is deleted or the ublk server exits?
And this seems like a UAPI change. Previously, the ublk server could
issue UBLK_CMD_START_DEV again if it failed, but now it must also
resubmit all the UBLK_IO_FETCH_REQ commands. This also means that
issuing UBLK_CMD_START_DEV after the ublk device has already been
started behaves like UBLK_CMD_STOP_DEV, which seems highly
unintuitive.

Best,
Caleb

>
> Cc: <stable@vger.kernel.org>
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>  drivers/block/ublk_drv.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index f6e5a0766721..2d6250d61a7b 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -2953,8 +2953,10 @@ static int ublk_ctrl_start_dev(struct ublk_device *ub,
>         if (wait_for_completion_interruptible(&ub->completion) != 0)
>                 return -EINTR;
>
> -       if (ub->ublksrv_tgid != ublksrv_pid)
> -               return -EINVAL;
> +       if (ub->ublksrv_tgid != ublksrv_pid) {
> +               ret = -EINVAL;
> +               goto out;
> +       }
>
>         mutex_lock(&ub->mutex);
>         if (ub->dev_info.state == UBLK_S_DEV_LIVE ||
> @@ -3017,6 +3019,9 @@ static int ublk_ctrl_start_dev(struct ublk_device *ub,
>                 put_disk(disk);
>  out_unlock:
>         mutex_unlock(&ub->mutex);
> +out:
> +       if (ret)
> +               ublk_cancel_dev(ub);
>         return ret;
>  }
>
> --
> 2.47.0
>

Re: [PATCH 1/2] ublk: cancel device on START_DEV failure

[Ming Lei]

On Mon, Jan 12, 2026 at 08:52:31AM -0800, Caleb Sander Mateos wrote:
> On Sun, Jan 11, 2026 at 8:12 PM Ming Lei <ming.lei@redhat.com> wrote:
> >
> > When ublk_ctrl_start_dev() fails after waiting for completion, the
> > device needs to be properly cancelled to prevent leaving it in an
> > inconsistent state. Without this, pending I/O commands may remain
> > uncompleted and the device cannot be cleanly removed.
> >
> > Add ublk_cancel_dev() call in the error path to ensure proper cleanup
> > when START_DEV fails.
> 
> It's not clear to me why the UBLK_IO_FETCH_REQ commands must be
> cancelled if UBLK_CMD_START_DEV fails. Wouldn't they get cancelled
> whenever the ublk device is deleted or the ublk server exits?

Good catch, DEL_DEV/STOP_DEV supposes to be capable of handling irrecoverable
START_DEV failure.

So this patch isn't needed.


Thanks, 
Ming