On Wed, Aug 18, 2021 at 5:07 AM Darrick J. Wong <djwong@xxxxxxxxxx> wrote: > > On Tue, Aug 17, 2021 at 03:44:16PM +0530, SelvaKumar S wrote: > > This started out as an attempt to support NVMe Simple Copy Command (SCC), > > and evolved during the RFC review process. > > > > The patchset, at this point, contains - > > 1. SCC support in NVMe driver > > 2. Block-layer infra for copy-offload operation > > 3. ioctl interface to user-space > > 4. copy-emulation infra in the block-layer > > 5. copy-offload plumbing to dm-kcopyd (thus creating couple of in-kernel > > users such as dm-clone) > > > > > > The SCC specification, i.e. TP4065a can be found in following link > > > > https://nvmexpress.org/wp-content/uploads/NVM-Express-1.4-Ratified-TPs.zip > > > > Simple copy is a copy offloading feature and can be used to copy multiple > > contiguous ranges (source_ranges) of LBA's to a single destination LBA > > within the device, reducing traffic between host and device. > > > > We define a block ioctl for copy and copy payload format similar to > > discard. For device supporting native simple copy, we attach the control > > information as payload to the bio and submit to the device. Copy emulation > > is implemented incase underlaying device does not support copy offload or > > based on sysfs choice. Copy emulation is done by reading each source range > > into buffer and writing it to the destination. > > Seems useful. Would you mind adapting the loop driver to call > copy_file_range (for files) so that anyone interested in making a > filesystem use this capability (cough) can write fstests? We are planning to look into copy_file_range plumbing after settling on the current series, which already became heavyweight for the first drop. Nitesh Shetty > > --D > > > At present this implementation does not support copy offload for stacked/dm > > devices, rather copy operation is completed through emulation. > > > > One of the in-kernel use case for copy-offload is implemented by this > > patchset. dm-kcopyd infra has been changed to leverage the copy-offload if > > it is natively available. Other future use-cases could be F2FS GC, BTRFS > > relocation/balance and copy_file_range. > > > > Following limits are added to queue limits and are exposed via sysfs to > > userspace, which user can use to form a payload. > > > > - copy_offload: > > configurable, can be used set emulation or copy offload > > 0 to disable copy offload, > > 1 to enable copy offloading support. Offload can be only > > enabled, if underlaying device supports offload > > > > - max_copy_sectors: > > total copy length supported by copy offload feature in device. > > 0 indicates copy offload is not supported. > > > > - max_copy_nr_ranges: > > maximum number of source range entries supported by copy offload > > feature in device > > > > - max_copy_range_sectors: > > maximum copy length per source range entry > > > > *blkdev_issue_copy* takes source bdev, no of sources, array of source > > ranges (in sectors), destination bdev and destination offset(in sectors). > > If both source and destination block devices are same and queue parameter > > copy_offload is 1, then copy is done through native copy offloading. > > Copy emulation is used in otherwise. > > > > Changes from RFC v5 > > > > 1. Handle copy larger than maximum copy limits > > 2. Create copy context and submit copy offload asynchronously > > 3. Remove BLKDEV_COPY_NOEMULATION opt-in option of copy offload and > > check for copy support before submission from dm and other layers > > 4. Allocate maximum possible allocatable buffer for copy emulation > > rather failing very large copy offload. > > 5. Fix copy_offload sysfs to be either have 0 or 1 > > > > Changes from RFC v4 > > > > 1. Extend dm-kcopyd to leverage copy-offload, while copying within the > > same device. The other approach was to have copy-emulation by moving > > dm-kcopyd to block layer. But it also required moving core dm-io infra, > > causing a massive churn across multiple dm-targets. > > 2. Remove export in bio_map_kern() > > 3. Change copy_offload sysfs to accept 0 or else > > 4. Rename copy support flag to QUEUE_FLAG_SIMPLE_COPY > > 5. Rename payload entries, add source bdev field to be used while > > partition remapping, remove copy_size > > 6. Change the blkdev_issue_copy() interface to accept destination and > > source values in sector rather in bytes > > 7. Add payload to bio using bio_map_kern() for copy_offload case > > 8. Add check to return error if one of the source range length is 0 > > 9. Add BLKDEV_COPY_NOEMULATION flag to allow user to not try copy > > emulation incase of copy offload is not supported. Caller can his use > > his existing copying logic to complete the io. > > 10. Bug fix copy checks and reduce size of rcu_lock() > > > > Changes from RFC v3 > > > > 1. gfp_flag fixes. > > 2. Export bio_map_kern() and use it to allocate and add pages to bio. > > 3. Move copy offload, reading to buf, writing from buf to separate functions > > 4. Send read bio of copy offload by chaining them and submit asynchronously > > 5. Add gendisk->part0 and part->bd_start_sect changes to blk_check_copy(). > > 6. Move single source range limit check to blk_check_copy() > > 7. Rename __blkdev_issue_copy() to blkdev_issue_copy and remove old helper. > > 8. Change blkdev_issue_copy() interface generic to accepts destination bdev > > to support XCOPY as well. > > 9. Add invalidate_kernel_vmap_range() after reading data for vmalloc'ed memory. > > 10. Fix buf allocoation logic to allocate buffer for the total size of copy. > > 11. Reword patch commit description. > > > > Changes from RFC v2 > > > > 1. Add emulation support for devices not supporting copy. > > 2. Add *copy_offload* sysfs entry to enable and disable copy_offload > > in devices supporting simple copy. > > 3. Remove simple copy support for stacked devices. > > > > Changes from RFC v1: > > > > 1. Fix memory leak in __blkdev_issue_copy > > 2. Unmark blk_check_copy inline > > 3. Fix line break in blk_check_copy_eod > > 4. Remove p checks and made code more readable > > 5. Don't use bio_set_op_attrs and remove op and set > > bi_opf directly > > 6. Use struct_size to calculate total_size > > 7. Fix partition remap of copy destination > > 8. Remove mcl,mssrl,msrc from nvme_ns > > 9. Initialize copy queue limits to 0 in nvme_config_copy > > 10. Remove return in QUEUE_FLAG_COPY check > > 11. Remove unused OCFS > > > > > > Nitesh Shetty (4): > > block: Introduce queue limits for copy-offload support > > block: copy offload support infrastructure > > block: Introduce a new ioctl for simple copy > > block: add emulation for simple copy > > > > SelvaKumar S (3): > > block: make bio_map_kern() non static > > nvme: add simple copy support > > dm kcopyd: add simple copy offload support > > > > block/blk-core.c | 84 ++++++++- > > block/blk-lib.c | 352 ++++++++++++++++++++++++++++++++++++++ > > block/blk-map.c | 2 +- > > block/blk-settings.c | 4 + > > block/blk-sysfs.c | 51 ++++++ > > block/blk-zoned.c | 1 + > > block/bounce.c | 1 + > > block/ioctl.c | 33 ++++ > > drivers/md/dm-kcopyd.c | 56 +++++- > > drivers/nvme/host/core.c | 83 +++++++++ > > drivers/nvme/host/trace.c | 19 ++ > > include/linux/bio.h | 1 + > > include/linux/blk_types.h | 20 +++ > > include/linux/blkdev.h | 21 +++ > > include/linux/nvme.h | 43 ++++- > > include/uapi/linux/fs.h | 20 +++ > > 16 files changed, 775 insertions(+), 16 deletions(-) > > > > -- > > 2.25.1 > >
