Spdk/module/bdev/raid/raid1.c

/*   SPDX-License-Identifier: BSD-3-Clause
 *   Copyright (C) 2022 Intel Corporation.
 *   All rights reserved.
 */

#include "bdev_raid.h"

#include "spdk/likely.h"
#include "spdk/log.h"

struct raid1_info {
	/* The parent raid bdev */
	struct raid_bdev *raid_bdev;
};

static void
raid1_bdev_io_completion(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
{
	struct raid_bdev_io *raid_io = cb_arg;

	spdk_bdev_free_io(bdev_io);

	raid_bdev_io_complete_part(raid_io, 1, success ?
				   SPDK_BDEV_IO_STATUS_SUCCESS :
				   SPDK_BDEV_IO_STATUS_FAILED);
}

static void raid1_submit_rw_request(struct raid_bdev_io *raid_io);

static void
_raid1_submit_rw_request(void *_raid_io)
{
	struct raid_bdev_io *raid_io = _raid_io;

	raid1_submit_rw_request(raid_io);
}

static int
raid1_submit_read_request(struct raid_bdev_io *raid_io)
{
	struct raid_bdev *raid_bdev = raid_io->raid_bdev;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(raid_io);
	uint8_t ch_idx = 0;
	struct raid_base_bdev_info *base_info = &raid_bdev->base_bdev_info[ch_idx];
	struct spdk_io_channel *base_ch = raid_io->raid_ch->base_channel[ch_idx];
	uint64_t pd_lba, pd_blocks;
	int ret;

	pd_lba = bdev_io->u.bdev.offset_blocks;
	pd_blocks = bdev_io->u.bdev.num_blocks;

	raid_io->base_bdev_io_remaining = 1;

	if (bdev_io->u.bdev.ext_opts != NULL) {
		ret = spdk_bdev_readv_blocks_ext(base_info->desc, base_ch,
						 bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
						 pd_lba, pd_blocks, raid1_bdev_io_completion,
						 raid_io, bdev_io->u.bdev.ext_opts);
	} else {
		ret = spdk_bdev_readv_blocks_with_md(base_info->desc, base_ch,
						     bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
						     bdev_io->u.bdev.md_buf,
						     pd_lba, pd_blocks,
						     raid1_bdev_io_completion, raid_io);
	}

	if (spdk_likely(ret == 0)) {
		raid_io->base_bdev_io_submitted++;
	} else if (spdk_unlikely(ret == -ENOMEM)) {
		raid_bdev_queue_io_wait(raid_io, base_info->bdev, base_ch,
					_raid1_submit_rw_request);
		return 0;
	}

	return ret;
}

static int
raid1_submit_write_request(struct raid_bdev_io *raid_io)
{
	struct raid_bdev *raid_bdev = raid_io->raid_bdev;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(raid_io);
	struct raid_base_bdev_info *base_info;
	struct spdk_io_channel *base_ch;
	uint64_t pd_lba, pd_blocks;
	uint16_t idx = raid_io->base_bdev_io_submitted;
	uint64_t base_bdev_io_not_submitted;
	int ret = 0;

	pd_lba = bdev_io->u.bdev.offset_blocks;
	pd_blocks = bdev_io->u.bdev.num_blocks;

	if (raid_io->base_bdev_io_submitted == 0) {
		raid_io->base_bdev_io_remaining = raid_bdev->num_base_bdevs;
	}

	for (; idx < raid_bdev->num_base_bdevs; idx++) {
		base_info = &raid_bdev->base_bdev_info[idx];
		base_ch = raid_io->raid_ch->base_channel[idx];

		if (bdev_io->u.bdev.ext_opts != NULL) {
			ret = spdk_bdev_writev_blocks_ext(base_info->desc, base_ch,
							  bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
							  pd_lba, pd_blocks, raid1_bdev_io_completion,
							  raid_io, bdev_io->u.bdev.ext_opts);
		} else {
			ret = spdk_bdev_writev_blocks_with_md(base_info->desc, base_ch,
							      bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
							      bdev_io->u.bdev.md_buf,
							      pd_lba, pd_blocks,
							      raid1_bdev_io_completion, raid_io);
		}

		if (spdk_unlikely(ret != 0)) {
			if (spdk_unlikely(ret == -ENOMEM)) {
				raid_bdev_queue_io_wait(raid_io, base_info->bdev, base_ch,
							_raid1_submit_rw_request);
				return 0;
			}

			base_bdev_io_not_submitted = raid_bdev->num_base_bdevs -
						     raid_io->base_bdev_io_submitted;
			raid_bdev_io_complete_part(raid_io, base_bdev_io_not_submitted,
						   SPDK_BDEV_IO_STATUS_FAILED);
			return 0;
		}

		raid_io->base_bdev_io_submitted++;
	}

	return ret;
}

static void
raid1_submit_rw_request(struct raid_bdev_io *raid_io)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(raid_io);
	int ret;

	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_READ:
		ret = raid1_submit_read_request(raid_io);
		break;
	case SPDK_BDEV_IO_TYPE_WRITE:
		ret = raid1_submit_write_request(raid_io);
		break;
	default:
		ret = -EINVAL;
		break;
	}

	if (spdk_unlikely(ret != 0)) {
		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
	}
}

static int
raid1_start(struct raid_bdev *raid_bdev)
{
	uint64_t min_blockcnt = UINT64_MAX;
	struct raid_base_bdev_info *base_info;
	struct raid1_info *r1info;

	r1info = calloc(1, sizeof(*r1info));
	if (!r1info) {
		SPDK_ERRLOG("Failed to allocate RAID1 info device structure\n");
		return -ENOMEM;
	}
	r1info->raid_bdev = raid_bdev;

	RAID_FOR_EACH_BASE_BDEV(raid_bdev, base_info) {
		min_blockcnt = spdk_min(min_blockcnt, base_info->bdev->blockcnt);
	}

	raid_bdev->bdev.blockcnt = min_blockcnt;
	raid_bdev->module_private = r1info;

	return 0;
}

static bool
raid1_stop(struct raid_bdev *raid_bdev)
{
	struct raid1_info *r1info = raid_bdev->module_private;

	free(r1info);

	return true;
}

static struct raid_bdev_module g_raid1_module = {
	.level = RAID1,
	.base_bdevs_min = 2,
	.base_bdevs_constraint = {CONSTRAINT_MIN_BASE_BDEVS_OPERATIONAL, 1},
	.start = raid1_start,
	.stop = raid1_stop,
	.submit_rw_request = raid1_submit_rw_request,
};
RAID_MODULE_REGISTER(&g_raid1_module)

SPDK_LOG_REGISTER_COMPONENT(bdev_raid1)