Re: [PATCH 1/6] block: introduce IOCTL for zone mgmt

["Javier González" <javier@javigon.com>]

On 26.06.2020 07:03, Damien Le Moal wrote:
>On 2020/06/26 15:51, Javier González wrote:
>> On 26.06.2020 06:37, Damien Le Moal wrote:
>>> On 2020/06/26 15:01, Javier González wrote:
>>>> On 26.06.2020 01:17, Damien Le Moal wrote:
>>>>> On 2020/06/25 21:22, Javier González wrote:
>>>>>> From: Javier González <javier.gonz@samsung.com>
>>>>>>
>>>>>> The current IOCTL interface for zone management is limited by struct
>>>>>> blk_zone_range, which is unfortunately not extensible. Specially, the
>>>>>> lack of flags is problematic for ZNS zoned devices.
>>>>>>
>>>>>> This new IOCTL is designed to be a superset of the current one, with
>>>>>> support for flags and bits for extensibility.
>>>>>>
>>>>>> Signed-off-by: Javier González <javier.gonz@samsung.com>
>>>>>> Signed-off-by: SelvaKumar S <selvakuma.s1@samsung.com>
>>>>>> Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
>>>>>> Signed-off-by: Nitesh Shetty <nj.shetty@samsung.com>
>>>>>> ---
>>>>>>  block/blk-zoned.c             | 56 ++++++++++++++++++++++++++++++++++-
>>>>>>  block/ioctl.c                 |  2 ++
>>>>>>  include/linux/blkdev.h        |  9 ++++++
>>>>>>  include/uapi/linux/blkzoned.h | 33 +++++++++++++++++++++
>>>>>>  4 files changed, 99 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/block/blk-zoned.c b/block/blk-zoned.c
>>>>>> index 81152a260354..e87c60004dc5 100644
>>>>>> --- a/block/blk-zoned.c
>>>>>> +++ b/block/blk-zoned.c
>>>>>> @@ -322,7 +322,7 @@ int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>>   * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
>>>>>>   * Called from blkdev_ioctl.
>>>>>>   */
>>>>>> -int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>> +int blkdev_zone_ops_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>>  			   unsigned int cmd, unsigned long arg)
>>>>>>  {
>>>>>>  	void __user *argp = (void __user *)arg;
>>>>>> @@ -370,6 +370,60 @@ int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>>  				GFP_KERNEL);
>>>>>>  }
>>>>>>
>>>>>> +/*
>>>>>> + * Zone management ioctl processing. Extension of blkdev_zone_ops_ioctl(), with
>>>>>> + * blk_zone_mgmt structure.
>>>>>> + *
>>>>>> + * Called from blkdev_ioctl.
>>>>>> + */
>>>>>> +int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>> +			   unsigned int cmd, unsigned long arg)
>>>>>> +{
>>>>>> +	void __user *argp = (void __user *)arg;
>>>>>> +	struct request_queue *q;
>>>>>> +	struct blk_zone_mgmt zmgmt;
>>>>>> +	enum req_opf op;
>>>>>> +
>>>>>> +	if (!argp)
>>>>>> +		return -EINVAL;
>>>>>> +
>>>>>> +	q = bdev_get_queue(bdev);
>>>>>> +	if (!q)
>>>>>> +		return -ENXIO;
>>>>>> +
>>>>>> +	if (!blk_queue_is_zoned(q))
>>>>>> +		return -ENOTTY;
>>>>>> +
>>>>>> +	if (!capable(CAP_SYS_ADMIN))
>>>>>> +		return -EACCES;
>>>>>> +
>>>>>> +	if (!(mode & FMODE_WRITE))
>>>>>> +		return -EBADF;
>>>>>> +
>>>>>> +	if (copy_from_user(&zmgmt, argp, sizeof(struct blk_zone_mgmt)))
>>>>>> +		return -EFAULT;
>>>>>> +
>>>>>> +	switch (zmgmt.action) {
>>>>>> +	case BLK_ZONE_MGMT_CLOSE:
>>>>>> +		op = REQ_OP_ZONE_CLOSE;
>>>>>> +		break;
>>>>>> +	case BLK_ZONE_MGMT_FINISH:
>>>>>> +		op = REQ_OP_ZONE_FINISH;
>>>>>> +		break;
>>>>>> +	case BLK_ZONE_MGMT_OPEN:
>>>>>> +		op = REQ_OP_ZONE_OPEN;
>>>>>> +		break;
>>>>>> +	case BLK_ZONE_MGMT_RESET:
>>>>>> +		op = REQ_OP_ZONE_RESET;
>>>>>> +		break;
>>>>>> +	default:
>>>>>> +		return -ENOTTY;
>>>>>> +	}
>>>>>> +
>>>>>> +	return blkdev_zone_mgmt(bdev, op, zmgmt.sector, zmgmt.nr_sectors,
>>>>>> +				GFP_KERNEL);
>>>>>> +}
>>>>>> +
>>>>>>  static inline unsigned long *blk_alloc_zone_bitmap(int node,
>>>>>>  						   unsigned int nr_zones)
>>>>>>  {
>>>>>> diff --git a/block/ioctl.c b/block/ioctl.c
>>>>>> index bdb3bbb253d9..0ea29754e7dd 100644
>>>>>> --- a/block/ioctl.c
>>>>>> +++ b/block/ioctl.c
>>>>>> @@ -514,6 +514,8 @@ static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>>  	case BLKOPENZONE:
>>>>>>  	case BLKCLOSEZONE:
>>>>>>  	case BLKFINISHZONE:
>>>>>> +		return blkdev_zone_ops_ioctl(bdev, mode, cmd, arg);
>>>>>> +	case BLKMGMTZONE:
>>>>>>  		return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
>>>>>>  	case BLKGETZONESZ:
>>>>>>  		return put_uint(argp, bdev_zone_sectors(bdev));
>>>>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
>>>>>> index 8fd900998b4e..bd8521f94dc4 100644
>>>>>> --- a/include/linux/blkdev.h
>>>>>> +++ b/include/linux/blkdev.h
>>>>>> @@ -368,6 +368,8 @@ int blk_revalidate_disk_zones(struct gendisk *disk,
>>>>>>
>>>>>>  extern int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>>  				     unsigned int cmd, unsigned long arg);
>>>>>> +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);
>>>>>>
>>>>>> @@ -385,6 +387,13 @@ static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
>>>>>>  	return -ENOTTY;
>>>>>>  }
>>>>>>
>>>>>> +
>>>>>> +static inline int blkdev_zone_ops_ioctl(struct block_device *bdev, fmode_t mode,
>>>>>> +					unsigned int cmd, unsigned long arg)
>>>>>> +{
>>>>>> +	return -ENOTTY;
>>>>>> +}
>>>>>> +
>>>>>>  static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
>>>>>>  					 fmode_t mode, unsigned int cmd,
>>>>>>  					 unsigned long arg)
>>>>>> diff --git a/include/uapi/linux/blkzoned.h b/include/uapi/linux/blkzoned.h
>>>>>> index 42c3366cc25f..07b5fde21d9f 100644
>>>>>> --- a/include/uapi/linux/blkzoned.h
>>>>>> +++ b/include/uapi/linux/blkzoned.h
>>>>>> @@ -142,6 +142,38 @@ struct blk_zone_range {
>>>>>>  	__u64		nr_sectors;
>>>>>>  };
>>>>>>
>>>>>> +/**
>>>>>> + * enum blk_zone_action - Zone state transitions managed from user-space
>>>>>> + *
>>>>>> + * @BLK_ZONE_MGMT_CLOSE: Transition to Closed state
>>>>>> + * @BLK_ZONE_MGMT_FINISH: Transition to Finish state
>>>>>> + * @BLK_ZONE_MGMT_OPEN: Transition to Open state
>>>>>> + * @BLK_ZONE_MGMT_RESET: Transition to Reset state
>>>>>> + */
>>>>>> +enum blk_zone_action {
>>>>>> +	BLK_ZONE_MGMT_CLOSE	= 0x1,
>>>>>> +	BLK_ZONE_MGMT_FINISH	= 0x2,
>>>>>> +	BLK_ZONE_MGMT_OPEN	= 0x3,
>>>>>> +	BLK_ZONE_MGMT_RESET	= 0x4,
>>>>>> +};
>>>>>> +
>>>>>> +/**
>>>>>> + * struct blk_zone_mgmt - Extended zoned management
>>>>>> + *
>>>>>> + * @action: Zone action as in described in enum blk_zone_action
>>>>>> + * @flags: Flags for the action
>>>>>> + * @sector: Starting sector of the first zone to operate on
>>>>>> + * @nr_sectors: Total number of sectors of all zones to operate on
>>>>>> + */
>>>>>> +struct blk_zone_mgmt {
>>>>>> +	__u8		action;
>>>>>> +	__u8		resv3[3];
>>>>>> +	__u32		flags;
>>>>>> +	__u64		sector;
>>>>>> +	__u64		nr_sectors;
>>>>>> +	__u64		resv31;
>>>>>> +};
>>>>>> +
>>>>>>  /**
>>>>>>   * Zoned block device ioctl's:
>>>>>>   *
>>>>>> @@ -166,5 +198,6 @@ struct blk_zone_range {
>>>>>>  #define BLKOPENZONE	_IOW(0x12, 134, struct blk_zone_range)
>>>>>>  #define BLKCLOSEZONE	_IOW(0x12, 135, struct blk_zone_range)
>>>>>>  #define BLKFINISHZONE	_IOW(0x12, 136, struct blk_zone_range)
>>>>>> +#define BLKMGMTZONE	_IOR(0x12, 137, struct blk_zone_mgmt)
>>>>>>
>>>>>>  #endif /* _UAPI_BLKZONED_H */
>>>>>>
>>>>>
>>>>> Without defining what the flags can be, it is hard to judge what will change
>>>> >from the current distinct ioctls.
>>>>>
>>>>
>>>> The first flag is the one to select all. Down the line we have other
>>>> modifiers that make sense, but it is true that it is public yet.
>>>
>>> You mean *not* public ?
>>
>> Yes...
>>
>>>
>>>>
>>>> Would you like to wait until then or is it an option to revise the IOCTL
>>>> now?
>>>
>>> Yes. Wait until it is actually needed. Adding code that has no users makes it
>>> impossible to test so not acceptable. As for the "all zones" flag, I already
>>> commented about it.
>>
>> Ok. We will have this in the backlog then.
>>
>> It would be great if you and Matias would like to comment on it if you
>> have some ideas on how to improve it. Happy to set a branch somewhere to
>> keep a patchset with this functionality somewhere.
>
>I sent a much simpler version of this using a REQ_ZONE_ALL flag too, but driven
>by the specified sector range. That allowed to do reset, open, close, finish all
>zones using a single command much more simply than your patch. But as Christoph
>commented, the only real use case interesting for this is reset all (e.g. for FS
>format). open, close and finish all zones have no user...

Yes. The use-case here is format too, which might take some time on very
large drives.
>
>Let's see first what kind of flags may be needed in the future, if at all. We
>can then cook something if needed.

Makes sense.

Thanks Damien!

Javier