1bae9a0aaSDmitry Fomichev // SPDX-License-Identifier: GPL-2.0-only 23b1a94c8SDamien Le Moal /* 33b1a94c8SDamien Le Moal * Copyright (C) 2017 Western Digital Corporation or its affiliates. 43b1a94c8SDamien Le Moal * 53b1a94c8SDamien Le Moal * This file is released under the GPL. 63b1a94c8SDamien Le Moal */ 73b1a94c8SDamien Le Moal 83b1a94c8SDamien Le Moal #include "dm-zoned.h" 93b1a94c8SDamien Le Moal 103b1a94c8SDamien Le Moal #include <linux/module.h> 113b1a94c8SDamien Le Moal #include <linux/crc32.h> 12bd976e52SDamien Le Moal #include <linux/sched/mm.h> 133b1a94c8SDamien Le Moal 143b1a94c8SDamien Le Moal #define DM_MSG_PREFIX "zoned metadata" 153b1a94c8SDamien Le Moal 163b1a94c8SDamien Le Moal /* 173b1a94c8SDamien Le Moal * Metadata version. 183b1a94c8SDamien Le Moal */ 19bd5c4031SHannes Reinecke #define DMZ_META_VER 2 203b1a94c8SDamien Le Moal 213b1a94c8SDamien Le Moal /* 223b1a94c8SDamien Le Moal * On-disk super block magic. 233b1a94c8SDamien Le Moal */ 243b1a94c8SDamien Le Moal #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \ 253b1a94c8SDamien Le Moal (((unsigned int)('Z')) << 16) | \ 263b1a94c8SDamien Le Moal (((unsigned int)('B')) << 8) | \ 273b1a94c8SDamien Le Moal ((unsigned int)('D'))) 283b1a94c8SDamien Le Moal 293b1a94c8SDamien Le Moal /* 303b1a94c8SDamien Le Moal * On disk super block. 313b1a94c8SDamien Le Moal * This uses only 512 B but uses on disk a full 4KB block. This block is 323b1a94c8SDamien Le Moal * followed on disk by the mapping table of chunks to zones and the bitmap 333b1a94c8SDamien Le Moal * blocks indicating zone block validity. 343b1a94c8SDamien Le Moal * The overall resulting metadata format is: 353b1a94c8SDamien Le Moal * (1) Super block (1 block) 363b1a94c8SDamien Le Moal * (2) Chunk mapping table (nr_map_blocks) 373b1a94c8SDamien Le Moal * (3) Bitmap blocks (nr_bitmap_blocks) 38ad1bd578SDmitry Fomichev * All metadata blocks are stored in conventional zones, starting from 393b1a94c8SDamien Le Moal * the first conventional zone found on disk. 403b1a94c8SDamien Le Moal */ 413b1a94c8SDamien Le Moal struct dmz_super { 423b1a94c8SDamien Le Moal /* Magic number */ 433b1a94c8SDamien Le Moal __le32 magic; /* 4 */ 443b1a94c8SDamien Le Moal 453b1a94c8SDamien Le Moal /* Metadata version number */ 463b1a94c8SDamien Le Moal __le32 version; /* 8 */ 473b1a94c8SDamien Le Moal 483b1a94c8SDamien Le Moal /* Generation number */ 493b1a94c8SDamien Le Moal __le64 gen; /* 16 */ 503b1a94c8SDamien Le Moal 513b1a94c8SDamien Le Moal /* This block number */ 523b1a94c8SDamien Le Moal __le64 sb_block; /* 24 */ 533b1a94c8SDamien Le Moal 543b1a94c8SDamien Le Moal /* The number of metadata blocks, including this super block */ 553b1a94c8SDamien Le Moal __le32 nr_meta_blocks; /* 28 */ 563b1a94c8SDamien Le Moal 573b1a94c8SDamien Le Moal /* The number of sequential zones reserved for reclaim */ 583b1a94c8SDamien Le Moal __le32 nr_reserved_seq; /* 32 */ 593b1a94c8SDamien Le Moal 603b1a94c8SDamien Le Moal /* The number of entries in the mapping table */ 613b1a94c8SDamien Le Moal __le32 nr_chunks; /* 36 */ 623b1a94c8SDamien Le Moal 633b1a94c8SDamien Le Moal /* The number of blocks used for the chunk mapping table */ 643b1a94c8SDamien Le Moal __le32 nr_map_blocks; /* 40 */ 653b1a94c8SDamien Le Moal 663b1a94c8SDamien Le Moal /* The number of blocks used for the block bitmaps */ 673b1a94c8SDamien Le Moal __le32 nr_bitmap_blocks; /* 44 */ 683b1a94c8SDamien Le Moal 693b1a94c8SDamien Le Moal /* Checksum */ 703b1a94c8SDamien Le Moal __le32 crc; /* 48 */ 713b1a94c8SDamien Le Moal 72bd5c4031SHannes Reinecke /* DM-Zoned label */ 73bd5c4031SHannes Reinecke u8 dmz_label[32]; /* 80 */ 74bd5c4031SHannes Reinecke 75bd5c4031SHannes Reinecke /* DM-Zoned UUID */ 76bd5c4031SHannes Reinecke u8 dmz_uuid[16]; /* 96 */ 77bd5c4031SHannes Reinecke 78bd5c4031SHannes Reinecke /* Device UUID */ 79bd5c4031SHannes Reinecke u8 dev_uuid[16]; /* 112 */ 80bd5c4031SHannes Reinecke 813b1a94c8SDamien Le Moal /* Padding to full 512B sector */ 82bd5c4031SHannes Reinecke u8 reserved[400]; /* 512 */ 833b1a94c8SDamien Le Moal }; 843b1a94c8SDamien Le Moal 853b1a94c8SDamien Le Moal /* 863b1a94c8SDamien Le Moal * Chunk mapping entry: entries are indexed by chunk number 873b1a94c8SDamien Le Moal * and give the zone ID (dzone_id) mapping the chunk on disk. 883b1a94c8SDamien Le Moal * This zone may be sequential or random. If it is a sequential 893b1a94c8SDamien Le Moal * zone, a second zone (bzone_id) used as a write buffer may 903b1a94c8SDamien Le Moal * also be specified. This second zone will always be a randomly 913b1a94c8SDamien Le Moal * writeable zone. 923b1a94c8SDamien Le Moal */ 933b1a94c8SDamien Le Moal struct dmz_map { 943b1a94c8SDamien Le Moal __le32 dzone_id; 953b1a94c8SDamien Le Moal __le32 bzone_id; 963b1a94c8SDamien Le Moal }; 973b1a94c8SDamien Le Moal 983b1a94c8SDamien Le Moal /* 993b1a94c8SDamien Le Moal * Chunk mapping table metadata: 512 8-bytes entries per 4KB block. 1003b1a94c8SDamien Le Moal */ 1013b1a94c8SDamien Le Moal #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map)) 1023b1a94c8SDamien Le Moal #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES)) 1033b1a94c8SDamien Le Moal #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1) 1043b1a94c8SDamien Le Moal #define DMZ_MAP_UNMAPPED UINT_MAX 1053b1a94c8SDamien Le Moal 1063b1a94c8SDamien Le Moal /* 1073b1a94c8SDamien Le Moal * Meta data block descriptor (for cached metadata blocks). 1083b1a94c8SDamien Le Moal */ 1093b1a94c8SDamien Le Moal struct dmz_mblock { 1103b1a94c8SDamien Le Moal struct rb_node node; 1113b1a94c8SDamien Le Moal struct list_head link; 1123b1a94c8SDamien Le Moal sector_t no; 11333c2865fSDamien Le Moal unsigned int ref; 1143b1a94c8SDamien Le Moal unsigned long state; 1153b1a94c8SDamien Le Moal struct page *page; 1163b1a94c8SDamien Le Moal void *data; 1173b1a94c8SDamien Le Moal }; 1183b1a94c8SDamien Le Moal 1193b1a94c8SDamien Le Moal /* 1203b1a94c8SDamien Le Moal * Metadata block state flags. 1213b1a94c8SDamien Le Moal */ 1223b1a94c8SDamien Le Moal enum { 1233b1a94c8SDamien Le Moal DMZ_META_DIRTY, 1243b1a94c8SDamien Le Moal DMZ_META_READING, 1253b1a94c8SDamien Le Moal DMZ_META_WRITING, 1263b1a94c8SDamien Le Moal DMZ_META_ERROR, 1273b1a94c8SDamien Le Moal }; 1283b1a94c8SDamien Le Moal 1293b1a94c8SDamien Le Moal /* 1303b1a94c8SDamien Le Moal * Super block information (one per metadata set). 1313b1a94c8SDamien Le Moal */ 1323b1a94c8SDamien Le Moal struct dmz_sb { 1333b1a94c8SDamien Le Moal sector_t block; 134bf28a3baSHannes Reinecke struct dmz_dev *dev; 1353b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 1363b1a94c8SDamien Le Moal struct dmz_super *sb; 137735bd7e4SHannes Reinecke struct dm_zone *zone; 1383b1a94c8SDamien Le Moal }; 1393b1a94c8SDamien Le Moal 1403b1a94c8SDamien Le Moal /* 1413b1a94c8SDamien Le Moal * In-memory metadata. 1423b1a94c8SDamien Le Moal */ 1433b1a94c8SDamien Le Moal struct dmz_metadata { 1443b1a94c8SDamien Le Moal struct dmz_dev *dev; 145bd5c4031SHannes Reinecke unsigned int nr_devs; 1463b1a94c8SDamien Le Moal 1472234e732SHannes Reinecke char devname[BDEVNAME_SIZE]; 148bd5c4031SHannes Reinecke char label[BDEVNAME_SIZE]; 149bd5c4031SHannes Reinecke uuid_t uuid; 1502234e732SHannes Reinecke 1513b1a94c8SDamien Le Moal sector_t zone_bitmap_size; 1523b1a94c8SDamien Le Moal unsigned int zone_nr_bitmap_blocks; 153b3996295SDmitry Fomichev unsigned int zone_bits_per_mblk; 1543b1a94c8SDamien Le Moal 15536820560SHannes Reinecke sector_t zone_nr_blocks; 15636820560SHannes Reinecke sector_t zone_nr_blocks_shift; 15736820560SHannes Reinecke 15836820560SHannes Reinecke sector_t zone_nr_sectors; 15936820560SHannes Reinecke sector_t zone_nr_sectors_shift; 16036820560SHannes Reinecke 1613b1a94c8SDamien Le Moal unsigned int nr_bitmap_blocks; 1623b1a94c8SDamien Le Moal unsigned int nr_map_blocks; 1633b1a94c8SDamien Le Moal 16436820560SHannes Reinecke unsigned int nr_zones; 1653b1a94c8SDamien Le Moal unsigned int nr_useable_zones; 1663b1a94c8SDamien Le Moal unsigned int nr_meta_blocks; 1673b1a94c8SDamien Le Moal unsigned int nr_meta_zones; 1683b1a94c8SDamien Le Moal unsigned int nr_data_zones; 16934f5affdSHannes Reinecke unsigned int nr_cache_zones; 1703b1a94c8SDamien Le Moal unsigned int nr_rnd_zones; 1713b1a94c8SDamien Le Moal unsigned int nr_reserved_seq; 1723b1a94c8SDamien Le Moal unsigned int nr_chunks; 1733b1a94c8SDamien Le Moal 1743b1a94c8SDamien Le Moal /* Zone information array */ 175a92fbc44SHannes Reinecke struct xarray zones; 1763b1a94c8SDamien Le Moal 1775d2c74f3SHannes Reinecke struct dmz_sb sb[2]; 1783b1a94c8SDamien Le Moal unsigned int mblk_primary; 179bd5c4031SHannes Reinecke unsigned int sb_version; 1803b1a94c8SDamien Le Moal u64 sb_gen; 1813b1a94c8SDamien Le Moal unsigned int min_nr_mblks; 1823b1a94c8SDamien Le Moal unsigned int max_nr_mblks; 1833b1a94c8SDamien Le Moal atomic_t nr_mblks; 1843b1a94c8SDamien Le Moal struct rw_semaphore mblk_sem; 1853b1a94c8SDamien Le Moal struct mutex mblk_flush_lock; 1863b1a94c8SDamien Le Moal spinlock_t mblk_lock; 1873b1a94c8SDamien Le Moal struct rb_root mblk_rbtree; 1883b1a94c8SDamien Le Moal struct list_head mblk_lru_list; 1893b1a94c8SDamien Le Moal struct list_head mblk_dirty_list; 1903b1a94c8SDamien Le Moal struct shrinker mblk_shrinker; 1913b1a94c8SDamien Le Moal 1923b1a94c8SDamien Le Moal /* Zone allocation management */ 1933b1a94c8SDamien Le Moal struct mutex map_lock; 1943b1a94c8SDamien Le Moal struct dmz_mblock **map_mblk; 1953b1a94c8SDamien Le Moal 19634f5affdSHannes Reinecke unsigned int nr_cache; 19734f5affdSHannes Reinecke atomic_t unmap_nr_cache; 19834f5affdSHannes Reinecke struct list_head unmap_cache_list; 19934f5affdSHannes Reinecke struct list_head map_cache_list; 20034f5affdSHannes Reinecke 2013b1a94c8SDamien Le Moal atomic_t nr_reserved_seq_zones; 2023b1a94c8SDamien Le Moal struct list_head reserved_seq_zones_list; 2033b1a94c8SDamien Le Moal 2043b1a94c8SDamien Le Moal wait_queue_head_t free_wq; 2053b1a94c8SDamien Le Moal }; 2063b1a94c8SDamien Le Moal 207ca1a7045SHannes Reinecke #define dmz_zmd_info(zmd, format, args...) \ 208bd5c4031SHannes Reinecke DMINFO("(%s): " format, (zmd)->label, ## args) 209ca1a7045SHannes Reinecke 210ca1a7045SHannes Reinecke #define dmz_zmd_err(zmd, format, args...) \ 211bd5c4031SHannes Reinecke DMERR("(%s): " format, (zmd)->label, ## args) 212ca1a7045SHannes Reinecke 213ca1a7045SHannes Reinecke #define dmz_zmd_warn(zmd, format, args...) \ 214bd5c4031SHannes Reinecke DMWARN("(%s): " format, (zmd)->label, ## args) 215ca1a7045SHannes Reinecke 216ca1a7045SHannes Reinecke #define dmz_zmd_debug(zmd, format, args...) \ 217bd5c4031SHannes Reinecke DMDEBUG("(%s): " format, (zmd)->label, ## args) 2183b1a94c8SDamien Le Moal /* 2193b1a94c8SDamien Le Moal * Various accessors 2203b1a94c8SDamien Le Moal */ 221bd5c4031SHannes Reinecke static unsigned int dmz_dev_zone_id(struct dmz_metadata *zmd, struct dm_zone *zone) 222bd5c4031SHannes Reinecke { 223bd5c4031SHannes Reinecke if (WARN_ON(!zone)) 224bd5c4031SHannes Reinecke return 0; 225bd5c4031SHannes Reinecke 2268f22272aSHannes Reinecke return zone->id - zone->dev->zone_offset; 227bd5c4031SHannes Reinecke } 228bd5c4031SHannes Reinecke 2293b1a94c8SDamien Le Moal sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone) 2303b1a94c8SDamien Le Moal { 231bd5c4031SHannes Reinecke unsigned int zone_id = dmz_dev_zone_id(zmd, zone); 232bd5c4031SHannes Reinecke 233bd5c4031SHannes Reinecke return (sector_t)zone_id << zmd->zone_nr_sectors_shift; 2343b1a94c8SDamien Le Moal } 2353b1a94c8SDamien Le Moal 2363b1a94c8SDamien Le Moal sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone) 2373b1a94c8SDamien Le Moal { 238bd5c4031SHannes Reinecke unsigned int zone_id = dmz_dev_zone_id(zmd, zone); 239bd5c4031SHannes Reinecke 240bd5c4031SHannes Reinecke return (sector_t)zone_id << zmd->zone_nr_blocks_shift; 2413b1a94c8SDamien Le Moal } 2423b1a94c8SDamien Le Moal 24336820560SHannes Reinecke unsigned int dmz_zone_nr_blocks(struct dmz_metadata *zmd) 24436820560SHannes Reinecke { 24536820560SHannes Reinecke return zmd->zone_nr_blocks; 24636820560SHannes Reinecke } 24736820560SHannes Reinecke 24836820560SHannes Reinecke unsigned int dmz_zone_nr_blocks_shift(struct dmz_metadata *zmd) 24936820560SHannes Reinecke { 25036820560SHannes Reinecke return zmd->zone_nr_blocks_shift; 25136820560SHannes Reinecke } 25236820560SHannes Reinecke 25336820560SHannes Reinecke unsigned int dmz_zone_nr_sectors(struct dmz_metadata *zmd) 25436820560SHannes Reinecke { 25536820560SHannes Reinecke return zmd->zone_nr_sectors; 25636820560SHannes Reinecke } 25736820560SHannes Reinecke 25836820560SHannes Reinecke unsigned int dmz_zone_nr_sectors_shift(struct dmz_metadata *zmd) 25936820560SHannes Reinecke { 26036820560SHannes Reinecke return zmd->zone_nr_sectors_shift; 26136820560SHannes Reinecke } 26236820560SHannes Reinecke 263bc3d5717SHannes Reinecke unsigned int dmz_nr_zones(struct dmz_metadata *zmd) 264bc3d5717SHannes Reinecke { 26536820560SHannes Reinecke return zmd->nr_zones; 266bc3d5717SHannes Reinecke } 267bc3d5717SHannes Reinecke 2683b1a94c8SDamien Le Moal unsigned int dmz_nr_chunks(struct dmz_metadata *zmd) 2693b1a94c8SDamien Le Moal { 2703b1a94c8SDamien Le Moal return zmd->nr_chunks; 2713b1a94c8SDamien Le Moal } 2723b1a94c8SDamien Le Moal 273bd82fdabSHannes Reinecke unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd, int idx) 2743b1a94c8SDamien Le Moal { 275bd82fdabSHannes Reinecke return zmd->dev[idx].nr_rnd; 2763b1a94c8SDamien Le Moal } 2773b1a94c8SDamien Le Moal 278bd82fdabSHannes Reinecke unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd, int idx) 2793b1a94c8SDamien Le Moal { 280bd82fdabSHannes Reinecke return atomic_read(&zmd->dev[idx].unmap_nr_rnd); 2813b1a94c8SDamien Le Moal } 2823b1a94c8SDamien Le Moal 28334f5affdSHannes Reinecke unsigned int dmz_nr_cache_zones(struct dmz_metadata *zmd) 28434f5affdSHannes Reinecke { 28534f5affdSHannes Reinecke return zmd->nr_cache; 28634f5affdSHannes Reinecke } 28734f5affdSHannes Reinecke 28834f5affdSHannes Reinecke unsigned int dmz_nr_unmap_cache_zones(struct dmz_metadata *zmd) 28934f5affdSHannes Reinecke { 29034f5affdSHannes Reinecke return atomic_read(&zmd->unmap_nr_cache); 29134f5affdSHannes Reinecke } 29234f5affdSHannes Reinecke 293bd82fdabSHannes Reinecke unsigned int dmz_nr_seq_zones(struct dmz_metadata *zmd, int idx) 294bc3d5717SHannes Reinecke { 295bd82fdabSHannes Reinecke return zmd->dev[idx].nr_seq; 296bc3d5717SHannes Reinecke } 297bc3d5717SHannes Reinecke 298bd82fdabSHannes Reinecke unsigned int dmz_nr_unmap_seq_zones(struct dmz_metadata *zmd, int idx) 299bc3d5717SHannes Reinecke { 300bd82fdabSHannes Reinecke return atomic_read(&zmd->dev[idx].unmap_nr_seq); 301bc3d5717SHannes Reinecke } 302bc3d5717SHannes Reinecke 303a92fbc44SHannes Reinecke static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id) 304a92fbc44SHannes Reinecke { 305a92fbc44SHannes Reinecke return xa_load(&zmd->zones, zone_id); 306a92fbc44SHannes Reinecke } 307a92fbc44SHannes Reinecke 308a92fbc44SHannes Reinecke static struct dm_zone *dmz_insert(struct dmz_metadata *zmd, 3098f22272aSHannes Reinecke unsigned int zone_id, struct dmz_dev *dev) 310a92fbc44SHannes Reinecke { 311a92fbc44SHannes Reinecke struct dm_zone *zone = kzalloc(sizeof(struct dm_zone), GFP_KERNEL); 312a92fbc44SHannes Reinecke 313a92fbc44SHannes Reinecke if (!zone) 314a92fbc44SHannes Reinecke return ERR_PTR(-ENOMEM); 315a92fbc44SHannes Reinecke 316a92fbc44SHannes Reinecke if (xa_insert(&zmd->zones, zone_id, zone, GFP_KERNEL)) { 317a92fbc44SHannes Reinecke kfree(zone); 318a92fbc44SHannes Reinecke return ERR_PTR(-EBUSY); 319a92fbc44SHannes Reinecke } 320a92fbc44SHannes Reinecke 321a92fbc44SHannes Reinecke INIT_LIST_HEAD(&zone->link); 322a92fbc44SHannes Reinecke atomic_set(&zone->refcount, 0); 323a92fbc44SHannes Reinecke zone->id = zone_id; 324a92fbc44SHannes Reinecke zone->chunk = DMZ_MAP_UNMAPPED; 3258f22272aSHannes Reinecke zone->dev = dev; 326a92fbc44SHannes Reinecke 327a92fbc44SHannes Reinecke return zone; 328a92fbc44SHannes Reinecke } 329a92fbc44SHannes Reinecke 3302234e732SHannes Reinecke const char *dmz_metadata_label(struct dmz_metadata *zmd) 3312234e732SHannes Reinecke { 332bd5c4031SHannes Reinecke return (const char *)zmd->label; 3332234e732SHannes Reinecke } 3342234e732SHannes Reinecke 335d0e21ce4SHannes Reinecke bool dmz_check_dev(struct dmz_metadata *zmd) 336d0e21ce4SHannes Reinecke { 337bd5c4031SHannes Reinecke unsigned int i; 338bd5c4031SHannes Reinecke 339bd5c4031SHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) { 340bd5c4031SHannes Reinecke if (!dmz_check_bdev(&zmd->dev[i])) 341bd5c4031SHannes Reinecke return false; 342bd5c4031SHannes Reinecke } 343bd5c4031SHannes Reinecke return true; 344d0e21ce4SHannes Reinecke } 345d0e21ce4SHannes Reinecke 346d0e21ce4SHannes Reinecke bool dmz_dev_is_dying(struct dmz_metadata *zmd) 347d0e21ce4SHannes Reinecke { 348bd5c4031SHannes Reinecke unsigned int i; 349bd5c4031SHannes Reinecke 350bd5c4031SHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) { 351bd5c4031SHannes Reinecke if (dmz_bdev_is_dying(&zmd->dev[i])) 352bd5c4031SHannes Reinecke return true; 353bd5c4031SHannes Reinecke } 354bd5c4031SHannes Reinecke return false; 355d0e21ce4SHannes Reinecke } 356d0e21ce4SHannes Reinecke 3573b1a94c8SDamien Le Moal /* 3583b1a94c8SDamien Le Moal * Lock/unlock mapping table. 3593b1a94c8SDamien Le Moal * The map lock also protects all the zone lists. 3603b1a94c8SDamien Le Moal */ 3613b1a94c8SDamien Le Moal void dmz_lock_map(struct dmz_metadata *zmd) 3623b1a94c8SDamien Le Moal { 3633b1a94c8SDamien Le Moal mutex_lock(&zmd->map_lock); 3643b1a94c8SDamien Le Moal } 3653b1a94c8SDamien Le Moal 3663b1a94c8SDamien Le Moal void dmz_unlock_map(struct dmz_metadata *zmd) 3673b1a94c8SDamien Le Moal { 3683b1a94c8SDamien Le Moal mutex_unlock(&zmd->map_lock); 3693b1a94c8SDamien Le Moal } 3703b1a94c8SDamien Le Moal 3713b1a94c8SDamien Le Moal /* 3723b1a94c8SDamien Le Moal * Lock/unlock metadata access. This is a "read" lock on a semaphore 3733b1a94c8SDamien Le Moal * that prevents metadata flush from running while metadata are being 3743b1a94c8SDamien Le Moal * modified. The actual metadata write mutual exclusion is achieved with 375ad1bd578SDmitry Fomichev * the map lock and zone state management (active and reclaim state are 3763b1a94c8SDamien Le Moal * mutually exclusive). 3773b1a94c8SDamien Le Moal */ 3783b1a94c8SDamien Le Moal void dmz_lock_metadata(struct dmz_metadata *zmd) 3793b1a94c8SDamien Le Moal { 3803b1a94c8SDamien Le Moal down_read(&zmd->mblk_sem); 3813b1a94c8SDamien Le Moal } 3823b1a94c8SDamien Le Moal 3833b1a94c8SDamien Le Moal void dmz_unlock_metadata(struct dmz_metadata *zmd) 3843b1a94c8SDamien Le Moal { 3853b1a94c8SDamien Le Moal up_read(&zmd->mblk_sem); 3863b1a94c8SDamien Le Moal } 3873b1a94c8SDamien Le Moal 3883b1a94c8SDamien Le Moal /* 3893b1a94c8SDamien Le Moal * Lock/unlock flush: prevent concurrent executions 3903b1a94c8SDamien Le Moal * of dmz_flush_metadata as well as metadata modification in reclaim 3913b1a94c8SDamien Le Moal * while flush is being executed. 3923b1a94c8SDamien Le Moal */ 3933b1a94c8SDamien Le Moal void dmz_lock_flush(struct dmz_metadata *zmd) 3943b1a94c8SDamien Le Moal { 3953b1a94c8SDamien Le Moal mutex_lock(&zmd->mblk_flush_lock); 3963b1a94c8SDamien Le Moal } 3973b1a94c8SDamien Le Moal 3983b1a94c8SDamien Le Moal void dmz_unlock_flush(struct dmz_metadata *zmd) 3993b1a94c8SDamien Le Moal { 4003b1a94c8SDamien Le Moal mutex_unlock(&zmd->mblk_flush_lock); 4013b1a94c8SDamien Le Moal } 4023b1a94c8SDamien Le Moal 4033b1a94c8SDamien Le Moal /* 4043b1a94c8SDamien Le Moal * Allocate a metadata block. 4053b1a94c8SDamien Le Moal */ 4063b1a94c8SDamien Le Moal static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd, 4073b1a94c8SDamien Le Moal sector_t mblk_no) 4083b1a94c8SDamien Le Moal { 4093b1a94c8SDamien Le Moal struct dmz_mblock *mblk = NULL; 4103b1a94c8SDamien Le Moal 4113b1a94c8SDamien Le Moal /* See if we can reuse cached blocks */ 4123b1a94c8SDamien Le Moal if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) { 4133b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 4143b1a94c8SDamien Le Moal mblk = list_first_entry_or_null(&zmd->mblk_lru_list, 4153b1a94c8SDamien Le Moal struct dmz_mblock, link); 4163b1a94c8SDamien Le Moal if (mblk) { 4173b1a94c8SDamien Le Moal list_del_init(&mblk->link); 4183b1a94c8SDamien Le Moal rb_erase(&mblk->node, &zmd->mblk_rbtree); 4193b1a94c8SDamien Le Moal mblk->no = mblk_no; 4203b1a94c8SDamien Le Moal } 4213b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 4223b1a94c8SDamien Le Moal if (mblk) 4233b1a94c8SDamien Le Moal return mblk; 4243b1a94c8SDamien Le Moal } 4253b1a94c8SDamien Le Moal 4263b1a94c8SDamien Le Moal /* Allocate a new block */ 4273b1a94c8SDamien Le Moal mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO); 4283b1a94c8SDamien Le Moal if (!mblk) 4293b1a94c8SDamien Le Moal return NULL; 4303b1a94c8SDamien Le Moal 4313b1a94c8SDamien Le Moal mblk->page = alloc_page(GFP_NOIO); 4323b1a94c8SDamien Le Moal if (!mblk->page) { 4333b1a94c8SDamien Le Moal kfree(mblk); 4343b1a94c8SDamien Le Moal return NULL; 4353b1a94c8SDamien Le Moal } 4363b1a94c8SDamien Le Moal 4373b1a94c8SDamien Le Moal RB_CLEAR_NODE(&mblk->node); 4383b1a94c8SDamien Le Moal INIT_LIST_HEAD(&mblk->link); 43933c2865fSDamien Le Moal mblk->ref = 0; 4403b1a94c8SDamien Le Moal mblk->state = 0; 4413b1a94c8SDamien Le Moal mblk->no = mblk_no; 4423b1a94c8SDamien Le Moal mblk->data = page_address(mblk->page); 4433b1a94c8SDamien Le Moal 4443b1a94c8SDamien Le Moal atomic_inc(&zmd->nr_mblks); 4453b1a94c8SDamien Le Moal 4463b1a94c8SDamien Le Moal return mblk; 4473b1a94c8SDamien Le Moal } 4483b1a94c8SDamien Le Moal 4493b1a94c8SDamien Le Moal /* 4503b1a94c8SDamien Le Moal * Free a metadata block. 4513b1a94c8SDamien Le Moal */ 4523b1a94c8SDamien Le Moal static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) 4533b1a94c8SDamien Le Moal { 4543b1a94c8SDamien Le Moal __free_pages(mblk->page, 0); 4553b1a94c8SDamien Le Moal kfree(mblk); 4563b1a94c8SDamien Le Moal 4573b1a94c8SDamien Le Moal atomic_dec(&zmd->nr_mblks); 4583b1a94c8SDamien Le Moal } 4593b1a94c8SDamien Le Moal 4603b1a94c8SDamien Le Moal /* 4613b1a94c8SDamien Le Moal * Insert a metadata block in the rbtree. 4623b1a94c8SDamien Le Moal */ 4633b1a94c8SDamien Le Moal static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) 4643b1a94c8SDamien Le Moal { 4653b1a94c8SDamien Le Moal struct rb_root *root = &zmd->mblk_rbtree; 4663b1a94c8SDamien Le Moal struct rb_node **new = &(root->rb_node), *parent = NULL; 4673b1a94c8SDamien Le Moal struct dmz_mblock *b; 4683b1a94c8SDamien Le Moal 4693b1a94c8SDamien Le Moal /* Figure out where to put the new node */ 4703b1a94c8SDamien Le Moal while (*new) { 4713b1a94c8SDamien Le Moal b = container_of(*new, struct dmz_mblock, node); 4723b1a94c8SDamien Le Moal parent = *new; 4733b1a94c8SDamien Le Moal new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right); 4743b1a94c8SDamien Le Moal } 4753b1a94c8SDamien Le Moal 4763b1a94c8SDamien Le Moal /* Add new node and rebalance tree */ 4773b1a94c8SDamien Le Moal rb_link_node(&mblk->node, parent, new); 4783b1a94c8SDamien Le Moal rb_insert_color(&mblk->node, root); 4793b1a94c8SDamien Le Moal } 4803b1a94c8SDamien Le Moal 4813b1a94c8SDamien Le Moal /* 4823d4e7383SDamien Le Moal * Lookup a metadata block in the rbtree. If the block is found, increment 4833d4e7383SDamien Le Moal * its reference count. 4843b1a94c8SDamien Le Moal */ 4853d4e7383SDamien Le Moal static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd, 4863b1a94c8SDamien Le Moal sector_t mblk_no) 4873b1a94c8SDamien Le Moal { 4883b1a94c8SDamien Le Moal struct rb_root *root = &zmd->mblk_rbtree; 4893b1a94c8SDamien Le Moal struct rb_node *node = root->rb_node; 4903b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 4913b1a94c8SDamien Le Moal 4923b1a94c8SDamien Le Moal while (node) { 4933b1a94c8SDamien Le Moal mblk = container_of(node, struct dmz_mblock, node); 4943d4e7383SDamien Le Moal if (mblk->no == mblk_no) { 4953d4e7383SDamien Le Moal /* 4963d4e7383SDamien Le Moal * If this is the first reference to the block, 4973d4e7383SDamien Le Moal * remove it from the LRU list. 4983d4e7383SDamien Le Moal */ 4993d4e7383SDamien Le Moal mblk->ref++; 5003d4e7383SDamien Le Moal if (mblk->ref == 1 && 5013d4e7383SDamien Le Moal !test_bit(DMZ_META_DIRTY, &mblk->state)) 5023d4e7383SDamien Le Moal list_del_init(&mblk->link); 5033b1a94c8SDamien Le Moal return mblk; 5043d4e7383SDamien Le Moal } 5053b1a94c8SDamien Le Moal node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right; 5063b1a94c8SDamien Le Moal } 5073b1a94c8SDamien Le Moal 5083b1a94c8SDamien Le Moal return NULL; 5093b1a94c8SDamien Le Moal } 5103b1a94c8SDamien Le Moal 5113b1a94c8SDamien Le Moal /* 5123b1a94c8SDamien Le Moal * Metadata block BIO end callback. 5133b1a94c8SDamien Le Moal */ 5143b1a94c8SDamien Le Moal static void dmz_mblock_bio_end_io(struct bio *bio) 5153b1a94c8SDamien Le Moal { 5163b1a94c8SDamien Le Moal struct dmz_mblock *mblk = bio->bi_private; 5173b1a94c8SDamien Le Moal int flag; 5183b1a94c8SDamien Le Moal 5193b1a94c8SDamien Le Moal if (bio->bi_status) 5203b1a94c8SDamien Le Moal set_bit(DMZ_META_ERROR, &mblk->state); 5213b1a94c8SDamien Le Moal 5223b1a94c8SDamien Le Moal if (bio_op(bio) == REQ_OP_WRITE) 5233b1a94c8SDamien Le Moal flag = DMZ_META_WRITING; 5243b1a94c8SDamien Le Moal else 5253b1a94c8SDamien Le Moal flag = DMZ_META_READING; 5263b1a94c8SDamien Le Moal 5273b1a94c8SDamien Le Moal clear_bit_unlock(flag, &mblk->state); 5283b1a94c8SDamien Le Moal smp_mb__after_atomic(); 5293b1a94c8SDamien Le Moal wake_up_bit(&mblk->state, flag); 5303b1a94c8SDamien Le Moal 5313b1a94c8SDamien Le Moal bio_put(bio); 5323b1a94c8SDamien Le Moal } 5333b1a94c8SDamien Le Moal 5343b1a94c8SDamien Le Moal /* 5353d4e7383SDamien Le Moal * Read an uncached metadata block from disk and add it to the cache. 5363b1a94c8SDamien Le Moal */ 5373d4e7383SDamien Le Moal static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd, 5383b1a94c8SDamien Le Moal sector_t mblk_no) 5393b1a94c8SDamien Le Moal { 5403d4e7383SDamien Le Moal struct dmz_mblock *mblk, *m; 5413b1a94c8SDamien Le Moal sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no; 542bf28a3baSHannes Reinecke struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev; 5433b1a94c8SDamien Le Moal struct bio *bio; 5443b1a94c8SDamien Le Moal 545bf28a3baSHannes Reinecke if (dmz_bdev_is_dying(dev)) 54675d66ffbSDmitry Fomichev return ERR_PTR(-EIO); 54775d66ffbSDmitry Fomichev 5483d4e7383SDamien Le Moal /* Get a new block and a BIO to read it */ 5493b1a94c8SDamien Le Moal mblk = dmz_alloc_mblock(zmd, mblk_no); 5503b1a94c8SDamien Le Moal if (!mblk) 55175d66ffbSDmitry Fomichev return ERR_PTR(-ENOMEM); 5523b1a94c8SDamien Le Moal 5533b1a94c8SDamien Le Moal bio = bio_alloc(GFP_NOIO, 1); 5543b1a94c8SDamien Le Moal if (!bio) { 5553b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 55675d66ffbSDmitry Fomichev return ERR_PTR(-ENOMEM); 5573b1a94c8SDamien Le Moal } 5583b1a94c8SDamien Le Moal 5593d4e7383SDamien Le Moal spin_lock(&zmd->mblk_lock); 5603d4e7383SDamien Le Moal 5613d4e7383SDamien Le Moal /* 5623d4e7383SDamien Le Moal * Make sure that another context did not start reading 5633d4e7383SDamien Le Moal * the block already. 5643d4e7383SDamien Le Moal */ 5653d4e7383SDamien Le Moal m = dmz_get_mblock_fast(zmd, mblk_no); 5663d4e7383SDamien Le Moal if (m) { 5673d4e7383SDamien Le Moal spin_unlock(&zmd->mblk_lock); 5683d4e7383SDamien Le Moal dmz_free_mblock(zmd, mblk); 5693d4e7383SDamien Le Moal bio_put(bio); 5703d4e7383SDamien Le Moal return m; 5713d4e7383SDamien Le Moal } 5723d4e7383SDamien Le Moal 5733d4e7383SDamien Le Moal mblk->ref++; 5743d4e7383SDamien Le Moal set_bit(DMZ_META_READING, &mblk->state); 5753d4e7383SDamien Le Moal dmz_insert_mblock(zmd, mblk); 5763d4e7383SDamien Le Moal 5773d4e7383SDamien Le Moal spin_unlock(&zmd->mblk_lock); 5783d4e7383SDamien Le Moal 5793d4e7383SDamien Le Moal /* Submit read BIO */ 5803b1a94c8SDamien Le Moal bio->bi_iter.bi_sector = dmz_blk2sect(block); 581bf28a3baSHannes Reinecke bio_set_dev(bio, dev->bdev); 5823b1a94c8SDamien Le Moal bio->bi_private = mblk; 5833b1a94c8SDamien Le Moal bio->bi_end_io = dmz_mblock_bio_end_io; 5843b1a94c8SDamien Le Moal bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO); 5853b1a94c8SDamien Le Moal bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); 5863b1a94c8SDamien Le Moal submit_bio(bio); 5873b1a94c8SDamien Le Moal 5883b1a94c8SDamien Le Moal return mblk; 5893b1a94c8SDamien Le Moal } 5903b1a94c8SDamien Le Moal 5913b1a94c8SDamien Le Moal /* 5923b1a94c8SDamien Le Moal * Free metadata blocks. 5933b1a94c8SDamien Le Moal */ 5943b1a94c8SDamien Le Moal static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd, 5953b1a94c8SDamien Le Moal unsigned long limit) 5963b1a94c8SDamien Le Moal { 5973b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 5983b1a94c8SDamien Le Moal unsigned long count = 0; 5993b1a94c8SDamien Le Moal 6003b1a94c8SDamien Le Moal if (!zmd->max_nr_mblks) 6013b1a94c8SDamien Le Moal return 0; 6023b1a94c8SDamien Le Moal 6033b1a94c8SDamien Le Moal while (!list_empty(&zmd->mblk_lru_list) && 6043b1a94c8SDamien Le Moal atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks && 6053b1a94c8SDamien Le Moal count < limit) { 6063b1a94c8SDamien Le Moal mblk = list_first_entry(&zmd->mblk_lru_list, 6073b1a94c8SDamien Le Moal struct dmz_mblock, link); 6083b1a94c8SDamien Le Moal list_del_init(&mblk->link); 6093b1a94c8SDamien Le Moal rb_erase(&mblk->node, &zmd->mblk_rbtree); 6103b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 6113b1a94c8SDamien Le Moal count++; 6123b1a94c8SDamien Le Moal } 6133b1a94c8SDamien Le Moal 6143b1a94c8SDamien Le Moal return count; 6153b1a94c8SDamien Le Moal } 6163b1a94c8SDamien Le Moal 6173b1a94c8SDamien Le Moal /* 6183b1a94c8SDamien Le Moal * For mblock shrinker: get the number of unused metadata blocks in the cache. 6193b1a94c8SDamien Le Moal */ 6203b1a94c8SDamien Le Moal static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink, 6213b1a94c8SDamien Le Moal struct shrink_control *sc) 6223b1a94c8SDamien Le Moal { 6233b1a94c8SDamien Le Moal struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker); 6243b1a94c8SDamien Le Moal 6253b1a94c8SDamien Le Moal return atomic_read(&zmd->nr_mblks); 6263b1a94c8SDamien Le Moal } 6273b1a94c8SDamien Le Moal 6283b1a94c8SDamien Le Moal /* 6293b1a94c8SDamien Le Moal * For mblock shrinker: scan unused metadata blocks and shrink the cache. 6303b1a94c8SDamien Le Moal */ 6313b1a94c8SDamien Le Moal static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink, 6323b1a94c8SDamien Le Moal struct shrink_control *sc) 6333b1a94c8SDamien Le Moal { 6343b1a94c8SDamien Le Moal struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker); 6353b1a94c8SDamien Le Moal unsigned long count; 6363b1a94c8SDamien Le Moal 6373b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 6383b1a94c8SDamien Le Moal count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan); 6393b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 6403b1a94c8SDamien Le Moal 6413b1a94c8SDamien Le Moal return count ? count : SHRINK_STOP; 6423b1a94c8SDamien Le Moal } 6433b1a94c8SDamien Le Moal 6443b1a94c8SDamien Le Moal /* 6453b1a94c8SDamien Le Moal * Release a metadata block. 6463b1a94c8SDamien Le Moal */ 6473b1a94c8SDamien Le Moal static void dmz_release_mblock(struct dmz_metadata *zmd, 6483b1a94c8SDamien Le Moal struct dmz_mblock *mblk) 6493b1a94c8SDamien Le Moal { 6503b1a94c8SDamien Le Moal 6513b1a94c8SDamien Le Moal if (!mblk) 6523b1a94c8SDamien Le Moal return; 6533b1a94c8SDamien Le Moal 6543b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 6553b1a94c8SDamien Le Moal 65633c2865fSDamien Le Moal mblk->ref--; 65733c2865fSDamien Le Moal if (mblk->ref == 0) { 6583b1a94c8SDamien Le Moal if (test_bit(DMZ_META_ERROR, &mblk->state)) { 6593b1a94c8SDamien Le Moal rb_erase(&mblk->node, &zmd->mblk_rbtree); 6603b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 6613b1a94c8SDamien Le Moal } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) { 6623b1a94c8SDamien Le Moal list_add_tail(&mblk->link, &zmd->mblk_lru_list); 6633b1a94c8SDamien Le Moal dmz_shrink_mblock_cache(zmd, 1); 6643b1a94c8SDamien Le Moal } 6653b1a94c8SDamien Le Moal } 6663b1a94c8SDamien Le Moal 6673b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 6683b1a94c8SDamien Le Moal } 6693b1a94c8SDamien Le Moal 6703b1a94c8SDamien Le Moal /* 6713b1a94c8SDamien Le Moal * Get a metadata block from the rbtree. If the block 6723b1a94c8SDamien Le Moal * is not present, read it from disk. 6733b1a94c8SDamien Le Moal */ 6743b1a94c8SDamien Le Moal static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd, 6753b1a94c8SDamien Le Moal sector_t mblk_no) 6763b1a94c8SDamien Le Moal { 6773b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 678bf28a3baSHannes Reinecke struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev; 6793b1a94c8SDamien Le Moal 6803b1a94c8SDamien Le Moal /* Check rbtree */ 6813b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 6823d4e7383SDamien Le Moal mblk = dmz_get_mblock_fast(zmd, mblk_no); 6833b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 6843b1a94c8SDamien Le Moal 6853b1a94c8SDamien Le Moal if (!mblk) { 6863b1a94c8SDamien Le Moal /* Cache miss: read the block from disk */ 6873d4e7383SDamien Le Moal mblk = dmz_get_mblock_slow(zmd, mblk_no); 68875d66ffbSDmitry Fomichev if (IS_ERR(mblk)) 68975d66ffbSDmitry Fomichev return mblk; 6903b1a94c8SDamien Le Moal } 6913b1a94c8SDamien Le Moal 6923b1a94c8SDamien Le Moal /* Wait for on-going read I/O and check for error */ 6933b1a94c8SDamien Le Moal wait_on_bit_io(&mblk->state, DMZ_META_READING, 6943b1a94c8SDamien Le Moal TASK_UNINTERRUPTIBLE); 6953b1a94c8SDamien Le Moal if (test_bit(DMZ_META_ERROR, &mblk->state)) { 6963b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 697bf28a3baSHannes Reinecke dmz_check_bdev(dev); 6983b1a94c8SDamien Le Moal return ERR_PTR(-EIO); 6993b1a94c8SDamien Le Moal } 7003b1a94c8SDamien Le Moal 7013b1a94c8SDamien Le Moal return mblk; 7023b1a94c8SDamien Le Moal } 7033b1a94c8SDamien Le Moal 7043b1a94c8SDamien Le Moal /* 7053b1a94c8SDamien Le Moal * Mark a metadata block dirty. 7063b1a94c8SDamien Le Moal */ 7073b1a94c8SDamien Le Moal static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) 7083b1a94c8SDamien Le Moal { 7093b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 7103b1a94c8SDamien Le Moal if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state)) 7113b1a94c8SDamien Le Moal list_add_tail(&mblk->link, &zmd->mblk_dirty_list); 7123b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 7133b1a94c8SDamien Le Moal } 7143b1a94c8SDamien Le Moal 7153b1a94c8SDamien Le Moal /* 7163b1a94c8SDamien Le Moal * Issue a metadata block write BIO. 7173b1a94c8SDamien Le Moal */ 71875d66ffbSDmitry Fomichev static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk, 7193b1a94c8SDamien Le Moal unsigned int set) 7203b1a94c8SDamien Le Moal { 721bf28a3baSHannes Reinecke struct dmz_dev *dev = zmd->sb[set].dev; 7223b1a94c8SDamien Le Moal sector_t block = zmd->sb[set].block + mblk->no; 7233b1a94c8SDamien Le Moal struct bio *bio; 7243b1a94c8SDamien Le Moal 725bf28a3baSHannes Reinecke if (dmz_bdev_is_dying(dev)) 72675d66ffbSDmitry Fomichev return -EIO; 72775d66ffbSDmitry Fomichev 7283b1a94c8SDamien Le Moal bio = bio_alloc(GFP_NOIO, 1); 7293b1a94c8SDamien Le Moal if (!bio) { 7303b1a94c8SDamien Le Moal set_bit(DMZ_META_ERROR, &mblk->state); 73175d66ffbSDmitry Fomichev return -ENOMEM; 7323b1a94c8SDamien Le Moal } 7333b1a94c8SDamien Le Moal 7343b1a94c8SDamien Le Moal set_bit(DMZ_META_WRITING, &mblk->state); 7353b1a94c8SDamien Le Moal 7363b1a94c8SDamien Le Moal bio->bi_iter.bi_sector = dmz_blk2sect(block); 737bf28a3baSHannes Reinecke bio_set_dev(bio, dev->bdev); 7383b1a94c8SDamien Le Moal bio->bi_private = mblk; 7393b1a94c8SDamien Le Moal bio->bi_end_io = dmz_mblock_bio_end_io; 7403b1a94c8SDamien Le Moal bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO); 7413b1a94c8SDamien Le Moal bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); 7423b1a94c8SDamien Le Moal submit_bio(bio); 74375d66ffbSDmitry Fomichev 74475d66ffbSDmitry Fomichev return 0; 7453b1a94c8SDamien Le Moal } 7463b1a94c8SDamien Le Moal 7473b1a94c8SDamien Le Moal /* 7483b1a94c8SDamien Le Moal * Read/write a metadata block. 7493b1a94c8SDamien Le Moal */ 750bf28a3baSHannes Reinecke static int dmz_rdwr_block(struct dmz_dev *dev, int op, 751bf28a3baSHannes Reinecke sector_t block, struct page *page) 7523b1a94c8SDamien Le Moal { 7533b1a94c8SDamien Le Moal struct bio *bio; 7543b1a94c8SDamien Le Moal int ret; 7553b1a94c8SDamien Le Moal 756bd5c4031SHannes Reinecke if (WARN_ON(!dev)) 757bd5c4031SHannes Reinecke return -EIO; 758bd5c4031SHannes Reinecke 759bf28a3baSHannes Reinecke if (dmz_bdev_is_dying(dev)) 76075d66ffbSDmitry Fomichev return -EIO; 76175d66ffbSDmitry Fomichev 7623b1a94c8SDamien Le Moal bio = bio_alloc(GFP_NOIO, 1); 7633b1a94c8SDamien Le Moal if (!bio) 7643b1a94c8SDamien Le Moal return -ENOMEM; 7653b1a94c8SDamien Le Moal 7663b1a94c8SDamien Le Moal bio->bi_iter.bi_sector = dmz_blk2sect(block); 767bf28a3baSHannes Reinecke bio_set_dev(bio, dev->bdev); 7683b1a94c8SDamien Le Moal bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO); 7693b1a94c8SDamien Le Moal bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0); 7703b1a94c8SDamien Le Moal ret = submit_bio_wait(bio); 7713b1a94c8SDamien Le Moal bio_put(bio); 7723b1a94c8SDamien Le Moal 773e7fad909SDmitry Fomichev if (ret) 774bf28a3baSHannes Reinecke dmz_check_bdev(dev); 7753b1a94c8SDamien Le Moal return ret; 7763b1a94c8SDamien Le Moal } 7773b1a94c8SDamien Le Moal 7783b1a94c8SDamien Le Moal /* 7793b1a94c8SDamien Le Moal * Write super block of the specified metadata set. 7803b1a94c8SDamien Le Moal */ 7813b1a94c8SDamien Le Moal static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set) 7823b1a94c8SDamien Le Moal { 7833b1a94c8SDamien Le Moal struct dmz_mblock *mblk = zmd->sb[set].mblk; 7843b1a94c8SDamien Le Moal struct dmz_super *sb = zmd->sb[set].sb; 785bf28a3baSHannes Reinecke struct dmz_dev *dev = zmd->sb[set].dev; 786bd5c4031SHannes Reinecke sector_t sb_block; 7873b1a94c8SDamien Le Moal u64 sb_gen = zmd->sb_gen + 1; 7883b1a94c8SDamien Le Moal int ret; 7893b1a94c8SDamien Le Moal 7903b1a94c8SDamien Le Moal sb->magic = cpu_to_le32(DMZ_MAGIC); 791bd5c4031SHannes Reinecke 792bd5c4031SHannes Reinecke sb->version = cpu_to_le32(zmd->sb_version); 793bd5c4031SHannes Reinecke if (zmd->sb_version > 1) { 794bd5c4031SHannes Reinecke BUILD_BUG_ON(UUID_SIZE != 16); 795bd5c4031SHannes Reinecke export_uuid(sb->dmz_uuid, &zmd->uuid); 796bd5c4031SHannes Reinecke memcpy(sb->dmz_label, zmd->label, BDEVNAME_SIZE); 797bd5c4031SHannes Reinecke export_uuid(sb->dev_uuid, &dev->uuid); 798bd5c4031SHannes Reinecke } 7993b1a94c8SDamien Le Moal 8003b1a94c8SDamien Le Moal sb->gen = cpu_to_le64(sb_gen); 8013b1a94c8SDamien Le Moal 802bd5c4031SHannes Reinecke /* 803bd5c4031SHannes Reinecke * The metadata always references the absolute block address, 804bd5c4031SHannes Reinecke * ie relative to the entire block range, not the per-device 805bd5c4031SHannes Reinecke * block address. 806bd5c4031SHannes Reinecke */ 807bd5c4031SHannes Reinecke sb_block = zmd->sb[set].zone->id << zmd->zone_nr_blocks_shift; 808bd5c4031SHannes Reinecke sb->sb_block = cpu_to_le64(sb_block); 8093b1a94c8SDamien Le Moal sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks); 8103b1a94c8SDamien Le Moal sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq); 8113b1a94c8SDamien Le Moal sb->nr_chunks = cpu_to_le32(zmd->nr_chunks); 8123b1a94c8SDamien Le Moal 8133b1a94c8SDamien Le Moal sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks); 8143b1a94c8SDamien Le Moal sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks); 8153b1a94c8SDamien Le Moal 8163b1a94c8SDamien Le Moal sb->crc = 0; 8173b1a94c8SDamien Le Moal sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE)); 8183b1a94c8SDamien Le Moal 819bd5c4031SHannes Reinecke ret = dmz_rdwr_block(dev, REQ_OP_WRITE, zmd->sb[set].block, 820bd5c4031SHannes Reinecke mblk->page); 8213b1a94c8SDamien Le Moal if (ret == 0) 822bf28a3baSHannes Reinecke ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); 8233b1a94c8SDamien Le Moal 8243b1a94c8SDamien Le Moal return ret; 8253b1a94c8SDamien Le Moal } 8263b1a94c8SDamien Le Moal 8273b1a94c8SDamien Le Moal /* 8283b1a94c8SDamien Le Moal * Write dirty metadata blocks to the specified set. 8293b1a94c8SDamien Le Moal */ 8303b1a94c8SDamien Le Moal static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd, 8313b1a94c8SDamien Le Moal struct list_head *write_list, 8323b1a94c8SDamien Le Moal unsigned int set) 8333b1a94c8SDamien Le Moal { 8343b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 835bf28a3baSHannes Reinecke struct dmz_dev *dev = zmd->sb[set].dev; 8363b1a94c8SDamien Le Moal struct blk_plug plug; 83775d66ffbSDmitry Fomichev int ret = 0, nr_mblks_submitted = 0; 8383b1a94c8SDamien Le Moal 8393b1a94c8SDamien Le Moal /* Issue writes */ 8403b1a94c8SDamien Le Moal blk_start_plug(&plug); 84175d66ffbSDmitry Fomichev list_for_each_entry(mblk, write_list, link) { 84275d66ffbSDmitry Fomichev ret = dmz_write_mblock(zmd, mblk, set); 84375d66ffbSDmitry Fomichev if (ret) 84475d66ffbSDmitry Fomichev break; 84575d66ffbSDmitry Fomichev nr_mblks_submitted++; 84675d66ffbSDmitry Fomichev } 8473b1a94c8SDamien Le Moal blk_finish_plug(&plug); 8483b1a94c8SDamien Le Moal 8493b1a94c8SDamien Le Moal /* Wait for completion */ 8503b1a94c8SDamien Le Moal list_for_each_entry(mblk, write_list, link) { 85175d66ffbSDmitry Fomichev if (!nr_mblks_submitted) 85275d66ffbSDmitry Fomichev break; 8533b1a94c8SDamien Le Moal wait_on_bit_io(&mblk->state, DMZ_META_WRITING, 8543b1a94c8SDamien Le Moal TASK_UNINTERRUPTIBLE); 8553b1a94c8SDamien Le Moal if (test_bit(DMZ_META_ERROR, &mblk->state)) { 8563b1a94c8SDamien Le Moal clear_bit(DMZ_META_ERROR, &mblk->state); 857bf28a3baSHannes Reinecke dmz_check_bdev(dev); 8583b1a94c8SDamien Le Moal ret = -EIO; 8593b1a94c8SDamien Le Moal } 86075d66ffbSDmitry Fomichev nr_mblks_submitted--; 8613b1a94c8SDamien Le Moal } 8623b1a94c8SDamien Le Moal 8633b1a94c8SDamien Le Moal /* Flush drive cache (this will also sync data) */ 8643b1a94c8SDamien Le Moal if (ret == 0) 865bf28a3baSHannes Reinecke ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); 8663b1a94c8SDamien Le Moal 8673b1a94c8SDamien Le Moal return ret; 8683b1a94c8SDamien Le Moal } 8693b1a94c8SDamien Le Moal 8703b1a94c8SDamien Le Moal /* 8713b1a94c8SDamien Le Moal * Log dirty metadata blocks. 8723b1a94c8SDamien Le Moal */ 8733b1a94c8SDamien Le Moal static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd, 8743b1a94c8SDamien Le Moal struct list_head *write_list) 8753b1a94c8SDamien Le Moal { 8763b1a94c8SDamien Le Moal unsigned int log_set = zmd->mblk_primary ^ 0x1; 8773b1a94c8SDamien Le Moal int ret; 8783b1a94c8SDamien Le Moal 8793b1a94c8SDamien Le Moal /* Write dirty blocks to the log */ 8803b1a94c8SDamien Le Moal ret = dmz_write_dirty_mblocks(zmd, write_list, log_set); 8813b1a94c8SDamien Le Moal if (ret) 8823b1a94c8SDamien Le Moal return ret; 8833b1a94c8SDamien Le Moal 8843b1a94c8SDamien Le Moal /* 8853b1a94c8SDamien Le Moal * No error so far: now validate the log by updating the 8863b1a94c8SDamien Le Moal * log index super block generation. 8873b1a94c8SDamien Le Moal */ 8883b1a94c8SDamien Le Moal ret = dmz_write_sb(zmd, log_set); 8893b1a94c8SDamien Le Moal if (ret) 8903b1a94c8SDamien Le Moal return ret; 8913b1a94c8SDamien Le Moal 8923b1a94c8SDamien Le Moal return 0; 8933b1a94c8SDamien Le Moal } 8943b1a94c8SDamien Le Moal 8953b1a94c8SDamien Le Moal /* 8963b1a94c8SDamien Le Moal * Flush dirty metadata blocks. 8973b1a94c8SDamien Le Moal */ 8983b1a94c8SDamien Le Moal int dmz_flush_metadata(struct dmz_metadata *zmd) 8993b1a94c8SDamien Le Moal { 9003b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 9013b1a94c8SDamien Le Moal struct list_head write_list; 902bf28a3baSHannes Reinecke struct dmz_dev *dev; 9033b1a94c8SDamien Le Moal int ret; 9043b1a94c8SDamien Le Moal 9053b1a94c8SDamien Le Moal if (WARN_ON(!zmd)) 9063b1a94c8SDamien Le Moal return 0; 9073b1a94c8SDamien Le Moal 9083b1a94c8SDamien Le Moal INIT_LIST_HEAD(&write_list); 9093b1a94c8SDamien Le Moal 9103b1a94c8SDamien Le Moal /* 9113b1a94c8SDamien Le Moal * Make sure that metadata blocks are stable before logging: take 9123b1a94c8SDamien Le Moal * the write lock on the metadata semaphore to prevent target BIOs 9133b1a94c8SDamien Le Moal * from modifying metadata. 9143b1a94c8SDamien Le Moal */ 9153b1a94c8SDamien Le Moal down_write(&zmd->mblk_sem); 916bf28a3baSHannes Reinecke dev = zmd->sb[zmd->mblk_primary].dev; 9173b1a94c8SDamien Le Moal 9183b1a94c8SDamien Le Moal /* 9193b1a94c8SDamien Le Moal * This is called from the target flush work and reclaim work. 9203b1a94c8SDamien Le Moal * Concurrent execution is not allowed. 9213b1a94c8SDamien Le Moal */ 9223b1a94c8SDamien Le Moal dmz_lock_flush(zmd); 9233b1a94c8SDamien Le Moal 924bf28a3baSHannes Reinecke if (dmz_bdev_is_dying(dev)) { 92575d66ffbSDmitry Fomichev ret = -EIO; 92675d66ffbSDmitry Fomichev goto out; 92775d66ffbSDmitry Fomichev } 92875d66ffbSDmitry Fomichev 9293b1a94c8SDamien Le Moal /* Get dirty blocks */ 9303b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 9313b1a94c8SDamien Le Moal list_splice_init(&zmd->mblk_dirty_list, &write_list); 9323b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 9333b1a94c8SDamien Le Moal 9343b1a94c8SDamien Le Moal /* If there are no dirty metadata blocks, just flush the device cache */ 9353b1a94c8SDamien Le Moal if (list_empty(&write_list)) { 936bf28a3baSHannes Reinecke ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); 937e7fad909SDmitry Fomichev goto err; 9383b1a94c8SDamien Le Moal } 9393b1a94c8SDamien Le Moal 9403b1a94c8SDamien Le Moal /* 9413b1a94c8SDamien Le Moal * The primary metadata set is still clean. Keep it this way until 9423b1a94c8SDamien Le Moal * all updates are successful in the secondary set. That is, use 9433b1a94c8SDamien Le Moal * the secondary set as a log. 9443b1a94c8SDamien Le Moal */ 9453b1a94c8SDamien Le Moal ret = dmz_log_dirty_mblocks(zmd, &write_list); 9463b1a94c8SDamien Le Moal if (ret) 947e7fad909SDmitry Fomichev goto err; 9483b1a94c8SDamien Le Moal 9493b1a94c8SDamien Le Moal /* 9503b1a94c8SDamien Le Moal * The log is on disk. It is now safe to update in place 9513b1a94c8SDamien Le Moal * in the primary metadata set. 9523b1a94c8SDamien Le Moal */ 9533b1a94c8SDamien Le Moal ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary); 9543b1a94c8SDamien Le Moal if (ret) 955e7fad909SDmitry Fomichev goto err; 9563b1a94c8SDamien Le Moal 9573b1a94c8SDamien Le Moal ret = dmz_write_sb(zmd, zmd->mblk_primary); 9583b1a94c8SDamien Le Moal if (ret) 959e7fad909SDmitry Fomichev goto err; 9603b1a94c8SDamien Le Moal 9613b1a94c8SDamien Le Moal while (!list_empty(&write_list)) { 9623b1a94c8SDamien Le Moal mblk = list_first_entry(&write_list, struct dmz_mblock, link); 9633b1a94c8SDamien Le Moal list_del_init(&mblk->link); 9643b1a94c8SDamien Le Moal 9653b1a94c8SDamien Le Moal spin_lock(&zmd->mblk_lock); 9663b1a94c8SDamien Le Moal clear_bit(DMZ_META_DIRTY, &mblk->state); 96733c2865fSDamien Le Moal if (mblk->ref == 0) 9683b1a94c8SDamien Le Moal list_add_tail(&mblk->link, &zmd->mblk_lru_list); 9693b1a94c8SDamien Le Moal spin_unlock(&zmd->mblk_lock); 9703b1a94c8SDamien Le Moal } 9713b1a94c8SDamien Le Moal 9723b1a94c8SDamien Le Moal zmd->sb_gen++; 9733b1a94c8SDamien Le Moal out: 9743b1a94c8SDamien Le Moal dmz_unlock_flush(zmd); 9753b1a94c8SDamien Le Moal up_write(&zmd->mblk_sem); 9763b1a94c8SDamien Le Moal 9773b1a94c8SDamien Le Moal return ret; 978e7fad909SDmitry Fomichev 979e7fad909SDmitry Fomichev err: 980e7fad909SDmitry Fomichev if (!list_empty(&write_list)) { 981e7fad909SDmitry Fomichev spin_lock(&zmd->mblk_lock); 982e7fad909SDmitry Fomichev list_splice(&write_list, &zmd->mblk_dirty_list); 983e7fad909SDmitry Fomichev spin_unlock(&zmd->mblk_lock); 984e7fad909SDmitry Fomichev } 985bf28a3baSHannes Reinecke if (!dmz_check_bdev(dev)) 986e7fad909SDmitry Fomichev ret = -EIO; 987e7fad909SDmitry Fomichev goto out; 9883b1a94c8SDamien Le Moal } 9893b1a94c8SDamien Le Moal 9903b1a94c8SDamien Le Moal /* 9913b1a94c8SDamien Le Moal * Check super block. 9923b1a94c8SDamien Le Moal */ 9935d2c74f3SHannes Reinecke static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_sb *dsb, 9945d2c74f3SHannes Reinecke bool tertiary) 9953b1a94c8SDamien Le Moal { 9965d2c74f3SHannes Reinecke struct dmz_super *sb = dsb->sb; 9975d2c74f3SHannes Reinecke struct dmz_dev *dev = dsb->dev; 9983b1a94c8SDamien Le Moal unsigned int nr_meta_zones, nr_data_zones; 9993b1a94c8SDamien Le Moal u32 crc, stored_crc; 10003b1a94c8SDamien Le Moal u64 gen; 10013b1a94c8SDamien Le Moal 1002bd5c4031SHannes Reinecke if (le32_to_cpu(sb->magic) != DMZ_MAGIC) { 1003bd5c4031SHannes Reinecke dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)", 1004bd5c4031SHannes Reinecke DMZ_MAGIC, le32_to_cpu(sb->magic)); 1005bd5c4031SHannes Reinecke return -ENXIO; 1006bd5c4031SHannes Reinecke } 1007bd5c4031SHannes Reinecke 1008bd5c4031SHannes Reinecke zmd->sb_version = le32_to_cpu(sb->version); 1009bd5c4031SHannes Reinecke if (zmd->sb_version > DMZ_META_VER) { 1010bd5c4031SHannes Reinecke dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)", 1011bd5c4031SHannes Reinecke DMZ_META_VER, zmd->sb_version); 1012bd5c4031SHannes Reinecke return -EINVAL; 1013bd5c4031SHannes Reinecke } 10145d2c74f3SHannes Reinecke if (zmd->sb_version < 2 && tertiary) { 1015bd5c4031SHannes Reinecke dmz_dev_err(dev, "Tertiary superblocks are not supported"); 1016bd5c4031SHannes Reinecke return -EINVAL; 1017bd5c4031SHannes Reinecke } 1018bd5c4031SHannes Reinecke 10193b1a94c8SDamien Le Moal gen = le64_to_cpu(sb->gen); 10203b1a94c8SDamien Le Moal stored_crc = le32_to_cpu(sb->crc); 10213b1a94c8SDamien Le Moal sb->crc = 0; 10223b1a94c8SDamien Le Moal crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE); 10233b1a94c8SDamien Le Moal if (crc != stored_crc) { 10243b1a94c8SDamien Le Moal dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)", 10253b1a94c8SDamien Le Moal crc, stored_crc); 10263b1a94c8SDamien Le Moal return -ENXIO; 10273b1a94c8SDamien Le Moal } 10283b1a94c8SDamien Le Moal 1029bd5c4031SHannes Reinecke if (zmd->sb_version > 1) { 1030bd5c4031SHannes Reinecke uuid_t sb_uuid; 1031bd5c4031SHannes Reinecke 1032bd5c4031SHannes Reinecke import_uuid(&sb_uuid, sb->dmz_uuid); 1033bd5c4031SHannes Reinecke if (uuid_is_null(&sb_uuid)) { 1034bd5c4031SHannes Reinecke dmz_dev_err(dev, "NULL DM-Zoned uuid"); 1035bd5c4031SHannes Reinecke return -ENXIO; 1036bd5c4031SHannes Reinecke } else if (uuid_is_null(&zmd->uuid)) { 1037bd5c4031SHannes Reinecke uuid_copy(&zmd->uuid, &sb_uuid); 1038bd5c4031SHannes Reinecke } else if (!uuid_equal(&zmd->uuid, &sb_uuid)) { 1039bd5c4031SHannes Reinecke dmz_dev_err(dev, "mismatching DM-Zoned uuid, " 1040bd5c4031SHannes Reinecke "is %pUl expected %pUl", 1041bd5c4031SHannes Reinecke &sb_uuid, &zmd->uuid); 1042bd5c4031SHannes Reinecke return -ENXIO; 1043bd5c4031SHannes Reinecke } 1044bd5c4031SHannes Reinecke if (!strlen(zmd->label)) 1045bd5c4031SHannes Reinecke memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE); 1046bd5c4031SHannes Reinecke else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) { 1047bd5c4031SHannes Reinecke dmz_dev_err(dev, "mismatching DM-Zoned label, " 1048bd5c4031SHannes Reinecke "is %s expected %s", 1049bd5c4031SHannes Reinecke sb->dmz_label, zmd->label); 1050bd5c4031SHannes Reinecke return -ENXIO; 1051bd5c4031SHannes Reinecke } 1052bd5c4031SHannes Reinecke import_uuid(&dev->uuid, sb->dev_uuid); 1053bd5c4031SHannes Reinecke if (uuid_is_null(&dev->uuid)) { 1054bd5c4031SHannes Reinecke dmz_dev_err(dev, "NULL device uuid"); 10553b1a94c8SDamien Le Moal return -ENXIO; 10563b1a94c8SDamien Le Moal } 10573b1a94c8SDamien Le Moal 10585d2c74f3SHannes Reinecke if (tertiary) { 1059bd5c4031SHannes Reinecke /* 1060bd5c4031SHannes Reinecke * Generation number should be 0, but it doesn't 1061bd5c4031SHannes Reinecke * really matter if it isn't. 1062bd5c4031SHannes Reinecke */ 1063bd5c4031SHannes Reinecke if (gen != 0) 1064bd5c4031SHannes Reinecke dmz_dev_warn(dev, "Invalid generation %llu", 1065bd5c4031SHannes Reinecke gen); 1066bd5c4031SHannes Reinecke return 0; 1067bd5c4031SHannes Reinecke } 10683b1a94c8SDamien Le Moal } 10693b1a94c8SDamien Le Moal 107036820560SHannes Reinecke nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1) 107136820560SHannes Reinecke >> zmd->zone_nr_blocks_shift; 10723b1a94c8SDamien Le Moal if (!nr_meta_zones || 10733b1a94c8SDamien Le Moal nr_meta_zones >= zmd->nr_rnd_zones) { 10743b1a94c8SDamien Le Moal dmz_dev_err(dev, "Invalid number of metadata blocks"); 10753b1a94c8SDamien Le Moal return -ENXIO; 10763b1a94c8SDamien Le Moal } 10773b1a94c8SDamien Le Moal 10783b1a94c8SDamien Le Moal if (!le32_to_cpu(sb->nr_reserved_seq) || 10793b1a94c8SDamien Le Moal le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) { 10803b1a94c8SDamien Le Moal dmz_dev_err(dev, "Invalid number of reserved sequential zones"); 10813b1a94c8SDamien Le Moal return -ENXIO; 10823b1a94c8SDamien Le Moal } 10833b1a94c8SDamien Le Moal 10843b1a94c8SDamien Le Moal nr_data_zones = zmd->nr_useable_zones - 10853b1a94c8SDamien Le Moal (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq)); 10863b1a94c8SDamien Le Moal if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) { 10873b1a94c8SDamien Le Moal dmz_dev_err(dev, "Invalid number of chunks %u / %u", 10883b1a94c8SDamien Le Moal le32_to_cpu(sb->nr_chunks), nr_data_zones); 10893b1a94c8SDamien Le Moal return -ENXIO; 10903b1a94c8SDamien Le Moal } 10913b1a94c8SDamien Le Moal 10923b1a94c8SDamien Le Moal /* OK */ 10933b1a94c8SDamien Le Moal zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks); 10943b1a94c8SDamien Le Moal zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq); 10953b1a94c8SDamien Le Moal zmd->nr_chunks = le32_to_cpu(sb->nr_chunks); 10963b1a94c8SDamien Le Moal zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks); 10973b1a94c8SDamien Le Moal zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks); 10983b1a94c8SDamien Le Moal zmd->nr_meta_zones = nr_meta_zones; 10993b1a94c8SDamien Le Moal zmd->nr_data_zones = nr_data_zones; 11003b1a94c8SDamien Le Moal 11013b1a94c8SDamien Le Moal return 0; 11023b1a94c8SDamien Le Moal } 11033b1a94c8SDamien Le Moal 11043b1a94c8SDamien Le Moal /* 11053b1a94c8SDamien Le Moal * Read the first or second super block from disk. 11063b1a94c8SDamien Le Moal */ 11075d2c74f3SHannes Reinecke static int dmz_read_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set) 11083b1a94c8SDamien Le Moal { 110935d0c96eSHannes Reinecke dmz_zmd_debug(zmd, "read superblock set %d dev %s block %llu", 11105d2c74f3SHannes Reinecke set, sb->dev->name, sb->block); 111135d0c96eSHannes Reinecke 11125d2c74f3SHannes Reinecke return dmz_rdwr_block(sb->dev, REQ_OP_READ, 11135d2c74f3SHannes Reinecke sb->block, sb->mblk->page); 11143b1a94c8SDamien Le Moal } 11153b1a94c8SDamien Le Moal 11163b1a94c8SDamien Le Moal /* 11173b1a94c8SDamien Le Moal * Determine the position of the secondary super blocks on disk. 11183b1a94c8SDamien Le Moal * This is used only if a corruption of the primary super block 11193b1a94c8SDamien Le Moal * is detected. 11203b1a94c8SDamien Le Moal */ 11213b1a94c8SDamien Le Moal static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd) 11223b1a94c8SDamien Le Moal { 112336820560SHannes Reinecke unsigned int zone_nr_blocks = zmd->zone_nr_blocks; 11243b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 1125a92fbc44SHannes Reinecke unsigned int zone_id = zmd->sb[0].zone->id; 11263b1a94c8SDamien Le Moal int i; 11273b1a94c8SDamien Le Moal 11283b1a94c8SDamien Le Moal /* Allocate a block */ 11293b1a94c8SDamien Le Moal mblk = dmz_alloc_mblock(zmd, 0); 11303b1a94c8SDamien Le Moal if (!mblk) 11313b1a94c8SDamien Le Moal return -ENOMEM; 11323b1a94c8SDamien Le Moal 11333b1a94c8SDamien Le Moal zmd->sb[1].mblk = mblk; 11343b1a94c8SDamien Le Moal zmd->sb[1].sb = mblk->data; 11353b1a94c8SDamien Le Moal 11363b1a94c8SDamien Le Moal /* Bad first super block: search for the second one */ 11373b1a94c8SDamien Le Moal zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks; 1138a92fbc44SHannes Reinecke zmd->sb[1].zone = dmz_get(zmd, zone_id + 1); 11391565929bSHannes Reinecke zmd->sb[1].dev = zmd->sb[0].dev; 1140a92fbc44SHannes Reinecke for (i = 1; i < zmd->nr_rnd_zones; i++) { 11415d2c74f3SHannes Reinecke if (dmz_read_sb(zmd, &zmd->sb[1], 1) != 0) 11423b1a94c8SDamien Le Moal break; 1143a92fbc44SHannes Reinecke if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC) 11443b1a94c8SDamien Le Moal return 0; 11453b1a94c8SDamien Le Moal zmd->sb[1].block += zone_nr_blocks; 1146a92fbc44SHannes Reinecke zmd->sb[1].zone = dmz_get(zmd, zone_id + i); 11473b1a94c8SDamien Le Moal } 11483b1a94c8SDamien Le Moal 11493b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 11503b1a94c8SDamien Le Moal zmd->sb[1].mblk = NULL; 1151735bd7e4SHannes Reinecke zmd->sb[1].zone = NULL; 1152bf28a3baSHannes Reinecke zmd->sb[1].dev = NULL; 11533b1a94c8SDamien Le Moal 11543b1a94c8SDamien Le Moal return -EIO; 11553b1a94c8SDamien Le Moal } 11563b1a94c8SDamien Le Moal 11573b1a94c8SDamien Le Moal /* 11585d2c74f3SHannes Reinecke * Read a super block from disk. 11593b1a94c8SDamien Le Moal */ 11605d2c74f3SHannes Reinecke static int dmz_get_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set) 11613b1a94c8SDamien Le Moal { 11623b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 11633b1a94c8SDamien Le Moal int ret; 11643b1a94c8SDamien Le Moal 11653b1a94c8SDamien Le Moal /* Allocate a block */ 11663b1a94c8SDamien Le Moal mblk = dmz_alloc_mblock(zmd, 0); 11673b1a94c8SDamien Le Moal if (!mblk) 11683b1a94c8SDamien Le Moal return -ENOMEM; 11693b1a94c8SDamien Le Moal 11705d2c74f3SHannes Reinecke sb->mblk = mblk; 11715d2c74f3SHannes Reinecke sb->sb = mblk->data; 11723b1a94c8SDamien Le Moal 11733b1a94c8SDamien Le Moal /* Read super block */ 11745d2c74f3SHannes Reinecke ret = dmz_read_sb(zmd, sb, set); 11753b1a94c8SDamien Le Moal if (ret) { 11763b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 11775d2c74f3SHannes Reinecke sb->mblk = NULL; 11783b1a94c8SDamien Le Moal return ret; 11793b1a94c8SDamien Le Moal } 11803b1a94c8SDamien Le Moal 11813b1a94c8SDamien Le Moal return 0; 11823b1a94c8SDamien Le Moal } 11833b1a94c8SDamien Le Moal 11843b1a94c8SDamien Le Moal /* 11853b1a94c8SDamien Le Moal * Recover a metadata set. 11863b1a94c8SDamien Le Moal */ 11873b1a94c8SDamien Le Moal static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set) 11883b1a94c8SDamien Le Moal { 11893b1a94c8SDamien Le Moal unsigned int src_set = dst_set ^ 0x1; 11903b1a94c8SDamien Le Moal struct page *page; 11913b1a94c8SDamien Le Moal int i, ret; 11923b1a94c8SDamien Le Moal 1193bf28a3baSHannes Reinecke dmz_dev_warn(zmd->sb[dst_set].dev, 1194bf28a3baSHannes Reinecke "Metadata set %u invalid: recovering", dst_set); 11953b1a94c8SDamien Le Moal 11963b1a94c8SDamien Le Moal if (dst_set == 0) 1197735bd7e4SHannes Reinecke zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone); 1198735bd7e4SHannes Reinecke else 1199735bd7e4SHannes Reinecke zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone); 12003b1a94c8SDamien Le Moal 12014218a955SDamien Le Moal page = alloc_page(GFP_NOIO); 12023b1a94c8SDamien Le Moal if (!page) 12033b1a94c8SDamien Le Moal return -ENOMEM; 12043b1a94c8SDamien Le Moal 12053b1a94c8SDamien Le Moal /* Copy metadata blocks */ 12063b1a94c8SDamien Le Moal for (i = 1; i < zmd->nr_meta_blocks; i++) { 1207bf28a3baSHannes Reinecke ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ, 12083b1a94c8SDamien Le Moal zmd->sb[src_set].block + i, page); 12093b1a94c8SDamien Le Moal if (ret) 12103b1a94c8SDamien Le Moal goto out; 1211bf28a3baSHannes Reinecke ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE, 12123b1a94c8SDamien Le Moal zmd->sb[dst_set].block + i, page); 12133b1a94c8SDamien Le Moal if (ret) 12143b1a94c8SDamien Le Moal goto out; 12153b1a94c8SDamien Le Moal } 12163b1a94c8SDamien Le Moal 12173b1a94c8SDamien Le Moal /* Finalize with the super block */ 12183b1a94c8SDamien Le Moal if (!zmd->sb[dst_set].mblk) { 12193b1a94c8SDamien Le Moal zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0); 12203b1a94c8SDamien Le Moal if (!zmd->sb[dst_set].mblk) { 12213b1a94c8SDamien Le Moal ret = -ENOMEM; 12223b1a94c8SDamien Le Moal goto out; 12233b1a94c8SDamien Le Moal } 12243b1a94c8SDamien Le Moal zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data; 12253b1a94c8SDamien Le Moal } 12263b1a94c8SDamien Le Moal 12273b1a94c8SDamien Le Moal ret = dmz_write_sb(zmd, dst_set); 12283b1a94c8SDamien Le Moal out: 12293b1a94c8SDamien Le Moal __free_pages(page, 0); 12303b1a94c8SDamien Le Moal 12313b1a94c8SDamien Le Moal return ret; 12323b1a94c8SDamien Le Moal } 12333b1a94c8SDamien Le Moal 12343b1a94c8SDamien Le Moal /* 12353b1a94c8SDamien Le Moal * Get super block from disk. 12363b1a94c8SDamien Le Moal */ 12373b1a94c8SDamien Le Moal static int dmz_load_sb(struct dmz_metadata *zmd) 12383b1a94c8SDamien Le Moal { 12393b1a94c8SDamien Le Moal bool sb_good[2] = {false, false}; 12403b1a94c8SDamien Le Moal u64 sb_gen[2] = {0, 0}; 12413b1a94c8SDamien Le Moal int ret; 12423b1a94c8SDamien Le Moal 1243735bd7e4SHannes Reinecke if (!zmd->sb[0].zone) { 1244ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Primary super block zone not set"); 1245735bd7e4SHannes Reinecke return -ENXIO; 1246735bd7e4SHannes Reinecke } 1247735bd7e4SHannes Reinecke 12483b1a94c8SDamien Le Moal /* Read and check the primary super block */ 1249735bd7e4SHannes Reinecke zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone); 12508f22272aSHannes Reinecke zmd->sb[0].dev = zmd->sb[0].zone->dev; 12515d2c74f3SHannes Reinecke ret = dmz_get_sb(zmd, &zmd->sb[0], 0); 12523b1a94c8SDamien Le Moal if (ret) { 1253bf28a3baSHannes Reinecke dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed"); 12543b1a94c8SDamien Le Moal return ret; 12553b1a94c8SDamien Le Moal } 12563b1a94c8SDamien Le Moal 12575d2c74f3SHannes Reinecke ret = dmz_check_sb(zmd, &zmd->sb[0], false); 12583b1a94c8SDamien Le Moal 12593b1a94c8SDamien Le Moal /* Read and check secondary super block */ 12603b1a94c8SDamien Le Moal if (ret == 0) { 12613b1a94c8SDamien Le Moal sb_good[0] = true; 1262a92fbc44SHannes Reinecke if (!zmd->sb[1].zone) { 1263a92fbc44SHannes Reinecke unsigned int zone_id = 1264a92fbc44SHannes Reinecke zmd->sb[0].zone->id + zmd->nr_meta_zones; 1265a92fbc44SHannes Reinecke 1266a92fbc44SHannes Reinecke zmd->sb[1].zone = dmz_get(zmd, zone_id); 1267a92fbc44SHannes Reinecke } 1268735bd7e4SHannes Reinecke zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone); 12691565929bSHannes Reinecke zmd->sb[1].dev = zmd->sb[0].dev; 12705d2c74f3SHannes Reinecke ret = dmz_get_sb(zmd, &zmd->sb[1], 1); 12713b1a94c8SDamien Le Moal } else 12723b1a94c8SDamien Le Moal ret = dmz_lookup_secondary_sb(zmd); 12733b1a94c8SDamien Le Moal 12743b1a94c8SDamien Le Moal if (ret) { 1275bf28a3baSHannes Reinecke dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed"); 12763b1a94c8SDamien Le Moal return ret; 12773b1a94c8SDamien Le Moal } 12783b1a94c8SDamien Le Moal 12795d2c74f3SHannes Reinecke ret = dmz_check_sb(zmd, &zmd->sb[1], false); 12803b1a94c8SDamien Le Moal if (ret == 0) 12813b1a94c8SDamien Le Moal sb_good[1] = true; 12823b1a94c8SDamien Le Moal 12833b1a94c8SDamien Le Moal /* Use highest generation sb first */ 12843b1a94c8SDamien Le Moal if (!sb_good[0] && !sb_good[1]) { 1285ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "No valid super block found"); 12863b1a94c8SDamien Le Moal return -EIO; 12873b1a94c8SDamien Le Moal } 12883b1a94c8SDamien Le Moal 12893b1a94c8SDamien Le Moal if (sb_good[0]) 12903b1a94c8SDamien Le Moal sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen); 1291bf28a3baSHannes Reinecke else { 12923b1a94c8SDamien Le Moal ret = dmz_recover_mblocks(zmd, 0); 1293bf28a3baSHannes Reinecke if (ret) { 1294bf28a3baSHannes Reinecke dmz_dev_err(zmd->sb[0].dev, 1295bf28a3baSHannes Reinecke "Recovery of superblock 0 failed"); 1296bf28a3baSHannes Reinecke return -EIO; 1297bf28a3baSHannes Reinecke } 1298bf28a3baSHannes Reinecke } 12993b1a94c8SDamien Le Moal 13003b1a94c8SDamien Le Moal if (sb_good[1]) 13013b1a94c8SDamien Le Moal sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen); 1302bf28a3baSHannes Reinecke else { 13033b1a94c8SDamien Le Moal ret = dmz_recover_mblocks(zmd, 1); 13043b1a94c8SDamien Le Moal 13053b1a94c8SDamien Le Moal if (ret) { 1306bf28a3baSHannes Reinecke dmz_dev_err(zmd->sb[1].dev, 1307bf28a3baSHannes Reinecke "Recovery of superblock 1 failed"); 13083b1a94c8SDamien Le Moal return -EIO; 13093b1a94c8SDamien Le Moal } 1310bf28a3baSHannes Reinecke } 13113b1a94c8SDamien Le Moal 13123b1a94c8SDamien Le Moal if (sb_gen[0] >= sb_gen[1]) { 13133b1a94c8SDamien Le Moal zmd->sb_gen = sb_gen[0]; 13143b1a94c8SDamien Le Moal zmd->mblk_primary = 0; 13153b1a94c8SDamien Le Moal } else { 13163b1a94c8SDamien Le Moal zmd->sb_gen = sb_gen[1]; 13173b1a94c8SDamien Le Moal zmd->mblk_primary = 1; 13183b1a94c8SDamien Le Moal } 13193b1a94c8SDamien Le Moal 1320bf28a3baSHannes Reinecke dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev, 1321bf28a3baSHannes Reinecke "Using super block %u (gen %llu)", 13223b1a94c8SDamien Le Moal zmd->mblk_primary, zmd->sb_gen); 13233b1a94c8SDamien Le Moal 13245d2c74f3SHannes Reinecke if (zmd->sb_version > 1) { 13255d2c74f3SHannes Reinecke int i; 13265d2c74f3SHannes Reinecke struct dmz_sb *sb; 13275d2c74f3SHannes Reinecke 13285d2c74f3SHannes Reinecke sb = kzalloc(sizeof(struct dmz_sb), GFP_KERNEL); 13295d2c74f3SHannes Reinecke if (!sb) 13305d2c74f3SHannes Reinecke return -ENOMEM; 13315d2c74f3SHannes Reinecke for (i = 1; i < zmd->nr_devs; i++) { 13325d2c74f3SHannes Reinecke sb->block = 0; 13335d2c74f3SHannes Reinecke sb->zone = dmz_get(zmd, zmd->dev[i].zone_offset); 13345d2c74f3SHannes Reinecke sb->dev = &zmd->dev[i]; 13355d2c74f3SHannes Reinecke if (!dmz_is_meta(sb->zone)) { 13365d2c74f3SHannes Reinecke dmz_dev_err(sb->dev, 13375d2c74f3SHannes Reinecke "Tertiary super block zone %u not marked as metadata zone", 13385d2c74f3SHannes Reinecke sb->zone->id); 13395d2c74f3SHannes Reinecke ret = -EINVAL; 13405d2c74f3SHannes Reinecke goto out_kfree; 13415d2c74f3SHannes Reinecke } 13425d2c74f3SHannes Reinecke ret = dmz_get_sb(zmd, sb, i + 1); 1343bd5c4031SHannes Reinecke if (ret) { 13445d2c74f3SHannes Reinecke dmz_dev_err(sb->dev, 1345bd5c4031SHannes Reinecke "Read tertiary super block failed"); 13465d2c74f3SHannes Reinecke dmz_free_mblock(zmd, sb->mblk); 13475d2c74f3SHannes Reinecke goto out_kfree; 1348bd5c4031SHannes Reinecke } 13495d2c74f3SHannes Reinecke ret = dmz_check_sb(zmd, sb, true); 13505d2c74f3SHannes Reinecke dmz_free_mblock(zmd, sb->mblk); 1351bd5c4031SHannes Reinecke if (ret == -EINVAL) 13525d2c74f3SHannes Reinecke goto out_kfree; 1353bd5c4031SHannes Reinecke } 13545d2c74f3SHannes Reinecke out_kfree: 13555d2c74f3SHannes Reinecke kfree(sb); 13565d2c74f3SHannes Reinecke } 13575d2c74f3SHannes Reinecke return ret; 13583b1a94c8SDamien Le Moal } 13593b1a94c8SDamien Le Moal 13603b1a94c8SDamien Le Moal /* 13613b1a94c8SDamien Le Moal * Initialize a zone descriptor. 13623b1a94c8SDamien Le Moal */ 1363bd5c4031SHannes Reinecke static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data) 13643b1a94c8SDamien Le Moal { 136518979819SHannes Reinecke struct dmz_dev *dev = data; 136618979819SHannes Reinecke struct dmz_metadata *zmd = dev->metadata; 1367bd5c4031SHannes Reinecke int idx = num + dev->zone_offset; 1368a92fbc44SHannes Reinecke struct dm_zone *zone; 1369a92fbc44SHannes Reinecke 13708f22272aSHannes Reinecke zone = dmz_insert(zmd, idx, dev); 1371a92fbc44SHannes Reinecke if (IS_ERR(zone)) 1372a92fbc44SHannes Reinecke return PTR_ERR(zone); 13733b1a94c8SDamien Le Moal 137436820560SHannes Reinecke if (blkz->len != zmd->zone_nr_sectors) { 1375bd5c4031SHannes Reinecke if (zmd->sb_version > 1) { 1376bd5c4031SHannes Reinecke /* Ignore the eventual runt (smaller) zone */ 1377bd5c4031SHannes Reinecke set_bit(DMZ_OFFLINE, &zone->flags); 1378bd5c4031SHannes Reinecke return 0; 1379bd5c4031SHannes Reinecke } else if (blkz->start + blkz->len == dev->capacity) 13803b1a94c8SDamien Le Moal return 0; 13813b1a94c8SDamien Le Moal return -ENXIO; 13823b1a94c8SDamien Le Moal } 13833b1a94c8SDamien Le Moal 1384d4100351SChristoph Hellwig switch (blkz->type) { 1385d4100351SChristoph Hellwig case BLK_ZONE_TYPE_CONVENTIONAL: 13863b1a94c8SDamien Le Moal set_bit(DMZ_RND, &zone->flags); 1387d4100351SChristoph Hellwig break; 1388d4100351SChristoph Hellwig case BLK_ZONE_TYPE_SEQWRITE_REQ: 1389d4100351SChristoph Hellwig case BLK_ZONE_TYPE_SEQWRITE_PREF: 13903b1a94c8SDamien Le Moal set_bit(DMZ_SEQ, &zone->flags); 1391d4100351SChristoph Hellwig break; 1392d4100351SChristoph Hellwig default: 13933b1a94c8SDamien Le Moal return -ENXIO; 1394d4100351SChristoph Hellwig } 13953b1a94c8SDamien Le Moal 13963b1a94c8SDamien Le Moal if (dmz_is_rnd(zone)) 13973b1a94c8SDamien Le Moal zone->wp_block = 0; 13983b1a94c8SDamien Le Moal else 13993b1a94c8SDamien Le Moal zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start); 14003b1a94c8SDamien Le Moal 1401d4100351SChristoph Hellwig if (blkz->cond == BLK_ZONE_COND_OFFLINE) 1402d4100351SChristoph Hellwig set_bit(DMZ_OFFLINE, &zone->flags); 1403d4100351SChristoph Hellwig else if (blkz->cond == BLK_ZONE_COND_READONLY) 1404d4100351SChristoph Hellwig set_bit(DMZ_READ_ONLY, &zone->flags); 1405d4100351SChristoph Hellwig else { 14063b1a94c8SDamien Le Moal zmd->nr_useable_zones++; 14073b1a94c8SDamien Le Moal if (dmz_is_rnd(zone)) { 14083b1a94c8SDamien Le Moal zmd->nr_rnd_zones++; 1409bd5c4031SHannes Reinecke if (zmd->nr_devs == 1 && !zmd->sb[0].zone) { 1410bd5c4031SHannes Reinecke /* Primary super block zone */ 1411735bd7e4SHannes Reinecke zmd->sb[0].zone = zone; 14123b1a94c8SDamien Le Moal } 14133b1a94c8SDamien Le Moal } 14145d2c74f3SHannes Reinecke if (zmd->nr_devs > 1 && num == 0) { 14155d2c74f3SHannes Reinecke /* 14165d2c74f3SHannes Reinecke * Tertiary superblock zones are always at the 14175d2c74f3SHannes Reinecke * start of the zoned devices, so mark them 14185d2c74f3SHannes Reinecke * as metadata zone. 14195d2c74f3SHannes Reinecke */ 14205d2c74f3SHannes Reinecke set_bit(DMZ_META, &zone->flags); 1421bd5c4031SHannes Reinecke } 14223b1a94c8SDamien Le Moal } 14233b1a94c8SDamien Le Moal return 0; 14243b1a94c8SDamien Le Moal } 14253b1a94c8SDamien Le Moal 1426a92fbc44SHannes Reinecke static int dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev) 1427bd5c4031SHannes Reinecke { 1428bd5c4031SHannes Reinecke int idx; 1429bd5c4031SHannes Reinecke sector_t zone_offset = 0; 1430bd5c4031SHannes Reinecke 1431bd5c4031SHannes Reinecke for(idx = 0; idx < dev->nr_zones; idx++) { 1432a92fbc44SHannes Reinecke struct dm_zone *zone; 1433bd5c4031SHannes Reinecke 14348f22272aSHannes Reinecke zone = dmz_insert(zmd, idx, dev); 1435a92fbc44SHannes Reinecke if (IS_ERR(zone)) 1436a92fbc44SHannes Reinecke return PTR_ERR(zone); 143734f5affdSHannes Reinecke set_bit(DMZ_CACHE, &zone->flags); 1438bd5c4031SHannes Reinecke zone->wp_block = 0; 143934f5affdSHannes Reinecke zmd->nr_cache_zones++; 1440bd5c4031SHannes Reinecke zmd->nr_useable_zones++; 1441bd5c4031SHannes Reinecke if (dev->capacity - zone_offset < zmd->zone_nr_sectors) { 1442bd5c4031SHannes Reinecke /* Disable runt zone */ 1443bd5c4031SHannes Reinecke set_bit(DMZ_OFFLINE, &zone->flags); 1444bd5c4031SHannes Reinecke break; 1445bd5c4031SHannes Reinecke } 1446bd5c4031SHannes Reinecke zone_offset += zmd->zone_nr_sectors; 1447bd5c4031SHannes Reinecke } 1448a92fbc44SHannes Reinecke return 0; 1449bd5c4031SHannes Reinecke } 1450bd5c4031SHannes Reinecke 14513b1a94c8SDamien Le Moal /* 14523b1a94c8SDamien Le Moal * Free zones descriptors. 14533b1a94c8SDamien Le Moal */ 14543b1a94c8SDamien Le Moal static void dmz_drop_zones(struct dmz_metadata *zmd) 14553b1a94c8SDamien Le Moal { 1456a92fbc44SHannes Reinecke int idx; 1457a92fbc44SHannes Reinecke 1458a92fbc44SHannes Reinecke for(idx = 0; idx < zmd->nr_zones; idx++) { 1459a92fbc44SHannes Reinecke struct dm_zone *zone = xa_load(&zmd->zones, idx); 1460a92fbc44SHannes Reinecke 1461a92fbc44SHannes Reinecke kfree(zone); 1462a92fbc44SHannes Reinecke xa_erase(&zmd->zones, idx); 1463a92fbc44SHannes Reinecke } 1464a92fbc44SHannes Reinecke xa_destroy(&zmd->zones); 14653b1a94c8SDamien Le Moal } 14663b1a94c8SDamien Le Moal 14673b1a94c8SDamien Le Moal /* 14683b1a94c8SDamien Le Moal * Allocate and initialize zone descriptors using the zone 14693b1a94c8SDamien Le Moal * information from disk. 14703b1a94c8SDamien Le Moal */ 14713b1a94c8SDamien Le Moal static int dmz_init_zones(struct dmz_metadata *zmd) 14723b1a94c8SDamien Le Moal { 1473bd5c4031SHannes Reinecke int i, ret; 1474bd5c4031SHannes Reinecke struct dmz_dev *zoned_dev = &zmd->dev[0]; 14753b1a94c8SDamien Le Moal 14763b1a94c8SDamien Le Moal /* Init */ 1477bd5c4031SHannes Reinecke zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors; 147836820560SHannes Reinecke zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors); 147936820560SHannes Reinecke zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors); 148036820560SHannes Reinecke zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks); 148136820560SHannes Reinecke zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3; 1482b3996295SDmitry Fomichev zmd->zone_nr_bitmap_blocks = 1483b3996295SDmitry Fomichev max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT); 148436820560SHannes Reinecke zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks, 1485b3996295SDmitry Fomichev DMZ_BLOCK_SIZE_BITS); 14863b1a94c8SDamien Le Moal 14873b1a94c8SDamien Le Moal /* Allocate zone array */ 1488bd5c4031SHannes Reinecke zmd->nr_zones = 0; 148918979819SHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) { 149018979819SHannes Reinecke struct dmz_dev *dev = &zmd->dev[i]; 149118979819SHannes Reinecke 149218979819SHannes Reinecke dev->metadata = zmd; 149318979819SHannes Reinecke zmd->nr_zones += dev->nr_zones; 1494bd82fdabSHannes Reinecke 1495bd82fdabSHannes Reinecke atomic_set(&dev->unmap_nr_rnd, 0); 1496bd82fdabSHannes Reinecke INIT_LIST_HEAD(&dev->unmap_rnd_list); 1497bd82fdabSHannes Reinecke INIT_LIST_HEAD(&dev->map_rnd_list); 1498bd82fdabSHannes Reinecke 1499bd82fdabSHannes Reinecke atomic_set(&dev->unmap_nr_seq, 0); 1500bd82fdabSHannes Reinecke INIT_LIST_HEAD(&dev->unmap_seq_list); 1501bd82fdabSHannes Reinecke INIT_LIST_HEAD(&dev->map_seq_list); 150218979819SHannes Reinecke } 1503bd5c4031SHannes Reinecke 1504bd5c4031SHannes Reinecke if (!zmd->nr_zones) { 1505bd5c4031SHannes Reinecke DMERR("(%s): No zones found", zmd->devname); 1506bd5c4031SHannes Reinecke return -ENXIO; 1507bd5c4031SHannes Reinecke } 1508a92fbc44SHannes Reinecke xa_init(&zmd->zones); 15093b1a94c8SDamien Le Moal 1510ae3c1f11SHannes Reinecke DMDEBUG("(%s): Using %zu B for zone information", 1511ca1a7045SHannes Reinecke zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones); 15123b1a94c8SDamien Le Moal 1513bd5c4031SHannes Reinecke if (zmd->nr_devs > 1) { 1514a92fbc44SHannes Reinecke ret = dmz_emulate_zones(zmd, &zmd->dev[0]); 1515a92fbc44SHannes Reinecke if (ret < 0) { 1516a92fbc44SHannes Reinecke DMDEBUG("(%s): Failed to emulate zones, error %d", 1517a92fbc44SHannes Reinecke zmd->devname, ret); 1518a92fbc44SHannes Reinecke dmz_drop_zones(zmd); 1519a92fbc44SHannes Reinecke return ret; 1520a92fbc44SHannes Reinecke } 1521a92fbc44SHannes Reinecke 1522bd5c4031SHannes Reinecke /* 1523bd5c4031SHannes Reinecke * Primary superblock zone is always at zone 0 when multiple 1524bd5c4031SHannes Reinecke * drives are present. 1525bd5c4031SHannes Reinecke */ 1526a92fbc44SHannes Reinecke zmd->sb[0].zone = dmz_get(zmd, 0); 1527bd5c4031SHannes Reinecke 15284dba1288SHannes Reinecke for (i = 1; i < zmd->nr_devs; i++) { 15294dba1288SHannes Reinecke zoned_dev = &zmd->dev[i]; 15304dba1288SHannes Reinecke 15314dba1288SHannes Reinecke ret = blkdev_report_zones(zoned_dev->bdev, 0, 15324dba1288SHannes Reinecke BLK_ALL_ZONES, 15334dba1288SHannes Reinecke dmz_init_zone, zoned_dev); 15344dba1288SHannes Reinecke if (ret < 0) { 15354dba1288SHannes Reinecke DMDEBUG("(%s): Failed to report zones, error %d", 15364dba1288SHannes Reinecke zmd->devname, ret); 15374dba1288SHannes Reinecke dmz_drop_zones(zmd); 15384dba1288SHannes Reinecke return ret; 15394dba1288SHannes Reinecke } 15404dba1288SHannes Reinecke } 15414dba1288SHannes Reinecke return 0; 1542bd5c4031SHannes Reinecke } 1543bd5c4031SHannes Reinecke 15443b1a94c8SDamien Le Moal /* 1545d4100351SChristoph Hellwig * Get zone information and initialize zone descriptors. At the same 1546d4100351SChristoph Hellwig * time, determine where the super block should be: first block of the 1547d4100351SChristoph Hellwig * first randomly writable zone. 15483b1a94c8SDamien Le Moal */ 1549bd5c4031SHannes Reinecke ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES, 155018979819SHannes Reinecke dmz_init_zone, zoned_dev); 1551d4100351SChristoph Hellwig if (ret < 0) { 1552bd5c4031SHannes Reinecke DMDEBUG("(%s): Failed to report zones, error %d", 1553bd5c4031SHannes Reinecke zmd->devname, ret); 15543b1a94c8SDamien Le Moal dmz_drop_zones(zmd); 15553b1a94c8SDamien Le Moal return ret; 15563b1a94c8SDamien Le Moal } 15573b1a94c8SDamien Le Moal 1558d4100351SChristoph Hellwig return 0; 1559d4100351SChristoph Hellwig } 1560d4100351SChristoph Hellwig 1561d4100351SChristoph Hellwig static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx, 1562d4100351SChristoph Hellwig void *data) 1563d4100351SChristoph Hellwig { 1564d4100351SChristoph Hellwig struct dm_zone *zone = data; 1565d4100351SChristoph Hellwig 1566d4100351SChristoph Hellwig clear_bit(DMZ_OFFLINE, &zone->flags); 1567d4100351SChristoph Hellwig clear_bit(DMZ_READ_ONLY, &zone->flags); 1568d4100351SChristoph Hellwig if (blkz->cond == BLK_ZONE_COND_OFFLINE) 1569d4100351SChristoph Hellwig set_bit(DMZ_OFFLINE, &zone->flags); 1570d4100351SChristoph Hellwig else if (blkz->cond == BLK_ZONE_COND_READONLY) 1571d4100351SChristoph Hellwig set_bit(DMZ_READ_ONLY, &zone->flags); 1572d4100351SChristoph Hellwig 1573d4100351SChristoph Hellwig if (dmz_is_seq(zone)) 1574d4100351SChristoph Hellwig zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start); 1575d4100351SChristoph Hellwig else 1576d4100351SChristoph Hellwig zone->wp_block = 0; 1577d4100351SChristoph Hellwig return 0; 1578d4100351SChristoph Hellwig } 1579d4100351SChristoph Hellwig 15803b1a94c8SDamien Le Moal /* 15813b1a94c8SDamien Le Moal * Update a zone information. 15823b1a94c8SDamien Le Moal */ 15833b1a94c8SDamien Le Moal static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 15843b1a94c8SDamien Le Moal { 15858f22272aSHannes Reinecke struct dmz_dev *dev = zone->dev; 1586bd976e52SDamien Le Moal unsigned int noio_flag; 15873b1a94c8SDamien Le Moal int ret; 15883b1a94c8SDamien Le Moal 1589bd5c4031SHannes Reinecke if (dev->flags & DMZ_BDEV_REGULAR) 1590bd5c4031SHannes Reinecke return 0; 1591bd5c4031SHannes Reinecke 1592bd976e52SDamien Le Moal /* 1593bd976e52SDamien Le Moal * Get zone information from disk. Since blkdev_report_zones() uses 1594bd976e52SDamien Le Moal * GFP_KERNEL by default for memory allocations, set the per-task 1595bd976e52SDamien Le Moal * PF_MEMALLOC_NOIO flag so that all allocations are done as if 1596bd976e52SDamien Le Moal * GFP_NOIO was specified. 1597bd976e52SDamien Le Moal */ 1598bd976e52SDamien Le Moal noio_flag = memalloc_noio_save(); 1599aa821c8dSHannes Reinecke ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1, 1600d4100351SChristoph Hellwig dmz_update_zone_cb, zone); 1601bd976e52SDamien Le Moal memalloc_noio_restore(noio_flag); 1602d4100351SChristoph Hellwig 1603d4100351SChristoph Hellwig if (ret == 0) 16047aedf75fSDamien Le Moal ret = -EIO; 1605d4100351SChristoph Hellwig if (ret < 0) { 1606aa821c8dSHannes Reinecke dmz_dev_err(dev, "Get zone %u report failed", 1607b7122873SHannes Reinecke zone->id); 1608aa821c8dSHannes Reinecke dmz_check_bdev(dev); 16093b1a94c8SDamien Le Moal return ret; 16103b1a94c8SDamien Le Moal } 16113b1a94c8SDamien Le Moal 16123b1a94c8SDamien Le Moal return 0; 16133b1a94c8SDamien Le Moal } 16143b1a94c8SDamien Le Moal 16153b1a94c8SDamien Le Moal /* 16163b1a94c8SDamien Le Moal * Check a zone write pointer position when the zone is marked 16173b1a94c8SDamien Le Moal * with the sequential write error flag. 16183b1a94c8SDamien Le Moal */ 16193b1a94c8SDamien Le Moal static int dmz_handle_seq_write_err(struct dmz_metadata *zmd, 16203b1a94c8SDamien Le Moal struct dm_zone *zone) 16213b1a94c8SDamien Le Moal { 16228f22272aSHannes Reinecke struct dmz_dev *dev = zone->dev; 16233b1a94c8SDamien Le Moal unsigned int wp = 0; 16243b1a94c8SDamien Le Moal int ret; 16253b1a94c8SDamien Le Moal 16263b1a94c8SDamien Le Moal wp = zone->wp_block; 16273b1a94c8SDamien Le Moal ret = dmz_update_zone(zmd, zone); 16283b1a94c8SDamien Le Moal if (ret) 16293b1a94c8SDamien Le Moal return ret; 16303b1a94c8SDamien Le Moal 1631aa821c8dSHannes Reinecke dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)", 1632b7122873SHannes Reinecke zone->id, zone->wp_block, wp); 16333b1a94c8SDamien Le Moal 16343b1a94c8SDamien Le Moal if (zone->wp_block < wp) { 16353b1a94c8SDamien Le Moal dmz_invalidate_blocks(zmd, zone, zone->wp_block, 16363b1a94c8SDamien Le Moal wp - zone->wp_block); 16373b1a94c8SDamien Le Moal } 16383b1a94c8SDamien Le Moal 16393b1a94c8SDamien Le Moal return 0; 16403b1a94c8SDamien Le Moal } 16413b1a94c8SDamien Le Moal 16423b1a94c8SDamien Le Moal /* 16433b1a94c8SDamien Le Moal * Reset a zone write pointer. 16443b1a94c8SDamien Le Moal */ 16453b1a94c8SDamien Le Moal static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 16463b1a94c8SDamien Le Moal { 16473b1a94c8SDamien Le Moal int ret; 16483b1a94c8SDamien Le Moal 16493b1a94c8SDamien Le Moal /* 16503b1a94c8SDamien Le Moal * Ignore offline zones, read only zones, 16513b1a94c8SDamien Le Moal * and conventional zones. 16523b1a94c8SDamien Le Moal */ 16533b1a94c8SDamien Le Moal if (dmz_is_offline(zone) || 16543b1a94c8SDamien Le Moal dmz_is_readonly(zone) || 16553b1a94c8SDamien Le Moal dmz_is_rnd(zone)) 16563b1a94c8SDamien Le Moal return 0; 16573b1a94c8SDamien Le Moal 16583b1a94c8SDamien Le Moal if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) { 16598f22272aSHannes Reinecke struct dmz_dev *dev = zone->dev; 16603b1a94c8SDamien Le Moal 16616c1b1da5SAjay Joshi ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET, 16623b1a94c8SDamien Le Moal dmz_start_sect(zmd, zone), 166336820560SHannes Reinecke zmd->zone_nr_sectors, GFP_NOIO); 16643b1a94c8SDamien Le Moal if (ret) { 16653b1a94c8SDamien Le Moal dmz_dev_err(dev, "Reset zone %u failed %d", 1666b7122873SHannes Reinecke zone->id, ret); 16673b1a94c8SDamien Le Moal return ret; 16683b1a94c8SDamien Le Moal } 16693b1a94c8SDamien Le Moal } 16703b1a94c8SDamien Le Moal 16713b1a94c8SDamien Le Moal /* Clear write error bit and rewind write pointer position */ 16723b1a94c8SDamien Le Moal clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags); 16733b1a94c8SDamien Le Moal zone->wp_block = 0; 16743b1a94c8SDamien Le Moal 16753b1a94c8SDamien Le Moal return 0; 16763b1a94c8SDamien Le Moal } 16773b1a94c8SDamien Le Moal 16783b1a94c8SDamien Le Moal static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone); 16793b1a94c8SDamien Le Moal 16803b1a94c8SDamien Le Moal /* 16813b1a94c8SDamien Le Moal * Initialize chunk mapping. 16823b1a94c8SDamien Le Moal */ 16833b1a94c8SDamien Le Moal static int dmz_load_mapping(struct dmz_metadata *zmd) 16843b1a94c8SDamien Le Moal { 16853b1a94c8SDamien Le Moal struct dm_zone *dzone, *bzone; 16863b1a94c8SDamien Le Moal struct dmz_mblock *dmap_mblk = NULL; 16873b1a94c8SDamien Le Moal struct dmz_map *dmap; 16883b1a94c8SDamien Le Moal unsigned int i = 0, e = 0, chunk = 0; 16893b1a94c8SDamien Le Moal unsigned int dzone_id; 16903b1a94c8SDamien Le Moal unsigned int bzone_id; 16913b1a94c8SDamien Le Moal 16923b1a94c8SDamien Le Moal /* Metadata block array for the chunk mapping table */ 16933b1a94c8SDamien Le Moal zmd->map_mblk = kcalloc(zmd->nr_map_blocks, 16943b1a94c8SDamien Le Moal sizeof(struct dmz_mblk *), GFP_KERNEL); 16953b1a94c8SDamien Le Moal if (!zmd->map_mblk) 16963b1a94c8SDamien Le Moal return -ENOMEM; 16973b1a94c8SDamien Le Moal 16983b1a94c8SDamien Le Moal /* Get chunk mapping table blocks and initialize zone mapping */ 16993b1a94c8SDamien Le Moal while (chunk < zmd->nr_chunks) { 17003b1a94c8SDamien Le Moal if (!dmap_mblk) { 17013b1a94c8SDamien Le Moal /* Get mapping block */ 17023b1a94c8SDamien Le Moal dmap_mblk = dmz_get_mblock(zmd, i + 1); 17033b1a94c8SDamien Le Moal if (IS_ERR(dmap_mblk)) 17043b1a94c8SDamien Le Moal return PTR_ERR(dmap_mblk); 17053b1a94c8SDamien Le Moal zmd->map_mblk[i] = dmap_mblk; 17063b1a94c8SDamien Le Moal dmap = (struct dmz_map *) dmap_mblk->data; 17073b1a94c8SDamien Le Moal i++; 17083b1a94c8SDamien Le Moal e = 0; 17093b1a94c8SDamien Le Moal } 17103b1a94c8SDamien Le Moal 17113b1a94c8SDamien Le Moal /* Check data zone */ 17123b1a94c8SDamien Le Moal dzone_id = le32_to_cpu(dmap[e].dzone_id); 17133b1a94c8SDamien Le Moal if (dzone_id == DMZ_MAP_UNMAPPED) 17143b1a94c8SDamien Le Moal goto next; 17153b1a94c8SDamien Le Moal 171636820560SHannes Reinecke if (dzone_id >= zmd->nr_zones) { 1717ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u", 17183b1a94c8SDamien Le Moal chunk, dzone_id); 17193b1a94c8SDamien Le Moal return -EIO; 17203b1a94c8SDamien Le Moal } 17213b1a94c8SDamien Le Moal 17223b1a94c8SDamien Le Moal dzone = dmz_get(zmd, dzone_id); 1723a92fbc44SHannes Reinecke if (!dzone) { 1724a92fbc44SHannes Reinecke dmz_zmd_err(zmd, "Chunk %u mapping: data zone %u not present", 1725a92fbc44SHannes Reinecke chunk, dzone_id); 1726a92fbc44SHannes Reinecke return -EIO; 1727a92fbc44SHannes Reinecke } 17283b1a94c8SDamien Le Moal set_bit(DMZ_DATA, &dzone->flags); 17293b1a94c8SDamien Le Moal dzone->chunk = chunk; 17303b1a94c8SDamien Le Moal dmz_get_zone_weight(zmd, dzone); 17313b1a94c8SDamien Le Moal 173234f5affdSHannes Reinecke if (dmz_is_cache(dzone)) 173334f5affdSHannes Reinecke list_add_tail(&dzone->link, &zmd->map_cache_list); 173434f5affdSHannes Reinecke else if (dmz_is_rnd(dzone)) 1735bd82fdabSHannes Reinecke list_add_tail(&dzone->link, &dzone->dev->map_rnd_list); 17363b1a94c8SDamien Le Moal else 1737bd82fdabSHannes Reinecke list_add_tail(&dzone->link, &dzone->dev->map_seq_list); 17383b1a94c8SDamien Le Moal 17393b1a94c8SDamien Le Moal /* Check buffer zone */ 17403b1a94c8SDamien Le Moal bzone_id = le32_to_cpu(dmap[e].bzone_id); 17413b1a94c8SDamien Le Moal if (bzone_id == DMZ_MAP_UNMAPPED) 17423b1a94c8SDamien Le Moal goto next; 17433b1a94c8SDamien Le Moal 174436820560SHannes Reinecke if (bzone_id >= zmd->nr_zones) { 1745ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u", 17463b1a94c8SDamien Le Moal chunk, bzone_id); 17473b1a94c8SDamien Le Moal return -EIO; 17483b1a94c8SDamien Le Moal } 17493b1a94c8SDamien Le Moal 17503b1a94c8SDamien Le Moal bzone = dmz_get(zmd, bzone_id); 1751a92fbc44SHannes Reinecke if (!bzone) { 1752a92fbc44SHannes Reinecke dmz_zmd_err(zmd, "Chunk %u mapping: buffer zone %u not present", 1753a92fbc44SHannes Reinecke chunk, bzone_id); 1754a92fbc44SHannes Reinecke return -EIO; 1755a92fbc44SHannes Reinecke } 175634f5affdSHannes Reinecke if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) { 1757ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u", 17583b1a94c8SDamien Le Moal chunk, bzone_id); 17593b1a94c8SDamien Le Moal return -EIO; 17603b1a94c8SDamien Le Moal } 17613b1a94c8SDamien Le Moal 17623b1a94c8SDamien Le Moal set_bit(DMZ_DATA, &bzone->flags); 17633b1a94c8SDamien Le Moal set_bit(DMZ_BUF, &bzone->flags); 17643b1a94c8SDamien Le Moal bzone->chunk = chunk; 17653b1a94c8SDamien Le Moal bzone->bzone = dzone; 17663b1a94c8SDamien Le Moal dzone->bzone = bzone; 17673b1a94c8SDamien Le Moal dmz_get_zone_weight(zmd, bzone); 176834f5affdSHannes Reinecke if (dmz_is_cache(bzone)) 176934f5affdSHannes Reinecke list_add_tail(&bzone->link, &zmd->map_cache_list); 177034f5affdSHannes Reinecke else 1771bd82fdabSHannes Reinecke list_add_tail(&bzone->link, &bzone->dev->map_rnd_list); 17723b1a94c8SDamien Le Moal next: 17733b1a94c8SDamien Le Moal chunk++; 17743b1a94c8SDamien Le Moal e++; 17753b1a94c8SDamien Le Moal if (e >= DMZ_MAP_ENTRIES) 17763b1a94c8SDamien Le Moal dmap_mblk = NULL; 17773b1a94c8SDamien Le Moal } 17783b1a94c8SDamien Le Moal 17793b1a94c8SDamien Le Moal /* 17803b1a94c8SDamien Le Moal * At this point, only meta zones and mapped data zones were 17813b1a94c8SDamien Le Moal * fully initialized. All remaining zones are unmapped data 17823b1a94c8SDamien Le Moal * zones. Finish initializing those here. 17833b1a94c8SDamien Le Moal */ 178436820560SHannes Reinecke for (i = 0; i < zmd->nr_zones; i++) { 17853b1a94c8SDamien Le Moal dzone = dmz_get(zmd, i); 1786a92fbc44SHannes Reinecke if (!dzone) 1787a92fbc44SHannes Reinecke continue; 17883b1a94c8SDamien Le Moal if (dmz_is_meta(dzone)) 17893b1a94c8SDamien Le Moal continue; 179034f5affdSHannes Reinecke if (dmz_is_offline(dzone)) 179134f5affdSHannes Reinecke continue; 17923b1a94c8SDamien Le Moal 179334f5affdSHannes Reinecke if (dmz_is_cache(dzone)) 179434f5affdSHannes Reinecke zmd->nr_cache++; 179534f5affdSHannes Reinecke else if (dmz_is_rnd(dzone)) 1796bd82fdabSHannes Reinecke dzone->dev->nr_rnd++; 17973b1a94c8SDamien Le Moal else 1798bd82fdabSHannes Reinecke dzone->dev->nr_seq++; 17993b1a94c8SDamien Le Moal 18003b1a94c8SDamien Le Moal if (dmz_is_data(dzone)) { 18013b1a94c8SDamien Le Moal /* Already initialized */ 18023b1a94c8SDamien Le Moal continue; 18033b1a94c8SDamien Le Moal } 18043b1a94c8SDamien Le Moal 18053b1a94c8SDamien Le Moal /* Unmapped data zone */ 18063b1a94c8SDamien Le Moal set_bit(DMZ_DATA, &dzone->flags); 18073b1a94c8SDamien Le Moal dzone->chunk = DMZ_MAP_UNMAPPED; 180834f5affdSHannes Reinecke if (dmz_is_cache(dzone)) { 180934f5affdSHannes Reinecke list_add_tail(&dzone->link, &zmd->unmap_cache_list); 181034f5affdSHannes Reinecke atomic_inc(&zmd->unmap_nr_cache); 181134f5affdSHannes Reinecke } else if (dmz_is_rnd(dzone)) { 1812bd82fdabSHannes Reinecke list_add_tail(&dzone->link, 1813bd82fdabSHannes Reinecke &dzone->dev->unmap_rnd_list); 1814bd82fdabSHannes Reinecke atomic_inc(&dzone->dev->unmap_nr_rnd); 18153b1a94c8SDamien Le Moal } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) { 18163b1a94c8SDamien Le Moal list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list); 1817aec67b4fSHannes Reinecke set_bit(DMZ_RESERVED, &dzone->flags); 18183b1a94c8SDamien Le Moal atomic_inc(&zmd->nr_reserved_seq_zones); 1819bd82fdabSHannes Reinecke dzone->dev->nr_seq--; 18203b1a94c8SDamien Le Moal } else { 1821bd82fdabSHannes Reinecke list_add_tail(&dzone->link, 1822bd82fdabSHannes Reinecke &dzone->dev->unmap_seq_list); 1823bd82fdabSHannes Reinecke atomic_inc(&dzone->dev->unmap_nr_seq); 18243b1a94c8SDamien Le Moal } 18253b1a94c8SDamien Le Moal } 18263b1a94c8SDamien Le Moal 18273b1a94c8SDamien Le Moal return 0; 18283b1a94c8SDamien Le Moal } 18293b1a94c8SDamien Le Moal 18303b1a94c8SDamien Le Moal /* 18313b1a94c8SDamien Le Moal * Set a data chunk mapping. 18323b1a94c8SDamien Le Moal */ 18333b1a94c8SDamien Le Moal static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, 18343b1a94c8SDamien Le Moal unsigned int dzone_id, unsigned int bzone_id) 18353b1a94c8SDamien Le Moal { 18363b1a94c8SDamien Le Moal struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; 18373b1a94c8SDamien Le Moal struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; 18383b1a94c8SDamien Le Moal int map_idx = chunk & DMZ_MAP_ENTRIES_MASK; 18393b1a94c8SDamien Le Moal 18403b1a94c8SDamien Le Moal dmap[map_idx].dzone_id = cpu_to_le32(dzone_id); 18413b1a94c8SDamien Le Moal dmap[map_idx].bzone_id = cpu_to_le32(bzone_id); 18423b1a94c8SDamien Le Moal dmz_dirty_mblock(zmd, dmap_mblk); 18433b1a94c8SDamien Le Moal } 18443b1a94c8SDamien Le Moal 18453b1a94c8SDamien Le Moal /* 18463b1a94c8SDamien Le Moal * The list of mapped zones is maintained in LRU order. 18473b1a94c8SDamien Le Moal * This rotates a zone at the end of its map list. 18483b1a94c8SDamien Le Moal */ 18493b1a94c8SDamien Le Moal static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 18503b1a94c8SDamien Le Moal { 18513b1a94c8SDamien Le Moal if (list_empty(&zone->link)) 18523b1a94c8SDamien Le Moal return; 18533b1a94c8SDamien Le Moal 18543b1a94c8SDamien Le Moal list_del_init(&zone->link); 18553b1a94c8SDamien Le Moal if (dmz_is_seq(zone)) { 18563b1a94c8SDamien Le Moal /* LRU rotate sequential zone */ 1857bd82fdabSHannes Reinecke list_add_tail(&zone->link, &zone->dev->map_seq_list); 185834f5affdSHannes Reinecke } else if (dmz_is_cache(zone)) { 185934f5affdSHannes Reinecke /* LRU rotate cache zone */ 186034f5affdSHannes Reinecke list_add_tail(&zone->link, &zmd->map_cache_list); 18613b1a94c8SDamien Le Moal } else { 18623b1a94c8SDamien Le Moal /* LRU rotate random zone */ 1863bd82fdabSHannes Reinecke list_add_tail(&zone->link, &zone->dev->map_rnd_list); 18643b1a94c8SDamien Le Moal } 18653b1a94c8SDamien Le Moal } 18663b1a94c8SDamien Le Moal 18673b1a94c8SDamien Le Moal /* 18683b1a94c8SDamien Le Moal * The list of mapped random zones is maintained 18693b1a94c8SDamien Le Moal * in LRU order. This rotates a zone at the end of the list. 18703b1a94c8SDamien Le Moal */ 18713b1a94c8SDamien Le Moal static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 18723b1a94c8SDamien Le Moal { 18733b1a94c8SDamien Le Moal __dmz_lru_zone(zmd, zone); 18743b1a94c8SDamien Le Moal if (zone->bzone) 18753b1a94c8SDamien Le Moal __dmz_lru_zone(zmd, zone->bzone); 18763b1a94c8SDamien Le Moal } 18773b1a94c8SDamien Le Moal 18783b1a94c8SDamien Le Moal /* 18793b1a94c8SDamien Le Moal * Wait for any zone to be freed. 18803b1a94c8SDamien Le Moal */ 18813b1a94c8SDamien Le Moal static void dmz_wait_for_free_zones(struct dmz_metadata *zmd) 18823b1a94c8SDamien Le Moal { 18833b1a94c8SDamien Le Moal DEFINE_WAIT(wait); 18843b1a94c8SDamien Le Moal 18853b1a94c8SDamien Le Moal prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE); 18863b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 18873b1a94c8SDamien Le Moal dmz_unlock_metadata(zmd); 18883b1a94c8SDamien Le Moal 18893b1a94c8SDamien Le Moal io_schedule_timeout(HZ); 18903b1a94c8SDamien Le Moal 18913b1a94c8SDamien Le Moal dmz_lock_metadata(zmd); 18923b1a94c8SDamien Le Moal dmz_lock_map(zmd); 18933b1a94c8SDamien Le Moal finish_wait(&zmd->free_wq, &wait); 18943b1a94c8SDamien Le Moal } 18953b1a94c8SDamien Le Moal 18963b1a94c8SDamien Le Moal /* 18973b1a94c8SDamien Le Moal * Lock a zone for reclaim (set the zone RECLAIM bit). 18983b1a94c8SDamien Le Moal * Returns false if the zone cannot be locked or if it is already locked 18993b1a94c8SDamien Le Moal * and 1 otherwise. 19003b1a94c8SDamien Le Moal */ 19013b1a94c8SDamien Le Moal int dmz_lock_zone_reclaim(struct dm_zone *zone) 19023b1a94c8SDamien Le Moal { 19033b1a94c8SDamien Le Moal /* Active zones cannot be reclaimed */ 19043b1a94c8SDamien Le Moal if (dmz_is_active(zone)) 19053b1a94c8SDamien Le Moal return 0; 19063b1a94c8SDamien Le Moal 19073b1a94c8SDamien Le Moal return !test_and_set_bit(DMZ_RECLAIM, &zone->flags); 19083b1a94c8SDamien Le Moal } 19093b1a94c8SDamien Le Moal 19103b1a94c8SDamien Le Moal /* 19113b1a94c8SDamien Le Moal * Clear a zone reclaim flag. 19123b1a94c8SDamien Le Moal */ 19133b1a94c8SDamien Le Moal void dmz_unlock_zone_reclaim(struct dm_zone *zone) 19143b1a94c8SDamien Le Moal { 19153b1a94c8SDamien Le Moal WARN_ON(dmz_is_active(zone)); 19163b1a94c8SDamien Le Moal WARN_ON(!dmz_in_reclaim(zone)); 19173b1a94c8SDamien Le Moal 19183b1a94c8SDamien Le Moal clear_bit_unlock(DMZ_RECLAIM, &zone->flags); 19193b1a94c8SDamien Le Moal smp_mb__after_atomic(); 19203b1a94c8SDamien Le Moal wake_up_bit(&zone->flags, DMZ_RECLAIM); 19213b1a94c8SDamien Le Moal } 19223b1a94c8SDamien Le Moal 19233b1a94c8SDamien Le Moal /* 19243b1a94c8SDamien Le Moal * Wait for a zone reclaim to complete. 19253b1a94c8SDamien Le Moal */ 19263b1a94c8SDamien Le Moal static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone) 19273b1a94c8SDamien Le Moal { 19283b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 19293b1a94c8SDamien Le Moal dmz_unlock_metadata(zmd); 1930a16b7deeSHannes Reinecke set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags); 19313b1a94c8SDamien Le Moal wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ); 1932a16b7deeSHannes Reinecke clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags); 19333b1a94c8SDamien Le Moal dmz_lock_metadata(zmd); 19343b1a94c8SDamien Le Moal dmz_lock_map(zmd); 19353b1a94c8SDamien Le Moal } 19363b1a94c8SDamien Le Moal 19373b1a94c8SDamien Le Moal /* 193834f5affdSHannes Reinecke * Select a cache or random write zone for reclaim. 19393b1a94c8SDamien Le Moal */ 194090a9b869SHannes Reinecke static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd, 194190a9b869SHannes Reinecke bool idle) 19423b1a94c8SDamien Le Moal { 19433b1a94c8SDamien Le Moal struct dm_zone *dzone = NULL; 19443b1a94c8SDamien Le Moal struct dm_zone *zone; 1945bd82fdabSHannes Reinecke struct list_head *zone_list; 19463b1a94c8SDamien Le Moal 194734f5affdSHannes Reinecke /* If we have cache zones select from the cache zone list */ 194890a9b869SHannes Reinecke if (zmd->nr_cache) { 194934f5affdSHannes Reinecke zone_list = &zmd->map_cache_list; 195090a9b869SHannes Reinecke /* Try to relaim random zones, too, when idle */ 1951bd82fdabSHannes Reinecke if (idle && list_empty(zone_list)) { 1952bd82fdabSHannes Reinecke int i; 1953bd82fdabSHannes Reinecke 1954bd82fdabSHannes Reinecke for (i = 1; i < zmd->nr_devs; i++) { 1955bd82fdabSHannes Reinecke zone_list = &zmd->dev[i].map_rnd_list; 1956bd82fdabSHannes Reinecke if (!list_empty(zone_list)) 1957bd82fdabSHannes Reinecke break; 1958bd82fdabSHannes Reinecke } 1959bd82fdabSHannes Reinecke } 1960bd82fdabSHannes Reinecke } else { 1961bd82fdabSHannes Reinecke /* Otherwise the random zones are on the first disk */ 1962bd82fdabSHannes Reinecke zone_list = &zmd->dev[0].map_rnd_list; 196390a9b869SHannes Reinecke } 19643b1a94c8SDamien Le Moal 196534f5affdSHannes Reinecke list_for_each_entry(zone, zone_list, link) { 19663b1a94c8SDamien Le Moal if (dmz_is_buf(zone)) 19673b1a94c8SDamien Le Moal dzone = zone->bzone; 19683b1a94c8SDamien Le Moal else 19693b1a94c8SDamien Le Moal dzone = zone; 19703b1a94c8SDamien Le Moal if (dmz_lock_zone_reclaim(dzone)) 19713b1a94c8SDamien Le Moal return dzone; 19723b1a94c8SDamien Le Moal } 19733b1a94c8SDamien Le Moal 1974489dc0f0SHannes Reinecke return NULL; 19753b1a94c8SDamien Le Moal } 19763b1a94c8SDamien Le Moal 19773b1a94c8SDamien Le Moal /* 19783b1a94c8SDamien Le Moal * Select a buffered sequential zone for reclaim. 19793b1a94c8SDamien Le Moal */ 19803b1a94c8SDamien Le Moal static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd) 19813b1a94c8SDamien Le Moal { 19823b1a94c8SDamien Le Moal struct dm_zone *zone; 1983bd82fdabSHannes Reinecke int i; 19843b1a94c8SDamien Le Moal 1985bd82fdabSHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) { 1986bd82fdabSHannes Reinecke struct dmz_dev *dev = &zmd->dev[i]; 1987bd82fdabSHannes Reinecke 1988bd82fdabSHannes Reinecke list_for_each_entry(zone, &dev->map_seq_list, link) { 19893b1a94c8SDamien Le Moal if (!zone->bzone) 19903b1a94c8SDamien Le Moal continue; 19913b1a94c8SDamien Le Moal if (dmz_lock_zone_reclaim(zone)) 19923b1a94c8SDamien Le Moal return zone; 19933b1a94c8SDamien Le Moal } 1994bd82fdabSHannes Reinecke } 19953b1a94c8SDamien Le Moal 1996489dc0f0SHannes Reinecke return NULL; 19973b1a94c8SDamien Le Moal } 19983b1a94c8SDamien Le Moal 19993b1a94c8SDamien Le Moal /* 20003b1a94c8SDamien Le Moal * Select a zone for reclaim. 20013b1a94c8SDamien Le Moal */ 200290a9b869SHannes Reinecke struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd, bool idle) 20033b1a94c8SDamien Le Moal { 20043b1a94c8SDamien Le Moal struct dm_zone *zone; 20053b1a94c8SDamien Le Moal 20063b1a94c8SDamien Le Moal /* 20073b1a94c8SDamien Le Moal * Search for a zone candidate to reclaim: 2 cases are possible. 20083b1a94c8SDamien Le Moal * (1) There is no free sequential zones. Then a random data zone 20093b1a94c8SDamien Le Moal * cannot be reclaimed. So choose a sequential zone to reclaim so 20103b1a94c8SDamien Le Moal * that afterward a random zone can be reclaimed. 20113b1a94c8SDamien Le Moal * (2) At least one free sequential zone is available, then choose 20123b1a94c8SDamien Le Moal * the oldest random zone (data or buffer) that can be locked. 20133b1a94c8SDamien Le Moal */ 20143b1a94c8SDamien Le Moal dmz_lock_map(zmd); 20153b1a94c8SDamien Le Moal if (list_empty(&zmd->reserved_seq_zones_list)) 20163b1a94c8SDamien Le Moal zone = dmz_get_seq_zone_for_reclaim(zmd); 20173b1a94c8SDamien Le Moal else 201890a9b869SHannes Reinecke zone = dmz_get_rnd_zone_for_reclaim(zmd, idle); 20193b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 20203b1a94c8SDamien Le Moal 20213b1a94c8SDamien Le Moal return zone; 20223b1a94c8SDamien Le Moal } 20233b1a94c8SDamien Le Moal 20243b1a94c8SDamien Le Moal /* 20253b1a94c8SDamien Le Moal * Get the zone mapping a chunk, if the chunk is mapped already. 20263b1a94c8SDamien Le Moal * If no mapping exist and the operation is WRITE, a zone is 20273b1a94c8SDamien Le Moal * allocated and used to map the chunk. 20283b1a94c8SDamien Le Moal * The zone returned will be set to the active state. 20293b1a94c8SDamien Le Moal */ 20303b1a94c8SDamien Le Moal struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op) 20313b1a94c8SDamien Le Moal { 20323b1a94c8SDamien Le Moal struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; 20333b1a94c8SDamien Le Moal struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; 20343b1a94c8SDamien Le Moal int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK; 20353b1a94c8SDamien Le Moal unsigned int dzone_id; 20363b1a94c8SDamien Le Moal struct dm_zone *dzone = NULL; 20373b1a94c8SDamien Le Moal int ret = 0; 203834f5affdSHannes Reinecke int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND; 20393b1a94c8SDamien Le Moal 20403b1a94c8SDamien Le Moal dmz_lock_map(zmd); 20413b1a94c8SDamien Le Moal again: 20423b1a94c8SDamien Le Moal /* Get the chunk mapping */ 20433b1a94c8SDamien Le Moal dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id); 20443b1a94c8SDamien Le Moal if (dzone_id == DMZ_MAP_UNMAPPED) { 20453b1a94c8SDamien Le Moal /* 20463b1a94c8SDamien Le Moal * Read or discard in unmapped chunks are fine. But for 20473b1a94c8SDamien Le Moal * writes, we need a mapping, so get one. 20483b1a94c8SDamien Le Moal */ 20493b1a94c8SDamien Le Moal if (op != REQ_OP_WRITE) 20503b1a94c8SDamien Le Moal goto out; 20513b1a94c8SDamien Le Moal 2052ad1bd578SDmitry Fomichev /* Allocate a random zone */ 205334f5affdSHannes Reinecke dzone = dmz_alloc_zone(zmd, alloc_flags); 20543b1a94c8SDamien Le Moal if (!dzone) { 2055d0e21ce4SHannes Reinecke if (dmz_dev_is_dying(zmd)) { 205675d66ffbSDmitry Fomichev dzone = ERR_PTR(-EIO); 205775d66ffbSDmitry Fomichev goto out; 205875d66ffbSDmitry Fomichev } 20593b1a94c8SDamien Le Moal dmz_wait_for_free_zones(zmd); 20603b1a94c8SDamien Le Moal goto again; 20613b1a94c8SDamien Le Moal } 20623b1a94c8SDamien Le Moal 20633b1a94c8SDamien Le Moal dmz_map_zone(zmd, dzone, chunk); 20643b1a94c8SDamien Le Moal 20653b1a94c8SDamien Le Moal } else { 20663b1a94c8SDamien Le Moal /* The chunk is already mapped: get the mapping zone */ 20673b1a94c8SDamien Le Moal dzone = dmz_get(zmd, dzone_id); 2068a92fbc44SHannes Reinecke if (!dzone) { 2069a92fbc44SHannes Reinecke dzone = ERR_PTR(-EIO); 2070a92fbc44SHannes Reinecke goto out; 2071a92fbc44SHannes Reinecke } 20723b1a94c8SDamien Le Moal if (dzone->chunk != chunk) { 20733b1a94c8SDamien Le Moal dzone = ERR_PTR(-EIO); 20743b1a94c8SDamien Le Moal goto out; 20753b1a94c8SDamien Le Moal } 20763b1a94c8SDamien Le Moal 20773b1a94c8SDamien Le Moal /* Repair write pointer if the sequential dzone has error */ 20783b1a94c8SDamien Le Moal if (dmz_seq_write_err(dzone)) { 20793b1a94c8SDamien Le Moal ret = dmz_handle_seq_write_err(zmd, dzone); 20803b1a94c8SDamien Le Moal if (ret) { 20813b1a94c8SDamien Le Moal dzone = ERR_PTR(-EIO); 20823b1a94c8SDamien Le Moal goto out; 20833b1a94c8SDamien Le Moal } 20843b1a94c8SDamien Le Moal clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags); 20853b1a94c8SDamien Le Moal } 20863b1a94c8SDamien Le Moal } 20873b1a94c8SDamien Le Moal 20883b1a94c8SDamien Le Moal /* 20893b1a94c8SDamien Le Moal * If the zone is being reclaimed, the chunk mapping may change 20903b1a94c8SDamien Le Moal * to a different zone. So wait for reclaim and retry. Otherwise, 20913b1a94c8SDamien Le Moal * activate the zone (this will prevent reclaim from touching it). 20923b1a94c8SDamien Le Moal */ 20933b1a94c8SDamien Le Moal if (dmz_in_reclaim(dzone)) { 20943b1a94c8SDamien Le Moal dmz_wait_for_reclaim(zmd, dzone); 20953b1a94c8SDamien Le Moal goto again; 20963b1a94c8SDamien Le Moal } 20973b1a94c8SDamien Le Moal dmz_activate_zone(dzone); 20983b1a94c8SDamien Le Moal dmz_lru_zone(zmd, dzone); 20993b1a94c8SDamien Le Moal out: 21003b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 21013b1a94c8SDamien Le Moal 21023b1a94c8SDamien Le Moal return dzone; 21033b1a94c8SDamien Le Moal } 21043b1a94c8SDamien Le Moal 21053b1a94c8SDamien Le Moal /* 21063b1a94c8SDamien Le Moal * Write and discard change the block validity of data zones and their buffer 21073b1a94c8SDamien Le Moal * zones. Check here that valid blocks are still present. If all blocks are 21083b1a94c8SDamien Le Moal * invalid, the zones can be unmapped on the fly without waiting for reclaim 21093b1a94c8SDamien Le Moal * to do it. 21103b1a94c8SDamien Le Moal */ 21113b1a94c8SDamien Le Moal void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone) 21123b1a94c8SDamien Le Moal { 21133b1a94c8SDamien Le Moal struct dm_zone *bzone; 21143b1a94c8SDamien Le Moal 21153b1a94c8SDamien Le Moal dmz_lock_map(zmd); 21163b1a94c8SDamien Le Moal 21173b1a94c8SDamien Le Moal bzone = dzone->bzone; 21183b1a94c8SDamien Le Moal if (bzone) { 21193b1a94c8SDamien Le Moal if (dmz_weight(bzone)) 21203b1a94c8SDamien Le Moal dmz_lru_zone(zmd, bzone); 21213b1a94c8SDamien Le Moal else { 21223b1a94c8SDamien Le Moal /* Empty buffer zone: reclaim it */ 21233b1a94c8SDamien Le Moal dmz_unmap_zone(zmd, bzone); 21243b1a94c8SDamien Le Moal dmz_free_zone(zmd, bzone); 21253b1a94c8SDamien Le Moal bzone = NULL; 21263b1a94c8SDamien Le Moal } 21273b1a94c8SDamien Le Moal } 21283b1a94c8SDamien Le Moal 21293b1a94c8SDamien Le Moal /* Deactivate the data zone */ 21303b1a94c8SDamien Le Moal dmz_deactivate_zone(dzone); 21313b1a94c8SDamien Le Moal if (dmz_is_active(dzone) || bzone || dmz_weight(dzone)) 21323b1a94c8SDamien Le Moal dmz_lru_zone(zmd, dzone); 21333b1a94c8SDamien Le Moal else { 21343b1a94c8SDamien Le Moal /* Unbuffered inactive empty data zone: reclaim it */ 21353b1a94c8SDamien Le Moal dmz_unmap_zone(zmd, dzone); 21363b1a94c8SDamien Le Moal dmz_free_zone(zmd, dzone); 21373b1a94c8SDamien Le Moal } 21383b1a94c8SDamien Le Moal 21393b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 21403b1a94c8SDamien Le Moal } 21413b1a94c8SDamien Le Moal 21423b1a94c8SDamien Le Moal /* 21433b1a94c8SDamien Le Moal * Allocate and map a random zone to buffer a chunk 21443b1a94c8SDamien Le Moal * already mapped to a sequential zone. 21453b1a94c8SDamien Le Moal */ 21463b1a94c8SDamien Le Moal struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd, 21473b1a94c8SDamien Le Moal struct dm_zone *dzone) 21483b1a94c8SDamien Le Moal { 21493b1a94c8SDamien Le Moal struct dm_zone *bzone; 215034f5affdSHannes Reinecke int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND; 21513b1a94c8SDamien Le Moal 21523b1a94c8SDamien Le Moal dmz_lock_map(zmd); 21533b1a94c8SDamien Le Moal again: 21543b1a94c8SDamien Le Moal bzone = dzone->bzone; 21553b1a94c8SDamien Le Moal if (bzone) 21563b1a94c8SDamien Le Moal goto out; 21573b1a94c8SDamien Le Moal 2158ad1bd578SDmitry Fomichev /* Allocate a random zone */ 215934f5affdSHannes Reinecke bzone = dmz_alloc_zone(zmd, alloc_flags); 21603b1a94c8SDamien Le Moal if (!bzone) { 2161d0e21ce4SHannes Reinecke if (dmz_dev_is_dying(zmd)) { 216275d66ffbSDmitry Fomichev bzone = ERR_PTR(-EIO); 216375d66ffbSDmitry Fomichev goto out; 216475d66ffbSDmitry Fomichev } 21653b1a94c8SDamien Le Moal dmz_wait_for_free_zones(zmd); 21663b1a94c8SDamien Le Moal goto again; 21673b1a94c8SDamien Le Moal } 21683b1a94c8SDamien Le Moal 21693b1a94c8SDamien Le Moal /* Update the chunk mapping */ 2170b7122873SHannes Reinecke dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id); 21713b1a94c8SDamien Le Moal 21723b1a94c8SDamien Le Moal set_bit(DMZ_BUF, &bzone->flags); 21733b1a94c8SDamien Le Moal bzone->chunk = dzone->chunk; 21743b1a94c8SDamien Le Moal bzone->bzone = dzone; 21753b1a94c8SDamien Le Moal dzone->bzone = bzone; 217634f5affdSHannes Reinecke if (dmz_is_cache(bzone)) 217734f5affdSHannes Reinecke list_add_tail(&bzone->link, &zmd->map_cache_list); 217834f5affdSHannes Reinecke else 2179bd82fdabSHannes Reinecke list_add_tail(&bzone->link, &bzone->dev->map_rnd_list); 21803b1a94c8SDamien Le Moal out: 21813b1a94c8SDamien Le Moal dmz_unlock_map(zmd); 21823b1a94c8SDamien Le Moal 21833b1a94c8SDamien Le Moal return bzone; 21843b1a94c8SDamien Le Moal } 21853b1a94c8SDamien Le Moal 21863b1a94c8SDamien Le Moal /* 21873b1a94c8SDamien Le Moal * Get an unmapped (free) zone. 21883b1a94c8SDamien Le Moal * This must be called with the mapping lock held. 21893b1a94c8SDamien Le Moal */ 21903b1a94c8SDamien Le Moal struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags) 21913b1a94c8SDamien Le Moal { 21923b1a94c8SDamien Le Moal struct list_head *list; 21933b1a94c8SDamien Le Moal struct dm_zone *zone; 2194bd82fdabSHannes Reinecke unsigned int dev_idx = 0; 21953b1a94c8SDamien Le Moal 2196bd82fdabSHannes Reinecke again: 219734f5affdSHannes Reinecke if (flags & DMZ_ALLOC_CACHE) 219834f5affdSHannes Reinecke list = &zmd->unmap_cache_list; 219934f5affdSHannes Reinecke else if (flags & DMZ_ALLOC_RND) 2200bd82fdabSHannes Reinecke list = &zmd->dev[dev_idx].unmap_rnd_list; 22013b1a94c8SDamien Le Moal else 2202bd82fdabSHannes Reinecke list = &zmd->dev[dev_idx].unmap_seq_list; 220334f5affdSHannes Reinecke 22043b1a94c8SDamien Le Moal if (list_empty(list)) { 22053b1a94c8SDamien Le Moal /* 220634f5affdSHannes Reinecke * No free zone: return NULL if this is for not reclaim. 22073b1a94c8SDamien Le Moal */ 220834f5affdSHannes Reinecke if (!(flags & DMZ_ALLOC_RECLAIM)) 22093b1a94c8SDamien Le Moal return NULL; 2210bd82fdabSHannes Reinecke if (dev_idx < zmd->nr_devs) { 2211bd82fdabSHannes Reinecke dev_idx++; 2212bd82fdabSHannes Reinecke goto again; 2213bd82fdabSHannes Reinecke } 2214bd82fdabSHannes Reinecke 221534f5affdSHannes Reinecke /* 221634f5affdSHannes Reinecke * Fallback to the reserved sequential zones 221734f5affdSHannes Reinecke */ 221834f5affdSHannes Reinecke zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list, 22193b1a94c8SDamien Le Moal struct dm_zone, link); 222034f5affdSHannes Reinecke if (zone) { 22213b1a94c8SDamien Le Moal list_del_init(&zone->link); 22223b1a94c8SDamien Le Moal atomic_dec(&zmd->nr_reserved_seq_zones); 222334f5affdSHannes Reinecke } 22243b1a94c8SDamien Le Moal return zone; 22253b1a94c8SDamien Le Moal } 22263b1a94c8SDamien Le Moal 22273b1a94c8SDamien Le Moal zone = list_first_entry(list, struct dm_zone, link); 22283b1a94c8SDamien Le Moal list_del_init(&zone->link); 22293b1a94c8SDamien Le Moal 223034f5affdSHannes Reinecke if (dmz_is_cache(zone)) 223134f5affdSHannes Reinecke atomic_dec(&zmd->unmap_nr_cache); 223234f5affdSHannes Reinecke else if (dmz_is_rnd(zone)) 2233bd82fdabSHannes Reinecke atomic_dec(&zone->dev->unmap_nr_rnd); 22343b1a94c8SDamien Le Moal else 2235bd82fdabSHannes Reinecke atomic_dec(&zone->dev->unmap_nr_seq); 22363b1a94c8SDamien Le Moal 22373b1a94c8SDamien Le Moal if (dmz_is_offline(zone)) { 2238ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "Zone %u is offline", zone->id); 22393b1a94c8SDamien Le Moal zone = NULL; 22403b1a94c8SDamien Le Moal goto again; 22413b1a94c8SDamien Le Moal } 2242dc076c83SHannes Reinecke if (dmz_is_meta(zone)) { 22438f22272aSHannes Reinecke dmz_zmd_warn(zmd, "Zone %u has metadata", zone->id); 2244dc076c83SHannes Reinecke zone = NULL; 2245dc076c83SHannes Reinecke goto again; 2246dc076c83SHannes Reinecke } 22473b1a94c8SDamien Le Moal return zone; 22483b1a94c8SDamien Le Moal } 22493b1a94c8SDamien Le Moal 22503b1a94c8SDamien Le Moal /* 22513b1a94c8SDamien Le Moal * Free a zone. 22523b1a94c8SDamien Le Moal * This must be called with the mapping lock held. 22533b1a94c8SDamien Le Moal */ 22543b1a94c8SDamien Le Moal void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 22553b1a94c8SDamien Le Moal { 22563b1a94c8SDamien Le Moal /* If this is a sequential zone, reset it */ 22573b1a94c8SDamien Le Moal if (dmz_is_seq(zone)) 22583b1a94c8SDamien Le Moal dmz_reset_zone(zmd, zone); 22593b1a94c8SDamien Le Moal 22603b1a94c8SDamien Le Moal /* Return the zone to its type unmap list */ 226134f5affdSHannes Reinecke if (dmz_is_cache(zone)) { 226234f5affdSHannes Reinecke list_add_tail(&zone->link, &zmd->unmap_cache_list); 226334f5affdSHannes Reinecke atomic_inc(&zmd->unmap_nr_cache); 226434f5affdSHannes Reinecke } else if (dmz_is_rnd(zone)) { 2265bd82fdabSHannes Reinecke list_add_tail(&zone->link, &zone->dev->unmap_rnd_list); 2266bd82fdabSHannes Reinecke atomic_inc(&zone->dev->unmap_nr_rnd); 2267aec67b4fSHannes Reinecke } else if (dmz_is_reserved(zone)) { 22683b1a94c8SDamien Le Moal list_add_tail(&zone->link, &zmd->reserved_seq_zones_list); 22693b1a94c8SDamien Le Moal atomic_inc(&zmd->nr_reserved_seq_zones); 22703b1a94c8SDamien Le Moal } else { 2271bd82fdabSHannes Reinecke list_add_tail(&zone->link, &zone->dev->unmap_seq_list); 2272bd82fdabSHannes Reinecke atomic_inc(&zone->dev->unmap_nr_seq); 22733b1a94c8SDamien Le Moal } 22743b1a94c8SDamien Le Moal 22753b1a94c8SDamien Le Moal wake_up_all(&zmd->free_wq); 22763b1a94c8SDamien Le Moal } 22773b1a94c8SDamien Le Moal 22783b1a94c8SDamien Le Moal /* 22793b1a94c8SDamien Le Moal * Map a chunk to a zone. 22803b1a94c8SDamien Le Moal * This must be called with the mapping lock held. 22813b1a94c8SDamien Le Moal */ 22823b1a94c8SDamien Le Moal void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone, 22833b1a94c8SDamien Le Moal unsigned int chunk) 22843b1a94c8SDamien Le Moal { 22853b1a94c8SDamien Le Moal /* Set the chunk mapping */ 2286b7122873SHannes Reinecke dmz_set_chunk_mapping(zmd, chunk, dzone->id, 22873b1a94c8SDamien Le Moal DMZ_MAP_UNMAPPED); 22883b1a94c8SDamien Le Moal dzone->chunk = chunk; 228934f5affdSHannes Reinecke if (dmz_is_cache(dzone)) 229034f5affdSHannes Reinecke list_add_tail(&dzone->link, &zmd->map_cache_list); 229134f5affdSHannes Reinecke else if (dmz_is_rnd(dzone)) 2292bd82fdabSHannes Reinecke list_add_tail(&dzone->link, &dzone->dev->map_rnd_list); 22933b1a94c8SDamien Le Moal else 2294bd82fdabSHannes Reinecke list_add_tail(&dzone->link, &dzone->dev->map_seq_list); 22953b1a94c8SDamien Le Moal } 22963b1a94c8SDamien Le Moal 22973b1a94c8SDamien Le Moal /* 22983b1a94c8SDamien Le Moal * Unmap a zone. 22993b1a94c8SDamien Le Moal * This must be called with the mapping lock held. 23003b1a94c8SDamien Le Moal */ 23013b1a94c8SDamien Le Moal void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone) 23023b1a94c8SDamien Le Moal { 23033b1a94c8SDamien Le Moal unsigned int chunk = zone->chunk; 23043b1a94c8SDamien Le Moal unsigned int dzone_id; 23053b1a94c8SDamien Le Moal 23063b1a94c8SDamien Le Moal if (chunk == DMZ_MAP_UNMAPPED) { 23073b1a94c8SDamien Le Moal /* Already unmapped */ 23083b1a94c8SDamien Le Moal return; 23093b1a94c8SDamien Le Moal } 23103b1a94c8SDamien Le Moal 23113b1a94c8SDamien Le Moal if (test_and_clear_bit(DMZ_BUF, &zone->flags)) { 23123b1a94c8SDamien Le Moal /* 23133b1a94c8SDamien Le Moal * Unmapping the chunk buffer zone: clear only 23143b1a94c8SDamien Le Moal * the chunk buffer mapping 23153b1a94c8SDamien Le Moal */ 2316b7122873SHannes Reinecke dzone_id = zone->bzone->id; 23173b1a94c8SDamien Le Moal zone->bzone->bzone = NULL; 23183b1a94c8SDamien Le Moal zone->bzone = NULL; 23193b1a94c8SDamien Le Moal 23203b1a94c8SDamien Le Moal } else { 23213b1a94c8SDamien Le Moal /* 23223b1a94c8SDamien Le Moal * Unmapping the chunk data zone: the zone must 23233b1a94c8SDamien Le Moal * not be buffered. 23243b1a94c8SDamien Le Moal */ 23253b1a94c8SDamien Le Moal if (WARN_ON(zone->bzone)) { 23263b1a94c8SDamien Le Moal zone->bzone->bzone = NULL; 23273b1a94c8SDamien Le Moal zone->bzone = NULL; 23283b1a94c8SDamien Le Moal } 23293b1a94c8SDamien Le Moal dzone_id = DMZ_MAP_UNMAPPED; 23303b1a94c8SDamien Le Moal } 23313b1a94c8SDamien Le Moal 23323b1a94c8SDamien Le Moal dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED); 23333b1a94c8SDamien Le Moal 23343b1a94c8SDamien Le Moal zone->chunk = DMZ_MAP_UNMAPPED; 23353b1a94c8SDamien Le Moal list_del_init(&zone->link); 23363b1a94c8SDamien Le Moal } 23373b1a94c8SDamien Le Moal 23383b1a94c8SDamien Le Moal /* 23393b1a94c8SDamien Le Moal * Set @nr_bits bits in @bitmap starting from @bit. 23403b1a94c8SDamien Le Moal * Return the number of bits changed from 0 to 1. 23413b1a94c8SDamien Le Moal */ 23423b1a94c8SDamien Le Moal static unsigned int dmz_set_bits(unsigned long *bitmap, 23433b1a94c8SDamien Le Moal unsigned int bit, unsigned int nr_bits) 23443b1a94c8SDamien Le Moal { 23453b1a94c8SDamien Le Moal unsigned long *addr; 23463b1a94c8SDamien Le Moal unsigned int end = bit + nr_bits; 23473b1a94c8SDamien Le Moal unsigned int n = 0; 23483b1a94c8SDamien Le Moal 23493b1a94c8SDamien Le Moal while (bit < end) { 23503b1a94c8SDamien Le Moal if (((bit & (BITS_PER_LONG - 1)) == 0) && 23513b1a94c8SDamien Le Moal ((end - bit) >= BITS_PER_LONG)) { 23523b1a94c8SDamien Le Moal /* Try to set the whole word at once */ 23533b1a94c8SDamien Le Moal addr = bitmap + BIT_WORD(bit); 23543b1a94c8SDamien Le Moal if (*addr == 0) { 23553b1a94c8SDamien Le Moal *addr = ULONG_MAX; 23563b1a94c8SDamien Le Moal n += BITS_PER_LONG; 23573b1a94c8SDamien Le Moal bit += BITS_PER_LONG; 23583b1a94c8SDamien Le Moal continue; 23593b1a94c8SDamien Le Moal } 23603b1a94c8SDamien Le Moal } 23613b1a94c8SDamien Le Moal 23623b1a94c8SDamien Le Moal if (!test_and_set_bit(bit, bitmap)) 23633b1a94c8SDamien Le Moal n++; 23643b1a94c8SDamien Le Moal bit++; 23653b1a94c8SDamien Le Moal } 23663b1a94c8SDamien Le Moal 23673b1a94c8SDamien Le Moal return n; 23683b1a94c8SDamien Le Moal } 23693b1a94c8SDamien Le Moal 23703b1a94c8SDamien Le Moal /* 23713b1a94c8SDamien Le Moal * Get the bitmap block storing the bit for chunk_block in zone. 23723b1a94c8SDamien Le Moal */ 23733b1a94c8SDamien Le Moal static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd, 23743b1a94c8SDamien Le Moal struct dm_zone *zone, 23753b1a94c8SDamien Le Moal sector_t chunk_block) 23763b1a94c8SDamien Le Moal { 23773b1a94c8SDamien Le Moal sector_t bitmap_block = 1 + zmd->nr_map_blocks + 2378b7122873SHannes Reinecke (sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) + 23793b1a94c8SDamien Le Moal (chunk_block >> DMZ_BLOCK_SHIFT_BITS); 23803b1a94c8SDamien Le Moal 23813b1a94c8SDamien Le Moal return dmz_get_mblock(zmd, bitmap_block); 23823b1a94c8SDamien Le Moal } 23833b1a94c8SDamien Le Moal 23843b1a94c8SDamien Le Moal /* 23853b1a94c8SDamien Le Moal * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone. 23863b1a94c8SDamien Le Moal */ 23873b1a94c8SDamien Le Moal int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone, 23883b1a94c8SDamien Le Moal struct dm_zone *to_zone) 23893b1a94c8SDamien Le Moal { 23903b1a94c8SDamien Le Moal struct dmz_mblock *from_mblk, *to_mblk; 23913b1a94c8SDamien Le Moal sector_t chunk_block = 0; 23923b1a94c8SDamien Le Moal 23933b1a94c8SDamien Le Moal /* Get the zones bitmap blocks */ 239436820560SHannes Reinecke while (chunk_block < zmd->zone_nr_blocks) { 23953b1a94c8SDamien Le Moal from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block); 23963b1a94c8SDamien Le Moal if (IS_ERR(from_mblk)) 23973b1a94c8SDamien Le Moal return PTR_ERR(from_mblk); 23983b1a94c8SDamien Le Moal to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block); 23993b1a94c8SDamien Le Moal if (IS_ERR(to_mblk)) { 24003b1a94c8SDamien Le Moal dmz_release_mblock(zmd, from_mblk); 24013b1a94c8SDamien Le Moal return PTR_ERR(to_mblk); 24023b1a94c8SDamien Le Moal } 24033b1a94c8SDamien Le Moal 24043b1a94c8SDamien Le Moal memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE); 24053b1a94c8SDamien Le Moal dmz_dirty_mblock(zmd, to_mblk); 24063b1a94c8SDamien Le Moal 24073b1a94c8SDamien Le Moal dmz_release_mblock(zmd, to_mblk); 24083b1a94c8SDamien Le Moal dmz_release_mblock(zmd, from_mblk); 24093b1a94c8SDamien Le Moal 2410b3996295SDmitry Fomichev chunk_block += zmd->zone_bits_per_mblk; 24113b1a94c8SDamien Le Moal } 24123b1a94c8SDamien Le Moal 24133b1a94c8SDamien Le Moal to_zone->weight = from_zone->weight; 24143b1a94c8SDamien Le Moal 24153b1a94c8SDamien Le Moal return 0; 24163b1a94c8SDamien Le Moal } 24173b1a94c8SDamien Le Moal 24183b1a94c8SDamien Le Moal /* 24193b1a94c8SDamien Le Moal * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone, 24203b1a94c8SDamien Le Moal * starting from chunk_block. 24213b1a94c8SDamien Le Moal */ 24223b1a94c8SDamien Le Moal int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone, 24233b1a94c8SDamien Le Moal struct dm_zone *to_zone, sector_t chunk_block) 24243b1a94c8SDamien Le Moal { 24253b1a94c8SDamien Le Moal unsigned int nr_blocks; 24263b1a94c8SDamien Le Moal int ret; 24273b1a94c8SDamien Le Moal 24283b1a94c8SDamien Le Moal /* Get the zones bitmap blocks */ 242936820560SHannes Reinecke while (chunk_block < zmd->zone_nr_blocks) { 24303b1a94c8SDamien Le Moal /* Get a valid region from the source zone */ 24313b1a94c8SDamien Le Moal ret = dmz_first_valid_block(zmd, from_zone, &chunk_block); 24323b1a94c8SDamien Le Moal if (ret <= 0) 24333b1a94c8SDamien Le Moal return ret; 24343b1a94c8SDamien Le Moal 24353b1a94c8SDamien Le Moal nr_blocks = ret; 24363b1a94c8SDamien Le Moal ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks); 24373b1a94c8SDamien Le Moal if (ret) 24383b1a94c8SDamien Le Moal return ret; 24393b1a94c8SDamien Le Moal 24403b1a94c8SDamien Le Moal chunk_block += nr_blocks; 24413b1a94c8SDamien Le Moal } 24423b1a94c8SDamien Le Moal 24433b1a94c8SDamien Le Moal return 0; 24443b1a94c8SDamien Le Moal } 24453b1a94c8SDamien Le Moal 24463b1a94c8SDamien Le Moal /* 24473b1a94c8SDamien Le Moal * Validate all the blocks in the range [block..block+nr_blocks-1]. 24483b1a94c8SDamien Le Moal */ 24493b1a94c8SDamien Le Moal int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone, 24503b1a94c8SDamien Le Moal sector_t chunk_block, unsigned int nr_blocks) 24513b1a94c8SDamien Le Moal { 24523b1a94c8SDamien Le Moal unsigned int count, bit, nr_bits; 245336820560SHannes Reinecke unsigned int zone_nr_blocks = zmd->zone_nr_blocks; 24543b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 24553b1a94c8SDamien Le Moal unsigned int n = 0; 24563b1a94c8SDamien Le Moal 2457ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks", 2458b7122873SHannes Reinecke zone->id, (unsigned long long)chunk_block, 24593b1a94c8SDamien Le Moal nr_blocks); 24603b1a94c8SDamien Le Moal 24613b1a94c8SDamien Le Moal WARN_ON(chunk_block + nr_blocks > zone_nr_blocks); 24623b1a94c8SDamien Le Moal 24633b1a94c8SDamien Le Moal while (nr_blocks) { 24643b1a94c8SDamien Le Moal /* Get bitmap block */ 24653b1a94c8SDamien Le Moal mblk = dmz_get_bitmap(zmd, zone, chunk_block); 24663b1a94c8SDamien Le Moal if (IS_ERR(mblk)) 24673b1a94c8SDamien Le Moal return PTR_ERR(mblk); 24683b1a94c8SDamien Le Moal 24693b1a94c8SDamien Le Moal /* Set bits */ 24703b1a94c8SDamien Le Moal bit = chunk_block & DMZ_BLOCK_MASK_BITS; 2471b3996295SDmitry Fomichev nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); 24723b1a94c8SDamien Le Moal 24733b1a94c8SDamien Le Moal count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits); 24743b1a94c8SDamien Le Moal if (count) { 24753b1a94c8SDamien Le Moal dmz_dirty_mblock(zmd, mblk); 24763b1a94c8SDamien Le Moal n += count; 24773b1a94c8SDamien Le Moal } 24783b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 24793b1a94c8SDamien Le Moal 24803b1a94c8SDamien Le Moal nr_blocks -= nr_bits; 24813b1a94c8SDamien Le Moal chunk_block += nr_bits; 24823b1a94c8SDamien Le Moal } 24833b1a94c8SDamien Le Moal 24843b1a94c8SDamien Le Moal if (likely(zone->weight + n <= zone_nr_blocks)) 24853b1a94c8SDamien Le Moal zone->weight += n; 24863b1a94c8SDamien Le Moal else { 2487ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u", 2488b7122873SHannes Reinecke zone->id, zone->weight, 24893b1a94c8SDamien Le Moal zone_nr_blocks - n); 24903b1a94c8SDamien Le Moal zone->weight = zone_nr_blocks; 24913b1a94c8SDamien Le Moal } 24923b1a94c8SDamien Le Moal 24933b1a94c8SDamien Le Moal return 0; 24943b1a94c8SDamien Le Moal } 24953b1a94c8SDamien Le Moal 24963b1a94c8SDamien Le Moal /* 24973b1a94c8SDamien Le Moal * Clear nr_bits bits in bitmap starting from bit. 24983b1a94c8SDamien Le Moal * Return the number of bits cleared. 24993b1a94c8SDamien Le Moal */ 25003b1a94c8SDamien Le Moal static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits) 25013b1a94c8SDamien Le Moal { 25023b1a94c8SDamien Le Moal unsigned long *addr; 25033b1a94c8SDamien Le Moal int end = bit + nr_bits; 25043b1a94c8SDamien Le Moal int n = 0; 25053b1a94c8SDamien Le Moal 25063b1a94c8SDamien Le Moal while (bit < end) { 25073b1a94c8SDamien Le Moal if (((bit & (BITS_PER_LONG - 1)) == 0) && 25083b1a94c8SDamien Le Moal ((end - bit) >= BITS_PER_LONG)) { 25093b1a94c8SDamien Le Moal /* Try to clear whole word at once */ 25103b1a94c8SDamien Le Moal addr = bitmap + BIT_WORD(bit); 25113b1a94c8SDamien Le Moal if (*addr == ULONG_MAX) { 25123b1a94c8SDamien Le Moal *addr = 0; 25133b1a94c8SDamien Le Moal n += BITS_PER_LONG; 25143b1a94c8SDamien Le Moal bit += BITS_PER_LONG; 25153b1a94c8SDamien Le Moal continue; 25163b1a94c8SDamien Le Moal } 25173b1a94c8SDamien Le Moal } 25183b1a94c8SDamien Le Moal 25193b1a94c8SDamien Le Moal if (test_and_clear_bit(bit, bitmap)) 25203b1a94c8SDamien Le Moal n++; 25213b1a94c8SDamien Le Moal bit++; 25223b1a94c8SDamien Le Moal } 25233b1a94c8SDamien Le Moal 25243b1a94c8SDamien Le Moal return n; 25253b1a94c8SDamien Le Moal } 25263b1a94c8SDamien Le Moal 25273b1a94c8SDamien Le Moal /* 25283b1a94c8SDamien Le Moal * Invalidate all the blocks in the range [block..block+nr_blocks-1]. 25293b1a94c8SDamien Le Moal */ 25303b1a94c8SDamien Le Moal int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone, 25313b1a94c8SDamien Le Moal sector_t chunk_block, unsigned int nr_blocks) 25323b1a94c8SDamien Le Moal { 25333b1a94c8SDamien Le Moal unsigned int count, bit, nr_bits; 25343b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 25353b1a94c8SDamien Le Moal unsigned int n = 0; 25363b1a94c8SDamien Le Moal 2537ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks", 2538b7122873SHannes Reinecke zone->id, (u64)chunk_block, nr_blocks); 25393b1a94c8SDamien Le Moal 254036820560SHannes Reinecke WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks); 25413b1a94c8SDamien Le Moal 25423b1a94c8SDamien Le Moal while (nr_blocks) { 25433b1a94c8SDamien Le Moal /* Get bitmap block */ 25443b1a94c8SDamien Le Moal mblk = dmz_get_bitmap(zmd, zone, chunk_block); 25453b1a94c8SDamien Le Moal if (IS_ERR(mblk)) 25463b1a94c8SDamien Le Moal return PTR_ERR(mblk); 25473b1a94c8SDamien Le Moal 25483b1a94c8SDamien Le Moal /* Clear bits */ 25493b1a94c8SDamien Le Moal bit = chunk_block & DMZ_BLOCK_MASK_BITS; 2550b3996295SDmitry Fomichev nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); 25513b1a94c8SDamien Le Moal 25523b1a94c8SDamien Le Moal count = dmz_clear_bits((unsigned long *)mblk->data, 25533b1a94c8SDamien Le Moal bit, nr_bits); 25543b1a94c8SDamien Le Moal if (count) { 25553b1a94c8SDamien Le Moal dmz_dirty_mblock(zmd, mblk); 25563b1a94c8SDamien Le Moal n += count; 25573b1a94c8SDamien Le Moal } 25583b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 25593b1a94c8SDamien Le Moal 25603b1a94c8SDamien Le Moal nr_blocks -= nr_bits; 25613b1a94c8SDamien Le Moal chunk_block += nr_bits; 25623b1a94c8SDamien Le Moal } 25633b1a94c8SDamien Le Moal 25643b1a94c8SDamien Le Moal if (zone->weight >= n) 25653b1a94c8SDamien Le Moal zone->weight -= n; 25663b1a94c8SDamien Le Moal else { 2567ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u", 2568b7122873SHannes Reinecke zone->id, zone->weight, n); 25693b1a94c8SDamien Le Moal zone->weight = 0; 25703b1a94c8SDamien Le Moal } 25713b1a94c8SDamien Le Moal 25723b1a94c8SDamien Le Moal return 0; 25733b1a94c8SDamien Le Moal } 25743b1a94c8SDamien Le Moal 25753b1a94c8SDamien Le Moal /* 25763b1a94c8SDamien Le Moal * Get a block bit value. 25773b1a94c8SDamien Le Moal */ 25783b1a94c8SDamien Le Moal static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone, 25793b1a94c8SDamien Le Moal sector_t chunk_block) 25803b1a94c8SDamien Le Moal { 25813b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 25823b1a94c8SDamien Le Moal int ret; 25833b1a94c8SDamien Le Moal 258436820560SHannes Reinecke WARN_ON(chunk_block >= zmd->zone_nr_blocks); 25853b1a94c8SDamien Le Moal 25863b1a94c8SDamien Le Moal /* Get bitmap block */ 25873b1a94c8SDamien Le Moal mblk = dmz_get_bitmap(zmd, zone, chunk_block); 25883b1a94c8SDamien Le Moal if (IS_ERR(mblk)) 25893b1a94c8SDamien Le Moal return PTR_ERR(mblk); 25903b1a94c8SDamien Le Moal 25913b1a94c8SDamien Le Moal /* Get offset */ 25923b1a94c8SDamien Le Moal ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS, 25933b1a94c8SDamien Le Moal (unsigned long *) mblk->data) != 0; 25943b1a94c8SDamien Le Moal 25953b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 25963b1a94c8SDamien Le Moal 25973b1a94c8SDamien Le Moal return ret; 25983b1a94c8SDamien Le Moal } 25993b1a94c8SDamien Le Moal 26003b1a94c8SDamien Le Moal /* 26013b1a94c8SDamien Le Moal * Return the number of blocks from chunk_block to the first block with a bit 26023b1a94c8SDamien Le Moal * value specified by set. Search at most nr_blocks blocks from chunk_block. 26033b1a94c8SDamien Le Moal */ 26043b1a94c8SDamien Le Moal static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone, 26053b1a94c8SDamien Le Moal sector_t chunk_block, unsigned int nr_blocks, 26063b1a94c8SDamien Le Moal int set) 26073b1a94c8SDamien Le Moal { 26083b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 26093b1a94c8SDamien Le Moal unsigned int bit, set_bit, nr_bits; 2610b3996295SDmitry Fomichev unsigned int zone_bits = zmd->zone_bits_per_mblk; 26113b1a94c8SDamien Le Moal unsigned long *bitmap; 26123b1a94c8SDamien Le Moal int n = 0; 26133b1a94c8SDamien Le Moal 261436820560SHannes Reinecke WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks); 26153b1a94c8SDamien Le Moal 26163b1a94c8SDamien Le Moal while (nr_blocks) { 26173b1a94c8SDamien Le Moal /* Get bitmap block */ 26183b1a94c8SDamien Le Moal mblk = dmz_get_bitmap(zmd, zone, chunk_block); 26193b1a94c8SDamien Le Moal if (IS_ERR(mblk)) 26203b1a94c8SDamien Le Moal return PTR_ERR(mblk); 26213b1a94c8SDamien Le Moal 26223b1a94c8SDamien Le Moal /* Get offset */ 26233b1a94c8SDamien Le Moal bitmap = (unsigned long *) mblk->data; 26243b1a94c8SDamien Le Moal bit = chunk_block & DMZ_BLOCK_MASK_BITS; 2625b3996295SDmitry Fomichev nr_bits = min(nr_blocks, zone_bits - bit); 26263b1a94c8SDamien Le Moal if (set) 2627b3996295SDmitry Fomichev set_bit = find_next_bit(bitmap, zone_bits, bit); 26283b1a94c8SDamien Le Moal else 2629b3996295SDmitry Fomichev set_bit = find_next_zero_bit(bitmap, zone_bits, bit); 26303b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 26313b1a94c8SDamien Le Moal 26323b1a94c8SDamien Le Moal n += set_bit - bit; 2633b3996295SDmitry Fomichev if (set_bit < zone_bits) 26343b1a94c8SDamien Le Moal break; 26353b1a94c8SDamien Le Moal 26363b1a94c8SDamien Le Moal nr_blocks -= nr_bits; 26373b1a94c8SDamien Le Moal chunk_block += nr_bits; 26383b1a94c8SDamien Le Moal } 26393b1a94c8SDamien Le Moal 26403b1a94c8SDamien Le Moal return n; 26413b1a94c8SDamien Le Moal } 26423b1a94c8SDamien Le Moal 26433b1a94c8SDamien Le Moal /* 26443b1a94c8SDamien Le Moal * Test if chunk_block is valid. If it is, the number of consecutive 26453b1a94c8SDamien Le Moal * valid blocks from chunk_block will be returned. 26463b1a94c8SDamien Le Moal */ 26473b1a94c8SDamien Le Moal int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone, 26483b1a94c8SDamien Le Moal sector_t chunk_block) 26493b1a94c8SDamien Le Moal { 26503b1a94c8SDamien Le Moal int valid; 26513b1a94c8SDamien Le Moal 26523b1a94c8SDamien Le Moal valid = dmz_test_block(zmd, zone, chunk_block); 26533b1a94c8SDamien Le Moal if (valid <= 0) 26543b1a94c8SDamien Le Moal return valid; 26553b1a94c8SDamien Le Moal 26563b1a94c8SDamien Le Moal /* The block is valid: get the number of valid blocks from block */ 26573b1a94c8SDamien Le Moal return dmz_to_next_set_block(zmd, zone, chunk_block, 265836820560SHannes Reinecke zmd->zone_nr_blocks - chunk_block, 0); 26593b1a94c8SDamien Le Moal } 26603b1a94c8SDamien Le Moal 26613b1a94c8SDamien Le Moal /* 26623b1a94c8SDamien Le Moal * Find the first valid block from @chunk_block in @zone. 26633b1a94c8SDamien Le Moal * If such a block is found, its number is returned using 26643b1a94c8SDamien Le Moal * @chunk_block and the total number of valid blocks from @chunk_block 26653b1a94c8SDamien Le Moal * is returned. 26663b1a94c8SDamien Le Moal */ 26673b1a94c8SDamien Le Moal int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone, 26683b1a94c8SDamien Le Moal sector_t *chunk_block) 26693b1a94c8SDamien Le Moal { 26703b1a94c8SDamien Le Moal sector_t start_block = *chunk_block; 26713b1a94c8SDamien Le Moal int ret; 26723b1a94c8SDamien Le Moal 26733b1a94c8SDamien Le Moal ret = dmz_to_next_set_block(zmd, zone, start_block, 267436820560SHannes Reinecke zmd->zone_nr_blocks - start_block, 1); 26753b1a94c8SDamien Le Moal if (ret < 0) 26763b1a94c8SDamien Le Moal return ret; 26773b1a94c8SDamien Le Moal 26783b1a94c8SDamien Le Moal start_block += ret; 26793b1a94c8SDamien Le Moal *chunk_block = start_block; 26803b1a94c8SDamien Le Moal 26813b1a94c8SDamien Le Moal return dmz_to_next_set_block(zmd, zone, start_block, 268236820560SHannes Reinecke zmd->zone_nr_blocks - start_block, 0); 26833b1a94c8SDamien Le Moal } 26843b1a94c8SDamien Le Moal 26853b1a94c8SDamien Le Moal /* 26863b1a94c8SDamien Le Moal * Count the number of bits set starting from bit up to bit + nr_bits - 1. 26873b1a94c8SDamien Le Moal */ 26883b1a94c8SDamien Le Moal static int dmz_count_bits(void *bitmap, int bit, int nr_bits) 26893b1a94c8SDamien Le Moal { 26903b1a94c8SDamien Le Moal unsigned long *addr; 26913b1a94c8SDamien Le Moal int end = bit + nr_bits; 26923b1a94c8SDamien Le Moal int n = 0; 26933b1a94c8SDamien Le Moal 26943b1a94c8SDamien Le Moal while (bit < end) { 26953b1a94c8SDamien Le Moal if (((bit & (BITS_PER_LONG - 1)) == 0) && 26963b1a94c8SDamien Le Moal ((end - bit) >= BITS_PER_LONG)) { 26973b1a94c8SDamien Le Moal addr = (unsigned long *)bitmap + BIT_WORD(bit); 26983b1a94c8SDamien Le Moal if (*addr == ULONG_MAX) { 26993b1a94c8SDamien Le Moal n += BITS_PER_LONG; 27003b1a94c8SDamien Le Moal bit += BITS_PER_LONG; 27013b1a94c8SDamien Le Moal continue; 27023b1a94c8SDamien Le Moal } 27033b1a94c8SDamien Le Moal } 27043b1a94c8SDamien Le Moal 27053b1a94c8SDamien Le Moal if (test_bit(bit, bitmap)) 27063b1a94c8SDamien Le Moal n++; 27073b1a94c8SDamien Le Moal bit++; 27083b1a94c8SDamien Le Moal } 27093b1a94c8SDamien Le Moal 27103b1a94c8SDamien Le Moal return n; 27113b1a94c8SDamien Le Moal } 27123b1a94c8SDamien Le Moal 27133b1a94c8SDamien Le Moal /* 27143b1a94c8SDamien Le Moal * Get a zone weight. 27153b1a94c8SDamien Le Moal */ 27163b1a94c8SDamien Le Moal static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone) 27173b1a94c8SDamien Le Moal { 27183b1a94c8SDamien Le Moal struct dmz_mblock *mblk; 27193b1a94c8SDamien Le Moal sector_t chunk_block = 0; 27203b1a94c8SDamien Le Moal unsigned int bit, nr_bits; 272136820560SHannes Reinecke unsigned int nr_blocks = zmd->zone_nr_blocks; 27223b1a94c8SDamien Le Moal void *bitmap; 27233b1a94c8SDamien Le Moal int n = 0; 27243b1a94c8SDamien Le Moal 27253b1a94c8SDamien Le Moal while (nr_blocks) { 27263b1a94c8SDamien Le Moal /* Get bitmap block */ 27273b1a94c8SDamien Le Moal mblk = dmz_get_bitmap(zmd, zone, chunk_block); 27283b1a94c8SDamien Le Moal if (IS_ERR(mblk)) { 27293b1a94c8SDamien Le Moal n = 0; 27303b1a94c8SDamien Le Moal break; 27313b1a94c8SDamien Le Moal } 27323b1a94c8SDamien Le Moal 27333b1a94c8SDamien Le Moal /* Count bits in this block */ 27343b1a94c8SDamien Le Moal bitmap = mblk->data; 27353b1a94c8SDamien Le Moal bit = chunk_block & DMZ_BLOCK_MASK_BITS; 2736b3996295SDmitry Fomichev nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); 27373b1a94c8SDamien Le Moal n += dmz_count_bits(bitmap, bit, nr_bits); 27383b1a94c8SDamien Le Moal 27393b1a94c8SDamien Le Moal dmz_release_mblock(zmd, mblk); 27403b1a94c8SDamien Le Moal 27413b1a94c8SDamien Le Moal nr_blocks -= nr_bits; 27423b1a94c8SDamien Le Moal chunk_block += nr_bits; 27433b1a94c8SDamien Le Moal } 27443b1a94c8SDamien Le Moal 27453b1a94c8SDamien Le Moal zone->weight = n; 27463b1a94c8SDamien Le Moal } 27473b1a94c8SDamien Le Moal 27483b1a94c8SDamien Le Moal /* 27493b1a94c8SDamien Le Moal * Cleanup the zoned metadata resources. 27503b1a94c8SDamien Le Moal */ 27513b1a94c8SDamien Le Moal static void dmz_cleanup_metadata(struct dmz_metadata *zmd) 27523b1a94c8SDamien Le Moal { 27533b1a94c8SDamien Le Moal struct rb_root *root; 27543b1a94c8SDamien Le Moal struct dmz_mblock *mblk, *next; 27553b1a94c8SDamien Le Moal int i; 27563b1a94c8SDamien Le Moal 27573b1a94c8SDamien Le Moal /* Release zone mapping resources */ 27583b1a94c8SDamien Le Moal if (zmd->map_mblk) { 27593b1a94c8SDamien Le Moal for (i = 0; i < zmd->nr_map_blocks; i++) 27603b1a94c8SDamien Le Moal dmz_release_mblock(zmd, zmd->map_mblk[i]); 27613b1a94c8SDamien Le Moal kfree(zmd->map_mblk); 27623b1a94c8SDamien Le Moal zmd->map_mblk = NULL; 27633b1a94c8SDamien Le Moal } 27643b1a94c8SDamien Le Moal 27653b1a94c8SDamien Le Moal /* Release super blocks */ 27663b1a94c8SDamien Le Moal for (i = 0; i < 2; i++) { 27673b1a94c8SDamien Le Moal if (zmd->sb[i].mblk) { 27683b1a94c8SDamien Le Moal dmz_free_mblock(zmd, zmd->sb[i].mblk); 27693b1a94c8SDamien Le Moal zmd->sb[i].mblk = NULL; 27703b1a94c8SDamien Le Moal } 27713b1a94c8SDamien Le Moal } 27723b1a94c8SDamien Le Moal 27733b1a94c8SDamien Le Moal /* Free cached blocks */ 27743b1a94c8SDamien Le Moal while (!list_empty(&zmd->mblk_dirty_list)) { 27753b1a94c8SDamien Le Moal mblk = list_first_entry(&zmd->mblk_dirty_list, 27763b1a94c8SDamien Le Moal struct dmz_mblock, link); 2777ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)", 277833c2865fSDamien Le Moal (u64)mblk->no, mblk->ref); 27793b1a94c8SDamien Le Moal list_del_init(&mblk->link); 27803b1a94c8SDamien Le Moal rb_erase(&mblk->node, &zmd->mblk_rbtree); 27813b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 27823b1a94c8SDamien Le Moal } 27833b1a94c8SDamien Le Moal 27843b1a94c8SDamien Le Moal while (!list_empty(&zmd->mblk_lru_list)) { 27853b1a94c8SDamien Le Moal mblk = list_first_entry(&zmd->mblk_lru_list, 27863b1a94c8SDamien Le Moal struct dmz_mblock, link); 27873b1a94c8SDamien Le Moal list_del_init(&mblk->link); 27883b1a94c8SDamien Le Moal rb_erase(&mblk->node, &zmd->mblk_rbtree); 27893b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 27903b1a94c8SDamien Le Moal } 27913b1a94c8SDamien Le Moal 27923b1a94c8SDamien Le Moal /* Sanity checks: the mblock rbtree should now be empty */ 27933b1a94c8SDamien Le Moal root = &zmd->mblk_rbtree; 27943b1a94c8SDamien Le Moal rbtree_postorder_for_each_entry_safe(mblk, next, root, node) { 2795ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree", 279633c2865fSDamien Le Moal (u64)mblk->no, mblk->ref); 279733c2865fSDamien Le Moal mblk->ref = 0; 27983b1a94c8SDamien Le Moal dmz_free_mblock(zmd, mblk); 27993b1a94c8SDamien Le Moal } 28003b1a94c8SDamien Le Moal 28013b1a94c8SDamien Le Moal /* Free the zone descriptors */ 28023b1a94c8SDamien Le Moal dmz_drop_zones(zmd); 2803d5ffebddSMike Snitzer 2804d5ffebddSMike Snitzer mutex_destroy(&zmd->mblk_flush_lock); 2805d5ffebddSMike Snitzer mutex_destroy(&zmd->map_lock); 28063b1a94c8SDamien Le Moal } 28073b1a94c8SDamien Le Moal 2808ca1a7045SHannes Reinecke static void dmz_print_dev(struct dmz_metadata *zmd, int num) 2809ca1a7045SHannes Reinecke { 2810ca1a7045SHannes Reinecke struct dmz_dev *dev = &zmd->dev[num]; 2811ca1a7045SHannes Reinecke 2812bd5c4031SHannes Reinecke if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE) 2813bd5c4031SHannes Reinecke dmz_dev_info(dev, "Regular block device"); 2814bd5c4031SHannes Reinecke else 2815ca1a7045SHannes Reinecke dmz_dev_info(dev, "Host-%s zoned block device", 2816ca1a7045SHannes Reinecke bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ? 2817ca1a7045SHannes Reinecke "aware" : "managed"); 2818bd5c4031SHannes Reinecke if (zmd->sb_version > 1) { 2819bd5c4031SHannes Reinecke sector_t sector_offset = 2820bd5c4031SHannes Reinecke dev->zone_offset << zmd->zone_nr_sectors_shift; 2821bd5c4031SHannes Reinecke 2822bd5c4031SHannes Reinecke dmz_dev_info(dev, " %llu 512-byte logical sectors (offset %llu)", 2823bd5c4031SHannes Reinecke (u64)dev->capacity, (u64)sector_offset); 2824bd5c4031SHannes Reinecke dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors (offset %llu)", 2825bd5c4031SHannes Reinecke dev->nr_zones, (u64)zmd->zone_nr_sectors, 2826bd5c4031SHannes Reinecke (u64)dev->zone_offset); 2827bd5c4031SHannes Reinecke } else { 2828ca1a7045SHannes Reinecke dmz_dev_info(dev, " %llu 512-byte logical sectors", 2829ca1a7045SHannes Reinecke (u64)dev->capacity); 2830ca1a7045SHannes Reinecke dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors", 2831ca1a7045SHannes Reinecke dev->nr_zones, (u64)zmd->zone_nr_sectors); 2832ca1a7045SHannes Reinecke } 2833bd5c4031SHannes Reinecke } 2834ca1a7045SHannes Reinecke 28353b1a94c8SDamien Le Moal /* 28363b1a94c8SDamien Le Moal * Initialize the zoned metadata. 28373b1a94c8SDamien Le Moal */ 2838bd5c4031SHannes Reinecke int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev, 2839bd5c4031SHannes Reinecke struct dmz_metadata **metadata, 28402234e732SHannes Reinecke const char *devname) 28413b1a94c8SDamien Le Moal { 28423b1a94c8SDamien Le Moal struct dmz_metadata *zmd; 2843b7122873SHannes Reinecke unsigned int i; 28443b1a94c8SDamien Le Moal struct dm_zone *zone; 28453b1a94c8SDamien Le Moal int ret; 28463b1a94c8SDamien Le Moal 28473b1a94c8SDamien Le Moal zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL); 28483b1a94c8SDamien Le Moal if (!zmd) 28493b1a94c8SDamien Le Moal return -ENOMEM; 28503b1a94c8SDamien Le Moal 28512234e732SHannes Reinecke strcpy(zmd->devname, devname); 28523b1a94c8SDamien Le Moal zmd->dev = dev; 2853bd5c4031SHannes Reinecke zmd->nr_devs = num_dev; 28543b1a94c8SDamien Le Moal zmd->mblk_rbtree = RB_ROOT; 28553b1a94c8SDamien Le Moal init_rwsem(&zmd->mblk_sem); 28563b1a94c8SDamien Le Moal mutex_init(&zmd->mblk_flush_lock); 28573b1a94c8SDamien Le Moal spin_lock_init(&zmd->mblk_lock); 28583b1a94c8SDamien Le Moal INIT_LIST_HEAD(&zmd->mblk_lru_list); 28593b1a94c8SDamien Le Moal INIT_LIST_HEAD(&zmd->mblk_dirty_list); 28603b1a94c8SDamien Le Moal 28613b1a94c8SDamien Le Moal mutex_init(&zmd->map_lock); 28623b1a94c8SDamien Le Moal 286334f5affdSHannes Reinecke atomic_set(&zmd->unmap_nr_cache, 0); 286434f5affdSHannes Reinecke INIT_LIST_HEAD(&zmd->unmap_cache_list); 286534f5affdSHannes Reinecke INIT_LIST_HEAD(&zmd->map_cache_list); 286634f5affdSHannes Reinecke 28673b1a94c8SDamien Le Moal atomic_set(&zmd->nr_reserved_seq_zones, 0); 28683b1a94c8SDamien Le Moal INIT_LIST_HEAD(&zmd->reserved_seq_zones_list); 28693b1a94c8SDamien Le Moal 28703b1a94c8SDamien Le Moal init_waitqueue_head(&zmd->free_wq); 28713b1a94c8SDamien Le Moal 28723b1a94c8SDamien Le Moal /* Initialize zone descriptors */ 28733b1a94c8SDamien Le Moal ret = dmz_init_zones(zmd); 28743b1a94c8SDamien Le Moal if (ret) 28753b1a94c8SDamien Le Moal goto err; 28763b1a94c8SDamien Le Moal 28773b1a94c8SDamien Le Moal /* Get super block */ 28783b1a94c8SDamien Le Moal ret = dmz_load_sb(zmd); 28793b1a94c8SDamien Le Moal if (ret) 28803b1a94c8SDamien Le Moal goto err; 28813b1a94c8SDamien Le Moal 28823b1a94c8SDamien Le Moal /* Set metadata zones starting from sb_zone */ 28833b1a94c8SDamien Le Moal for (i = 0; i < zmd->nr_meta_zones << 1; i++) { 2884735bd7e4SHannes Reinecke zone = dmz_get(zmd, zmd->sb[0].zone->id + i); 2885a92fbc44SHannes Reinecke if (!zone) { 2886a92fbc44SHannes Reinecke dmz_zmd_err(zmd, 2887a92fbc44SHannes Reinecke "metadata zone %u not present", i); 2888a92fbc44SHannes Reinecke ret = -ENXIO; 2889a92fbc44SHannes Reinecke goto err; 2890a92fbc44SHannes Reinecke } 289134f5affdSHannes Reinecke if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) { 2892bd5c4031SHannes Reinecke dmz_zmd_err(zmd, 2893bd5c4031SHannes Reinecke "metadata zone %d is not random", i); 2894bd5c4031SHannes Reinecke ret = -ENXIO; 28953b1a94c8SDamien Le Moal goto err; 2896bd5c4031SHannes Reinecke } 28973b1a94c8SDamien Le Moal set_bit(DMZ_META, &zone->flags); 28983b1a94c8SDamien Le Moal } 28993b1a94c8SDamien Le Moal /* Load mapping table */ 29003b1a94c8SDamien Le Moal ret = dmz_load_mapping(zmd); 29013b1a94c8SDamien Le Moal if (ret) 29023b1a94c8SDamien Le Moal goto err; 29033b1a94c8SDamien Le Moal 29043b1a94c8SDamien Le Moal /* 29053b1a94c8SDamien Le Moal * Cache size boundaries: allow at least 2 super blocks, the chunk map 29063b1a94c8SDamien Le Moal * blocks and enough blocks to be able to cache the bitmap blocks of 29073b1a94c8SDamien Le Moal * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow 29083b1a94c8SDamien Le Moal * the cache to add 512 more metadata blocks. 29093b1a94c8SDamien Le Moal */ 29103b1a94c8SDamien Le Moal zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16; 29113b1a94c8SDamien Le Moal zmd->max_nr_mblks = zmd->min_nr_mblks + 512; 29123b1a94c8SDamien Le Moal zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count; 29133b1a94c8SDamien Le Moal zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan; 29143b1a94c8SDamien Le Moal zmd->mblk_shrinker.seeks = DEFAULT_SEEKS; 29153b1a94c8SDamien Le Moal 29163b1a94c8SDamien Le Moal /* Metadata cache shrinker */ 29173b1a94c8SDamien Le Moal ret = register_shrinker(&zmd->mblk_shrinker); 29183b1a94c8SDamien Le Moal if (ret) { 2919ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Register metadata cache shrinker failed"); 29203b1a94c8SDamien Le Moal goto err; 29213b1a94c8SDamien Le Moal } 29223b1a94c8SDamien Le Moal 2923bd5c4031SHannes Reinecke dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version); 2924bd5c4031SHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) 2925bd5c4031SHannes Reinecke dmz_print_dev(zmd, i); 29263b1a94c8SDamien Le Moal 2927ca1a7045SHannes Reinecke dmz_zmd_info(zmd, " %u zones of %llu 512-byte logical sectors", 2928ca1a7045SHannes Reinecke zmd->nr_zones, (u64)zmd->zone_nr_sectors); 2929ae3c1f11SHannes Reinecke dmz_zmd_debug(zmd, " %u metadata zones", 2930ca1a7045SHannes Reinecke zmd->nr_meta_zones * 2); 2931ae3c1f11SHannes Reinecke dmz_zmd_debug(zmd, " %u data zones for %u chunks", 2932ca1a7045SHannes Reinecke zmd->nr_data_zones, zmd->nr_chunks); 293334f5affdSHannes Reinecke dmz_zmd_debug(zmd, " %u cache zones (%u unmapped)", 293434f5affdSHannes Reinecke zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache)); 2935bd82fdabSHannes Reinecke for (i = 0; i < zmd->nr_devs; i++) { 2936ae3c1f11SHannes Reinecke dmz_zmd_debug(zmd, " %u random zones (%u unmapped)", 2937bd82fdabSHannes Reinecke dmz_nr_rnd_zones(zmd, i), 2938bd82fdabSHannes Reinecke dmz_nr_unmap_rnd_zones(zmd, i)); 2939ae3c1f11SHannes Reinecke dmz_zmd_debug(zmd, " %u sequential zones (%u unmapped)", 2940bd82fdabSHannes Reinecke dmz_nr_seq_zones(zmd, i), 2941bd82fdabSHannes Reinecke dmz_nr_unmap_seq_zones(zmd, i)); 2942bd82fdabSHannes Reinecke } 2943ae3c1f11SHannes Reinecke dmz_zmd_debug(zmd, " %u reserved sequential data zones", 2944ca1a7045SHannes Reinecke zmd->nr_reserved_seq); 2945ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, "Format:"); 2946ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)", 29473b1a94c8SDamien Le Moal zmd->nr_meta_blocks, zmd->max_nr_mblks); 2948ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, " %u data zone mapping blocks", 29493b1a94c8SDamien Le Moal zmd->nr_map_blocks); 2950ca1a7045SHannes Reinecke dmz_zmd_debug(zmd, " %u bitmap blocks", 29513b1a94c8SDamien Le Moal zmd->nr_bitmap_blocks); 29523b1a94c8SDamien Le Moal 29533b1a94c8SDamien Le Moal *metadata = zmd; 29543b1a94c8SDamien Le Moal 29553b1a94c8SDamien Le Moal return 0; 29563b1a94c8SDamien Le Moal err: 29573b1a94c8SDamien Le Moal dmz_cleanup_metadata(zmd); 29583b1a94c8SDamien Le Moal kfree(zmd); 29593b1a94c8SDamien Le Moal *metadata = NULL; 29603b1a94c8SDamien Le Moal 29613b1a94c8SDamien Le Moal return ret; 29623b1a94c8SDamien Le Moal } 29633b1a94c8SDamien Le Moal 29643b1a94c8SDamien Le Moal /* 29653b1a94c8SDamien Le Moal * Cleanup the zoned metadata resources. 29663b1a94c8SDamien Le Moal */ 29673b1a94c8SDamien Le Moal void dmz_dtr_metadata(struct dmz_metadata *zmd) 29683b1a94c8SDamien Le Moal { 29693b1a94c8SDamien Le Moal unregister_shrinker(&zmd->mblk_shrinker); 29703b1a94c8SDamien Le Moal dmz_cleanup_metadata(zmd); 29713b1a94c8SDamien Le Moal kfree(zmd); 29723b1a94c8SDamien Le Moal } 29733b1a94c8SDamien Le Moal 29743b1a94c8SDamien Le Moal /* 29753b1a94c8SDamien Le Moal * Check zone information on resume. 29763b1a94c8SDamien Le Moal */ 29773b1a94c8SDamien Le Moal int dmz_resume_metadata(struct dmz_metadata *zmd) 29783b1a94c8SDamien Le Moal { 29793b1a94c8SDamien Le Moal struct dm_zone *zone; 29803b1a94c8SDamien Le Moal sector_t wp_block; 29813b1a94c8SDamien Le Moal unsigned int i; 29823b1a94c8SDamien Le Moal int ret; 29833b1a94c8SDamien Le Moal 29843b1a94c8SDamien Le Moal /* Check zones */ 298536820560SHannes Reinecke for (i = 0; i < zmd->nr_zones; i++) { 29863b1a94c8SDamien Le Moal zone = dmz_get(zmd, i); 29873b1a94c8SDamien Le Moal if (!zone) { 2988ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Unable to get zone %u", i); 29893b1a94c8SDamien Le Moal return -EIO; 29903b1a94c8SDamien Le Moal } 29913b1a94c8SDamien Le Moal wp_block = zone->wp_block; 29923b1a94c8SDamien Le Moal 29933b1a94c8SDamien Le Moal ret = dmz_update_zone(zmd, zone); 29943b1a94c8SDamien Le Moal if (ret) { 2995ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Broken zone %u", i); 29963b1a94c8SDamien Le Moal return ret; 29973b1a94c8SDamien Le Moal } 29983b1a94c8SDamien Le Moal 29993b1a94c8SDamien Le Moal if (dmz_is_offline(zone)) { 3000ca1a7045SHannes Reinecke dmz_zmd_warn(zmd, "Zone %u is offline", i); 30013b1a94c8SDamien Le Moal continue; 30023b1a94c8SDamien Le Moal } 30033b1a94c8SDamien Le Moal 30043b1a94c8SDamien Le Moal /* Check write pointer */ 30053b1a94c8SDamien Le Moal if (!dmz_is_seq(zone)) 30063b1a94c8SDamien Le Moal zone->wp_block = 0; 30073b1a94c8SDamien Le Moal else if (zone->wp_block != wp_block) { 3008ca1a7045SHannes Reinecke dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)", 30093b1a94c8SDamien Le Moal i, (u64)zone->wp_block, (u64)wp_block); 30103b1a94c8SDamien Le Moal zone->wp_block = wp_block; 30113b1a94c8SDamien Le Moal dmz_invalidate_blocks(zmd, zone, zone->wp_block, 301236820560SHannes Reinecke zmd->zone_nr_blocks - zone->wp_block); 30133b1a94c8SDamien Le Moal } 30143b1a94c8SDamien Le Moal } 30153b1a94c8SDamien Le Moal 30163b1a94c8SDamien Le Moal return 0; 30173b1a94c8SDamien Le Moal } 3018