From: Damien Le Moal <damien.lemoal@xxxxxxx> The dm-zoned device mapper target provides transparent write access to zoned block devices (ZBC and ZAC compliant block devices). dm-zoned hides to the device user (a file system or an application doing raw block device accesses) any constraint imposed on write requests by the device, equivalent to a drive-managed zoned block device model. Write requests are processed using a combination of on-disk buffering using the device conventional zones and direct in-place processing for requests aligned to a zone sequential write pointer position. A background reclaim process implemented using dm_kcopyd_copy ensures that conventional zones are always available for executing unaligned write requests. The reclaim process overhead is minimized by managing buffer zones in a least-recently-written order and first targeting the oldest buffer zones. Doing so, blocks under regular write access (such as metadata blocks of a file system) remain stored in conventional zones, resulting in no apparent overhead. dm-zoned implementation focus on simplicity and on minimizing overhead (CPU, memory and storage overhead). For a 10TB host-managed disk with 256 MB zones, dm-zoned memory usage per disk instance is at most 4.5 MB and as little as 5 zones will be used internally for storing metadata and performing buffer zone reclaim operations. This is achieved using zone level indirection rather than a full block indirection system for managing block movement between zones. dm-zoned primary target is host-managed zoned block devices but it can also be used with host-aware device models to mitigate potential device-side performance degradation due to excessive random writing. dm-zoned target devices can be formatted and checked using the dmzadm utility available at: https://github.com/hgst/dm-zoned-tools Signed-off-by: Damien Le Moal <damien.lemoal@xxxxxxx> Reviewed-by: Hannes Reinecke <hare@xxxxxxxx> Reviewed-by: Bart Van Assche <bart.vanassche@xxxxxxxxxxx> --- Documentation/device-mapper/dm-zoned.txt | 154 +++ drivers/md/Kconfig | 17 + drivers/md/Makefile | 2 + drivers/md/dm-zoned-io.c | 999 ++++++++++++++ drivers/md/dm-zoned-metadata.c | 2225 ++++++++++++++++++++++++++++++ drivers/md/dm-zoned-reclaim.c | 535 +++++++ drivers/md/dm-zoned.h | 530 +++++++ 7 files changed, 4462 insertions(+) create mode 100644 Documentation/device-mapper/dm-zoned.txt create mode 100644 drivers/md/dm-zoned-io.c create mode 100644 drivers/md/dm-zoned-metadata.c create mode 100644 drivers/md/dm-zoned-reclaim.c create mode 100644 drivers/md/dm-zoned.h diff --git a/Documentation/device-mapper/dm-zoned.txt b/Documentation/device-mapper/dm-zoned.txt new file mode 100644 index 0000000..d41f597 --- /dev/null +++ b/Documentation/device-mapper/dm-zoned.txt @@ -0,0 +1,154 @@ +dm-zoned +======== + +The dm-zoned device mapper exposes a zoned block device (ZBC and ZAC compliant +devices) as a regular block device without any write pattern constraint. In +effect, it implements a drive-managed zoned block device which hides to the +user (a file system or an application doing raw block device accesses) the +sequential write constraints of host-managed zoned block devices and can +mitigate the potential device-side performance degradation due to excessive +random writes on host-aware zoned block devices. + +For a more detailed description of the zoned block device models and +their constraints see (for SCSI devices): + +http://www.t10.org/drafts.htm#ZBC_Family + +and (for ATA devices): + +http://www.t13.org/Documents/UploadedDocuments/docs2015/ +di537r05-Zoned_Device_ATA_Command_Set_ZAC.pdf + +dm-zoned implementation is simple and minimizes system overhead (CPU and +memory usage as well as storage capacity loss). For a 10TB host-manmaged disk +with 256 MB zones, dm-zoned memory usage per disk instance is at most 4.5 MB +and as little as 5 zones will be used internally for storing metadata and +performaing reclaim operations. + +dm-zoned targte devices can be formatted and checked using the dmzadm utility +available at: + +https://github.com/hgst/dm-zoned-tools + +Algorithm +========= + +dm-zoned implements an on-disk buffering scheme to handle non-sequential write +accesses to the sequential zones of a zoned block device. Conventional zones +are used for caching as well as for storing internal metadata. + +The zones of the device are separated into 2 types: + +1) Metadata zones: these are conventional zones used to store metadata. +Metadata zones are not reported as useable capacity to the user. + +2) Data zones: all remaining zones, the vast majority of which will be +sequential zones used exclusively to store user data. The conventional zones +of the device may be used also for buffering user random writes. Data in these +zones may be directly mapped to the conventional zone, but later moved to a +sequential zone after so that the conventional zone can be reused for buffering +incoming random writes. + +dm-zoned exposes a logical device with a sector size of 4096 bytes, +irrespectively of the physical sector size of the backend zoned block device +being used. This allows reducing the amount of metadata needed to manage valid +blocks (blocks written). + +The on-disk metadata format is as follows: + +1) The first block of the first convnetional zone found contains the +super block which describes the amount and position on disk of metadata blocks. + +2) Following the super block, a set of blocks is used to describe the mapping +of the logical device blocks. The mapping is done per chunk of blocks, with +the chunk size equal to the zoned block device size. The mapping table is +indexed by chunk number and each mapping entry indicates the zone number of +the device storing the chunk of data. Each mapping entry may also indicate if +the zone number of a conventional zone used to buffer random modification to +the data zone. + +3) A set of blocks used to store bitmaps indicating the validity of blocks in +the data zones follows the mapping table. A valid block is defined as a block +that was writen and not discarded. For a buffered data chunk, a block is +always valid only in the data zone mapping the chunk or in the buffer zone of +the chunk. + +For a logical chunk mapped to a conventional zone, all write operations are +processed by directly writing to the zone. If the mapping zone is a +sequential zone, the write operation is processed directly only and only if +the write offset within the logical chunk is equal to the write pointer offset +within of the sequential data zone (i.e. the write operation is aligned on the +zone write pointer). Otherwise, write operations are processed indirectly +using a buffer zone. In such case, an unused conventional zone is allocated +and assigned to the chunk being accessed. Writing a block to the buffer zone +of a chunk will automatically invalidate the same block in the sequential zone +mapping the chunk. If all blocks of the sequential zone become invalid, the +zone is freed and the chunk buffer zone becomes the primary zone mapping the +chunk, resulting is native random write performance similar to a regular +block device. + +Read operations are processed according to the block validity information +provided by the bitmaps. Valid blocks are read either from the sequential zone +mapping a chunk, or if the chunk is buffered, from the buffer zone assigned. +If the accessed chunk has no mapping, or the accessed blocks are invalid, the +read buffer is zeroed and the read operation terminated. + +After some time, the limited number of convnetional zones available may be +exhausted (all used to map chunks or buffer sequential zones) and unaligned +writes to unbuffered chunks become impossible. To avoid such situation, a +reclaim process regularly scans used conventional zones and try to reclaim +the least recently used ones copying the valid blocks of the buffer zone +to a free sequential zone. Once the copy completes, the chunk mapping is +updated to point to the sequential zone and the buffer zone freed for reuse. + +Metadata Protection +=================== + +To protect metadata against corruption in case of sudden power loss or system +crash, 2 sets of metadata zones are used. One set, the primary set, is used as +the main metadata region, while the secondary set is used as a staging area. +Modified metadata are first written to the secondary set and validated by +updating the super block in the secondary set, indicating using a generation +counter that this set contains the newest metadata. Once this operation +completes, updates in place of metadata blocks can be done in the primary +metadata set, ensuring that one of the set is always consistent (all +modifications committed or none at all). Flush operations are used as a commit +point. Upon reception of a flush request, metadata modification activity is +temporarily blocked (for both incoming BIO processing and reclaim process) and +all dirty metadata blocks staged and updated. Normal operation is then resumed. +Metadata flush thus only temporarily delays write and discard requests. Read +requests can be concurrently processed while metadata flush is being executed. + +Usage +===== + +A zoned block device must first be formatted using the dmzadm tool. This will +analyze the device zone configuration, determine where to place the metadata +sets on the device and initialize the metadata sets. + +Ex: + +dmzadm --format /dev/sdxx + +For a formatted device, the target can be created normally with the dmsetup +utility. The only parameter that dm-zoned requires is the device name. + +Example scripts +=============== + +[[ +#!/bin/sh + +if [ $# -ne 1 ]; then + echo "Usage: $0 <Zoned device path>" + exit 1 +fi + +dev="${1}" +shift + +modprobe dm-zoned + +echo "0 `blockdev --getsize ${dev}` dm-zoned ${dev}" | dmsetup create dmz-`basename ${dev}` +]] + diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig index 906103c..a081b3c 100644 --- a/drivers/md/Kconfig +++ b/drivers/md/Kconfig @@ -521,6 +521,23 @@ config DM_INTEGRITY To compile this code as a module, choose M here: the module will be called dm-integrity. +config DM_ZONED + tristate "Drive-managed zoned block device target support" + depends on BLK_DEV_DM + depends on BLK_DEV_ZONED + ---help--- + This device-mapper target takes a host-managed or host-aware zoned + block device and expose most of its capacity as a regular block + device (drive-managed zoned block device) without any write + constraint. This is mainly intended for use with file systems that + do not natively support zoned block devices but still want to + benefit from the increased capacity offered by SMR disks. Other uses + by applications using raw block devices (for example object stores) + is also possible. + + To compile this code as a module, choose M here: the module will + be called dm-zoned. + If unsure, say N. endif # MD diff --git a/drivers/md/Makefile b/drivers/md/Makefile index 913720b..f77cc3b 100644 --- a/drivers/md/Makefile +++ b/drivers/md/Makefile @@ -20,6 +20,7 @@ dm-era-y += dm-era-target.o dm-verity-y += dm-verity-target.o md-mod-y += md.o bitmap.o raid456-y += raid5.o raid5-cache.o raid5-ppl.o +dm-zoned-y += dm-zoned-io.o dm-zoned-metadata.o dm-zoned-reclaim.o # Note: link order is important. All raid personalities # and must come before md.o, as they each initialise @@ -60,6 +61,7 @@ obj-$(CONFIG_DM_CACHE_SMQ) += dm-cache-smq.o obj-$(CONFIG_DM_ERA) += dm-era.o obj-$(CONFIG_DM_LOG_WRITES) += dm-log-writes.o obj-$(CONFIG_DM_INTEGRITY) += dm-integrity.o +obj-$(CONFIG_DM_ZONED) += dm-zoned.o ifeq ($(CONFIG_DM_UEVENT),y) dm-mod-objs += dm-uevent.o diff --git a/drivers/md/dm-zoned-io.c b/drivers/md/dm-zoned-io.c new file mode 100644 index 0000000..0119896 --- /dev/null +++ b/drivers/md/dm-zoned-io.c @@ -0,0 +1,999 @@ +/* + * Drive-managed zoned block device target + * Copyright (C) 2017 Western Digital Corporation or its affiliates. + * + * Written by: Damien Le Moal <damien.lemoal@xxxxxxx> + * + * This software is distributed under the terms of the GNU General Public + * License version 2, or any later version, "as is," without technical + * support, and WITHOUT ANY WARRANTY, without even the implied warranty + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. + */ + +#include <linux/module.h> + +#include "dm-zoned.h" + +/* + * Target BIO completion. + */ +static inline void dmz_bio_end(struct bio *bio, int err) +{ + struct dmz_bioctx *bioctx = + dm_per_bio_data(bio, sizeof(struct dmz_bioctx)); + + if (atomic_dec_and_test(&bioctx->ref)) { + struct dmz_target *dmz = bioctx->target; + + /* User BIO Completed */ + if (bioctx->zone) + dmz_deactivate_zone(dmz, bioctx->zone); + atomic_dec(&dmz->bio_count); + bio->bi_error = bioctx->error; + bio_endio(bio); + } +} + +/* + * Partial/internal BIO completion callback. + * This terminates the user target BIO when there + * are no more references to its context. + */ +static void dmz_bio_end_io(struct bio *bio) +{ + struct dmz_bioctx *bioctx = bio->bi_private; + int err = bio->bi_error; + + if (err) { + struct dm_zone *zone = bioctx->zone; + + bioctx->error = err; + if (bio_op(bio) == REQ_OP_WRITE && + dmz_is_seq(zone)) + set_bit(DMZ_SEQ_WRITE_ERR, &zone->flags); + } + + dmz_bio_end(bioctx->bio, err); + + bio_put(bio); + +} + +/* + * Issue a BIO to a zone. The BIO may only partially process the + * original target BIO. + */ +static int dmz_submit_bio(struct dmz_target *dmz, struct dm_zone *zone, + struct bio *dmz_bio, + sector_t chunk_block, unsigned int nr_blocks) +{ + struct dmz_bioctx *bioctx + = dm_per_bio_data(dmz_bio, sizeof(struct dmz_bioctx)); + unsigned int nr_sectors = dmz_blk2sect(nr_blocks); + unsigned int size = nr_sectors << SECTOR_SHIFT; + struct bio *clone; + + clone = bio_clone_fast(dmz_bio, GFP_NOIO, dmz->bio_set); + if (!clone) + return -ENOMEM; + + /* Setup the clone */ + clone->bi_bdev = dmz->zbd; + clone->bi_opf = dmz_bio->bi_opf; + clone->bi_iter.bi_sector = + dmz_start_sect(dmz, zone) + dmz_blk2sect(chunk_block); + clone->bi_iter.bi_size = size; + clone->bi_end_io = dmz_bio_end_io; + clone->bi_private = bioctx; + + bio_advance(dmz_bio, size); + + /* Submit the clone */ + atomic_inc(&bioctx->ref); + generic_make_request(clone); + + return 0; +} + +/* + * Zero out pages of discarded blocks accessed by a read BIO. + */ +static void dmz_handle_read_zero(struct dmz_target *dmz, struct bio *bio, + sector_t chunk_block, unsigned int nr_blocks) +{ + unsigned int size = nr_blocks << DMZ_BLOCK_SHIFT; + + dmz_dev_debug(dmz, + "=> ZERO READ chunk %llu -> block %llu, %u blocks\n", + (unsigned long long)dmz_bio_chunk(dmz, bio), + (unsigned long long)chunk_block, + nr_blocks); + + /* Clear nr_blocks */ + swap(bio->bi_iter.bi_size, size); + zero_fill_bio(bio); + swap(bio->bi_iter.bi_size, size); + + bio_advance(bio, size); +} + +/* + * Process a read BIO. + */ +static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone, + struct bio *bio) +{ + sector_t block = dmz_bio_block(bio); + unsigned int nr_blocks = dmz_bio_blocks(bio); + sector_t chunk_block = dmz_chunk_block(dmz, block); + sector_t end_block = chunk_block + nr_blocks; + struct dm_zone *rzone, *bzone; + int ret; + + /* Read into unmapped chunks need only zeroing the BIO buffer */ + if (!zone) { + dmz_handle_read_zero(dmz, bio, chunk_block, nr_blocks); + return 0; + } + + dmz_dev_debug(dmz, "READ %s zone %u, block %llu, %u blocks\n", + (dmz_is_rnd(zone) ? "RND" : "SEQ"), dmz_id(dmz, zone), + (unsigned long long)chunk_block, nr_blocks); + + /* Check block validity to determine the read location */ + bzone = zone->bzone; + while (chunk_block < end_block) { + + nr_blocks = 0; + if (dmz_is_rnd(zone) + || chunk_block < zone->wp_block) { + /* Test block validity in the data zone */ + ret = dmz_block_valid(dmz, zone, chunk_block); + if (ret < 0) + return ret; + if (ret > 0) { + /* Read data zone blocks */ + nr_blocks = ret; + rzone = zone; + } + } + + /* + * No valid blocks found in the data zone. + * Check the buffer zone, if there is one. + */ + if (!nr_blocks && bzone) { + ret = dmz_block_valid(dmz, bzone, chunk_block); + if (ret < 0) + return ret; + if (ret > 0) { + /* Read buffer zone blocks */ + nr_blocks = ret; + rzone = bzone; + } + } + + if (nr_blocks) { + + /* Valid blocks found: read them */ + nr_blocks = min_t(unsigned int, nr_blocks, + end_block - chunk_block); + + dmz_dev_debug(dmz, + "=> %s READ zone %u, block %llu, %u blocks\n", + (dmz_is_buf(rzone) ? "BUF" : "DATA"), + dmz_id(dmz, rzone), + (unsigned long long)chunk_block, + nr_blocks); + + ret = dmz_submit_bio(dmz, rzone, bio, + chunk_block, nr_blocks); + if (ret) + return ret; + chunk_block += nr_blocks; + + } else { + + /* No valid block: zeroout the current BIO block */ + dmz_handle_read_zero(dmz, bio, chunk_block, 1); + chunk_block++; + + } + + } + + return 0; +} + +/* + * Write blocks directly in a data zone, at the write pointer. + * If a buffer zone is assigned, invalidate the blocks written + * in place. + */ +static int dmz_handle_direct_write(struct dmz_target *dmz, + struct dm_zone *zone, struct bio *bio, + sector_t chunk_block, + unsigned int nr_blocks) +{ + struct dm_zone *bzone = zone->bzone; + int ret; + + dmz_dev_debug(dmz, "WRITE %s zone %u, block %llu, %u blocks\n", + (dmz_is_rnd(zone) ? "RND" : "SEQ"), dmz_id(dmz, zone), + (unsigned long long)chunk_block, nr_blocks); + + if (dmz_is_readonly(zone)) + return -EROFS; + + /* Submit write */ + ret = dmz_submit_bio(dmz, zone, bio, chunk_block, nr_blocks); + if (ret) + return -EIO; + + if (dmz_is_seq(zone)) + zone->wp_block += nr_blocks; + + /* + * Validate the blocks in the data zone and invalidate + * in the buffer zone, if there is one. + */ + ret = dmz_validate_blocks(dmz, zone, chunk_block, nr_blocks); + if (ret == 0 && bzone) + ret = dmz_invalidate_blocks(dmz, bzone, chunk_block, nr_blocks); + + return ret; +} + +/* + * Write blocks in the buffer zone of @zone. + * If no buffer zone is assigned yet, get one. + * Called with @zone write locked. + */ +static int dmz_handle_buffered_write(struct dmz_target *dmz, + struct dm_zone *zone, struct bio *bio, + sector_t chunk_block, + unsigned int nr_blocks) +{ + struct dm_zone *bzone = zone->bzone; + int ret; + + if (!bzone) { + /* Get a buffer zone */ + bzone = dmz_get_chunk_buffer(dmz, zone); + if (!bzone) + return -ENOSPC; + } + + dmz_dev_debug(dmz, "WRITE BUF zone %u, block %llu, %u blocks\n", + dmz_id(dmz, bzone), (unsigned long long)chunk_block, + nr_blocks); + + if (dmz_is_readonly(bzone)) + return -EROFS; + + /* Submit write */ + ret = dmz_submit_bio(dmz, bzone, bio, chunk_block, nr_blocks); + if (ret) + return -EIO; + + /* + * Validate the blocks in the buffer zone + * and invalidate in the data zone. + */ + ret = dmz_validate_blocks(dmz, bzone, chunk_block, nr_blocks); + if (ret == 0 && chunk_block < zone->wp_block) + ret = dmz_invalidate_blocks(dmz, zone, chunk_block, nr_blocks); + + return ret; +} + +/* + * Process a write BIO. + */ +static int dmz_handle_write(struct dmz_target *dmz, struct dm_zone *zone, + struct bio *bio) +{ + sector_t block = dmz_bio_block(bio); + unsigned int nr_blocks = dmz_bio_blocks(bio); + sector_t chunk_block = dmz_chunk_block(dmz, block); + + if (!zone) + return -ENOSPC; + + if (dmz_is_rnd(zone) || + chunk_block == zone->wp_block) + /* + * zone is a random zone, or it is a sequential zone + * and the BIO is aligned to the zone write pointer: + * direct write the zone. + */ + return dmz_handle_direct_write(dmz, zone, bio, + chunk_block, nr_blocks); + + /* + * This is an unaligned write in a sequential zone: + * use buffered write. + */ + return dmz_handle_buffered_write(dmz, zone, bio, + chunk_block, nr_blocks); +} + +/* + * Process a discard BIO. + */ +static int dmz_handle_discard(struct dmz_target *dmz, struct dm_zone *zone, + struct bio *bio) +{ + sector_t block = dmz_bio_block(bio); + unsigned int nr_blocks = dmz_bio_blocks(bio); + sector_t chunk_block = dmz_chunk_block(dmz, block); + int ret = 0; + + /* For unmapped chunks, there is nothing to do */ + if (!zone) + return 0; + + if (dmz_is_readonly(zone)) + return -EROFS; + + dmz_dev_debug(dmz, + "DISCARD chunk %llu -> zone %u, block %llu, %u blocks\n", + (unsigned long long)dmz_bio_chunk(dmz, bio), + dmz_id(dmz, zone), + (unsigned long long)chunk_block, nr_blocks); + + /* + * Invalidate blocks in the data zone and its + * buffer zone if one is mapped. + */ + if (dmz_is_rnd(zone) || + chunk_block < zone->wp_block) + ret = dmz_invalidate_blocks(dmz, zone, + chunk_block, nr_blocks); + if (ret == 0 && zone->bzone) + ret = dmz_invalidate_blocks(dmz, zone->bzone, + chunk_block, nr_blocks); + + return ret; +} + +/* + * Process a BIO. + */ +static void dmz_handle_bio(struct dmz_target *dmz, struct dm_chunk_work *cw, + struct bio *bio) +{ + struct dmz_bioctx *bioctx = + dm_per_bio_data(bio, sizeof(struct dmz_bioctx)); + struct dm_zone *zone; + int ret; + + down_read(&dmz->mblk_sem); + + /* + * Get the data zone mapping the chunk. There may be no + * mapping for read and discard. If a mapping is obtained, + + the zone returned will be set to active state. + */ + zone = dmz_get_chunk_mapping(dmz, dmz_bio_chunk(dmz, bio), + bio_op(bio)); + if (IS_ERR(zone)) { + dmz_bio_end(bio, PTR_ERR(zone)); + goto out; + } + + /* Process the BIO */ + if (zone) { + dmz_activate_zone(dmz, zone); + bioctx->zone = zone; + } + + switch (bio_op(bio)) { + case REQ_OP_READ: + ret = dmz_handle_read(dmz, zone, bio); + break; + case REQ_OP_WRITE: + ret = dmz_handle_write(dmz, zone, bio); + break; + case REQ_OP_DISCARD: + case REQ_OP_WRITE_ZEROES: + ret = dmz_handle_discard(dmz, zone, bio); + break; + default: + dmz_dev_err(dmz, + "Unsupported BIO operation 0x%x\n", + bio_op(bio)); + ret = -EIO; + } + + dmz_bio_end(bio, ret); + + /* + * Release the chunk mapping. This will check that the mapping + * is still valid, that is, that the zone used still has valid blocks. + */ + if (zone) + dmz_put_chunk_mapping(dmz, zone); + +out: + up_read(&dmz->mblk_sem); +} + +/* + * Increment a chunk reference counter. + */ +static inline void dmz_get_chunk_work(struct dm_chunk_work *cw) +{ + atomic_inc(&cw->refcount); +} + +/* + * Decrement a chunk work reference count and + * free it if it becomes 0. + */ +static void dmz_put_chunk_work(struct dm_chunk_work *cw) +{ + if (atomic_dec_and_test(&cw->refcount)) { + atomic_dec(&cw->target->nr_active_chunks); + radix_tree_delete(&cw->target->chunk_rxtree, cw->chunk); + kfree(cw); + } +} + +/* + * Chunk BIO work function. + */ +static void dmz_chunk_work(struct work_struct *work) +{ + struct dm_chunk_work *cw = + container_of(work, struct dm_chunk_work, work); + struct dmz_target *dmz = cw->target; + struct bio *bio; + + mutex_lock(&dmz->chunk_lock); + + /* Process the chunk BIOs */ + while ((bio = bio_list_pop(&cw->bio_list))) { + + mutex_unlock(&dmz->chunk_lock); + dmz_handle_bio(dmz, cw, bio); + mutex_lock(&dmz->chunk_lock); + + dmz_put_chunk_work(cw); + + } + + /* + * Queueing the work added one to the work refcount. + * So drop this here. + */ + dmz_put_chunk_work(cw); + + mutex_unlock(&dmz->chunk_lock); +} + +/* + * Flush work. + */ +static void dmz_flush_work(struct work_struct *work) +{ + struct dmz_target *dmz = + container_of(work, struct dmz_target, flush_work.work); + struct bio *bio; + int ret; + + /* Flush metablocks */ + ret = dmz_flush_mblocks(dmz); + + /* Process queued flush requests */ + while (1) { + + spin_lock(&dmz->flush_lock); + bio = bio_list_pop(&dmz->flush_list); + spin_unlock(&dmz->flush_lock); + + if (!bio) + break; + + /* Do flush */ + dmz_bio_end(bio, ret); + + } + + queue_delayed_work(dmz->flush_wq, &dmz->flush_work, + DMZ_FLUSH_PERIOD); +} + +/* + * Get a chunk work and start it to process a new BIO. + * If the BIO chunk has no work yet, create one. + */ +static void dmz_queue_chunk_work(struct dmz_target *dmz, + struct bio *bio) +{ + unsigned int chunk = dmz_bio_chunk(dmz, bio); + struct dm_chunk_work *cw; + + mutex_lock(&dmz->chunk_lock); + + /* Get the BIO chunk work. If one is not active yet, create one */ + cw = radix_tree_lookup(&dmz->chunk_rxtree, chunk); + if (!cw) { + int ret; + + /* Create a new chunk work */ + cw = kmalloc(sizeof(struct dm_chunk_work), GFP_NOFS); + if (!cw) + goto out; + + INIT_WORK(&cw->work, dmz_chunk_work); + atomic_set(&cw->refcount, 0); + cw->target = dmz; + cw->chunk = chunk; + bio_list_init(&cw->bio_list); + + ret = radix_tree_insert(&dmz->chunk_rxtree, chunk, cw); + if (unlikely(ret != 0)) { + kfree(cw); + cw = NULL; + goto out; + } + + atomic_inc(&dmz->nr_active_chunks); + } + + bio_list_add(&cw->bio_list, bio); + dmz_get_chunk_work(cw); + + if (queue_work(dmz->chunk_wq, &cw->work)) + dmz_get_chunk_work(cw); + +out: + mutex_unlock(&dmz->chunk_lock); +} + +/* + * Process a new BIO. + */ +static int dmz_map(struct dm_target *ti, struct bio *bio) +{ + struct dmz_target *dmz = ti->private; + struct dmz_bioctx *bioctx + = dm_per_bio_data(bio, sizeof(struct dmz_bioctx)); + sector_t sector = bio->bi_iter.bi_sector; + unsigned int nr_sectors = bio_sectors(bio); + sector_t chunk_sector; + + dmz_dev_debug(dmz, + "BIO sector %llu + %u => chunk %llu, block %llu, %u blocks\n", + (u64)sector, nr_sectors, + (u64)dmz_bio_chunk(dmz, bio), + (u64)dmz_chunk_block(dmz, dmz_bio_block(bio)), + (unsigned int)dmz_bio_blocks(bio)); + + bio->bi_bdev = dmz->zbd; + + if (!nr_sectors && + (bio_op(bio) != REQ_OP_FLUSH) && + (bio_op(bio) != REQ_OP_WRITE)) { + bio->bi_bdev = dmz->zbd; + return DM_MAPIO_REMAPPED; + } + + /* The BIO should be block aligned */ + if ((nr_sectors & DMZ_BLOCK_SECTORS_MASK) || + (sector & DMZ_BLOCK_SECTORS_MASK)) { + dmz_dev_err(dmz, + "Unaligned BIO sector %llu, len %u\n", + (u64)sector, + nr_sectors); + return -EIO; + } + + /* Initialize the BIO context */ + bioctx->target = dmz; + bioctx->zone = NULL; + bioctx->bio = bio; + atomic_set(&bioctx->ref, 1); + bioctx->error = 0; + + atomic_inc(&dmz->bio_count); + dmz->atime = jiffies; + + /* Set the BIO pending in the flush list */ + if (bio_op(bio) == REQ_OP_FLUSH || + (!nr_sectors && bio_op(bio) == REQ_OP_WRITE)) { + spin_lock(&dmz->flush_lock); + bio_list_add(&dmz->flush_list, bio); + spin_unlock(&dmz->flush_lock); + dmz_trigger_flush(dmz); + return DM_MAPIO_SUBMITTED; + } + + /* Split zone BIOs to fit entirely into a zone */ + chunk_sector = dmz_chunk_sector(dmz, sector); + if (chunk_sector + nr_sectors > dmz->zone_nr_sectors) + dm_accept_partial_bio(bio, + dmz->zone_nr_sectors - chunk_sector); + + /* Now ready to handle this BIO */ + dmz_queue_chunk_work(dmz, bio); + + return DM_MAPIO_SUBMITTED; +} + +/* + * Setup target. + */ +static int dmz_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + struct dmz_target *dmz; + int ret; + + /* Check arguments */ + if (argc != 1) { + ti->error = "Invalid argument count"; + return -EINVAL; + } + + /* Allocate and initialize the target descriptor */ + dmz = kzalloc(sizeof(struct dmz_target), GFP_KERNEL); + if (!dmz) { + ti->error = "Allocate target descriptor failed"; + return -ENOMEM; + } + + /* Get the target device */ + ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), + &dmz->ddev); + if (ret != 0) { + ti->error = "Get target device failed"; + goto err; + } + + dmz->zbd = dmz->ddev->bdev; + if (!bdev_is_zoned(dmz->zbd)) { + ti->error = "Not a zoned block device"; + ret = -EINVAL; + goto err; + } + + dmz->zbd_capacity = i_size_read(dmz->zbd->bd_inode) >> SECTOR_SHIFT; + if (ti->begin || (ti->len != dmz->zbd_capacity)) { + ti->error = "Partial mapping not supported"; + ret = -EINVAL; + goto err; + } + + (void)bdevname(dmz->zbd, dmz->zbd_name); + dmz->zbdq = bdev_get_queue(dmz->zbd); + + dmz->mblk_rbtree = RB_ROOT; + init_rwsem(&dmz->mblk_sem); + spin_lock_init(&dmz->mblk_lock); + INIT_LIST_HEAD(&dmz->mblk_lru_list); + INIT_LIST_HEAD(&dmz->mblk_dirty_list); + + mutex_init(&dmz->map_lock); + atomic_set(&dmz->dz_unmap_nr_rnd, 0); + INIT_LIST_HEAD(&dmz->dz_unmap_rnd_list); + INIT_LIST_HEAD(&dmz->dz_map_rnd_list); + + atomic_set(&dmz->dz_unmap_nr_seq, 0); + INIT_LIST_HEAD(&dmz->dz_unmap_seq_list); + INIT_LIST_HEAD(&dmz->dz_map_seq_list); + + init_waitqueue_head(&dmz->dz_free_wq); + + atomic_set(&dmz->nr_reclaim_seq_zones, 0); + INIT_LIST_HEAD(&dmz->reclaim_seq_zones_list); + + ret = dmz_init_meta(dmz); + if (ret != 0) { + ti->error = "Metadata initialization failed"; + goto err; + } + + /* Set target (no write same support) */ + ti->private = dmz; + ti->max_io_len = dmz->zone_nr_sectors << 9; + ti->num_flush_bios = 1; + ti->num_discard_bios = 1; + ti->num_write_zeroes_bios = 1; + ti->per_io_data_size = sizeof(struct dmz_bioctx); + ti->flush_supported = true; + ti->discards_supported = true; + ti->split_discard_bios = true; + + /* The exposed capacity is the number of chunks that can be mapped */ + ti->len = dmz->nr_chunks * dmz->zone_nr_sectors; + + /* Zone BIO */ + atomic_set(&dmz->bio_count, 0); + dmz->atime = jiffies; + dmz->bio_set = bioset_create_nobvec(DMZ_MIN_BIOS, 0); + if (!dmz->bio_set) { + ti->error = "Create BIO set failed"; + ret = -ENOMEM; + goto err; + } + + /* Chunk BIO work */ + mutex_init(&dmz->chunk_lock); + atomic_set(&dmz->nr_active_chunks, 0); + INIT_RADIX_TREE(&dmz->chunk_rxtree, GFP_NOFS); + dmz->chunk_wq = alloc_workqueue("dmz_cwq_%s", + WQ_MEM_RECLAIM | WQ_UNBOUND, + 0, dmz->zbd_name); + if (!dmz->chunk_wq) { + ti->error = "Create chunk workqueue failed"; + ret = -ENOMEM; + goto err; + } + + /* Flush work */ + spin_lock_init(&dmz->flush_lock); + bio_list_init(&dmz->flush_list); + INIT_DELAYED_WORK(&dmz->flush_work, dmz_flush_work); + dmz->flush_wq = alloc_ordered_workqueue("dmz_fwq_%s", WQ_MEM_RECLAIM, + dmz->zbd_name); + if (!dmz->flush_wq) { + ti->error = "Create flush workqueue failed"; + ret = -ENOMEM; + goto err; + } + mod_delayed_work(dmz->flush_wq, &dmz->flush_work, DMZ_FLUSH_PERIOD); + + /* Reclaim kcopyd client */ + dmz->reclaim_kc = dm_kcopyd_client_create(&dmz->reclaim_throttle); + if (IS_ERR(dmz->reclaim_kc)) { + ti->error = "Create kcopyd client failed"; + ret = PTR_ERR(dmz->reclaim_kc); + dmz->reclaim_kc = NULL; + goto err; + } + + /* Reclaim work */ + INIT_DELAYED_WORK(&dmz->reclaim_work, dmz_reclaim_work); + dmz->reclaim_wq = alloc_ordered_workqueue("dmz_rwq_%s", WQ_MEM_RECLAIM, + dmz->zbd_name); + if (!dmz->reclaim_wq) { + ti->error = "Create reclaim workqueue failed"; + ret = -ENOMEM; + goto err; + } + + dmz_dev_info(dmz, + "Target device: %llu 512-byte logical sectors (%llu blocks)\n", + (unsigned long long)ti->len, + (unsigned long long)dmz_sect2blk(ti->len)); + + dmz_trigger_reclaim(dmz); + + return 0; + +err: + if (dmz->ddev) { + if (dmz->reclaim_wq) + destroy_workqueue(dmz->reclaim_wq); + if (dmz->reclaim_kc) + dm_kcopyd_client_destroy(dmz->reclaim_kc); + if (dmz->flush_wq) + destroy_workqueue(dmz->flush_wq); + if (dmz->chunk_wq) + destroy_workqueue(dmz->chunk_wq); + if (dmz->bio_set) + bioset_free(dmz->bio_set); + dmz_cleanup_meta(dmz); + dm_put_device(ti, dmz->ddev); + } + + kfree(dmz); + + return ret; + +} + +/* + * Cleanup target. + */ +static void dmz_dtr(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + dmz_dev_info(dmz, "Removing target device\n"); + + flush_workqueue(dmz->chunk_wq); + destroy_workqueue(dmz->chunk_wq); + + cancel_delayed_work_sync(&dmz->reclaim_work); + destroy_workqueue(dmz->reclaim_wq); + dm_kcopyd_client_destroy(dmz->reclaim_kc); + + cancel_delayed_work_sync(&dmz->flush_work); + destroy_workqueue(dmz->flush_wq); + + dmz_flush_mblocks(dmz); + + bioset_free(dmz->bio_set); + + dmz_cleanup_meta(dmz); + + dm_put_device(ti, dmz->ddev); + + kfree(dmz); +} + +/* + * Setup target request queue limits. + */ +static void dmz_io_hints(struct dm_target *ti, + struct queue_limits *limits) +{ + struct dmz_target *dmz = ti->private; + unsigned int chunk_sectors = dmz->zone_nr_sectors; + + limits->logical_block_size = DMZ_BLOCK_SIZE; + limits->physical_block_size = DMZ_BLOCK_SIZE; + + blk_limits_io_min(limits, DMZ_BLOCK_SIZE); + blk_limits_io_opt(limits, DMZ_BLOCK_SIZE); + + limits->discard_alignment = DMZ_BLOCK_SIZE; + limits->discard_granularity = DMZ_BLOCK_SIZE; + limits->max_discard_sectors = chunk_sectors; + limits->max_hw_discard_sectors = chunk_sectors; + limits->max_write_zeroes_sectors = chunk_sectors; + + /* FS hint to try to align to the device zone size */ + limits->chunk_sectors = chunk_sectors; + limits->max_sectors = chunk_sectors; + + /* We are exposing a drive-managed zone model */ + limits->zoned = BLK_ZONED_NONE; +} + +/* + * Pass on ioctl to the backend device. + */ +static int dmz_prepare_ioctl(struct dm_target *ti, + struct block_device **bdev, fmode_t *mode) +{ + struct dmz_target *dmz = ti->private; + + *bdev = dmz->zbd; + + return 0; +} + +/* + * Stop reclaim before suspend. + */ +static void dmz_presuspend(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + dmz_dev_debug(dmz, "Pre-suspend\n"); + + /* Enter suspend state */ + set_bit(DMZ_SUSPENDED, &dmz->flags); + smp_mb__after_atomic(); + + /* Stop reclaim */ + cancel_delayed_work_sync(&dmz->reclaim_work); +} + +/* + * Restart reclaim if suspend failed. + */ +static void dmz_presuspend_undo(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + dmz_dev_debug(dmz, "Pre-suspend undo\n"); + + /* Clear suspend state */ + clear_bit_unlock(DMZ_SUSPENDED, &dmz->flags); + smp_mb__after_atomic(); + + /* Restart reclaim */ + mod_delayed_work(dmz->reclaim_wq, &dmz->reclaim_work, 0); +} + +/* + * Stop works and flush on suspend. + */ +static void dmz_postsuspend(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + dmz_dev_debug(dmz, "Post-suspend\n"); + + /* Stop works */ + flush_workqueue(dmz->chunk_wq); + flush_workqueue(dmz->flush_wq); +} + +/* + * Refresh zone information before resuming. + */ +static int dmz_preresume(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + if (!test_bit(DMZ_SUSPENDED, &dmz->flags)) + return 0; + + dmz_dev_debug(dmz, "Pre-resume\n"); + + /* Refresh zone information */ + return dmz_resume_meta(dmz); +} + +/* + * Resume. + */ +static void dmz_resume(struct dm_target *ti) +{ + struct dmz_target *dmz = ti->private; + + if (!test_bit(DMZ_SUSPENDED, &dmz->flags)) + return; + + dmz_dev_debug(dmz, "Resume\n"); + + /* Clear suspend state */ + clear_bit_unlock(DMZ_SUSPENDED, &dmz->flags); + smp_mb__after_atomic(); + + /* Restart reclaim */ + mod_delayed_work(dmz->reclaim_wq, &dmz->reclaim_work, 0); + +} + +static int dmz_iterate_devices(struct dm_target *ti, + iterate_devices_callout_fn fn, void *data) +{ + struct dmz_target *dmz = ti->private; + sector_t offset = dmz->zbd_capacity - + ((sector_t)dmz->nr_chunks * dmz->zone_nr_sectors); + + return fn(ti, dmz->ddev, offset, ti->len, data); +} + +static struct target_type dmz_type = { + .name = "dm-zoned", + .version = {1, 0, 0}, + .features = DM_TARGET_SINGLETON | DM_TARGET_ZONED_HM, + .module = THIS_MODULE, + .ctr = dmz_ctr, + .dtr = dmz_dtr, + .map = dmz_map, + .io_hints = dmz_io_hints, + .prepare_ioctl = dmz_prepare_ioctl, + .presuspend = dmz_presuspend, + .presuspend_undo = dmz_presuspend_undo, + .postsuspend = dmz_postsuspend, + .preresume = dmz_preresume, + .resume = dmz_resume, + .iterate_devices = dmz_iterate_devices, +}; + +static int __init dmz_init(void) +{ + dmz_info("Zoned block device target (C) Western Digital\n"); + + return dm_register_target(&dmz_type); +} + +static void __exit dmz_exit(void) +{ + dm_unregister_target(&dmz_type); +} + +module_init(dmz_init); +module_exit(dmz_exit); + +MODULE_DESCRIPTION(DM_NAME " target for zoned block devices"); +MODULE_AUTHOR("Damien Le Moal <damien.lemoal@xxxxxxx>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-zoned-metadata.c b/drivers/md/dm-zoned-metadata.c new file mode 100644 index 0000000..7440b47 --- /dev/null +++ b/drivers/md/dm-zoned-metadata.c @@ -0,0 +1,2225 @@ +/* + * Drive-managed zoned block device target + * Copyright (C) 2017 Western Digital Corporation or its affiliates. + * + * Written by: Damien Le Moal <damien.lemoal@xxxxxxx> + * + * This software is distributed under the terms of the GNU General Public + * License version 2, or any later version, "as is," without technical + * support, and WITHOUT ANY WARRANTY, without even the implied warranty + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. + */ + +#include <linux/module.h> +#include <linux/crc32.h> + +#include "dm-zoned.h" + +/* + * Allocate a metadata block. + */ +static struct dmz_mblock *dmz_alloc_mblock(struct dmz_target *dmz, + sector_t mblk_no) +{ + struct dmz_mblock *mblk = NULL; + + /* See if we can reuse cached blocks */ + if (dmz->max_nr_mblks && + atomic_read(&dmz->nr_mblks) > dmz->max_nr_mblks) { + + spin_lock(&dmz->mblk_lock); + + if (list_empty(&dmz->mblk_lru_list) && + !list_empty(&dmz->mblk_dirty_list)) + /* Cleanup dirty blocks */ + dmz_trigger_flush(dmz); + + mblk = list_first_entry_or_null(&dmz->mblk_lru_list, + struct dmz_mblock, link); + if (mblk) { + list_del_init(&mblk->link); + rb_erase(&mblk->node, &dmz->mblk_rbtree); + mblk->no = mblk_no; + } + + spin_unlock(&dmz->mblk_lock); + + if (mblk) + return mblk; + } + + /* Allocate a new block */ + mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO); + if (!mblk) + return NULL; + + mblk->page = alloc_page(GFP_NOIO); + if (!mblk->page) { + kfree(mblk); + return NULL; + } + + RB_CLEAR_NODE(&mblk->node); + INIT_LIST_HEAD(&mblk->link); + atomic_set(&mblk->ref, 0); + mblk->state = 0; + mblk->no = mblk_no; + mblk->data = page_address(mblk->page); + + atomic_inc(&dmz->nr_mblks); + + return mblk; +} + +/* + * Free a metadata block. + */ +static void dmz_free_mblock(struct dmz_target *dmz, struct dmz_mblock *mblk) +{ + __free_pages(mblk->page, 0); + kfree(mblk); + + atomic_dec(&dmz->nr_mblks); +} + +/* + * Insert a metadata block in the rbtree. + */ +static void dmz_insert_mblock(struct dmz_target *dmz, + struct dmz_mblock *mblk) +{ + struct rb_root *root = &dmz->mblk_rbtree; + struct rb_node **new = &(root->rb_node), *parent = NULL; + struct dmz_mblock *b; + + /* Figure out where to put the new node */ + while (*new) { + b = container_of(*new, struct dmz_mblock, node); + parent = *new; + new = (b->no < mblk->no) ? + &((*new)->rb_left) : &((*new)->rb_right); + } + + /* Add new node and rebalance tree */ + rb_link_node(&mblk->node, parent, new); + rb_insert_color(&mblk->node, root); +} + +/* + * Lookup a metadata block in the rbtree. + */ +static struct dmz_mblock *dmz_lookup_mblock(struct dmz_target *dmz, + sector_t mblk_no) +{ + struct rb_root *root = &dmz->mblk_rbtree; + struct rb_node *node = root->rb_node; + struct dmz_mblock *mblk; + + while (node) { + mblk = container_of(node, struct dmz_mblock, node); + if (mblk->no == mblk_no) + return mblk; + node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right; + } + + return NULL; +} + +/* + * Metadata block BIO end callback. + */ +static void dmz_mblock_bio_end_io(struct bio *bio) +{ + struct dmz_mblock *mblk = bio->bi_private; + int flag; + + if (bio->bi_error) + set_bit(DMZ_META_ERROR, &mblk->state); + + if (bio_op(bio) == REQ_OP_WRITE) + flag = DMZ_META_WRITING; + else + flag = DMZ_META_READING; + + clear_bit_unlock(flag, &mblk->state); + smp_mb__after_atomic(); + wake_up_bit(&mblk->state, flag); + + bio_put(bio); +} + +/* + * Read a metadata block from disk. + */ +static struct dmz_mblock *dmz_fetch_mblock(struct dmz_target *dmz, + sector_t mblk_no) +{ + struct dmz_mblock *mblk; + sector_t block = dmz->sb[dmz->mblk_primary].block + mblk_no; + struct bio *bio; + + /* Get block and insert it */ + mblk = dmz_alloc_mblock(dmz, mblk_no); + if (!mblk) + return NULL; + + spin_lock(&dmz->mblk_lock); + atomic_inc(&mblk->ref); + set_bit(DMZ_META_READING, &mblk->state); + dmz_insert_mblock(dmz, mblk); + spin_unlock(&dmz->mblk_lock); + + bio = bio_alloc(GFP_NOIO, 1); + if (!bio) { + dmz_free_mblock(dmz, mblk); + return NULL; + } + + bio->bi_iter.bi_sector = dmz_blk2sect(block); + bio->bi_bdev = dmz->zbd; + bio->bi_private = mblk; + bio->bi_end_io = dmz_mblock_bio_end_io; + bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO); + bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); + submit_bio(bio); + + return mblk; +} + +/* + * Free metadata blocks. + */ +static unsigned long dmz_shrink_mblock_cache(struct dmz_target *dmz, + unsigned long limit) +{ + struct dmz_mblock *mblk; + unsigned long count = 0; + + if (!dmz->max_nr_mblks) + return 0; + + while (!list_empty(&dmz->mblk_lru_list) && + atomic_read(&dmz->nr_mblks) > dmz->min_nr_mblks && + count < limit) { + mblk = list_first_entry(&dmz->mblk_lru_list, + struct dmz_mblock, link); + list_del_init(&mblk->link); + rb_erase(&mblk->node, &dmz->mblk_rbtree); + dmz_free_mblock(dmz, mblk); + count++; + } + + return count; +} + +/* + * For mblock shrinker: get the number of unused metadata blocks in the cache. + */ +static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink, + struct shrink_control *sc) +{ + struct dmz_target *dmz = + container_of(shrink, struct dmz_target, mblk_shrinker); + + return atomic_read(&dmz->nr_mblks); +} + +/* + * For mblock shrinker: scan unused metadata blocks and shrink the cache. + */ +static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink, + struct shrink_control *sc) +{ + struct dmz_target *dmz = + container_of(shrink, struct dmz_target, mblk_shrinker); + unsigned long count; + + spin_lock(&dmz->mblk_lock); + count = dmz_shrink_mblock_cache(dmz, sc->nr_to_scan); + spin_unlock(&dmz->mblk_lock); + + return count ? count : SHRINK_STOP; +} + +/* + * Release a metadata block. + */ +static void dmz_release_mblock(struct dmz_target *dmz, struct dmz_mblock *mblk) +{ + + if (!mblk) + return; + + spin_lock(&dmz->mblk_lock); + + if (atomic_dec_and_test(&mblk->ref)) { + if (test_bit(DMZ_META_ERROR, &mblk->state)) { + rb_erase(&mblk->node, &dmz->mblk_rbtree); + dmz_free_mblock(dmz, mblk); + } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) { + list_add_tail(&mblk->link, &dmz->mblk_lru_list); + dmz_shrink_mblock_cache(dmz, 1); + } + } + + spin_unlock(&dmz->mblk_lock); +} + +/* + * Get a metadata block from the rbtree. If the block + * is not present, read it from disk. + */ +static struct dmz_mblock *dmz_get_mblock(struct dmz_target *dmz, + sector_t mblk_no) +{ + struct dmz_mblock *mblk; + + /* Check rbtree */ + spin_lock(&dmz->mblk_lock); + mblk = dmz_lookup_mblock(dmz, mblk_no); + if (mblk) { + /* Cache hit: remove block from LRU list */ + if (atomic_inc_return(&mblk->ref) == 1 && + !test_bit(DMZ_META_DIRTY, &mblk->state)) + list_del_init(&mblk->link); + } + spin_unlock(&dmz->mblk_lock); + + if (!mblk) { + /* Cache miss: read the block from disk */ + mblk = dmz_fetch_mblock(dmz, mblk_no); + if (!mblk) + return ERR_PTR(-ENOMEM); + } + + /* Wait for on-going read I/O and check for error */ + wait_on_bit_io(&mblk->state, DMZ_META_READING, + TASK_UNINTERRUPTIBLE); + if (test_bit(DMZ_META_ERROR, &mblk->state)) { + dmz_release_mblock(dmz, mblk); + return ERR_PTR(-EIO); + } + + return mblk; +} + +/* + * Mark a metadata block dirty. + */ +static void dmz_dirty_mblock(struct dmz_target *dmz, struct dmz_mblock *mblk) +{ + spin_lock(&dmz->mblk_lock); + if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state)) + list_add_tail(&mblk->link, &dmz->mblk_dirty_list); + spin_unlock(&dmz->mblk_lock); +} + +/* + * Issue a metadata block write BIO. + */ +static void dmz_write_mblock(struct dmz_target *dmz, struct dmz_mblock *mblk, + unsigned int set) +{ + sector_t block = dmz->sb[set].block + mblk->no; + struct bio *bio; + + bio = bio_alloc(GFP_NOIO, 1); + if (!bio) { + set_bit(DMZ_META_ERROR, &mblk->state); + return; + } + + set_bit(DMZ_META_WRITING, &mblk->state); + + bio->bi_iter.bi_sector = dmz_blk2sect(block); + bio->bi_bdev = dmz->zbd; + bio->bi_private = mblk; + bio->bi_end_io = dmz_mblock_bio_end_io; + bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO); + bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); + submit_bio(bio); +} + +/* + * Sync read/write a block. + */ +static int dmz_rdwr_block_sync(struct dmz_target *dmz, int op, sector_t block, + struct page *page) +{ + struct bio *bio; + int ret; + + bio = bio_alloc(GFP_NOIO, 1); + if (!bio) + return -ENOMEM; + + bio->bi_iter.bi_sector = dmz_blk2sect(block); + bio->bi_bdev = dmz->zbd; + bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO); + bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0); + ret = submit_bio_wait(bio); + bio_put(bio); + + return ret; +} + +/* + * Write super block of the specified metadata set. + */ +static int dmz_write_sb(struct dmz_target *dmz, unsigned int set) +{ + sector_t block = dmz->sb[set].block; + struct dmz_mblock *mblk = dmz->sb[set].mblk; + struct dmz_super *sb = dmz->sb[set].sb; + u64 sb_gen = dmz->sb_gen + 1; + int ret; + + sb->magic = cpu_to_le32(DMZ_MAGIC); + sb->version = cpu_to_le32(DMZ_META_VER); + + sb->gen = cpu_to_le64(sb_gen); + + sb->sb_block = cpu_to_le64(block); + sb->nr_meta_blocks = cpu_to_le32(dmz->nr_meta_blocks); + sb->nr_reserved_seq = cpu_to_le32(dmz->nr_reserved_seq); + sb->nr_chunks = cpu_to_le32(dmz->nr_chunks); + + sb->nr_map_blocks = cpu_to_le32(dmz->nr_map_blocks); + sb->nr_bitmap_blocks = cpu_to_le32(dmz->nr_bitmap_blocks); + + sb->crc = 0; + sb->crc = cpu_to_le32(crc32_le(sb_gen, + (unsigned char *)sb, DMZ_BLOCK_SIZE)); + + ret = dmz_rdwr_block_sync(dmz, REQ_OP_WRITE, block, mblk->page); + if (ret == 0) + ret = blkdev_issue_flush(dmz->zbd, GFP_KERNEL, NULL); + + return ret; +} + +/* + * Write dirty metadata blocks to the specified set. + */ +static int dmz_write_dirty_mblocks(struct dmz_target *dmz, + struct list_head *write_list, + unsigned int set) +{ + struct dmz_mblock *mblk; + struct blk_plug plug; + int ret = 0; + + /* Issue writes */ + blk_start_plug(&plug); + list_for_each_entry(mblk, write_list, link) + dmz_write_mblock(dmz, mblk, set); + blk_finish_plug(&plug); + + /* Wait for completion */ + list_for_each_entry(mblk, write_list, link) { + wait_on_bit_io(&mblk->state, DMZ_META_WRITING, + TASK_UNINTERRUPTIBLE); + if (test_bit(DMZ_META_ERROR, &mblk->state)) { + dmz_dev_err(dmz, "Write metablock %u/%llu failed\n", + set, (u64)mblk->no); + clear_bit(DMZ_META_ERROR, &mblk->state); + ret = -EIO; + } + } + + /* Flush drive cache (this will also sync data) */ + if (ret == 0) + ret = blkdev_issue_flush(dmz->zbd, GFP_KERNEL, NULL); + + return ret; +} + +/* + * Log dirty metadata blocks. + */ +static int dmz_log_dirty_mblocks(struct dmz_target *dmz, + struct list_head *write_list) +{ + unsigned int log_set = dmz->mblk_primary ^ 0x1; + int ret; + + dmz_dev_debug(dmz, "Log metadata to set %u, gen %llu\n", + log_set, dmz->sb_gen + 1); + + /* Write dirty blocks to the log */ + ret = dmz_write_dirty_mblocks(dmz, write_list, log_set); + if (ret) + return ret; + + /* + * No error so far: now validate the log by updating the + * log index super block generation. + */ + ret = dmz_write_sb(dmz, log_set); + if (ret) + return ret; + + return 0; +} + +/* + * Flush dirty metadata blocks. + */ +int dmz_flush_mblocks(struct dmz_target *dmz) +{ + struct dmz_mblock *mblk; + struct list_head write_list; + int ret; + + INIT_LIST_HEAD(&write_list); + + /* + * Prevent BIOs to zones and reclaim. This ensure exclusive + * access to metadata. + */ + down_write(&dmz->mblk_sem); + + /* If there are no dirty metadata blocks, just flush the device cache */ + if (list_empty(&dmz->mblk_dirty_list)) { + ret = blkdev_issue_flush(dmz->zbd, GFP_KERNEL, NULL); + goto out; + } + + /* + * The primary metadata set is still clean. Keep it this way until + * all updates are successful in the secondary set. That is, use + * the secondary set as a log. + */ + list_splice_init(&dmz->mblk_dirty_list, &write_list); + ret = dmz_log_dirty_mblocks(dmz, &write_list); + if (ret) + goto out; + + /* + * The log is on disk. It is now safe to update in place + * in the primary metadata set. + */ + dmz_dev_debug(dmz, "Commit metadata to set %u, gen %llu\n", + dmz->mblk_primary, dmz->sb_gen + 1); + ret = dmz_write_dirty_mblocks(dmz, &write_list, dmz->mblk_primary); + if (ret) + goto out; + + ret = dmz_write_sb(dmz, dmz->mblk_primary); + if (ret) + goto out; + + while (!list_empty(&write_list)) { + mblk = list_first_entry(&write_list, + struct dmz_mblock, link); + list_del_init(&mblk->link); + + clear_bit(DMZ_META_DIRTY, &mblk->state); + if (atomic_read(&mblk->ref) == 0) + list_add_tail(&mblk->link, &dmz->mblk_lru_list); + + } + + dmz->sb_gen++; + +out: + if (ret && !list_empty(&write_list)) + list_splice(&write_list, &dmz->mblk_dirty_list); + + up_write(&dmz->mblk_sem); + + return ret; +} + +/* + * Check super block. + */ +static int dmz_check_sb(struct dmz_target *dmz, struct dmz_super *sb) +{ + unsigned int nr_meta_zones, nr_data_zones; + u32 crc, stored_crc; + u64 gen; + + gen = le64_to_cpu(sb->gen); + stored_crc = le32_to_cpu(sb->crc); + sb->crc = 0; + crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE); + if (crc != stored_crc) { + dmz_dev_err(dmz, + "Invalid checksum (needed 0x%08x, got 0x%08x)\n", + crc, stored_crc); + return -ENXIO; + } + + if (le32_to_cpu(sb->magic) != DMZ_MAGIC) { + dmz_dev_err(dmz, + "Invalid meta magic (need 0x%08x, got 0x%08x)\n", + DMZ_MAGIC, le32_to_cpu(sb->magic)); + return -ENXIO; + } + + if (le32_to_cpu(sb->version) != DMZ_META_VER) { + dmz_dev_err(dmz, "Invalid meta version (need %d, got %d)\n", + DMZ_META_VER, le32_to_cpu(sb->version)); + return -ENXIO; + } + + nr_meta_zones = + (le32_to_cpu(sb->nr_meta_blocks) + dmz->zone_nr_blocks - 1) + >> dmz->zone_nr_blocks_shift; + if (!nr_meta_zones || + nr_meta_zones >= dmz->nr_rnd_zones) { + dmz_dev_err(dmz, "Invalid number of metadata blocks\n"); + return -ENXIO; + } + + if (!le32_to_cpu(sb->nr_reserved_seq) || + le32_to_cpu(sb->nr_reserved_seq) >= + (dmz->nr_useable_zones - nr_meta_zones)) { + dmz_dev_err(dmz, + "Invalid number of reserved sequential zones\n"); + return -ENXIO; + } + + nr_data_zones = dmz->nr_useable_zones - + (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq)); + if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) { + dmz_dev_err(dmz, "Invalid number of chunks %u / %u\n", + le32_to_cpu(sb->nr_chunks), nr_data_zones); + return -ENXIO; + } + + /* OK */ + dmz->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks); + dmz->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq); + dmz->nr_chunks = le32_to_cpu(sb->nr_chunks); + dmz->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks); + dmz->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks); + dmz->nr_meta_zones = nr_meta_zones; + dmz->nr_data_zones = nr_data_zones; + + return 0; +} + +/* + * Read the first or second super block from disk. + */ +static int dmz_read_sb(struct dmz_target *dmz, unsigned int set) +{ + return dmz_rdwr_block_sync(dmz, REQ_OP_READ, + dmz->sb[set].block, + dmz->sb[set].mblk->page); +} + +/* + * Determine the position of the secondary super blocks on disk. + * This is used only if a corruption of the primary super block + * is detected. + */ +static int dmz_lookup_secondary_sb(struct dmz_target *dmz) +{ + struct dmz_mblock *mblk; + int i; + + /* Allocate a block */ + mblk = dmz_alloc_mblock(dmz, 0); + if (!mblk) + return -ENOMEM; + + dmz->sb[1].mblk = mblk; + dmz->sb[1].sb = mblk->data; + + /* Bad first super block: search for the second one */ + dmz->sb[1].block = dmz->sb[0].block + dmz->zone_nr_blocks; + for (i = 0; i < dmz->nr_rnd_zones - 1; i++) { + if (dmz_read_sb(dmz, 1) != 0) + break; + if (le32_to_cpu(dmz->sb[1].sb->magic) == DMZ_MAGIC) + return 0; + dmz->sb[1].block += dmz->zone_nr_blocks; + } + + dmz_free_mblock(dmz, mblk); + dmz->sb[1].mblk = NULL; + + return -EIO; +} + +/* + * Read the first or second super block from disk. + */ +static int dmz_get_sb(struct dmz_target *dmz, unsigned int set) +{ + struct dmz_mblock *mblk; + int ret; + + /* Allocate a block */ + mblk = dmz_alloc_mblock(dmz, 0); + if (!mblk) + return -ENOMEM; + + dmz->sb[set].mblk = mblk; + dmz->sb[set].sb = mblk->data; + + /* Read super block */ + ret = dmz_read_sb(dmz, set); + if (ret) { + dmz_free_mblock(dmz, mblk); + dmz->sb[set].mblk = NULL; + return ret; + } + + return 0; +} + +/* + * Recover a metadata set. + */ +static int dmz_recover_mblocks(struct dmz_target *dmz, unsigned int dst_set) +{ + unsigned int src_set = dst_set ^ 0x1; + struct page *page; + int i, ret; + + dmz_dev_warn(dmz, "Metadata set %u invalid: recovering\n", + dst_set); + + if (dst_set == 0) + dmz->sb[0].block = dmz_start_block(dmz, dmz->sb_zone); + else + dmz->sb[1].block = dmz->sb[0].block + + (dmz->nr_meta_zones * dmz->zone_nr_blocks); + + page = alloc_page(GFP_KERNEL); + if (!page) + return -ENOMEM; + + /* Copy metadata blocks */ + for (i = 1; i < dmz->nr_meta_blocks; i++) { + ret = dmz_rdwr_block_sync(dmz, REQ_OP_READ, + dmz->sb[src_set].block + i, + page); + if (ret) + goto out; + ret = dmz_rdwr_block_sync(dmz, REQ_OP_WRITE, + dmz->sb[dst_set].block + i, + page); + if (ret) + goto out; + } + + /* Finalize with the super block */ + if (!dmz->sb[dst_set].mblk) { + dmz->sb[dst_set].mblk = dmz_alloc_mblock(dmz, 0); + if (!dmz->sb[dst_set].mblk) { + ret = -ENOMEM; + goto out; + } + dmz->sb[dst_set].sb = dmz->sb[dst_set].mblk->data; + } + + ret = dmz_write_sb(dmz, dst_set); + +out: + __free_pages(page, 0); + + return ret; +} + +/* + * Get super block from disk. + */ +static int dmz_load_sb(struct dmz_target *dmz) +{ + bool sb_good[2] = {false, false}; + u64 sb_gen[2] = {0, 0}; + int ret; + + /* Read and check the primary super block */ + dmz->sb[0].block = dmz_start_block(dmz, dmz->sb_zone); + ret = dmz_get_sb(dmz, 0); + if (ret) { + dmz_dev_err(dmz, "Read primary super block failed\n"); + return ret; + } + + ret = dmz_check_sb(dmz, dmz->sb[0].sb); + + /* Read and check secondary super block */ + if (ret == 0) { + sb_good[0] = true; + dmz->sb[1].block = dmz->sb[0].block + + (dmz->nr_meta_zones * dmz->zone_nr_blocks); + ret = dmz_get_sb(dmz, 1); + } else { + ret = dmz_lookup_secondary_sb(dmz); + } + if (ret) { + dmz_dev_err(dmz, "Read secondary super block failed\n"); + return ret; + } + + ret = dmz_check_sb(dmz, dmz->sb[1].sb); + if (ret == 0) + sb_good[1] = true; + + /* Use highest generation sb first */ + if (!sb_good[0] && !sb_good[1]) { + dmz_dev_err(dmz, "No valid super block found\n"); + return -EIO; + } + + if (sb_good[0]) + sb_gen[0] = le64_to_cpu(dmz->sb[0].sb->gen); + else + ret = dmz_recover_mblocks(dmz, 0); + + if (sb_good[1]) + sb_gen[1] = le64_to_cpu(dmz->sb[1].sb->gen); + else + ret = dmz_recover_mblocks(dmz, 1); + + if (ret) { + dmz_dev_err(dmz, "Recovery failed\n"); + return -EIO; + } + + if (sb_gen[0] >= sb_gen[1]) { + dmz->sb_gen = sb_gen[0]; + dmz->mblk_primary = 0; + } else { + dmz->sb_gen = sb_gen[1]; + dmz->mblk_primary = 1; + } + + dmz_dev_debug(dmz, "Using super block %u (gen %llu)\n", + dmz->mblk_primary, dmz->sb_gen); + + return 0; +} + +/* + * Initialize a zone descriptor. + */ +static int dmz_init_zone(struct dmz_target *dmz, struct dm_zone *zone, + struct blk_zone *blkz) +{ + + /* Ignore the eventual last runt (smaller) zone */ + if (blkz->len != dmz->zone_nr_sectors) { + if (blkz->start + blkz->len == dmz->zbd_capacity) + return 0; + return -ENXIO; + } + + INIT_LIST_HEAD(&zone->link); + atomic_set(&zone->refcount, 0); + zone->chunk = DMZ_MAP_UNMAPPED; + + if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) { + set_bit(DMZ_RND, &zone->flags); + dmz->nr_rnd_zones++; + } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ || + blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) { + set_bit(DMZ_SEQ, &zone->flags); + } else { + return -ENXIO; + } + + if (blkz->cond == BLK_ZONE_COND_OFFLINE) + set_bit(DMZ_OFFLINE, &zone->flags); + else if (blkz->cond == BLK_ZONE_COND_READONLY) + set_bit(DMZ_READ_ONLY, &zone->flags); + + if (dmz_is_rnd(zone)) + zone->wp_block = 0; + else + zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start); + + if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) { + dmz->nr_useable_zones++; + if (dmz_is_rnd(zone)) { + dmz->nr_rnd_zones++; + if (!dmz->sb_zone) { + /* Super block zone */ + dmz->sb_zone = zone; + } + } + } + + return 0; +} + +/* + * Free zones descriptors. + */ +static void dmz_drop_zones(struct dmz_target *dmz) +{ + kfree(dmz->zones); + dmz->zones = NULL; +} + +/* + * Allocate and initialize zone descriptors using the zone + * information from disk. + */ +static int dmz_init_zones(struct dmz_target *dmz) +{ + struct dm_zone *zone; + struct blk_zone *blkz; + unsigned int nr_blkz; + sector_t sector = 0; + int i, ret = 0; + + /* Init */ + dmz->zone_nr_sectors = dmz->zbdq->limits.chunk_sectors; + dmz->zone_nr_sectors_shift = ilog2(dmz->zone_nr_sectors); + + dmz->zone_nr_blocks = dmz_sect2blk(dmz->zone_nr_sectors); + dmz->zone_nr_blocks_shift = ilog2(dmz->zone_nr_blocks); + + dmz->zone_bitmap_size = dmz->zone_nr_blocks >> 3; + dmz->zone_nr_bitmap_blocks = + dmz->zone_bitmap_size >> DMZ_BLOCK_SHIFT; + + dmz->nr_zones = (dmz->zbd_capacity + dmz->zone_nr_sectors - 1) + >> dmz->zone_nr_sectors_shift; + + /* Allocate zone array */ + dmz->zones = kcalloc(dmz->nr_zones, sizeof(struct dm_zone), GFP_KERNEL); + if (!dmz->zones) + return -ENOMEM; + + dmz_dev_info(dmz, "Using %zu B for zone information\n", + sizeof(struct dm_zone) * dmz->nr_zones); + + /* Get zone information */ + nr_blkz = DMZ_REPORT_NR_ZONES; + blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL); + if (!blkz) { + ret = -ENOMEM; + goto out; + } + + /* + * Get zone information and initialize zone descriptors. + * At the same time, determine where the super block + * should be: first block of the first randomly writable + * zone. + */ + zone = dmz->zones; + while (sector < dmz->zbd_capacity) { + + /* Get zone information */ + nr_blkz = DMZ_REPORT_NR_ZONES; + ret = blkdev_report_zones(dmz->zbd, sector, + blkz, &nr_blkz, + GFP_KERNEL); + if (ret) { + dmz_dev_err(dmz, "Report zones failed %d\n", ret); + goto out; + } + + /* Process report */ + for (i = 0; i < nr_blkz; i++) { + ret = dmz_init_zone(dmz, zone, &blkz[i]); + if (ret) + goto out; + sector += dmz->zone_nr_sectors; + zone++; + } + + } + + /* The entire zone configuration of the disk should now be known */ + if (sector < dmz->zbd_capacity) { + dmz_dev_err(dmz, "Failed to get zone information\n"); + ret = -ENXIO; + goto out; + } + +out: + kfree(blkz); + + if (ret) + dmz_drop_zones(dmz); + + return ret; +} + +/* + * Update a zone information. + */ +static int dmz_update_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + unsigned int nr_blkz = 1; + struct blk_zone blkz; + int ret; + + /* Get zone information from disk */ + ret = blkdev_report_zones(dmz->zbd, dmz_start_sect(dmz, zone), + &blkz, &nr_blkz, + GFP_KERNEL); + if (ret) { + dmz_dev_err(dmz, "Get zone %u report failed\n", + dmz_id(dmz, zone)); + return ret; + } + + clear_bit(DMZ_OFFLINE, &zone->flags); + clear_bit(DMZ_READ_ONLY, &zone->flags); + if (blkz.cond == BLK_ZONE_COND_OFFLINE) + set_bit(DMZ_OFFLINE, &zone->flags); + else if (blkz.cond == BLK_ZONE_COND_READONLY) + set_bit(DMZ_READ_ONLY, &zone->flags); + + if (dmz_is_seq(zone)) + zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start); + else + zone->wp_block = 0; + + return 0; +} + +/* + * Check a zone write pointer position when the zone is marked + * with the sequential write error flag. + */ +static int dmz_handle_seq_write_err(struct dmz_target *dmz, + struct dm_zone *zone) +{ + unsigned int wp = 0; + int ret = 0; + + wp = zone->wp_block; + ret = dmz_update_zone(dmz, zone); + if (ret != 0) + return ret; + + dmz_dev_warn(dmz, "Processing zone %u write error (zone wp %u/%u)\n", + dmz_id(dmz, zone), zone->wp_block, wp); + + if (zone->wp_block < wp) + dmz_invalidate_blocks(dmz, zone, + zone->wp_block, + wp - zone->wp_block); + + return 0; +} + +/* + * Check zone information after a resume. + */ +static int dmz_check_zones(struct dmz_target *dmz) +{ + struct dm_zone *zone; + sector_t wp_block; + unsigned int i; + int ret; + + /* Check zones */ + for (i = 0; i < dmz->nr_zones; i++) { + + zone = dmz_get(dmz, i); + if (!zone) { + dmz_dev_err(dmz, "Unable to get zone %u\n", i); + return -EIO; + } + + wp_block = zone->wp_block; + + ret = dmz_update_zone(dmz, zone); + if (ret) { + dmz_dev_err(dmz, "Broken zone %u\n", i); + return ret; + } + + if (dmz_is_offline(zone)) { + dmz_dev_warn(dmz, "Zone %u is offline\n", i); + continue; + } + + /* Check write pointer */ + if (!dmz_is_seq(zone)) + zone->wp_block = 0; + else if (zone->wp_block != wp_block) { + dmz_dev_err(dmz, "Zone %u: Invalid wp (%llu / %llu)\n", + i, (u64)zone->wp_block, (u64)wp_block); + zone->wp_block = wp_block; + dmz_invalidate_blocks(dmz, zone, zone->wp_block, + dmz->zone_nr_blocks - zone->wp_block); + } + + } + + return 0; +} + +/* + * Reset a zone write pointer. + */ +static int dmz_reset_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + int ret; + + /* + * Ignore offline zones, read only zones, + * and conventional zones. + */ + if (dmz_is_offline(zone) || + dmz_is_readonly(zone) || + dmz_is_rnd(zone)) + return 0; + + if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) { + ret = blkdev_reset_zones(dmz->zbd, + dmz_start_sect(dmz, zone), + dmz->zone_nr_sectors, + GFP_KERNEL); + if (ret) { + dmz_dev_err(dmz, "Reset zone %u failed %d\n", + dmz_id(dmz, zone), ret); + return ret; + } + } + + /* Clear write error bit and rewind write pointer position */ + clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags); + zone->wp_block = 0; + + return 0; +} + +static void dmz_get_zone_weight(struct dmz_target *dmz, struct dm_zone *zone); + +/* + * Initialize chunk mapping. + */ +static int dmz_load_mapping(struct dmz_target *dmz) +{ + struct dm_zone *dzone, *bzone; + struct dmz_mblock *dmap_mblk = NULL; + struct dmz_map *dmap; + unsigned int i = 0, e = 0, chunk = 0; + unsigned int dzone_id; + unsigned int bzone_id; + + /* Metadata block array for the chunk mapping table */ + dmz->dz_map_mblk = kcalloc(dmz->nr_map_blocks, + sizeof(struct dmz_mblk *), + GFP_KERNEL); + if (!dmz->dz_map_mblk) + return -ENOMEM; + + /* Get chunk mapping table blocks and initialize zone mapping */ + while (chunk < dmz->nr_chunks) { + + if (!dmap_mblk) { + /* Get mapping block */ + dmap_mblk = dmz_get_mblock(dmz, i + 1); + if (IS_ERR(dmap_mblk)) + return PTR_ERR(dmap_mblk); + dmz->dz_map_mblk[i] = dmap_mblk; + dmap = (struct dmz_map *) dmap_mblk->data; + i++; + e = 0; + } + + /* Check data zone */ + dzone_id = le32_to_cpu(dmap[e].dzone_id); + if (dzone_id == DMZ_MAP_UNMAPPED) + goto next; + + if (dzone_id >= dmz->nr_zones) { + dmz_dev_err(dmz, + "Chunk %u mapping: invalid data zone ID %u\n", + chunk, dzone_id); + return -EIO; + } + + dzone = dmz_get(dmz, dzone_id); + set_bit(DMZ_DATA, &dzone->flags); + dzone->chunk = chunk; + dmz_get_zone_weight(dmz, dzone); + + if (dmz_is_rnd(dzone)) + list_add_tail(&dzone->link, &dmz->dz_map_rnd_list); + else + list_add_tail(&dzone->link, &dmz->dz_map_seq_list); + + /* Check buffer zone */ + bzone_id = le32_to_cpu(dmap[e].bzone_id); + if (bzone_id == DMZ_MAP_UNMAPPED) + goto next; + + if (bzone_id >= dmz->nr_zones) { + dmz_dev_err(dmz, + "Chunk %u mapping: invalid buffer zone ID %u\n", + chunk, bzone_id); + return -EIO; + } + + bzone = dmz_get(dmz, bzone_id); + if (!dmz_is_rnd(bzone)) { + dmz_dev_err(dmz, + "Chunk %u mapping: invalid buffer zone %u\n", + chunk, bzone_id); + return -EIO; + } + + set_bit(DMZ_DATA, &bzone->flags); + set_bit(DMZ_BUF, &bzone->flags); + bzone->chunk = chunk; + bzone->bzone = dzone; + dzone->bzone = bzone; + dmz_get_zone_weight(dmz, bzone); + list_add_tail(&bzone->link, &dmz->dz_map_rnd_list); + +next: + chunk++; + e++; + if (e >= DMZ_MAP_ENTRIES) + dmap_mblk = NULL; + + } + + /* + * At this point, only meta zones and mapped data zones were + * fully initialized. All remaining zones are unmapped data + * zones. Finish initializing those here. + */ + for (i = 0; i < dmz->nr_zones; i++) { + + dzone = dmz_get(dmz, i); + if (dmz_is_meta(dzone)) + continue; + + if (dmz_is_rnd(dzone)) + dmz->dz_nr_rnd++; + else + dmz->dz_nr_seq++; + + if (dmz_is_data(dzone)) + /* Already initialized */ + continue; + + /* Unmapped data zone */ + set_bit(DMZ_DATA, &dzone->flags); + dzone->chunk = DMZ_MAP_UNMAPPED; + if (dmz_is_rnd(dzone)) { + list_add_tail(&dzone->link, + &dmz->dz_unmap_rnd_list); + atomic_inc(&dmz->dz_unmap_nr_rnd); + } else if (atomic_read(&dmz->nr_reclaim_seq_zones) < + dmz->nr_reserved_seq) { + list_add_tail(&dzone->link, + &dmz->reclaim_seq_zones_list); + atomic_inc(&dmz->nr_reclaim_seq_zones); + dmz->dz_nr_seq--; + } else { + list_add_tail(&dzone->link, + &dmz->dz_unmap_seq_list); + atomic_inc(&dmz->dz_unmap_nr_seq); + } + } + + return 0; +} + +/* + * Set a data chunk mapping. + */ +static void dmz_set_chunk_mapping(struct dmz_target *dmz, + unsigned int chunk, + unsigned int dzone_id, + unsigned int bzone_id) +{ + struct dmz_mblock *dmap_mblk = + dmz->dz_map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; + struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; + int map_idx = chunk & DMZ_MAP_ENTRIES_MASK; + + dmap[map_idx].dzone_id = cpu_to_le32(dzone_id); + dmap[map_idx].bzone_id = cpu_to_le32(bzone_id); + dmz_dirty_mblock(dmz, dmap_mblk); +} + +/* + * The list of mapped zones is maintained in LRU order. + * This rotates a zone at the end of its map list. + */ +static void __dmz_lru_zone(struct dmz_target *dmz, + struct dm_zone *zone) +{ + if (list_empty(&zone->link)) + return; + + list_del_init(&zone->link); + if (dmz_is_seq(zone)) + /* LRU rotate sequential zone */ + list_add_tail(&zone->link, &dmz->dz_map_seq_list); + else + /* LRU rotate random zone */ + list_add_tail(&zone->link, &dmz->dz_map_rnd_list); +} + +/* + * The list of mapped random zones is maintained + * in LRU order. This rotates a zone at the end of the list. + */ +static void dmz_lru_zone(struct dmz_target *dmz, + struct dm_zone *zone) +{ + __dmz_lru_zone(dmz, zone); + if (zone->bzone) + __dmz_lru_zone(dmz, zone->bzone); +} + +/* + * Wait for any zone to be freed. + */ +static void dmz_wait_for_free_zones(struct dmz_target *dmz) +{ + DEFINE_WAIT(wait); + + dmz_trigger_reclaim(dmz); + + prepare_to_wait(&dmz->dz_free_wq, &wait, TASK_UNINTERRUPTIBLE); + dmz_unlock_map(dmz); + up_read(&dmz->mblk_sem); + + io_schedule_timeout(HZ); + + down_read(&dmz->mblk_sem); + dmz_lock_map(dmz); + finish_wait(&dmz->dz_free_wq, &wait); +} + +/* + * Wait for a zone reclaim to complete. + */ +static void dmz_wait_for_reclaim(struct dmz_target *dmz, + struct dm_zone *zone) +{ + dmz_unlock_map(dmz); + wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, + TASK_UNINTERRUPTIBLE, + HZ); + dmz_lock_map(dmz); +} + +/* + * Activate a zone (increment its reference count). + */ +void dmz_activate_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + set_bit(DMZ_ACTIVE, &zone->flags); + atomic_inc(&zone->refcount); +} + +/* + * Deactivate a zone. This decrement the zone reference counter + * and clears the active state of the zone once the count reaches 0, + * indicating that all BIOs to the zone have completed. Returns + * true if the zone was deactivated. + */ +void dmz_deactivate_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + if (atomic_dec_and_test(&zone->refcount)) { + WARN_ON(!test_bit(DMZ_ACTIVE, &zone->flags)); + clear_bit_unlock(DMZ_ACTIVE, &zone->flags); + smp_mb__after_atomic(); + } +} + +/* + * Get the zone mapping a chunk, if the chunk is mapped already. + * If no mapping exist and the operation is WRITE, a zone is + * allocated and used to map the chunk. + * The zone returned will be set to the active state. + */ +struct dm_zone *dmz_get_chunk_mapping(struct dmz_target *dmz, + unsigned int chunk, int op) +{ + struct dmz_mblock *dmap_mblk = + dmz->dz_map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; + struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; + int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK; + unsigned int dzone_id; + struct dm_zone *dzone = NULL; + int ret = 0; + + dmz_lock_map(dmz); + +again: + + /* Get the chunk mapping */ + dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id); + if (dzone_id == DMZ_MAP_UNMAPPED) { + + /* + * Read or discard in unmapped chunks are fine. But for + * writes, we need a mapping, so get one. + */ + if (op != REQ_OP_WRITE) + goto out; + + /* Alloate a random zone */ + dzone = dmz_alloc_zone(dmz, DMZ_ALLOC_RND); + if (!dzone) { + dmz_wait_for_free_zones(dmz); + goto again; + } + + dmz_map_zone(dmz, dzone, chunk); + + } else { + + /* The chunk is already mapped: get the mapping zone */ + dzone = dmz_get(dmz, dzone_id); + if (dzone->chunk != chunk) { + dzone = ERR_PTR(-EIO); + goto out; + } + + /* Repair write pointer if the sequential dzone has error */ + if (dmz_seq_write_err(dzone)) { + ret = dmz_handle_seq_write_err(dmz, dzone); + if (ret) { + dzone = ERR_PTR(-EIO); + goto out; + } + clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags); + } + } + + /* + * If the zone is being reclaimed, the chunk mapping may change + * to a different zone. So wait for reclaim and retry. Otherwise, + * activate the zone (this will prevent reclaim from touching it). + */ + if (dmz_in_reclaim(dzone)) { + dmz_wait_for_reclaim(dmz, dzone); + goto again; + } + dmz_activate_zone(dmz, dzone); + dmz_lru_zone(dmz, dzone); + +out: + dmz_unlock_map(dmz); + + return dzone; +} + +/* + * Write and discard change the block validity of data zones and their buffer + * zones. Check here that valid blocks are still present. If all blocks are + * invalid, the zones can be unmapped on the fly without waiting for reclaim + * to do it. + */ +void dmz_put_chunk_mapping(struct dmz_target *dmz, struct dm_zone *dzone) +{ + struct dm_zone *bzone; + + dmz_lock_map(dmz); + + bzone = dzone->bzone; + if (bzone) { + if (dmz_weight(bzone)) { + dmz_lru_zone(dmz, bzone); + } else { + /* Empty buffer zone: reclaim it */ + dmz_unmap_zone(dmz, bzone); + dmz_free_zone(dmz, bzone); + bzone = NULL; + } + } + + /* Deactivate the data zone */ + dmz_deactivate_zone(dmz, dzone); + if (dmz_is_active(dzone) || bzone || dmz_weight(dzone)) { + dmz_lru_zone(dmz, dzone); + } else { + /* Unbuffered inactive empty data zone: reclaim it */ + dmz_unmap_zone(dmz, dzone); + dmz_free_zone(dmz, dzone); + } + + dmz_unlock_map(dmz); +} + +/* + * Allocate and map a random zone to buffer a chunk + * already mapped to a sequential zone. + */ +struct dm_zone *dmz_get_chunk_buffer(struct dmz_target *dmz, + struct dm_zone *dzone) +{ + struct dm_zone *bzone; + unsigned int chunk; + + dmz_lock_map(dmz); + + chunk = dzone->chunk; + + /* Alloate a random zone */ + do { + bzone = dmz_alloc_zone(dmz, DMZ_ALLOC_RND); + if (!bzone) + dmz_wait_for_free_zones(dmz); + } while (!bzone); + + /* Update the chunk mapping */ + dmz_set_chunk_mapping(dmz, chunk, + dmz_id(dmz, dzone), + dmz_id(dmz, bzone)); + + set_bit(DMZ_BUF, &bzone->flags); + bzone->chunk = chunk; + bzone->bzone = dzone; + dzone->bzone = bzone; + list_add_tail(&bzone->link, &dmz->dz_map_rnd_list); + + dmz_unlock_map(dmz); + + return bzone; +} + +/* + * Get an unmapped (free) zone. + * This must be called with the mapping lock held. + */ +struct dm_zone *dmz_alloc_zone(struct dmz_target *dmz, unsigned long flags) +{ + struct list_head *list; + struct dm_zone *zone; + + if (flags & DMZ_ALLOC_RND) + list = &dmz->dz_unmap_rnd_list; + else + list = &dmz->dz_unmap_seq_list; + +again: + if (list_empty(list)) { + + /* + * No free zone: if this is for reclaim, allow using the + * reserved sequential zones. + */ + if (!(flags & DMZ_ALLOC_RECLAIM) || + list_empty(&dmz->reclaim_seq_zones_list)) + return NULL; + + zone = list_first_entry(&dmz->reclaim_seq_zones_list, + struct dm_zone, link); + list_del_init(&zone->link); + atomic_dec(&dmz->nr_reclaim_seq_zones); + return zone; + + } + + zone = list_first_entry(list, struct dm_zone, link); + list_del_init(&zone->link); + + if (dmz_is_rnd(zone)) + atomic_dec(&dmz->dz_unmap_nr_rnd); + else + atomic_dec(&dmz->dz_unmap_nr_seq); + + if (dmz_is_offline(zone)) { + dmz_dev_warn(dmz, "Zone %u is offline\n", + dmz_id(dmz, zone)); + zone = NULL; + goto again; + } + + if (dmz_should_reclaim(dmz)) + dmz_trigger_reclaim(dmz); + + return zone; +} + +/* + * Free a zone. + * This must be called with the mapping lock held. + */ +void dmz_free_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + + + + /* If this is a sequential zone, reset it */ + if (dmz_is_seq(zone)) + dmz_reset_zone(dmz, zone); + + /* Return the zone to its type unmap list */ + if (dmz_is_rnd(zone)) { + list_add_tail(&zone->link, &dmz->dz_unmap_rnd_list); + atomic_inc(&dmz->dz_unmap_nr_rnd); + } else if (atomic_read(&dmz->nr_reclaim_seq_zones) < + dmz->nr_reserved_seq) { + list_add_tail(&zone->link, &dmz->reclaim_seq_zones_list); + atomic_inc(&dmz->nr_reclaim_seq_zones); + } else { + list_add_tail(&zone->link, &dmz->dz_unmap_seq_list); + atomic_inc(&dmz->dz_unmap_nr_seq); + } + + wake_up_all(&dmz->dz_free_wq); +} + +/* + * Map a chunk to a zone. + * This must be called with the mapping lock held. + */ +void dmz_map_zone(struct dmz_target *dmz, struct dm_zone *dzone, + unsigned int chunk) +{ + + /* Set the chunk mapping */ + dmz_set_chunk_mapping(dmz, chunk, + dmz_id(dmz, dzone), + DMZ_MAP_UNMAPPED); + dzone->chunk = chunk; + if (dmz_is_rnd(dzone)) + list_add_tail(&dzone->link, &dmz->dz_map_rnd_list); + else + list_add_tail(&dzone->link, &dmz->dz_map_seq_list); +} + +/* + * Unmap a zone. + * This must be called with the mapping lock held. + */ +void dmz_unmap_zone(struct dmz_target *dmz, struct dm_zone *zone) +{ + unsigned int chunk = zone->chunk; + unsigned int dzone_id; + + if (chunk == DMZ_MAP_UNMAPPED) + /* Already unmapped */ + return; + + if (test_and_clear_bit(DMZ_BUF, &zone->flags)) { + + /* + * Unmapping the chunk buffer zone: clear only + * the chunk buffer mapping + */ + dzone_id = dmz_id(dmz, zone->bzone); + zone->bzone->bzone = NULL; + zone->bzone = NULL; + + } else { + + /* + * Unmapping the chunk data zone: the zone must + * not be buffered. + */ + if (WARN_ON(zone->bzone)) { + zone->bzone->bzone = NULL; + zone->bzone = NULL; + } + dzone_id = DMZ_MAP_UNMAPPED; + + } + + dmz_set_chunk_mapping(dmz, chunk, dzone_id, + DMZ_MAP_UNMAPPED); + + zone->chunk = DMZ_MAP_UNMAPPED; + list_del_init(&zone->link); +} + +/* + * Set @nr_bits bits in @bitmap starting from @bit. + * Return the number of bits changed from 0 to 1. + */ +static unsigned int dmz_set_bits(unsigned long *bitmap, + unsigned int bit, unsigned int nr_bits) +{ + unsigned long *addr; + unsigned int end = bit + nr_bits; + unsigned int n = 0; + + while (bit < end) { + + if (((bit & (BITS_PER_LONG - 1)) == 0) && + ((end - bit) >= BITS_PER_LONG)) { + /* Try to set the whole word at once */ + addr = bitmap + BIT_WORD(bit); + if (*addr == 0) { + *addr = ULONG_MAX; + n += BITS_PER_LONG; + bit += BITS_PER_LONG; + continue; + } + } + + if (!test_and_set_bit(bit, bitmap)) + n++; + bit++; + + } + + return n; + +} + +/* + * Get the bitmap block storing the bit for chunk_block in zone. + */ +static struct dmz_mblock *dmz_get_bitmap(struct dmz_target *dmz, + struct dm_zone *zone, + sector_t chunk_block) +{ + sector_t bitmap_block = 1 + dmz->nr_map_blocks + + (sector_t)(dmz_id(dmz, zone) * dmz->zone_nr_bitmap_blocks) + + (chunk_block >> DMZ_BLOCK_SHIFT_BITS); + + return dmz_get_mblock(dmz, bitmap_block); +} + +/* + * Copy the bitmap of from_zone to the bitmap of to_zone. + */ +int dmz_valid_copy(struct dmz_target *dmz, struct dm_zone *from_zone, + struct dm_zone *to_zone) +{ + struct dmz_mblock *from_mblk, *to_mblk; + sector_t chunk_block = 0; + + /* Get the zones bitmap blocks */ + while (chunk_block < dmz->zone_nr_blocks) { + + from_mblk = dmz_get_bitmap(dmz, from_zone, chunk_block); + if (IS_ERR(from_mblk)) + return PTR_ERR(from_mblk); + to_mblk = dmz_get_bitmap(dmz, to_zone, chunk_block); + if (IS_ERR(to_mblk)) { + dmz_release_mblock(dmz, from_mblk); + return PTR_ERR(to_mblk); + } + + memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE); + dmz_dirty_mblock(dmz, to_mblk); + + dmz_release_mblock(dmz, to_mblk); + dmz_release_mblock(dmz, from_mblk); + + chunk_block += DMZ_BLOCK_SIZE_BITS; + + } + + to_zone->weight = from_zone->weight; + + return 0; +} + +/* + * Merge the valid blocks of from_zone into the bitmap of to_zone. + */ +int dmz_valid_merge(struct dmz_target *dmz, struct dm_zone *from_zone, + struct dm_zone *to_zone, sector_t chunk_block) +{ + unsigned int nr_blocks; + int ret; + + /* Get the zones bitmap blocks */ + while (chunk_block < dmz->zone_nr_blocks) { + + /* Get a valid region from the source zone */ + ret = dmz_first_valid_block(dmz, from_zone, &chunk_block); + if (ret < 0) + return ret; + + /* Are we done ? */ + nr_blocks = ret; + if (!nr_blocks) + return 0; + + ret = dmz_validate_blocks(dmz, to_zone, chunk_block, nr_blocks); + if (ret != 0) + return ret; + + chunk_block += nr_blocks; + + } + + return 0; +} + +/* + * Validate all the blocks in the range [block..block+nr_blocks-1]. + */ +int dmz_validate_blocks(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block, unsigned int nr_blocks) +{ + unsigned int count, bit, nr_bits; + struct dmz_mblock *mblk; + unsigned int n = 0; + + dmz_dev_debug(dmz, "=> VALIDATE zone %u, block %llu, %u blocks\n", + dmz_id(dmz, zone), (u64)chunk_block, nr_blocks); + + WARN_ON(chunk_block + nr_blocks > dmz->zone_nr_blocks); + + while (nr_blocks) { + + /* Get bitmap block */ + mblk = dmz_get_bitmap(dmz, zone, chunk_block); + if (IS_ERR(mblk)) + return PTR_ERR(mblk); + + /* Set bits */ + bit = chunk_block & DMZ_BLOCK_MASK_BITS; + nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit); + + count = dmz_set_bits((unsigned long *) mblk->data, + bit, nr_bits); + if (count) { + dmz_dirty_mblock(dmz, mblk); + n += count; + } + dmz_release_mblock(dmz, mblk); + + nr_blocks -= nr_bits; + chunk_block += nr_bits; + + } + + if (likely(zone->weight + n <= dmz->zone_nr_blocks)) { + zone->weight += n; + } else { + dmz_dev_warn(dmz, "Zone %u: weight %u should be <= %llu\n", + dmz_id(dmz, zone), zone->weight, + (u64)dmz->zone_nr_blocks - n); + zone->weight = dmz->zone_nr_blocks; + } + + return 0; +} + +/* + * Clear nr_bits bits in bitmap starting from bit. + * Return the number of bits cleared. + */ +static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits) +{ + unsigned long *addr; + int end = bit + nr_bits; + int n = 0; + + while (bit < end) { + + if (((bit & (BITS_PER_LONG - 1)) == 0) && + ((end - bit) >= BITS_PER_LONG)) { + /* Try to clear whole word at once */ + addr = bitmap + BIT_WORD(bit); + if (*addr == ULONG_MAX) { + *addr = 0; + n += BITS_PER_LONG; + bit += BITS_PER_LONG; + continue; + } + } + + if (test_and_clear_bit(bit, bitmap)) + n++; + bit++; + + } + + return n; + +} + +/* + * Invalidate all the blocks in the range [block..block+nr_blocks-1]. + */ +int dmz_invalidate_blocks(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block, unsigned int nr_blocks) +{ + unsigned int count, bit, nr_bits; + struct dmz_mblock *mblk; + unsigned int n = 0; + + dmz_dev_debug(dmz, "=> INVALIDATE zone %u, block %llu, %u blocks\n", + dmz_id(dmz, zone), (u64)chunk_block, nr_blocks); + + WARN_ON(chunk_block + nr_blocks > dmz->zone_nr_blocks); + + while (nr_blocks) { + + /* Get bitmap block */ + mblk = dmz_get_bitmap(dmz, zone, chunk_block); + if (IS_ERR(mblk)) + return PTR_ERR(mblk); + + /* Clear bits */ + bit = chunk_block & DMZ_BLOCK_MASK_BITS; + nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit); + + count = dmz_clear_bits((unsigned long *) mblk->data, + bit, nr_bits); + if (count) { + dmz_dirty_mblock(dmz, mblk); + n += count; + } + dmz_release_mblock(dmz, mblk); + + nr_blocks -= nr_bits; + chunk_block += nr_bits; + + } + + if (zone->weight >= n) { + zone->weight -= n; + } else { + dmz_dev_warn(dmz, "Zone %u: weight %u should be >= %u\n", + dmz_id(dmz, zone), zone->weight, n); + zone->weight = 0; + } + + return 0; +} + +/* + * Get a block bit value. + */ +static int dmz_test_block(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block) +{ + struct dmz_mblock *mblk; + int ret; + + WARN_ON(chunk_block >= dmz->zone_nr_blocks); + + /* Get bitmap block */ + mblk = dmz_get_bitmap(dmz, zone, chunk_block); + if (IS_ERR(mblk)) + return PTR_ERR(mblk); + + /* Get offset */ + ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS, + (unsigned long *) mblk->data) != 0; + + dmz_release_mblock(dmz, mblk); + + return ret; +} + +/* + * Return the number of blocks from chunk_block to the first block with a bit + * value specified by set. Search at most nr_blocks blocks from chunk_block. + */ +static int dmz_to_next_set_block(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block, unsigned int nr_blocks, + int set) +{ + struct dmz_mblock *mblk; + unsigned int bit, set_bit, nr_bits; + unsigned long *bitmap; + int n = 0; + + WARN_ON(chunk_block + nr_blocks > dmz->zone_nr_blocks); + + while (nr_blocks) { + + /* Get bitmap block */ + mblk = dmz_get_bitmap(dmz, zone, chunk_block); + if (IS_ERR(mblk)) + return PTR_ERR(mblk); + + /* Get offset */ + bitmap = (unsigned long *) mblk->data; + bit = chunk_block & DMZ_BLOCK_MASK_BITS; + nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit); + if (set) + set_bit = find_next_bit(bitmap, + DMZ_BLOCK_SIZE_BITS, + bit); + else + set_bit = find_next_zero_bit(bitmap, + DMZ_BLOCK_SIZE_BITS, + bit); + dmz_release_mblock(dmz, mblk); + + n += set_bit - bit; + if (set_bit < DMZ_BLOCK_SIZE_BITS) + break; + + nr_blocks -= nr_bits; + chunk_block += nr_bits; + + } + + return n; +} + +/* + * Test if chunk_block is valid. If it is, the number of consecutive + * valid blocks from chunk_block will be returned. + */ +int dmz_block_valid(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block) +{ + int valid; + + /* Test block */ + valid = dmz_test_block(dmz, zone, chunk_block); + if (valid <= 0) + return valid; + + /* The block is valid: get the number of valid blocks from block */ + return dmz_to_next_set_block(dmz, zone, chunk_block, + dmz->zone_nr_blocks - chunk_block, + 0); +} + +/* + * Find the first valid block from @chunk_block in @zone. + * If such a block is found, its number is returned using + * @chunk_block and the total number of valid blocks from @chunk_block + * is returned. + */ +int dmz_first_valid_block(struct dmz_target *dmz, struct dm_zone *zone, + sector_t *chunk_block) +{ + sector_t start_block = *chunk_block; + int ret; + + ret = dmz_to_next_set_block(dmz, zone, start_block, + dmz->zone_nr_blocks - start_block, 1); + if (ret < 0) + return ret; + + start_block += ret; + *chunk_block = start_block; + + return dmz_to_next_set_block(dmz, zone, start_block, + dmz->zone_nr_blocks - start_block, 0); +} + +/* + * Count the number of bits set starting from bit up to bit + nr_bits - 1. + */ +static int dmz_count_bits(void *bitmap, int bit, int nr_bits) +{ + unsigned long *addr; + int end = bit + nr_bits; + int n = 0; + + while (bit < end) { + + if (((bit & (BITS_PER_LONG - 1)) == 0) && + ((end - bit) >= BITS_PER_LONG)) { + addr = (unsigned long *)bitmap + BIT_WORD(bit); + if (*addr == ULONG_MAX) { + n += BITS_PER_LONG; + bit += BITS_PER_LONG; + continue; + } + } + + if (test_bit(bit, bitmap)) + n++; + bit++; + + } + + return n; + +} + +/* + * Get a zone weight. + */ +static void dmz_get_zone_weight(struct dmz_target *dmz, struct dm_zone *zone) +{ + struct dmz_mblock *mblk; + sector_t chunk_block = 0; + unsigned int bit, nr_bits; + unsigned int nr_blocks = dmz->zone_nr_blocks; + void *bitmap; + int n = 0; + + while (nr_blocks) { + + /* Get bitmap block */ + mblk = dmz_get_bitmap(dmz, zone, chunk_block); + if (IS_ERR(mblk)) { + n = 0; + break; + } + + /* Count bits in this block */ + bitmap = mblk->data; + bit = chunk_block & DMZ_BLOCK_MASK_BITS; + nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit); + n += dmz_count_bits(bitmap, bit, nr_bits); + + dmz_release_mblock(dmz, mblk); + + nr_blocks -= nr_bits; + chunk_block += nr_bits; + + } + + zone->weight = n; +} + +/* + * Initialize the target metadata. + */ +int dmz_init_meta(struct dmz_target *dmz) +{ + unsigned int i, zid; + struct dm_zone *zone; + int ret; + + /* Initialize zone descriptors */ + ret = dmz_init_zones(dmz); + if (ret) + return ret; + + /* Get super block */ + ret = dmz_load_sb(dmz); + if (ret) + goto out; + + /* Set metadata zones starting from sb_zone */ + zid = dmz_id(dmz, dmz->sb_zone); + for (i = 0; i < dmz->nr_meta_zones << 1; i++) { + zone = dmz_get(dmz, zid + i); + if (!dmz_is_rnd(zone)) + return -ENXIO; + set_bit(DMZ_META, &zone->flags); + } + + /* Load mapping table */ + ret = dmz_load_mapping(dmz); + if (ret) + goto out; + + /* + * Cache size boundaries: allow at least 2 super blocks, the chunk map + * blocks and enough blocks to be able to cache the bitmap blocks of + * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow + * the cache to add 512 more metadata blocks. + */ + dmz->min_nr_mblks = 2 + dmz->nr_map_blocks + + dmz->zone_nr_bitmap_blocks * 16; + dmz->max_nr_mblks = dmz->min_nr_mblks + 512; + dmz->mblk_shrinker.count_objects = dmz_mblock_shrinker_count; + dmz->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan; + dmz->mblk_shrinker.seeks = DEFAULT_SEEKS; + ret = register_shrinker(&dmz->mblk_shrinker); + if (ret) + goto out; + + dmz_dev_info(dmz, "Host-%s zoned block device\n", + bdev_zoned_model(dmz->zbd) == BLK_ZONED_HA ? + "aware" : "managed"); + dmz_dev_info(dmz, " %llu 512-byte logical sectors\n", + (u64)dmz->nr_zones + << dmz->zone_nr_sectors_shift); + dmz_dev_info(dmz, " %u zones of %llu 512-byte logical sectors\n", + dmz->nr_zones, + (u64)dmz->zone_nr_sectors); + dmz_dev_info(dmz, " %u metadata zones\n", + dmz->nr_meta_zones * 2); + dmz_dev_info(dmz, " %u data zones for %u chunks\n", + dmz->nr_data_zones, + dmz->nr_chunks); + dmz_dev_info(dmz, " %u random zones (%u unmapped)\n", + dmz->dz_nr_rnd, + atomic_read(&dmz->dz_unmap_nr_rnd)); + dmz_dev_info(dmz, " %u sequential zones (%u unmapped)\n", + dmz->dz_nr_seq, + atomic_read(&dmz->dz_unmap_nr_seq)); + dmz_dev_info(dmz, " %u reserved sequential data zones\n", + dmz->nr_reserved_seq); + + dmz_dev_debug(dmz, "Format:\n"); + dmz_dev_debug(dmz, "%u metadata blocks per set (%u max cache)\n", + dmz->nr_meta_blocks, + dmz->max_nr_mblks); + dmz_dev_debug(dmz, " %u data zone mapping blocks\n", + dmz->nr_map_blocks); + dmz_dev_debug(dmz, " %u bitmap blocks\n", + dmz->nr_bitmap_blocks); + +out: + if (ret) + dmz_cleanup_meta(dmz); + + return ret; +} + +/* + * Cleanup the target metadata resources. + */ +void dmz_cleanup_meta(struct dmz_target *dmz) +{ + struct rb_root *root = &dmz->mblk_rbtree; + struct dmz_mblock *mblk, *next; + int i; + + /* Release zone mapping resources */ + if (dmz->dz_map_mblk) { + for (i = 0; i < dmz->nr_map_blocks; i++) + dmz_release_mblock(dmz, dmz->dz_map_mblk[i]); + kfree(dmz->dz_map_mblk); + dmz->dz_map_mblk = NULL; + } + + /* Release super blocks */ + for (i = 0; i < 2; i++) { + if (dmz->sb[i].mblk) { + dmz_free_mblock(dmz, dmz->sb[i].mblk); + dmz->sb[i].mblk = NULL; + } + } + + /* Free cached blocks */ + while (!list_empty(&dmz->mblk_dirty_list)) { + mblk = list_first_entry(&dmz->mblk_dirty_list, + struct dmz_mblock, link); + dmz_dev_warn(dmz, "mblock %llu still in dirty list (ref %u)\n", + (u64)mblk->no, + atomic_read(&mblk->ref)); + list_del_init(&mblk->link); + rb_erase(&mblk->node, &dmz->mblk_rbtree); + dmz_free_mblock(dmz, mblk); + } + + while (!list_empty(&dmz->mblk_lru_list)) { + mblk = list_first_entry(&dmz->mblk_lru_list, + struct dmz_mblock, link); + list_del_init(&mblk->link); + rb_erase(&mblk->node, &dmz->mblk_rbtree); + dmz_free_mblock(dmz, mblk); + } + + /* Sanity checks: the mblock rbtree should now be empty */ + rbtree_postorder_for_each_entry_safe(mblk, next, root, node) { + dmz_dev_warn(dmz, "mblock %llu ref %u still in rbtree\n", + (u64)mblk->no, + atomic_read(&mblk->ref)); + atomic_set(&mblk->ref, 0); + dmz_free_mblock(dmz, mblk); + } + + /* Free the zone descriptors */ + dmz_drop_zones(dmz); +} + +/* + * Check metadata on resume. + */ +int dmz_resume_meta(struct dmz_target *dmz) +{ + return dmz_check_zones(dmz); +} + diff --git a/drivers/md/dm-zoned-reclaim.c b/drivers/md/dm-zoned-reclaim.c new file mode 100644 index 0000000..aa76692 --- /dev/null +++ b/drivers/md/dm-zoned-reclaim.c @@ -0,0 +1,535 @@ +/* + * Drive-managed zoned block device target + * Copyright (C) 2017 Western Digital Corporation or its affiliates. + * + * Written by: Damien Le Moal <damien.lemoal@xxxxxxx> + * + * This software is distributed under the terms of the GNU General PUBLIC + * License version 2, or any later version, "as is," without technical + * support, and WITHOUT ANY WARRANTY, without even the implied warranty + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. + */ + +#include <linux/module.h> + +#include "dm-zoned.h" + +/* + * Align a sequential zone write pointer to chunk_block. + */ +static int dmz_reclaim_align_wp(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block) +{ + sector_t wp_block = zone->wp_block; + unsigned int nr_blocks; + int ret; + + if (wp_block > chunk_block) + return -EIO; + + /* + * Zeroout the space between the write + * pointer and the requested position. + */ + nr_blocks = chunk_block - zone->wp_block; + if (!nr_blocks) + return 0; + + ret = blkdev_issue_zeroout(dmz->zbd, + dmz_start_sect(dmz, zone) + dmz_blk2sect(wp_block), + dmz_blk2sect(nr_blocks), + GFP_NOFS, false); + if (ret) { + dmz_dev_err(dmz, + "Align zone %u wp %llu to +%u blocks failed %d\n", + dmz_id(dmz, zone), + (unsigned long long)wp_block, + nr_blocks, + ret); + return ret; + } + + zone->wp_block += nr_blocks; + + return 0; +} + +/* + * dm_kcopyd_copy notification. + */ +static void dmz_reclaim_copy_end(int read_err, unsigned long write_err, + void *context) +{ + struct dmz_target *dmz = context; + + if (read_err || write_err) + dmz->reclaim_err = -EIO; + else + dmz->reclaim_err = 0; + + clear_bit_unlock(DMZ_RECLAIM_COPY, &dmz->flags); + smp_mb__after_atomic(); + wake_up_bit(&dmz->flags, DMZ_RECLAIM_COPY); +} + +/* + * Copy valid blocks of src_zone into dst_zone. + */ +static int dmz_reclaim_copy(struct dmz_target *dmz, + struct dm_zone *src_zone, struct dm_zone *dst_zone) +{ + struct dm_io_region src, dst; + sector_t block = 0, end_block; + sector_t nr_blocks; + sector_t src_zone_block; + sector_t dst_zone_block; + unsigned long flags = 0; + int ret; + + if (dmz_is_seq(src_zone)) + end_block = src_zone->wp_block; + else + end_block = dmz->zone_nr_blocks; + src_zone_block = dmz_start_block(dmz, src_zone); + dst_zone_block = dmz_start_block(dmz, dst_zone); + + if (dmz_is_seq(dst_zone)) + set_bit(DM_KCOPYD_WRITE_SEQ, &flags); + + while (block < end_block) { + + /* Get a valid region from the source zone */ + ret = dmz_first_valid_block(dmz, src_zone, &block); + if (ret < 0) + return ret; + + /* Are we done ? */ + nr_blocks = ret; + if (!nr_blocks) + return 0; + + /* + * If we are writing in a sequential zone, we must make sure + * that writes are sequential. So Zero out any eventual hole + * between writes. + */ + if (dmz_is_seq(dst_zone)) { + ret = dmz_reclaim_align_wp(dmz, dst_zone, block); + if (ret) + return ret; + } + + src.bdev = dmz->zbd; + src.sector = dmz_blk2sect(src_zone_block + block); + src.count = dmz_blk2sect(nr_blocks); + + dst.bdev = dmz->zbd; + dst.sector = dmz_blk2sect(dst_zone_block + block); + dst.count = src.count; + + dmz_dev_debug(dmz, + "Reclaim: Copy %s zone %u, block %llu+%llu to %s zone %u\n", + dmz_is_rnd(src_zone) ? "RND" : "SEQ", + dmz_id(dmz, src_zone), + (unsigned long long)block, + (unsigned long long)nr_blocks, + dmz_is_rnd(dst_zone) ? "RND" : "SEQ", + dmz_id(dmz, dst_zone)); + + /* Copy the valid region */ + set_bit(DMZ_RECLAIM_COPY, &dmz->flags); + ret = dm_kcopyd_copy(dmz->reclaim_kc, &src, 1, &dst, flags, + dmz_reclaim_copy_end, dmz); + if (ret != 0) + return ret; + + /* Wait for copy to complete */ + wait_on_bit_io(&dmz->flags, DMZ_RECLAIM_COPY, + TASK_UNINTERRUPTIBLE); + if (dmz->reclaim_err) + return dmz->reclaim_err; + + if (dmz_is_seq(dst_zone)) + dst_zone->wp_block += nr_blocks; + + block += nr_blocks; + + } + + return 0; +} + +/* + * Clear a zone reclaim flag. + */ +static inline void dmz_reclaim_put_zone(struct dmz_target *dmz, + struct dm_zone *zone) +{ + WARN_ON(dmz_is_active(zone)); + WARN_ON(!dmz_in_reclaim(zone)); + + clear_bit_unlock(DMZ_RECLAIM, &zone->flags); + smp_mb__after_atomic(); + wake_up_bit(&zone->flags, DMZ_RECLAIM); +} + +/* + * Move valid blocks of dzone buffer zone into dzone (after its write pointer) + * and free the buffer zone. + */ +static int dmz_reclaim_buf(struct dmz_target *dmz, struct dm_zone *dzone) +{ + struct dm_zone *bzone = dzone->bzone; + sector_t chunk_block = dzone->wp_block; + int ret; + + dmz_dev_debug(dmz, + "Chunk %u, move buf zone %u (weight %u) to data zone %u (weight %u)\n", + dzone->chunk, dmz_id(dmz, bzone), dmz_weight(bzone), + dmz_id(dmz, dzone), dmz_weight(dzone)); + + /* Flush data zone into the buffer zone */ + ret = dmz_reclaim_copy(dmz, bzone, dzone); + if (ret < 0) + return ret; + + down_read(&dmz->mblk_sem); + + /* Validate copied blocks */ + ret = dmz_valid_merge(dmz, bzone, dzone, chunk_block); + if (ret == 0) { + /* Free the buffer zone */ + dmz_invalidate_zone(dmz, bzone); + dmz_lock_map(dmz); + dmz_unmap_zone(dmz, bzone); + dmz_reclaim_put_zone(dmz, dzone); + dmz_free_zone(dmz, bzone); + dmz_unlock_map(dmz); + } + + up_read(&dmz->mblk_sem); + + return 0; +} + +/* + * Merge valid blocks of dzone into its buffer zone and free dzone. + */ +static int dmz_reclaim_seq_data(struct dmz_target *dmz, struct dm_zone *dzone) +{ + unsigned int chunk = dzone->chunk; + struct dm_zone *bzone = dzone->bzone; + int ret = 0; + + dmz_dev_debug(dmz, + "Chunk %u, move data zone %u (weight %u) to buf zone %u (weight %u)\n", + chunk, dmz_id(dmz, dzone), dmz_weight(dzone), + dmz_id(dmz, bzone), dmz_weight(bzone)); + + /* Flush data zone into the buffer zone */ + ret = dmz_reclaim_copy(dmz, dzone, bzone); + if (ret < 0) + return ret; + + down_read(&dmz->mblk_sem); + + /* Validate copied blocks */ + ret = dmz_valid_merge(dmz, dzone, bzone, 0); + if (ret == 0) { + /* + * Free the data zone and remap the chunk to + * the buffer zone. + */ + dmz_invalidate_zone(dmz, dzone); + dmz_lock_map(dmz); + dmz_unmap_zone(dmz, bzone); + dmz_unmap_zone(dmz, dzone); + dmz_reclaim_put_zone(dmz, dzone); + dmz_free_zone(dmz, dzone); + dmz_map_zone(dmz, bzone, chunk); + dmz_unlock_map(dmz); + } + + up_read(&dmz->mblk_sem); + + return 0; +} + +/* + * Move valid blocks of the random data zone dzone into a free sequential zone. + * Once blocks are moved, remap the zone chunk to the sequential zone. + */ +static int dmz_reclaim_rnd_data(struct dmz_target *dmz, struct dm_zone *dzone) +{ + unsigned int chunk = dzone->chunk; + struct dm_zone *szone = NULL; + int ret; + + /* Get a free sequential zone */ + dmz_lock_map(dmz); + szone = dmz_alloc_zone(dmz, DMZ_ALLOC_RECLAIM); + dmz_unlock_map(dmz); + if (!szone) + return -ENOSPC; + + dmz_dev_debug(dmz, + "Chunk %u, move rnd zone %u (weight %u) to seq zone %u\n", + chunk, dmz_id(dmz, dzone), dmz_weight(dzone), + dmz_id(dmz, szone)); + + /* Flush the random data zone into the sequential zone */ + ret = dmz_reclaim_copy(dmz, dzone, szone); + + down_read(&dmz->mblk_sem); + + if (ret == 0) + /* Validate copied blocks */ + ret = dmz_valid_copy(dmz, dzone, szone); + + if (ret) { + /* Free the sequential zone */ + dmz_lock_map(dmz); + dmz_free_zone(dmz, szone); + dmz_unlock_map(dmz); + } else { + /* Free the data zone and remap the chunk */ + dmz_invalidate_zone(dmz, dzone); + dmz_lock_map(dmz); + dmz_unmap_zone(dmz, dzone); + dmz_reclaim_put_zone(dmz, dzone); + dmz_free_zone(dmz, dzone); + dmz_map_zone(dmz, szone, chunk); + dmz_unlock_map(dmz); + } + + up_read(&dmz->mblk_sem); + + return 0; +} + +/* + * Reclaim an empty zone. + */ +static void dmz_reclaim_empty(struct dmz_target *dmz, struct dm_zone *dzone) +{ + down_read(&dmz->mblk_sem); + dmz_lock_map(dmz); + dmz_unmap_zone(dmz, dzone); + dmz_reclaim_put_zone(dmz, dzone); + dmz_free_zone(dmz, dzone); + dmz_unlock_map(dmz); + up_read(&dmz->mblk_sem); +} + +/* + * Lock a zone for reclaim. Returns 0 if the zone cannot be locked or if it is + * already locked and 1 otherwise. + */ +static inline int dmz_reclaim_lock_zone(struct dmz_target *dmz, + struct dm_zone *zone) +{ + /* Active zones cannot be reclaimed */ + if (dmz_is_active(zone)) + return 0; + + return !test_and_set_bit(DMZ_RECLAIM, &zone->flags); +} + +/* + * Select a random zone for reclaim. + */ +static struct dm_zone *dmz_reclaim_get_rnd_zone(struct dmz_target *dmz) +{ + struct dm_zone *dzone = NULL; + struct dm_zone *zone; + + if (list_empty(&dmz->dz_map_rnd_list)) + return NULL; + + list_for_each_entry(zone, &dmz->dz_map_rnd_list, link) { + if (dmz_is_buf(zone)) + dzone = zone->bzone; + else + dzone = zone; + if (dmz_reclaim_lock_zone(dmz, dzone)) + return dzone; + } + + return NULL; +} + +/* + * Select a buffered sequential zone for reclaim. + */ +static struct dm_zone *dmz_reclaim_get_seq_zone(struct dmz_target *dmz) +{ + struct dm_zone *zone; + + if (list_empty(&dmz->dz_map_seq_list)) + return NULL; + + list_for_each_entry(zone, &dmz->dz_map_seq_list, link) { + if (!zone->bzone) + continue; + if (dmz_reclaim_lock_zone(dmz, zone)) + return zone; + } + + return NULL; +} + +/* + * Select a zone for reclaim. + */ +static struct dm_zone *dmz_reclaim_get_zone(struct dmz_target *dmz) +{ + struct dm_zone *zone = NULL; + + /* + * Search for a zone candidate to reclaim: 2 cases are possible. + * (1) There is no free sequential zones. Then a random data zone + * cannot be reclaimed. So choose a sequential zone to reclaim so + * that afterward a random zone can be reclaimed. + * (2) At least one free sequential zone is available, then choose + * the oldest random zone (data or buffer) that can be locked. + */ + dmz_lock_map(dmz); + if (list_empty(&dmz->reclaim_seq_zones_list)) + zone = dmz_reclaim_get_seq_zone(dmz); + else + zone = dmz_reclaim_get_rnd_zone(dmz); + dmz_unlock_map(dmz); + + return zone; +} + +/* + * Find a reclaim candidate zone and reclaim it. + */ +static void dmz_reclaim(struct dmz_target *dmz) +{ + struct dm_zone *dzone; + struct dm_zone *rzone; + unsigned long start; + int ret; + + /* Get a data zone */ + dzone = dmz_reclaim_get_zone(dmz); + if (!dzone) + return; + + start = jiffies; + + if (dmz_is_rnd(dzone)) { + + rzone = dzone; + if (!dmz_weight(dzone)) { + /* Empty zone */ + dmz_reclaim_empty(dmz, dzone); + ret = 0; + } else { + /* + * Reclaim the random data zone by moving its + * valid data blocks to a free sequential zone. + */ + ret = dmz_reclaim_rnd_data(dmz, dzone); + } + + } else { + + struct dm_zone *bzone = dzone->bzone; + sector_t chunk_block = 0; + + ret = dmz_first_valid_block(dmz, bzone, &chunk_block); + if (ret < 0) + goto out; + + if (chunk_block >= dzone->wp_block) { + /* + * Valid blocks in the buffer zone are after + * the data zone write pointer: copy them there. + */ + ret = dmz_reclaim_buf(dmz, dzone); + rzone = bzone; + } else { + /* + * Reclaim the data zone by merging it into the + * buffer zone so that the buffer zone itself can + * be later reclaimed. + */ + ret = dmz_reclaim_seq_data(dmz, dzone); + rzone = dzone; + } + + } + +out: + if (ret) { + dmz_reclaim_put_zone(dmz, dzone); + return; + } + + dmz_dev_debug(dmz, "Reclaimed zone %u in %u ms\n", + dmz_id(dmz, rzone), jiffies_to_msecs(jiffies - start)); + + dmz_trigger_flush(dmz); +} + +/* + * Zone reclaim work. + */ +void dmz_reclaim_work(struct work_struct *work) +{ + struct dmz_target *dmz = + container_of(work, struct dmz_target, reclaim_work.work); + unsigned long next_reclaim = DMZ_RECLAIM_PERIOD; + unsigned int unmap_nr_rnd = atomic_read(&dmz->dz_unmap_nr_rnd); + unsigned int throttle, unmap_perc; + + /* If there are still plenty of random zones, do not reclaim */ + unmap_perc = unmap_nr_rnd * 100 / dmz->dz_nr_rnd; + if (unmap_perc >= DMZ_RECLAIM_HIGH_FREE_RND) + goto out; + + /* + * If we are not idle and still have unmapped random zones, + * do not reclaim. + */ + if (!dmz_idle(dmz) && unmap_perc > DMZ_RECLAIM_LOW_FREE_RND) + goto out; + + /* + * We need to start reclaiming random zones: set up zone copy + * throttling to either go fast if we are very low on random zones + * and slower if there are still some free random zones to avoid + * as much as possible to negatively impact the user workload. + */ + if (dmz_idle(dmz) || + unmap_nr_rnd < atomic_read(&dmz->nr_active_chunks)) + /* Idle or very low: go fast */ + throttle = 100; + else + /* Busy but we still have some random zone: go slower */ + throttle = min(75U, 100U - unmap_perc / 2); + dmz->reclaim_throttle.throttle = throttle; + + dmz_dev_debug(dmz, + "Reclaim (%u): %s (%u BIOs, %u active chunks), %u%% free rnd zones (%u/%u)\n", + dmz->reclaim_throttle.throttle, + (dmz_idle(dmz) ? "Idle" : "Busy"), + atomic_read(&dmz->bio_count), + atomic_read(&dmz->nr_active_chunks), + unmap_perc, + unmap_nr_rnd, dmz->dz_nr_rnd); + + dmz_reclaim(dmz); + + if ((dmz_should_reclaim(dmz) + && atomic_read(&dmz->nr_reclaim_seq_zones))) + /* Run again immmediately */ + next_reclaim = 0; + +out: + dmz_schedule_reclaim(dmz, next_reclaim); +} + diff --git a/drivers/md/dm-zoned.h b/drivers/md/dm-zoned.h new file mode 100644 index 0000000..bdcd538 --- /dev/null +++ b/drivers/md/dm-zoned.h @@ -0,0 +1,530 @@ +/* + * Drive-managed zoned block device target + * Copyright (C) 2017 Western Digital Corporation or its affiliates. + * + * This software is distributed under the terms of the GNU General Public + * License version 2, or any later version, "as is," without technical + * support, and WITHOUT ANY WARRANTY, without even the implied warranty + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. + */ +#include <linux/types.h> +#include <linux/blkdev.h> +#include <linux/device-mapper.h> +#include <linux/dm-kcopyd.h> +#include <linux/list.h> +#include <linux/spinlock.h> +#include <linux/mutex.h> +#include <linux/workqueue.h> +#include <linux/rwsem.h> +#include <linux/rbtree.h> +#include <linux/radix-tree.h> +#include <linux/shrinker.h> + +#ifndef __DM_ZONED_H__ +#define __DM_ZONED_H__ + +/* + * Metadata version. + */ +#define DMZ_META_VER 1 + +/* + * On-disk super block magic. + */ +#define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \ + (((unsigned int)('Z')) << 16) | \ + (((unsigned int)('B')) << 8) | \ + ((unsigned int)('D'))) + +/* + * On disk super block. + * This uses only 512 B but uses on disk a full 4KB block. This block is + * followed on disk by the mapping table of chunks to zones and the bitmap + * blocks indicating zone block validity. + * The overall resulting metadata format is: + * (1) Super block (1 block) + * (2) Chunk mapping table (nr_map_blocks) + * (3) Bitmap blocks (nr_bitmap_blocks) + * All metadata blocks are stored in conventional zones, starting from the + * the first conventional zone found on disk. + */ +struct dmz_super { + + /* Magic number */ + __le32 magic; /* 4 */ + + /* Metadata version number */ + __le32 version; /* 8 */ + + /* Generation number */ + __le64 gen; /* 16 */ + + /* This block number */ + __le64 sb_block; /* 24 */ + + /* The number of metadata blocks, including this super block */ + __le32 nr_meta_blocks; /* 28 */ + + /* The number of sequential zones reserved for reclaim */ + __le32 nr_reserved_seq; /* 32 */ + + /* The number of entries in the mapping table */ + __le32 nr_chunks; /* 36 */ + + /* The number of blocks used for the chunk mapping table */ + __le32 nr_map_blocks; /* 40 */ + + /* The number of blocks used for the block bitmaps */ + __le32 nr_bitmap_blocks; /* 44 */ + + /* Checksum */ + __le32 crc; /* 48 */ + + /* Padding to full 512B sector */ + u8 reserved[464]; /* 512 */ + +}; + +/* + * Chunk mapping entry: entries are indexed by chunk number + * and give the zone ID (dzone_id) mapping the chunk on disk. + * This zone may be sequential or random. If it is a sequential + * zone, a second zone (bzone_id) used as a write buffer may + * also be specified. This second zone will always be a randomly + * writeable zone. + */ +struct dmz_map { + __le32 dzone_id; + __le32 bzone_id; +}; + +/* + * dm-zoned creates block devices with 4KB blocks, always. + */ +#define DMZ_BLOCK_SHIFT 12 +#define DMZ_BLOCK_SIZE (1 << DMZ_BLOCK_SHIFT) +#define DMZ_BLOCK_MASK (DMZ_BLOCK_SIZE - 1) + +#define DMZ_BLOCK_SHIFT_BITS (DMZ_BLOCK_SHIFT + 3) +#define DMZ_BLOCK_SIZE_BITS (1 << DMZ_BLOCK_SHIFT_BITS) +#define DMZ_BLOCK_MASK_BITS (DMZ_BLOCK_SIZE_BITS - 1) + +#define DMZ_BLOCK_SECTORS_SHIFT (DMZ_BLOCK_SHIFT - SECTOR_SHIFT) +#define DMZ_BLOCK_SECTORS (DMZ_BLOCK_SIZE >> SECTOR_SHIFT) +#define DMZ_BLOCK_SECTORS_MASK (DMZ_BLOCK_SECTORS - 1) + +/* + * Chunk mapping table metadata: 512 8-bytes entries per 4KB block. + */ +#define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map)) +#define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES)) +#define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1) +#define DMZ_MAP_UNMAPPED UINT_MAX + +/* + * Block <-> 512B sector conversion. + */ +#define dmz_blk2sect(b) ((b) << DMZ_BLOCK_SECTORS_SHIFT) +#define dmz_sect2blk(s) ((s) >> DMZ_BLOCK_SECTORS_SHIFT) + +#define DMZ_MIN_BIOS 8192 + +/* + * The size of a zone report in number of zones. + * This results in 4096*64B=256KB report zones commands. + */ +#define DMZ_REPORT_NR_ZONES 4096 + +/* + * Zone flags. + */ +enum { + + /* Zone write type */ + DMZ_RND, + DMZ_SEQ, + + /* Zone critical condition */ + DMZ_OFFLINE, + DMZ_READ_ONLY, + + /* How the zone is being used */ + DMZ_META, + DMZ_DATA, + DMZ_BUF, + + /* Zone internal state */ + DMZ_ACTIVE, + DMZ_RECLAIM, + DMZ_SEQ_WRITE_ERR, + +}; + +/* + * Zone descriptor. + */ +struct dm_zone { + + /* For listing the zone depending on its state */ + struct list_head link; + + /* Zone type and state */ + unsigned long flags; + + /* Zone activation reference count */ + atomic_t refcount; + + /* Zone write pointer block (relative to the zone start block) */ + unsigned int wp_block; + + /* Zone weight (number of valid blocks in the zone) */ + unsigned int weight; + + /* The chunk that the zone maps */ + unsigned int chunk; + + /* + * For a sequential data zone, pointer to the random zone + * used as a buffer for processing unaligned writes. + * For a buffer zone, this points back to the data zone. + */ + struct dm_zone *bzone; + +}; + +/* + * Meta data block descriptor (for cached metadata blocks). + */ +struct dmz_mblock { + + struct rb_node node; + struct list_head link; + sector_t no; + atomic_t ref; + unsigned long state; + struct page *page; + void *data; + +}; + +/* + * Super block information (one per metadata set). + */ +struct dmz_sb { + sector_t block; + struct dmz_mblock *mblk; + struct dmz_super *sb; +}; + +/* + * Metadata block state flags. + */ +enum { + DMZ_META_DIRTY, + DMZ_META_READING, + DMZ_META_WRITING, + DMZ_META_ERROR, +}; + +/* + * Target flags. + */ +enum { + DMZ_RECLAIM_COPY, + DMZ_SUSPENDED, +}; + +/* + * Target descriptor. + */ +struct dmz_target { + + struct dm_dev *ddev; + + /* Zoned block device information */ + char zbd_name[BDEVNAME_SIZE]; + struct block_device *zbd; + sector_t zbd_capacity; + struct request_queue *zbdq; + unsigned long flags; + + unsigned int nr_zones; + unsigned int nr_useable_zones; + unsigned int nr_meta_blocks; + unsigned int nr_meta_zones; + unsigned int nr_data_zones; + unsigned int nr_rnd_zones; + unsigned int nr_reserved_seq; + unsigned int nr_chunks; + + sector_t zone_nr_sectors; + unsigned int zone_nr_sectors_shift; + + sector_t zone_nr_blocks; + sector_t zone_nr_blocks_shift; + + sector_t zone_bitmap_size; + unsigned int zone_nr_bitmap_blocks; + + unsigned int nr_bitmap_blocks; + unsigned int nr_map_blocks; + + /* Zone information array */ + struct dm_zone *zones; + + /* For metadata handling */ + struct dm_zone *sb_zone; + struct dmz_sb sb[2]; + unsigned int mblk_primary; + u64 sb_gen; + unsigned int min_nr_mblks; + unsigned int max_nr_mblks; + atomic_t nr_mblks; + struct rw_semaphore mblk_sem; + spinlock_t mblk_lock; + struct rb_root mblk_rbtree; + struct list_head mblk_lru_list; + struct list_head mblk_dirty_list; + struct shrinker mblk_shrinker; + + /* Zone allocation management */ + struct mutex map_lock; + struct dmz_mblock **dz_map_mblk; + unsigned int dz_nr_rnd; + atomic_t dz_unmap_nr_rnd; + struct list_head dz_unmap_rnd_list; + struct list_head dz_map_rnd_list; + + unsigned int dz_nr_seq; + atomic_t dz_unmap_nr_seq; + struct list_head dz_unmap_seq_list; + struct list_head dz_map_seq_list; + + wait_queue_head_t dz_free_wq; + + /* For chunk work */ + struct mutex chunk_lock; + struct radix_tree_root chunk_rxtree; + struct workqueue_struct *chunk_wq; + atomic_t nr_active_chunks; + + /* For chunk BIOs to zones */ + struct bio_set *bio_set; + atomic_t bio_count; + unsigned long atime; + + /* For flush */ + spinlock_t flush_lock; + struct bio_list flush_list; + struct delayed_work flush_work; + struct workqueue_struct *flush_wq; + + /* For reclaim */ + struct delayed_work reclaim_work; + struct workqueue_struct *reclaim_wq; + atomic_t nr_reclaim_seq_zones; + struct list_head reclaim_seq_zones_list; + struct dm_kcopyd_client *reclaim_kc; + struct dm_kcopyd_throttle reclaim_throttle; + int reclaim_err; + +}; + +/* + * Chunk work descriptor. + */ +struct dm_chunk_work { + struct work_struct work; + atomic_t refcount; + struct dmz_target *target; + unsigned int chunk; + struct bio_list bio_list; +}; + +#define dmz_id(dmz, z) ((unsigned int)((z) - (dmz)->zones)) +#define dmz_get(dmz, z) (&(dmz)->zones[z]) +#define dmz_start_sect(dmz, z) (dmz_id(dmz, z) << (dmz)->zone_nr_sectors_shift) +#define dmz_start_block(dmz, z) (dmz_id(dmz, z) << (dmz)->zone_nr_blocks_shift) +#define dmz_is_rnd(z) test_bit(DMZ_RND, &(z)->flags) +#define dmz_is_seq(z) test_bit(DMZ_SEQ, &(z)->flags) +#define dmz_is_empty(z) ((z)->wp_block == 0) +#define dmz_is_offline(z) test_bit(DMZ_OFFLINE, &(z)->flags) +#define dmz_is_readonly(z) test_bit(DMZ_READ_ONLY, &(z)->flags) +#define dmz_is_active(z) test_bit(DMZ_ACTIVE, &(z)->flags) +#define dmz_in_reclaim(z) test_bit(DMZ_RECLAIM, &(z)->flags) +#define dmz_seq_write_err(z) test_bit(DMZ_SEQ_WRITE_ERR, &(z)->flags) + +#define dmz_is_meta(z) test_bit(DMZ_META, &(z)->flags) +#define dmz_is_buf(z) test_bit(DMZ_BUF, &(z)->flags) +#define dmz_is_data(z) test_bit(DMZ_DATA, &(z)->flags) + +#define dmz_weight(z) ((z)->weight) + +#define dmz_chunk_sector(dmz, s) ((s) & ((dmz)->zone_nr_sectors - 1)) +#define dmz_chunk_block(dmz, b) ((b) & ((dmz)->zone_nr_blocks - 1)) + +#define dmz_bio_block(bio) dmz_sect2blk((bio)->bi_iter.bi_sector) +#define dmz_bio_blocks(bio) dmz_sect2blk(bio_sectors(bio)) +#define dmz_bio_chunk(dmz, bio) ((bio)->bi_iter.bi_sector >> \ + (dmz)->zone_nr_sectors_shift) + +#define dmz_lock_map(dmz) mutex_lock(&(dmz)->map_lock) +#define dmz_unlock_map(dmz) mutex_unlock(&(dmz)->map_lock) + +/* + * Flush intervals (seconds). + */ +#define DMZ_FLUSH_PERIOD (10 * HZ) + +/* + * Trigger flush. + */ +static inline void dmz_trigger_flush(struct dmz_target *dmz) +{ + mod_delayed_work(dmz->flush_wq, &dmz->flush_work, 0); +} + +/* + * Number of seconds without BIO to consider the target device idle. + */ +#define DMZ_IDLE_PERIOD (10UL * HZ) + +/* + * Zone reclaim check period. + */ +#define DMZ_RECLAIM_PERIOD (HZ) + +/* + * Percentage of unmapped (free) random zones below which reclaim starts + * even if the device is not idle. + */ +#define DMZ_RECLAIM_LOW_FREE_RND 50 + +/* + * Percentage of unmapped (free) random zones above which reclaim stops + * * even if the device is idle. + */ +#define DMZ_RECLAIM_HIGH_FREE_RND 75 + +/* + * Test if the target device is idle. + */ +static inline int dmz_idle(struct dmz_target *dmz) +{ + return atomic_read(&(dmz)->bio_count) == 0 && + time_is_before_jiffies(dmz->atime + DMZ_IDLE_PERIOD); +} + +/* + * Test if triggerring reclaim is necessary. + */ +static inline bool dmz_should_reclaim(struct dmz_target *dmz) +{ + unsigned int unmap_rnd = atomic_read(&dmz->dz_unmap_nr_rnd); + + if (dmz_idle(dmz) && unmap_rnd < dmz->dz_nr_rnd) + return true; + + /* Percentage of unmappped random zones low ? */ + return ((unmap_rnd * 100) / dmz->dz_nr_rnd) <= DMZ_RECLAIM_LOW_FREE_RND; +} + +/* + * Schedule reclaim (delay in jiffies). + */ +static inline void dmz_schedule_reclaim(struct dmz_target *dmz, + unsigned long delay) +{ + mod_delayed_work(dmz->reclaim_wq, &dmz->reclaim_work, delay); +} + +/* + * Trigger reclaim. + */ +static inline void dmz_trigger_reclaim(struct dmz_target *dmz) +{ + dmz_schedule_reclaim(dmz, 0); +} + +extern void dmz_reclaim_work(struct work_struct *work); + +/* + * Zone BIO context. + */ +struct dmz_bioctx { + struct dmz_target *target; + struct dm_zone *zone; + struct bio *bio; + atomic_t ref; + int error; +}; + +#define dmz_info(format, args...) \ + pr_info("dm-zoned: " format, \ + ## args) + +#define dmz_dev_info(target, format, args...) \ + pr_info("dm-zoned (%s): " format, \ + (dmz)->zbd_name, ## args) + +#define dmz_dev_err(dmz, format, args...) \ + pr_err("dm-zoned (%s): " format, \ + (dmz)->zbd_name, ## args) + +#define dmz_dev_warn(dmz, format, args...) \ + pr_warn("dm-zoned (%s): " format, \ + (dmz)->zbd_name, ## args) + +#define dmz_dev_debug(dmz, format, args...) \ + pr_debug("dm-zoned (%s): " format, \ + (dmz)->zbd_name, ## args) + +extern int dmz_init_meta(struct dmz_target *dmz); +extern int dmz_resume_meta(struct dmz_target *dmz); +extern void dmz_cleanup_meta(struct dmz_target *dmz); + +extern int dmz_flush_mblocks(struct dmz_target *dmz); + +#define DMZ_ALLOC_RND 0x01 +#define DMZ_ALLOC_RECLAIM 0x02 + +struct dm_zone *dmz_alloc_zone(struct dmz_target *dmz, unsigned long flags); +extern void dmz_free_zone(struct dmz_target *dmz, struct dm_zone *zone); + +extern void dmz_map_zone(struct dmz_target *dmz, struct dm_zone *zone, + unsigned int chunk); +extern void dmz_unmap_zone(struct dmz_target *dmz, struct dm_zone *zone); + +extern void dmz_activate_zone(struct dmz_target *dmz, struct dm_zone *zone); +extern void dmz_deactivate_zone(struct dmz_target *dmz, struct dm_zone *zone); + +extern struct dm_zone *dmz_get_chunk_mapping(struct dmz_target *dmz, + unsigned int chunk, int op); +extern void dmz_put_chunk_mapping(struct dmz_target *dmz, + struct dm_zone *zone); + +extern struct dm_zone *dmz_get_chunk_buffer(struct dmz_target *dmz, + struct dm_zone *dzone); + +extern int dmz_valid_copy(struct dmz_target *dmz, struct dm_zone *from_zone, + struct dm_zone *to_zone); +extern int dmz_valid_merge(struct dmz_target *dmz, struct dm_zone *from_zone, + struct dm_zone *to_zone, sector_t chunk_block); + +extern int dmz_validate_blocks(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block, unsigned int nr_blocks); +extern int dmz_invalidate_blocks(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block, unsigned int nr_blocks); +static inline int dmz_invalidate_zone(struct dmz_target *dmz, + struct dm_zone *zone) +{ + return dmz_invalidate_blocks(dmz, zone, 0, dmz->zone_nr_blocks); +} + +extern int dmz_block_valid(struct dmz_target *dmz, struct dm_zone *zone, + sector_t chunk_block); + +extern int dmz_first_valid_block(struct dmz_target *dmz, struct dm_zone *zone, + sector_t *chunk_block); + +#endif /* __DM_ZONED_H__ */ -- 2.9.3 -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
