Commit 3f9cbe51 authored by Seth Howell's avatar Seth Howell Committed by Daniel Verkamp
Browse files

blob: add write_zeroes support 



Unmap does not guarantee that erased blocks will return all zeroes.
using write_zeroes when unmapping metadata gives the
desired behavior for a blob.

Only metadata pages will be cleared with write_zeroes in this patch;
blob data clusters will still call unmap.  This behavior may be made
configurable in a later patch (to allow the user to request zeroing of
clusters rather than just unmapping).

Change-Id: I1b210abac110867ce703bcfdeb634eb45aa9d5c9
Signed-off-by: default avatarSeth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/372004


Tested-by: default avatarSPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: default avatarDaniel Verkamp <daniel.verkamp@intel.com>
parent 3fd7f28d

include/spdk/blob.h

+4 −0
Original line number Diff line number Diff line
@@ -132,6 +132,10 @@ struct spdk_bs_dev { 
	void (*flush)(struct spdk_bs_dev *dev, struct spdk_io_channel *channel,

		      struct spdk_bs_dev_cb_args *cb_args);



	void (*write_zeroes)(struct spdk_bs_dev *dev, struct spdk_io_channel *channel,

			     uint64_t lba, uint32_t lba_count,

			     struct spdk_bs_dev_cb_args *cb_args);



	void (*unmap)(struct spdk_bs_dev *dev, struct spdk_io_channel *channel,

		      uint64_t lba, uint32_t lba_count,

		      struct spdk_bs_dev_cb_args *cb_args);

lib/blob/bdev/blob_bdev.c

+14 −0
Original line number Diff line number Diff line
@@ -123,6 +123,19 @@ bdev_blob_writev(struct spdk_bs_dev *dev, struct spdk_io_channel *channel, 
	}

}



static void

bdev_blob_write_zeroes(struct spdk_bs_dev *dev, struct spdk_io_channel *channel, uint64_t lba,

		       uint32_t lba_count, struct spdk_bs_dev_cb_args *cb_args)

{

	int rc;



	rc = spdk_bdev_write_zeroes_blocks(__get_desc(dev), channel, lba,

					   lba_count, bdev_blob_io_complete, cb_args);

	if (rc) {

		cb_args->cb_fn(cb_args->channel, cb_args->cb_arg, rc);

	}

}



static void

bdev_blob_unmap(struct spdk_bs_dev *dev, struct spdk_io_channel *channel, uint64_t lba,

		uint32_t lba_count, struct spdk_bs_dev_cb_args *cb_args)
@@ -213,6 +226,7 @@ spdk_bdev_create_bs_dev(struct spdk_bdev *bdev, spdk_bdev_remove_cb_t remove_cb, 
	b->bs_dev.write = bdev_blob_write;

	b->bs_dev.readv = bdev_blob_readv;

	b->bs_dev.writev = bdev_blob_writev;

	b->bs_dev.write_zeroes = bdev_blob_write_zeroes;

	b->bs_dev.unmap = bdev_blob_unmap;



	return &b->bs_dev;

lib/blob/blobstore.c

+13 −13
Original line number Diff line number Diff line
@@ -760,7 +760,7 @@ _spdk_blob_persist_unmap_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bse 
}



static void

_spdk_blob_persist_unmap_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)

_spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)

{

	struct spdk_blob_persist_ctx	*ctx = cb_arg;

	struct spdk_blob 		*blob = ctx->blob;
@@ -769,7 +769,7 @@ _spdk_blob_persist_unmap_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bs 


	/* This loop starts at 1 because the first page is special and handled

	 * below. The pages (except the first) are never written in place,

	 * so any pages in the clean list must be unmapped.

	 * so any pages in the clean list must be zeroed.

	 */

	for (i = 1; i < blob->clean.num_pages; i++) {

		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
@@ -787,7 +787,7 @@ _spdk_blob_persist_unmap_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bs 
}



static void

_spdk_blob_persist_unmap_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)

_spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)

{

	struct spdk_blob_persist_ctx 	*ctx = cb_arg;

	struct spdk_blob 		*blob = ctx->blob;
@@ -797,21 +797,21 @@ _spdk_blob_persist_unmap_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrn 
	spdk_bs_batch_t			*batch;

	size_t				i;



	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_unmap_pages_cpl, ctx);

	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);



	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);



	/* This loop starts at 1 because the first page is special and handled

	 * below. The pages (except the first) are never written in place,

	 * so any pages in the clean list must be unmapped.

	 * so any pages in the clean list must be zeroed.

	 */

	for (i = 1; i < blob->clean.num_pages; i++) {

		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);



		spdk_bs_batch_unmap(batch, lba, lba_count);

		spdk_bs_batch_write_zeroes(batch, lba, lba_count);

	}



	/* The first page will only be unmapped if this is a delete. */

	/* The first page will only be zeroed if this is a delete. */

	if (blob->active.num_pages == 0) {

		uint32_t page_num;
@@ -819,7 +819,7 @@ _spdk_blob_persist_unmap_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrn 
		page_num = _spdk_bs_blobid_to_page(blob->id);

		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);



		spdk_bs_batch_unmap(batch, lba, lba_count);

		spdk_bs_batch_write_zeroes(batch, lba, lba_count);

	}



	spdk_bs_batch_close(batch);
@@ -837,7 +837,7 @@ _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bs 


	if (blob->active.num_pages == 0) {

		/* Move on to the next step */

		_spdk_blob_persist_unmap_pages(seq, ctx, 0);

		_spdk_blob_persist_zero_pages(seq, ctx, 0);

		return;

	}
@@ -848,7 +848,7 @@ _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bs 
	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));



	spdk_bs_sequence_write(seq, page, lba, lba_count,

			       _spdk_blob_persist_unmap_pages, ctx);

			       _spdk_blob_persist_zero_pages, ctx);

}



static void
@@ -994,7 +994,7 @@ _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 
		 * Immediately jump to the clean up routine. */

		assert(blob->clean.num_pages > 0);

		ctx->idx = blob->clean.num_pages - 1;

		_spdk_blob_persist_unmap_pages(seq, ctx, 0);

		_spdk_blob_persist_zero_pages(seq, ctx, 0);

		return;



	}
@@ -2177,8 +2177,8 @@ spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 
		return;

	}



	/* TRIM the entire device */

	spdk_bs_sequence_unmap(seq, 0, bs->dev->blockcnt, _spdk_bs_init_trim_cpl, ctx);

	/* Zero the entire device */

	spdk_bs_sequence_write_zeroes(seq, 0, bs->dev->blockcnt, _spdk_bs_init_trim_cpl, ctx);

}



/* END spdk_bs_init */

lib/blob/request.c

+31 −0
Original line number Diff line number Diff line
@@ -226,6 +226,23 @@ spdk_bs_sequence_unmap(spdk_bs_sequence_t *seq, 
			    &set->cb_args);

}



void

spdk_bs_sequence_write_zeroes(spdk_bs_sequence_t *seq,

			      uint64_t lba, uint32_t lba_count,

			      spdk_bs_sequence_cpl cb_fn, void *cb_arg)

{

	struct spdk_bs_request_set      *set = (struct spdk_bs_request_set *)seq;

	struct spdk_bs_channel       *channel = set->channel;



	SPDK_DEBUGLOG(SPDK_TRACE_BLOB_RW, "writing zeroes to %u blocks at LBA %lu\n", lba_count, lba);



	set->u.sequence.cb_fn = cb_fn;

	set->u.sequence.cb_arg = cb_arg;



	channel->dev->write_zeroes(channel->dev, channel->dev_channel, lba, lba_count,

				   &set->cb_args);

}



void

spdk_bs_sequence_finish(spdk_bs_sequence_t *seq, int bserrno)

{
@@ -342,6 +359,20 @@ spdk_bs_batch_unmap(spdk_bs_batch_t *batch, 
			    &set->cb_args);

}



void

spdk_bs_batch_write_zeroes(spdk_bs_batch_t *batch,

			   uint64_t lba, uint32_t lba_count)

{

	struct spdk_bs_request_set	*set = (struct spdk_bs_request_set *)batch;

	struct spdk_bs_channel		*channel = set->channel;



	SPDK_DEBUGLOG(SPDK_TRACE_BLOB_RW, "Zeroing %u blocks at LBA %lu\n", lba_count, lba);



	set->u.batch.outstanding_ops++;

	channel->dev->write_zeroes(channel->dev, channel->dev_channel, lba, lba_count,

				   &set->cb_args);

}



void

spdk_bs_batch_close(spdk_bs_batch_t *batch)

{

lib/blob/request.h

+7 −0
Original line number Diff line number Diff line
@@ -155,6 +155,10 @@ void spdk_bs_sequence_unmap(spdk_bs_sequence_t *seq, 
			    uint64_t lba, uint32_t lba_count,

			    spdk_bs_sequence_cpl cb_fn, void *cb_arg);



void spdk_bs_sequence_write_zeroes(spdk_bs_sequence_t *seq,

				   uint64_t lba, uint32_t lba_count,

				   spdk_bs_sequence_cpl cb_fn, void *cb_arg);



void spdk_bs_sequence_finish(spdk_bs_sequence_t *seq, int bserrno);



spdk_bs_batch_t *spdk_bs_batch_open(struct spdk_io_channel *channel,
@@ -171,6 +175,9 @@ void spdk_bs_batch_flush(spdk_bs_batch_t *batch); 
void spdk_bs_batch_unmap(spdk_bs_batch_t *batch,

			 uint64_t lba, uint32_t lba_count);



void spdk_bs_batch_write_zeroes(spdk_bs_batch_t *batch,

				uint64_t lba, uint32_t lba_count);



void spdk_bs_batch_close(spdk_bs_batch_t *batch);



spdk_bs_batch_t *spdk_bs_sequence_to_batch(spdk_bs_sequence_t *seq,