| Message ID | 20200310094653.33257-1-johannes.thumshirn@wdc.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Introduce Zone Append for writing to zoned block devices | expand |
On Tue, Mar 10, 2020 at 06:46:42PM +0900, Johannes Thumshirn wrote: > For null_blk the emulation is way simpler, as null_blk's zoned block > device emulation support already caches the write pointer position, so we > only need to report the position back to the upper layers. Additional > caching is not needed here. > > Testing has been conducted by translating RWF_APPEND DIOs into > REQ_OP_ZONE_APPEND commands in the block device's direct I/O function and > injecting errors by bypassing the block layer interface and directly > writing to the disc via the SCSI generic interface. We really need a user of this to be useful upstream. Didn't you plan to look into converting zonefs/iomap to use it? Without that it is at best a RFC. Even better would be converting zonefs and the f2fs zoned code so that can get rid of the old per-zone serialization in the I/O scheduler entirely.
On 2020/03/11 1:42, Christoph Hellwig wrote: > On Tue, Mar 10, 2020 at 06:46:42PM +0900, Johannes Thumshirn wrote: >> For null_blk the emulation is way simpler, as null_blk's zoned block >> device emulation support already caches the write pointer position, so we >> only need to report the position back to the upper layers. Additional >> caching is not needed here. >> >> Testing has been conducted by translating RWF_APPEND DIOs into >> REQ_OP_ZONE_APPEND commands in the block device's direct I/O function and >> injecting errors by bypassing the block layer interface and directly >> writing to the disc via the SCSI generic interface. > > We really need a user of this to be useful upstream. Didn't you plan > to look into converting zonefs/iomap to use it? Without that it is > at best a RFC. Even better would be converting zonefs and the f2fs > zoned code so that can get rid of the old per-zone serialization in > the I/O scheduler entirely. I do not think we can get rid of it entirely as it is needed for applications using regular writes on raw zoned block devices. But the zone write locking will be completely bypassed for zone append writes issued by file systems.
On Wed, Mar 11, 2020 at 12:37:33AM +0000, Damien Le Moal wrote: > I do not think we can get rid of it entirely as it is needed for applications > using regular writes on raw zoned block devices. But the zone write locking will > be completely bypassed for zone append writes issued by file systems. But applications that are aware of zones should not be sending multiple write commands to a zone anyway. We certainly can't use zone write locking for nvme if we want to be able to use multiple queues.
On 2020/03/11 15:25, Christoph Hellwig wrote: > On Wed, Mar 11, 2020 at 12:37:33AM +0000, Damien Le Moal wrote: >> I do not think we can get rid of it entirely as it is needed for applications >> using regular writes on raw zoned block devices. But the zone write locking will >> be completely bypassed for zone append writes issued by file systems. > > But applications that are aware of zones should not be sending multiple > write commands to a zone anyway. We certainly can't use zone write > locking for nvme if we want to be able to use multiple queues. > True, and that is the main use case I am seeing in the field. However, even for this to work properly, we will also need to have a special bio_add_page() function for regular writes to zones, similarly to zone append, to ensure that a large BIO does not become multiple requests, won't we ? Otherwise, a write bio submit will generate multiple requests that may get reordered on dispatch and on requeue (on SAS or on SATA). Furthermore, we already have aio supported. Customers in the field use that with fio libaio engine to test drives and for applications development. So I am afraid that removing the zone write locking now would break user space, no ? For nvme, we want to allow the "none" elevator as the default rather than mq-deadline which is now the default for all zoned block devices. This is a very simple change to the default elevator selection we can add based on the nonrot queue flag.
On 10/03/2020 17:42, Christoph Hellwig wrote: [...] > We really need a user of this to be useful upstream. Didn't you plan > to look into converting zonefs/iomap to use it? Without that it is > at best a RFC. Even better would be converting zonefs and the f2fs > zoned code so that can get rid of the old per-zone serialization in > the I/O scheduler entirely. Yes I'm right now working on iomap/zonefs support for zone append.
The upcoming NVMe ZNS Specification will define a new type of write command for zoned block devices, zone append. When when writing to a zoned block device using zone append, the start sector of the write is pointing at the start LBA of the zone to write to. Upon completion the block device will respond with the position the data has been placed in the zone. This from a high level perspective can be seen like a file system's block allocator, where the user writes to a file and the file-system takes care of the data placement on the device. In order to fully exploit the new zone append command in file-systems and other interfaces above the block layer, we choose to emulate zone append in SCSI and null_blk. This way we can have a single write path for both file-systems and other interfaces above the block-layer, like io_uring on zoned block devices, without having to care too much about the underlying characteristics of the device itself. The emulation works by providing a cache of each zone's write pointer, so zone append issued to the disk can be translated to a write with a starting LBA of the write pointer. This LBA is used as input zone number for the write pointer lookup in the zone write pointer offset cache and the cached offset is then added to the LBA to get the actual position to write the data. In SCSI we then turn the REQ_OP_ZONE_APPEND request into a WRITE(16) command. Upon successful completion of the WRITE(16), the cache will be updated to the new write pointer location and the written sector will be noted in the request. On error the cache entry will be marked as invalid and on the next write an update of the write pointer will be scheduled, before issuing the actual write. In order to reduce memory consumption, the only cached item is the offset of the write pointer from the start of the zone, everything else can be calculated. On an example drive with 52156 zones, the additional memory consumption of the cache is thus 52156 * 4 = 208624 Bytes or 51 4k Byte pages. The performance impact is neglectable for a spinning drive. For null_blk the emulation is way simpler, as null_blk's zoned block device emulation support already caches the write pointer position, so we only need to report the position back to the upper layers. Additional caching is not needed here. Testing has been conducted by translating RWF_APPEND DIOs into REQ_OP_ZONE_APPEND commands in the block device's direct I/O function and injecting errors by bypassing the block layer interface and directly writing to the disc via the SCSI generic interface. The whole series is relative to Jens' block-5.6 branch 14afc5936197 ("block, bfq: fix overwrite of bfq_group pointer in bfq_find_set_group()"). Damien Le Moal (2): null_blk: Support REQ_OP_ZONE_APPEND block: Introduce zone write pointer offset caching Johannes Thumshirn (8): block: provide fallbacks for blk_queue_zone_is_seq and blk_queue_zone_no block: introduce bio_add_append_page block: introduce BLK_STS_ZONE_RESOURCE block: introduce blk_req_zone_write_trylock block: factor out requeue handling from dispatch code block: delay un-dispatchable request scsi: sd_zbc: factor out sanity checks for zoned commands scsi: sd_zbc: emulate ZONE_APPEND commands Keith Busch (1): block: Introduce REQ_OP_ZONE_APPEND block/bio.c | 41 +++- block/blk-core.c | 49 +++++ block/blk-map.c | 2 +- block/blk-mq.c | 54 +++++- block/blk-settings.c | 16 ++ block/blk-sysfs.c | 15 +- block/blk-zoned.c | 83 +++++++- block/blk.h | 4 +- drivers/block/null_blk_main.c | 9 +- drivers/block/null_blk_zoned.c | 21 +- drivers/scsi/scsi_lib.c | 1 + drivers/scsi/sd.c | 28 ++- drivers/scsi/sd.h | 35 +++- drivers/scsi/sd_zbc.c | 344 +++++++++++++++++++++++++++++++-- include/linux/bio.h | 3 +- include/linux/blk_types.h | 14 ++ include/linux/blkdev.h | 42 +++- 17 files changed, 701 insertions(+), 60 deletions(-)