Re: [PATCH] ublk: Add UBLK_U_CMD_SET_SIZE

[Ming Lei <ming.lei@redhat.com>]

Hello Jared,

On Thu, Apr 03, 2025 at 12:37:11PM +0000, Jared Holzman wrote:
> Apologies if this is a dup, but I am not seeing the original mail on the mailing list archive.

I guess it is because the patch is sent as html, instead of plain test,
please follow the patch submission guide:

https://www.kernel.org/doc/Documentation/process/submitting-patches.rst

> 
> ________________________________
> From: Jared Holzman <jholzman@nvidia.com>
> Sent: Monday, 31 March 2025 4:54 PM
> To: linux-block@vger.kernel.org <linux-block@vger.kernel.org>
> Cc: ming.lei@redhat.com <ming.lei@redhat.com>; Omri Mann <omri@nvidia.com>; Ofer Oshri <ofer@nvidia.com>; Omri Levi <omril@nvidia.com>; Jared Holzman <jholzman@nvidia.com>
> Subject: [PATCH] ublk: Add UBLK_U_CMD_SET_SIZE
> 
> From: Omri Mann <omri@nvidia.com>
> 
> Currently ublk only allows the size of the ublkb block device to be
> set via UBLK_CMD_SET_PARAMS before UBLK_CMD_START_DEV is triggered.
> 
> This does not provide support for extendable user-space block devices
> without having to stop and restart the underlying ublkb block device
> causing IO interruption.

The requirement is reasonable.

> 
> This patch adds a new ublk command UBLK_U_CMD_SET_SIZE to allow the
> ublk block device to be resized on-the-fly.

Looks CMD_SET_SIZE is not generic enough, maybe UBLK_CMD_UPDATE_PARAMS
can be added for support any parameter update by allowing to do it
when device is in LIVE state.

> 
> Signed-off-by: Omri Mann <omri@nvidia.com>
> ---
>  Documentation/block/ublk.rst  |  5 +++++
>  drivers/block/ublk_drv.c      | 26 +++++++++++++++++++++++++-
>  include/uapi/linux/ublk_cmd.h |  7 +++++++
>  3 files changed, 37 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/block/ublk.rst b/Documentation/block/ublk.rst
> index 1e0e7358e14a..7eca87a66b9c 100644
> --- a/Documentation/block/ublk.rst
> +++ b/Documentation/block/ublk.rst
> @@ -139,6 +139,11 @@ managing and controlling ublk devices with help of several control commands:
>    set up the per-queue context efficiently, such as bind affine CPUs with IO
>    pthread and try to allocate buffers in IO thread context.
> 
> +- ``UBLK_F_SET_SIZE``
> +
> +  Allows changing the size of the block device after it has started. Useful for
> +  userspace implementations that allow extending the underlying block device.
> +
>  - ``UBLK_CMD_GET_DEV_INFO``
> 
>    For retrieving device info via ``ublksrv_ctrl_dev_info``. It is the server's
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index c060da409ed8..ab6364475b9c 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -64,7 +64,8 @@
>                  | UBLK_F_CMD_IOCTL_ENCODE \
>                  | UBLK_F_USER_COPY \
>                  | UBLK_F_ZONED \
> -               | UBLK_F_USER_RECOVERY_FAIL_IO)
> +               | UBLK_F_USER_RECOVERY_FAIL_IO \
> +               | UBLK_F_SET_SIZE)
> 
>  #define UBLK_F_ALL_RECOVERY_FLAGS (UBLK_F_USER_RECOVERY \
>                  | UBLK_F_USER_RECOVERY_REISSUE \
> @@ -2917,6 +2918,25 @@ static int ublk_ctrl_get_features(struct io_uring_cmd *cmd)
>          return 0;
>  }
> 
> +static int ublk_ctrl_set_size(struct ublk_device *ub,
> +               struct io_uring_cmd *cmd)
> +{
> +       const struct ublksrv_ctrl_cmd *header = io_uring_sqe_cmd(cmd->sqe);
> +       struct ublk_param_basic *p = &ub->params.basic;
> +       size_t new_size = (int)header->data[0];
> +       int ret = 0;
> +       unsigned int memflags;
> +
> +       p->dev_sectors = new_size;
> +
> +       memflags = blk_mq_freeze_queue(ub->ub_disk->queue);
> +       mutex_lock(&ub->mutex);
> +       set_capacity_and_notify(ub->ub_disk, p->dev_sectors);
> +       mutex_unlock(&ub->mutex);
> +       blk_mq_unfreeze_queue(ub->ub_disk->queue, memflags);

Actually if it is just for updating device size, queue freeze isn't needed,
because bio_check_eod() is called without grabbing ->q_usage_counter, but
for updating other parameters, freezing queue is often needed.


Thanks,
Ming