On 2020/06/26 4:48, Javier González wrote:
> On 25.06.2020 16:12, Matias Bjørling wrote:
>> On 25/06/2020 14.21, Javier González wrote:
>>> From: Javier González <javier.gonz@xxxxxxxxxxx>
>>>
>>> Add support for offline transition on the zoned block device using the
>>> new zone management IOCTL
>>>
>>> Signed-off-by: Javier González <javier.gonz@xxxxxxxxxxx>
>>> Signed-off-by: SelvaKumar S <selvakuma.s1@xxxxxxxxxxx>
>>> Signed-off-by: Kanchan Joshi <joshi.k@xxxxxxxxxxx>
>>> Signed-off-by: Nitesh Shetty <nj.shetty@xxxxxxxxxxx>
>>> ---
>>> block/blk-core.c | 2 ++
>>> block/blk-zoned.c | 3 +++
>>> drivers/nvme/host/core.c | 3 +++
>>> include/linux/blk_types.h | 3 +++
>>> include/linux/blkdev.h | 1 -
>>> include/uapi/linux/blkzoned.h | 1 +
>>> 6 files changed, 12 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/block/blk-core.c b/block/blk-core.c
>>> index 03252af8c82c..589cbdacc5ec 100644
>>> --- a/block/blk-core.c
>>> +++ b/block/blk-core.c
>>> @@ -140,6 +140,7 @@ static const char *const blk_op_name[] = {
>>> REQ_OP_NAME(ZONE_CLOSE),
>>> REQ_OP_NAME(ZONE_FINISH),
>>> REQ_OP_NAME(ZONE_APPEND),
>>> + REQ_OP_NAME(ZONE_OFFLINE),
>>> REQ_OP_NAME(WRITE_SAME),
>>> REQ_OP_NAME(WRITE_ZEROES),
>>> REQ_OP_NAME(SCSI_IN),
>>> @@ -1030,6 +1031,7 @@ generic_make_request_checks(struct bio *bio)
>>> case REQ_OP_ZONE_OPEN:
>>> case REQ_OP_ZONE_CLOSE:
>>> case REQ_OP_ZONE_FINISH:
>>> + case REQ_OP_ZONE_OFFLINE:
>>> if (!blk_queue_is_zoned(q))
>>> goto not_supported;
>>> break;
>>> diff --git a/block/blk-zoned.c b/block/blk-zoned.c
>>> index 29194388a1bb..704fc15813d1 100644
>>> --- a/block/blk-zoned.c
>>> +++ b/block/blk-zoned.c
>>> @@ -416,6 +416,9 @@ int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
>>> case BLK_ZONE_MGMT_RESET:
>>> op = REQ_OP_ZONE_RESET;
>>> break;
>>> + case BLK_ZONE_MGMT_OFFLINE:
>>> + op = REQ_OP_ZONE_OFFLINE;
>>> + break;
>>> default:
>>> return -ENOTTY;
>>> }
>>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>>> index f1215523792b..5b95c81d2a2d 100644
>>> --- a/drivers/nvme/host/core.c
>>> +++ b/drivers/nvme/host/core.c
>>> @@ -776,6 +776,9 @@ blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
>>> case REQ_OP_ZONE_FINISH:
>>> ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
>>> break;
>>> + case REQ_OP_ZONE_OFFLINE:
>>> + ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OFFLINE);
>>> + break;
>>> case REQ_OP_WRITE_ZEROES:
>>> ret = nvme_setup_write_zeroes(ns, req, cmd);
>>> break;
>>> diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
>>> index 16b57fb2b99c..b3921263c3dd 100644
>>> --- a/include/linux/blk_types.h
>>> +++ b/include/linux/blk_types.h
>>> @@ -316,6 +316,8 @@ enum req_opf {
>>> REQ_OP_ZONE_FINISH = 12,
>>> /* write data at the current zone write pointer */
>>> REQ_OP_ZONE_APPEND = 13,
>>> + /* Transition a zone to offline */
>>> + REQ_OP_ZONE_OFFLINE = 14,
>>> /* SCSI passthrough using struct scsi_request */
>>> REQ_OP_SCSI_IN = 32,
>>> @@ -456,6 +458,7 @@ static inline bool op_is_zone_mgmt(enum req_opf op)
>>> case REQ_OP_ZONE_OPEN:
>>> case REQ_OP_ZONE_CLOSE:
>>> case REQ_OP_ZONE_FINISH:
>>> + case REQ_OP_ZONE_OFFLINE:
>>> return true;
>>> default:
>>> return false;
>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
>>> index bd8521f94dc4..8308d8a3720b 100644
>>> --- a/include/linux/blkdev.h
>>> +++ b/include/linux/blkdev.h
>>> @@ -372,7 +372,6 @@ extern int blkdev_zone_ops_ioctl(struct block_device *bdev, fmode_t mode,
>>> unsigned int cmd, unsigned long arg);
>>> extern int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
>>> unsigned int cmd, unsigned long arg);
>>> -
>>> #else /* CONFIG_BLK_DEV_ZONED */
>>> static inline unsigned int blkdev_nr_zones(struct gendisk *disk)
>>> diff --git a/include/uapi/linux/blkzoned.h b/include/uapi/linux/blkzoned.h
>>> index a8c89fe58f97..d0978ee10fc7 100644
>>> --- a/include/uapi/linux/blkzoned.h
>>> +++ b/include/uapi/linux/blkzoned.h
>>> @@ -155,6 +155,7 @@ enum blk_zone_action {
>>> BLK_ZONE_MGMT_FINISH = 0x2,
>>> BLK_ZONE_MGMT_OPEN = 0x3,
>>> BLK_ZONE_MGMT_RESET = 0x4,
>>> + BLK_ZONE_MGMT_OFFLINE = 0x5,
>>> };
>>> /**
>>
>> I am not sure this makes sense to expose through the kernel zone api.
>> One of the goals of the kernel zone API is to be a layer that provides
>> an unified zone model across SMR HDDs and ZNS SSDs. The offline zone
>> operation, as defined in the ZNS specification, does not have an
>> equivalent in SMR HDDs (ZAC/ZBC).
>>
>> This is different from the Zone Capacity change, where the zone
>> capacity simply was zone size for SMR HDDs. Making it easy to support.
>> That is not the same for ZAC/ZBC, that does not offer the offline
>> operation to transition zones in read only state to offline state.
>
> I agree that an unified interface is desirable. However, the truth is
> that ZAC/ZBC are different, and will differ more and more and time goes
> by. We can deal with the differences at the driver level or with checks
> at the API level, but limiting ZNS with ZAC/ZBC is a hard constraint.
As long as you keep ZNS namespace report itself as being "host-managed" like
ZBC/ZAC disks, we need the consistency and common interface. If you break that,
the meaning of the zoned model queue attribute disappears and an application or
in-kernel user cannot rely on this model anymore to know how the drive will behave.
> Note too that I chose to only support this particular transition on the
> new management IOCTL to avoid confusion for existing ZAC/ZBC users.
>
> It would be good to clarify what is the plan for kernel APIs moving
> forward, as I believe there is a general desire to support new ZNS
> features, which will not necessarily be replicated in SMR drives.
What the drive is supposed to support and its behavior is determined by the
zoned model. ZNS standard was written so that most things have an equivalent
with ZBC/ZAC, e.g. the zone state machine is nearly identical. Differences are
either emulated (e.g. zone append scsi emulation), or not supported (e.g. zone
capacity change) so that the kernel follows the same pattern and maintains a
coherent behavior between device protocols for the host-managed model.
Think of a file system, or any other in-kernel user. If they have to change
their code based on the device type (NVMe vs SCSI), then the zoned block device
interface is broken. Right now, that is not the case, everything works equally
well on ZNS and SCSI, modulo the need by a user for conventional zones that ZNS
do not define. But that is still consistent with the host-managed model since
conventional zones are optional.
For this particular patch, there is currently no in-kernel user, and it is not
clear how this will be useful to applications. At least please clarify this. And
most likely, similarly to discard etc operations that are optional, having a
sysfs attribute and in-kernel API indicating if the drive supports offlining
zones will be needed. Otherwise, the caller will have to play with error codes
to understand if the drive does not support the command or if it is supported
but the command failed. Not nice. Better to know before issuing the command.
>
> Javier
>
--
Damien Le Moal
Western Digital Research
