blob: Optimize copy-on-write flow for clusters backed by zeroes device

Writing to unallocated cluster triggers copy-on-write sequence. If
this cluster is backed by zeroes device we can skip the copy part. For
a simple thin provisioned volume copy this shortcut is already
implemented because `blob->parent_id == SPDK_BLOBID_INVALID`. But this
will not work for thin provisioned volumes created from snapshot. In
this case we need to traverse the whole stack of underlying
`spdk_bs_dev` devices for specific cluster to check if it is zeroes
backed.

This patch adds `is_zeroes` operation to `spdk_bs_dev`. For zeroes
device it always returns 'true', for real bdev (`blob_bs_dev`) always
returns false, for another layer of `blob_bs_dev` does lba conversion
and forwards to backing device.

In blobstore's cluster copy flow we check if cluster is backed by
zeroes device and skip copy part if it is.

Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I640773ac78f8f466b96e96a34c3a6c3c91f87dab
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/13446
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>

CHANGELOG.md

@@ -27,6 +27,9 @@ For now we are using hard-coded PSK and only support TLS 1.3
 Reserve space for used_cluster bitmap. The reserved space could be used for blobstore growing
 in the future.
 
+Added `is_zeroes` operation to `spdk_bs_dev`. It allows to detect if logical blocks are backed
+by zeroes device and do a shortcut in copy-on-write flow by excluding copy part from zeroes device.
+
 ### lvol
 
 Add num_md_pages_per_cluster_ratio parameter to the bdev_lvol_create_lvstore RPC.

include/spdk/blob.h

@@ -1,7 +1,7 @@
 /*   SPDX-License-Identifier: BSD-3-Clause
  *   Copyright (c) Intel Corporation.
  *   All rights reserved.
- *   Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ *   Copyright (c) 2021, 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  */
 
 /** \file
@@ -195,6 +195,8 @@ struct spdk_bs_dev {
 
 	struct spdk_bdev *(*get_base_bdev)(struct spdk_bs_dev *dev);
 
+	bool (*is_zeroes)(struct spdk_bs_dev *dev, uint64_t lba, uint64_t lba_count);
+
 	uint64_t	blockcnt;
 	uint32_t	blocklen; /* In bytes */
 };

lib/blob/blob_bs_dev.c

@@ -117,6 +117,24 @@ blob_bs_dev_destroy(struct spdk_bs_dev *bs_dev)
 	spdk_blob_close(b->blob, blob_bs_dev_destroy_cpl, b);
 }
 
+static bool
+blob_bs_is_zeroes(struct spdk_bs_dev *dev, uint64_t lba, uint64_t lba_count)
+{
+	struct spdk_blob_bs_dev *b = (struct spdk_blob_bs_dev *)dev;
+	struct spdk_blob *blob = b->blob;
+
+	assert(lba == bs_cluster_to_lba(blob->bs, bs_lba_to_cluster(blob->bs, lba)));
+	assert(lba_count == bs_dev_byte_to_lba(dev, blob->bs->cluster_sz));
+
+	if (bs_io_unit_is_allocated(blob, lba)) {
+		return false;
+	}
+
+	assert(blob->back_bs_dev != NULL);
+	return blob->back_bs_dev->is_zeroes(blob->back_bs_dev,
+					    bs_io_unit_to_back_dev_lba(blob, lba),
+					    bs_io_unit_to_back_dev_lba(blob, lba_count));
+}
 
 struct spdk_bs_dev *
 bs_create_blob_bs_dev(struct spdk_blob *blob)
@@ -142,6 +160,7 @@ bs_create_blob_bs_dev(struct spdk_blob *blob)
 	b->bs_dev.readv_ext = blob_bs_dev_readv_ext;
 	b->bs_dev.write_zeroes = blob_bs_dev_write_zeroes;
 	b->bs_dev.unmap = blob_bs_dev_unmap;
+	b->bs_dev.is_zeroes = blob_bs_is_zeroes;
 	b->blob = blob;
 
 	return &b->bs_dev;

lib/blob/blobstore.c

@@ -2399,6 +2399,7 @@ bs_allocate_and_copy_cluster(struct spdk_blob *blob,
 	struct spdk_blob_copy_cluster_ctx *ctx;
 	uint32_t cluster_start_page;
 	uint32_t cluster_number;
+	bool is_zeroes;
 	int rc;
 
 	ch = spdk_io_channel_get_ctx(_ch);
@@ -2431,7 +2432,10 @@ bs_allocate_and_copy_cluster(struct spdk_blob *blob,
 	ctx->new_cluster_page = ch->new_cluster_page;
 	memset(ctx->new_cluster_page, 0, SPDK_BS_PAGE_SIZE);
 
-	if (blob->parent_id != SPDK_BLOBID_INVALID) {
+	is_zeroes = blob->back_bs_dev->is_zeroes(blob->back_bs_dev,
+			bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
+			bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz));
+	if (blob->parent_id != SPDK_BLOBID_INVALID && !is_zeroes) {
 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
 		if (!ctx->buf) {
@@ -2472,7 +2476,7 @@ bs_allocate_and_copy_cluster(struct spdk_blob *blob,
 	/* Queue the user op to block other incoming operations */
 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
 
-	if (blob->parent_id != SPDK_BLOBID_INVALID) {
+	if (blob->parent_id != SPDK_BLOBID_INVALID && !is_zeroes) {
 		/* Read cluster from backing device */
 		bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
 					bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),

lib/blob/zeroes.c

@@ -118,6 +118,12 @@ zeroes_unmap(struct spdk_bs_dev *dev, struct spdk_io_channel *channel,
 	assert(false);
 }
 
+static bool
+zeroes_is_zeroes(struct spdk_bs_dev *dev, uint64_t lba, uint64_t lba_count)
+{
+	return true;
+}
+
 static struct spdk_bs_dev g_zeroes_bs_dev = {
 	.blockcnt = UINT64_MAX,
 	.blocklen = 512,
@@ -132,6 +138,7 @@ static struct spdk_bs_dev g_zeroes_bs_dev = {
 	.writev_ext = zeroes_writev_ext,
 	.write_zeroes = zeroes_write_zeroes,
 	.unmap = zeroes_unmap,
+	.is_zeroes = zeroes_is_zeroes,
 };
 
 struct spdk_bs_dev *

lib/blobfs/Makefile

@@ -6,7 +6,7 @@
 SPDK_ROOT_DIR := $(abspath $(CURDIR)/../..)
 include $(SPDK_ROOT_DIR)/mk/spdk.common.mk
 
-SO_VER := 6
+SO_VER := 7
 SO_MINOR := 0
 
 C_SRCS = blobfs.c tree.c

module/blob/bdev/Makefile

@@ -6,7 +6,7 @@
 SPDK_ROOT_DIR := $(abspath $(CURDIR)/../../..)
 include $(SPDK_ROOT_DIR)/mk/spdk.common.mk
 
-SO_VER := 7
+SO_VER := 8
 SO_MINOR := 0
 
 C_SRCS = blob_bdev.c

module/blob/bdev/blob_bdev.c

@@ -1,6 +1,7 @@
 /*   SPDX-License-Identifier: BSD-3-Clause
  *   Copyright (c) Intel Corporation.
  *   All rights reserved.
+ *   Copyright (c) 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  */
 
 #include "spdk/stdinc.h"
@@ -353,6 +354,12 @@ bdev_blob_get_base_bdev(struct spdk_bs_dev *bs_dev)
 	return __get_bdev(bs_dev);
 }
 
+static bool
+bdev_blob_is_zeroes(struct spdk_bs_dev *dev, uint64_t lba, uint64_t lba_count)
+{
+	return false;
+}
+
 static void
 blob_bdev_init(struct blob_bdev *b, struct spdk_bdev_desc *desc)
 {
@@ -377,6 +384,7 @@ blob_bdev_init(struct blob_bdev *b, struct spdk_bdev_desc *desc)
 	b->bs_dev.write_zeroes = bdev_blob_write_zeroes;
 	b->bs_dev.unmap = bdev_blob_unmap;
 	b->bs_dev.get_base_bdev = bdev_blob_get_base_bdev;
+	b->bs_dev.is_zeroes = bdev_blob_is_zeroes;
 }
 
 int