On Wed, Aug 31, 2022 at 11:51:33PM +0800, ZiyangZhang wrote:
> Define some macros for recovery feature. Especially define a new state:
> UBLK_S_DEV_RECOVERING which implies the ublk_device is recovering.
>
> UBLK_F_USER_RECOVERY implies that:
> (1) ublk_drv enables recovery feature. It won't let monitor_work to
> automatically abort rqs and release the device. Instead, it waits
> for user's START_USER_RECOVERY ctrl-cmd.
>
> (2) In monitor_work after a crash, ublk_drv ends(aborts) rqs issued to
> userspace(ublksrv) before crash.
>
> (3) In task work and ublk_queue_rq() after a crash, ublk_drv requeues
> rqs dispatched after crash.
>
> UBLK_F_USER_RECOVERY_REISSUE implies that:
> (1) everything UBLK_F_USER_RECOVERY implies except
> (2) ublk_drv requeues rqs issued to userspace(ublksrv) before crash.
>
> UBLK_F_USER_RECOVERY_REISSUE is designed for backends which:
> (1) tolerate double-writes because we may issue the same rq twice.
> (2) cannot let frontend users get I/O error, such as a RDONLY system.
>
> Signed-off-by: ZiyangZhang <ZiyangZhang@xxxxxxxxxxxxxxxxx>
> ---
> drivers/block/ublk_drv.c | 31 ++++++++++++++++++++++++++++++-
> include/uapi/linux/ublk_cmd.h | 7 +++++++
> 2 files changed, 37 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index 0c6db0978ed0..0c3d32e8d686 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -49,7 +49,9 @@
> /* All UBLK_F_* have to be included into UBLK_F_ALL */
> #define UBLK_F_ALL (UBLK_F_SUPPORT_ZERO_COPY \
> | UBLK_F_URING_CMD_COMP_IN_TASK \
> - | UBLK_F_NEED_GET_DATA)
> + | UBLK_F_NEED_GET_DATA \
> + | UBLK_F_USER_RECOVERY \
> + | UBLK_F_USER_RECOVERY_REISSUE)
>
> /* All UBLK_PARAM_TYPE_* should be included here */
> #define UBLK_PARAM_TYPE_ALL (UBLK_PARAM_TYPE_BASIC | UBLK_PARAM_TYPE_DISCARD)
> @@ -323,6 +325,33 @@ static inline int ublk_queue_cmd_buf_size(struct ublk_device *ub, int q_id)
> PAGE_SIZE);
> }
>
> +/*
> + * TODO: UBLK_F_USER_RECOVERY should be a flag for device, not for queue,
> + * since "some queues are aborted while others are recoverd" is really weird.
> + */
> +static inline bool ublk_can_use_recovery(struct ublk_device *ub)
> +{
> + struct ublk_queue *ubq = ublk_get_queue(ub, 0);
This way is too tricky, just wondering why you don't passe ubq to
ublk_can_use_recovery()?
Thanks,
Ming
