Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2003 Sistina Software Limited. |
| 3 | * |
| 4 | * This file is released under the GPL. |
| 5 | */ |
| 6 | |
| 7 | #include "dm.h" |
| 8 | #include "dm-bio-list.h" |
| 9 | #include "dm-io.h" |
| 10 | #include "dm-log.h" |
| 11 | #include "kcopyd.h" |
| 12 | |
| 13 | #include <linux/ctype.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/mempool.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/pagemap.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/time.h> |
| 20 | #include <linux/vmalloc.h> |
| 21 | #include <linux/workqueue.h> |
vignesh babu | 6f3c3f0 | 2007-10-19 22:38:44 +0100 | [diff] [blame^] | 22 | #include <linux/log2.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 24 | #define DM_MSG_PREFIX "raid1" |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 25 | #define DM_IO_PAGES 64 |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 26 | |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 27 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 28 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 29 | |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 30 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | /*----------------------------------------------------------------- |
| 33 | * Region hash |
| 34 | * |
| 35 | * The mirror splits itself up into discrete regions. Each |
| 36 | * region can be in one of three states: clean, dirty, |
| 37 | * nosync. There is no need to put clean regions in the hash. |
| 38 | * |
| 39 | * In addition to being present in the hash table a region _may_ |
| 40 | * be present on one of three lists. |
| 41 | * |
| 42 | * clean_regions: Regions on this list have no io pending to |
| 43 | * them, they are in sync, we are no longer interested in them, |
| 44 | * they are dull. rh_update_states() will remove them from the |
| 45 | * hash table. |
| 46 | * |
| 47 | * quiesced_regions: These regions have been spun down, ready |
| 48 | * for recovery. rh_recovery_start() will remove regions from |
| 49 | * this list and hand them to kmirrord, which will schedule the |
| 50 | * recovery io with kcopyd. |
| 51 | * |
| 52 | * recovered_regions: Regions that kcopyd has successfully |
| 53 | * recovered. rh_update_states() will now schedule any delayed |
| 54 | * io, up the recovery_count, and remove the region from the |
| 55 | * hash. |
| 56 | * |
| 57 | * There are 2 locks: |
| 58 | * A rw spin lock 'hash_lock' protects just the hash table, |
| 59 | * this is never held in write mode from interrupt context, |
| 60 | * which I believe means that we only have to disable irqs when |
| 61 | * doing a write lock. |
| 62 | * |
| 63 | * An ordinary spin lock 'region_lock' that protects the three |
| 64 | * lists in the region_hash, with the 'state', 'list' and |
| 65 | * 'bhs_delayed' fields of the regions. This is used from irq |
| 66 | * context, so all other uses will have to suspend local irqs. |
| 67 | *---------------------------------------------------------------*/ |
| 68 | struct mirror_set; |
| 69 | struct region_hash { |
| 70 | struct mirror_set *ms; |
| 71 | uint32_t region_size; |
| 72 | unsigned region_shift; |
| 73 | |
| 74 | /* holds persistent region state */ |
| 75 | struct dirty_log *log; |
| 76 | |
| 77 | /* hash table */ |
| 78 | rwlock_t hash_lock; |
| 79 | mempool_t *region_pool; |
| 80 | unsigned int mask; |
| 81 | unsigned int nr_buckets; |
| 82 | struct list_head *buckets; |
| 83 | |
| 84 | spinlock_t region_lock; |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 85 | atomic_t recovery_in_flight; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | struct semaphore recovery_count; |
| 87 | struct list_head clean_regions; |
| 88 | struct list_head quiesced_regions; |
| 89 | struct list_head recovered_regions; |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 90 | struct list_head failed_recovered_regions; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | }; |
| 92 | |
| 93 | enum { |
| 94 | RH_CLEAN, |
| 95 | RH_DIRTY, |
| 96 | RH_NOSYNC, |
| 97 | RH_RECOVERING |
| 98 | }; |
| 99 | |
| 100 | struct region { |
| 101 | struct region_hash *rh; /* FIXME: can we get rid of this ? */ |
| 102 | region_t key; |
| 103 | int state; |
| 104 | |
| 105 | struct list_head hash_list; |
| 106 | struct list_head list; |
| 107 | |
| 108 | atomic_t pending; |
| 109 | struct bio_list delayed_bios; |
| 110 | }; |
| 111 | |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 112 | |
| 113 | /*----------------------------------------------------------------- |
| 114 | * Mirror set structures. |
| 115 | *---------------------------------------------------------------*/ |
| 116 | struct mirror { |
| 117 | atomic_t error_count; |
| 118 | struct dm_dev *dev; |
| 119 | sector_t offset; |
| 120 | }; |
| 121 | |
| 122 | struct mirror_set { |
| 123 | struct dm_target *ti; |
| 124 | struct list_head list; |
| 125 | struct region_hash rh; |
| 126 | struct kcopyd_client *kcopyd_client; |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 127 | uint64_t features; |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 128 | |
| 129 | spinlock_t lock; /* protects the next two lists */ |
| 130 | struct bio_list reads; |
| 131 | struct bio_list writes; |
| 132 | |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 133 | struct dm_io_client *io_client; |
| 134 | |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 135 | /* recovery */ |
| 136 | region_t nr_regions; |
| 137 | int in_sync; |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 138 | int log_failure; |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 139 | |
| 140 | struct mirror *default_mirror; /* Default mirror */ |
| 141 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 142 | struct workqueue_struct *kmirrord_wq; |
| 143 | struct work_struct kmirrord_work; |
| 144 | |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 145 | unsigned int nr_mirrors; |
| 146 | struct mirror mirror[0]; |
| 147 | }; |
| 148 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | /* |
| 150 | * Conversion fns |
| 151 | */ |
| 152 | static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio) |
| 153 | { |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 154 | return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | } |
| 156 | |
| 157 | static inline sector_t region_to_sector(struct region_hash *rh, region_t region) |
| 158 | { |
| 159 | return region << rh->region_shift; |
| 160 | } |
| 161 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 162 | static void wake(struct mirror_set *ms) |
| 163 | { |
| 164 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); |
| 165 | } |
| 166 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | /* FIXME move this */ |
| 168 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw); |
| 169 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | #define MIN_REGIONS 64 |
| 171 | #define MAX_RECOVERY 1 |
| 172 | static int rh_init(struct region_hash *rh, struct mirror_set *ms, |
| 173 | struct dirty_log *log, uint32_t region_size, |
| 174 | region_t nr_regions) |
| 175 | { |
| 176 | unsigned int nr_buckets, max_buckets; |
| 177 | size_t i; |
| 178 | |
| 179 | /* |
| 180 | * Calculate a suitable number of buckets for our hash |
| 181 | * table. |
| 182 | */ |
| 183 | max_buckets = nr_regions >> 6; |
| 184 | for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1) |
| 185 | ; |
| 186 | nr_buckets >>= 1; |
| 187 | |
| 188 | rh->ms = ms; |
| 189 | rh->log = log; |
| 190 | rh->region_size = region_size; |
| 191 | rh->region_shift = ffs(region_size) - 1; |
| 192 | rwlock_init(&rh->hash_lock); |
| 193 | rh->mask = nr_buckets - 1; |
| 194 | rh->nr_buckets = nr_buckets; |
| 195 | |
| 196 | rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets)); |
| 197 | if (!rh->buckets) { |
| 198 | DMERR("unable to allocate region hash memory"); |
| 199 | return -ENOMEM; |
| 200 | } |
| 201 | |
| 202 | for (i = 0; i < nr_buckets; i++) |
| 203 | INIT_LIST_HEAD(rh->buckets + i); |
| 204 | |
| 205 | spin_lock_init(&rh->region_lock); |
| 206 | sema_init(&rh->recovery_count, 0); |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 207 | atomic_set(&rh->recovery_in_flight, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | INIT_LIST_HEAD(&rh->clean_regions); |
| 209 | INIT_LIST_HEAD(&rh->quiesced_regions); |
| 210 | INIT_LIST_HEAD(&rh->recovered_regions); |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 211 | INIT_LIST_HEAD(&rh->failed_recovered_regions); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | |
Matthew Dobson | 0eaae62a | 2006-03-26 01:37:47 -0800 | [diff] [blame] | 213 | rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS, |
| 214 | sizeof(struct region)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | if (!rh->region_pool) { |
| 216 | vfree(rh->buckets); |
| 217 | rh->buckets = NULL; |
| 218 | return -ENOMEM; |
| 219 | } |
| 220 | |
| 221 | return 0; |
| 222 | } |
| 223 | |
| 224 | static void rh_exit(struct region_hash *rh) |
| 225 | { |
| 226 | unsigned int h; |
| 227 | struct region *reg, *nreg; |
| 228 | |
| 229 | BUG_ON(!list_empty(&rh->quiesced_regions)); |
| 230 | for (h = 0; h < rh->nr_buckets; h++) { |
| 231 | list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) { |
| 232 | BUG_ON(atomic_read(®->pending)); |
| 233 | mempool_free(reg, rh->region_pool); |
| 234 | } |
| 235 | } |
| 236 | |
| 237 | if (rh->log) |
| 238 | dm_destroy_dirty_log(rh->log); |
| 239 | if (rh->region_pool) |
| 240 | mempool_destroy(rh->region_pool); |
| 241 | vfree(rh->buckets); |
| 242 | } |
| 243 | |
| 244 | #define RH_HASH_MULT 2654435387U |
| 245 | |
| 246 | static inline unsigned int rh_hash(struct region_hash *rh, region_t region) |
| 247 | { |
| 248 | return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask; |
| 249 | } |
| 250 | |
| 251 | static struct region *__rh_lookup(struct region_hash *rh, region_t region) |
| 252 | { |
| 253 | struct region *reg; |
| 254 | |
| 255 | list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list) |
| 256 | if (reg->key == region) |
| 257 | return reg; |
| 258 | |
| 259 | return NULL; |
| 260 | } |
| 261 | |
| 262 | static void __rh_insert(struct region_hash *rh, struct region *reg) |
| 263 | { |
| 264 | unsigned int h = rh_hash(rh, reg->key); |
| 265 | list_add(®->hash_list, rh->buckets + h); |
| 266 | } |
| 267 | |
| 268 | static struct region *__rh_alloc(struct region_hash *rh, region_t region) |
| 269 | { |
| 270 | struct region *reg, *nreg; |
| 271 | |
| 272 | read_unlock(&rh->hash_lock); |
Daniel Kobras | c06aad8 | 2006-08-27 01:23:24 -0700 | [diff] [blame] | 273 | nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC); |
| 274 | if (unlikely(!nreg)) |
| 275 | nreg = kmalloc(sizeof(struct region), GFP_NOIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | nreg->state = rh->log->type->in_sync(rh->log, region, 1) ? |
| 277 | RH_CLEAN : RH_NOSYNC; |
| 278 | nreg->rh = rh; |
| 279 | nreg->key = region; |
| 280 | |
| 281 | INIT_LIST_HEAD(&nreg->list); |
| 282 | |
| 283 | atomic_set(&nreg->pending, 0); |
| 284 | bio_list_init(&nreg->delayed_bios); |
| 285 | write_lock_irq(&rh->hash_lock); |
| 286 | |
| 287 | reg = __rh_lookup(rh, region); |
| 288 | if (reg) |
| 289 | /* we lost the race */ |
| 290 | mempool_free(nreg, rh->region_pool); |
| 291 | |
| 292 | else { |
| 293 | __rh_insert(rh, nreg); |
| 294 | if (nreg->state == RH_CLEAN) { |
| 295 | spin_lock(&rh->region_lock); |
| 296 | list_add(&nreg->list, &rh->clean_regions); |
| 297 | spin_unlock(&rh->region_lock); |
| 298 | } |
| 299 | reg = nreg; |
| 300 | } |
| 301 | write_unlock_irq(&rh->hash_lock); |
| 302 | read_lock(&rh->hash_lock); |
| 303 | |
| 304 | return reg; |
| 305 | } |
| 306 | |
| 307 | static inline struct region *__rh_find(struct region_hash *rh, region_t region) |
| 308 | { |
| 309 | struct region *reg; |
| 310 | |
| 311 | reg = __rh_lookup(rh, region); |
| 312 | if (!reg) |
| 313 | reg = __rh_alloc(rh, region); |
| 314 | |
| 315 | return reg; |
| 316 | } |
| 317 | |
| 318 | static int rh_state(struct region_hash *rh, region_t region, int may_block) |
| 319 | { |
| 320 | int r; |
| 321 | struct region *reg; |
| 322 | |
| 323 | read_lock(&rh->hash_lock); |
| 324 | reg = __rh_lookup(rh, region); |
| 325 | read_unlock(&rh->hash_lock); |
| 326 | |
| 327 | if (reg) |
| 328 | return reg->state; |
| 329 | |
| 330 | /* |
| 331 | * The region wasn't in the hash, so we fall back to the |
| 332 | * dirty log. |
| 333 | */ |
| 334 | r = rh->log->type->in_sync(rh->log, region, may_block); |
| 335 | |
| 336 | /* |
| 337 | * Any error from the dirty log (eg. -EWOULDBLOCK) gets |
| 338 | * taken as a RH_NOSYNC |
| 339 | */ |
| 340 | return r == 1 ? RH_CLEAN : RH_NOSYNC; |
| 341 | } |
| 342 | |
| 343 | static inline int rh_in_sync(struct region_hash *rh, |
| 344 | region_t region, int may_block) |
| 345 | { |
| 346 | int state = rh_state(rh, region, may_block); |
| 347 | return state == RH_CLEAN || state == RH_DIRTY; |
| 348 | } |
| 349 | |
| 350 | static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list) |
| 351 | { |
| 352 | struct bio *bio; |
| 353 | |
| 354 | while ((bio = bio_list_pop(bio_list))) { |
| 355 | queue_bio(ms, bio, WRITE); |
| 356 | } |
| 357 | } |
| 358 | |
Jonathan E Brassow | f3ee6b2 | 2006-12-08 02:41:11 -0800 | [diff] [blame] | 359 | static void complete_resync_work(struct region *reg, int success) |
| 360 | { |
| 361 | struct region_hash *rh = reg->rh; |
| 362 | |
| 363 | rh->log->type->set_region_sync(rh->log, reg->key, success); |
| 364 | dispatch_bios(rh->ms, ®->delayed_bios); |
| 365 | if (atomic_dec_and_test(&rh->recovery_in_flight)) |
| 366 | wake_up_all(&_kmirrord_recovery_stopped); |
| 367 | up(&rh->recovery_count); |
| 368 | } |
| 369 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | static void rh_update_states(struct region_hash *rh) |
| 371 | { |
| 372 | struct region *reg, *next; |
| 373 | |
| 374 | LIST_HEAD(clean); |
| 375 | LIST_HEAD(recovered); |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 376 | LIST_HEAD(failed_recovered); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | |
| 378 | /* |
| 379 | * Quickly grab the lists. |
| 380 | */ |
| 381 | write_lock_irq(&rh->hash_lock); |
| 382 | spin_lock(&rh->region_lock); |
| 383 | if (!list_empty(&rh->clean_regions)) { |
| 384 | list_splice(&rh->clean_regions, &clean); |
| 385 | INIT_LIST_HEAD(&rh->clean_regions); |
| 386 | |
Jonathan Brassow | 943317e | 2007-07-12 17:28:25 +0100 | [diff] [blame] | 387 | list_for_each_entry(reg, &clean, list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | list_del(®->hash_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | } |
| 390 | |
| 391 | if (!list_empty(&rh->recovered_regions)) { |
| 392 | list_splice(&rh->recovered_regions, &recovered); |
| 393 | INIT_LIST_HEAD(&rh->recovered_regions); |
| 394 | |
| 395 | list_for_each_entry (reg, &recovered, list) |
| 396 | list_del(®->hash_list); |
| 397 | } |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 398 | |
| 399 | if (!list_empty(&rh->failed_recovered_regions)) { |
| 400 | list_splice(&rh->failed_recovered_regions, &failed_recovered); |
| 401 | INIT_LIST_HEAD(&rh->failed_recovered_regions); |
| 402 | |
| 403 | list_for_each_entry(reg, &failed_recovered, list) |
| 404 | list_del(®->hash_list); |
| 405 | } |
| 406 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | spin_unlock(&rh->region_lock); |
| 408 | write_unlock_irq(&rh->hash_lock); |
| 409 | |
| 410 | /* |
| 411 | * All the regions on the recovered and clean lists have |
| 412 | * now been pulled out of the system, so no need to do |
| 413 | * any more locking. |
| 414 | */ |
| 415 | list_for_each_entry_safe (reg, next, &recovered, list) { |
| 416 | rh->log->type->clear_region(rh->log, reg->key); |
Jonathan E Brassow | f3ee6b2 | 2006-12-08 02:41:11 -0800 | [diff] [blame] | 417 | complete_resync_work(reg, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | mempool_free(reg, rh->region_pool); |
| 419 | } |
| 420 | |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 421 | list_for_each_entry_safe(reg, next, &failed_recovered, list) { |
| 422 | complete_resync_work(reg, errors_handled(rh->ms) ? 0 : 1); |
| 423 | mempool_free(reg, rh->region_pool); |
| 424 | } |
| 425 | |
Jonathan Brassow | 943317e | 2007-07-12 17:28:25 +0100 | [diff] [blame] | 426 | list_for_each_entry_safe(reg, next, &clean, list) { |
| 427 | rh->log->type->clear_region(rh->log, reg->key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | mempool_free(reg, rh->region_pool); |
Jonathan Brassow | 943317e | 2007-07-12 17:28:25 +0100 | [diff] [blame] | 429 | } |
| 430 | |
| 431 | rh->log->type->flush(rh->log); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | } |
| 433 | |
| 434 | static void rh_inc(struct region_hash *rh, region_t region) |
| 435 | { |
| 436 | struct region *reg; |
| 437 | |
| 438 | read_lock(&rh->hash_lock); |
| 439 | reg = __rh_find(rh, region); |
Jun'ichi Nomura | 844e8d9 | 2005-09-09 16:23:42 -0700 | [diff] [blame] | 440 | |
Jonathan E Brassow | 7692c5d | 2005-11-21 21:32:37 -0800 | [diff] [blame] | 441 | spin_lock_irq(&rh->region_lock); |
Jun'ichi Nomura | 844e8d9 | 2005-09-09 16:23:42 -0700 | [diff] [blame] | 442 | atomic_inc(®->pending); |
| 443 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | if (reg->state == RH_CLEAN) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | reg->state = RH_DIRTY; |
| 446 | list_del_init(®->list); /* take off the clean list */ |
Jonathan E Brassow | 7692c5d | 2005-11-21 21:32:37 -0800 | [diff] [blame] | 447 | spin_unlock_irq(&rh->region_lock); |
| 448 | |
| 449 | rh->log->type->mark_region(rh->log, reg->key); |
| 450 | } else |
| 451 | spin_unlock_irq(&rh->region_lock); |
| 452 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | read_unlock(&rh->hash_lock); |
| 455 | } |
| 456 | |
| 457 | static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios) |
| 458 | { |
| 459 | struct bio *bio; |
| 460 | |
| 461 | for (bio = bios->head; bio; bio = bio->bi_next) |
| 462 | rh_inc(rh, bio_to_region(rh, bio)); |
| 463 | } |
| 464 | |
| 465 | static void rh_dec(struct region_hash *rh, region_t region) |
| 466 | { |
| 467 | unsigned long flags; |
| 468 | struct region *reg; |
| 469 | int should_wake = 0; |
| 470 | |
| 471 | read_lock(&rh->hash_lock); |
| 472 | reg = __rh_lookup(rh, region); |
| 473 | read_unlock(&rh->hash_lock); |
| 474 | |
Jonathan E Brassow | 7692c5d | 2005-11-21 21:32:37 -0800 | [diff] [blame] | 475 | spin_lock_irqsave(&rh->region_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | if (atomic_dec_and_test(®->pending)) { |
Jun'ichi Nomura | 930d332 | 2006-03-27 01:17:47 -0800 | [diff] [blame] | 477 | /* |
| 478 | * There is no pending I/O for this region. |
| 479 | * We can move the region to corresponding list for next action. |
| 480 | * At this point, the region is not yet connected to any list. |
| 481 | * |
| 482 | * If the state is RH_NOSYNC, the region should be kept off |
| 483 | * from clean list. |
| 484 | * The hash entry for RH_NOSYNC will remain in memory |
| 485 | * until the region is recovered or the map is reloaded. |
| 486 | */ |
| 487 | |
| 488 | /* do nothing for RH_NOSYNC */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | if (reg->state == RH_RECOVERING) { |
| 490 | list_add_tail(®->list, &rh->quiesced_regions); |
Jun'ichi Nomura | 930d332 | 2006-03-27 01:17:47 -0800 | [diff] [blame] | 491 | } else if (reg->state == RH_DIRTY) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | reg->state = RH_CLEAN; |
| 493 | list_add(®->list, &rh->clean_regions); |
| 494 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | should_wake = 1; |
| 496 | } |
Jonathan E Brassow | 7692c5d | 2005-11-21 21:32:37 -0800 | [diff] [blame] | 497 | spin_unlock_irqrestore(&rh->region_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | |
| 499 | if (should_wake) |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 500 | wake(rh->ms); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | } |
| 502 | |
| 503 | /* |
| 504 | * Starts quiescing a region in preparation for recovery. |
| 505 | */ |
| 506 | static int __rh_recovery_prepare(struct region_hash *rh) |
| 507 | { |
| 508 | int r; |
| 509 | struct region *reg; |
| 510 | region_t region; |
| 511 | |
| 512 | /* |
| 513 | * Ask the dirty log what's next. |
| 514 | */ |
| 515 | r = rh->log->type->get_resync_work(rh->log, ®ion); |
| 516 | if (r <= 0) |
| 517 | return r; |
| 518 | |
| 519 | /* |
| 520 | * Get this region, and start it quiescing by setting the |
| 521 | * recovering flag. |
| 522 | */ |
| 523 | read_lock(&rh->hash_lock); |
| 524 | reg = __rh_find(rh, region); |
| 525 | read_unlock(&rh->hash_lock); |
| 526 | |
| 527 | spin_lock_irq(&rh->region_lock); |
| 528 | reg->state = RH_RECOVERING; |
| 529 | |
| 530 | /* Already quiesced ? */ |
| 531 | if (atomic_read(®->pending)) |
| 532 | list_del_init(®->list); |
Akinobu Mita | 179e091 | 2006-06-26 00:24:41 -0700 | [diff] [blame] | 533 | else |
| 534 | list_move(®->list, &rh->quiesced_regions); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | spin_unlock_irq(&rh->region_lock); |
| 537 | |
| 538 | return 1; |
| 539 | } |
| 540 | |
| 541 | static void rh_recovery_prepare(struct region_hash *rh) |
| 542 | { |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 543 | /* Extra reference to avoid race with rh_stop_recovery */ |
| 544 | atomic_inc(&rh->recovery_in_flight); |
| 545 | |
| 546 | while (!down_trylock(&rh->recovery_count)) { |
| 547 | atomic_inc(&rh->recovery_in_flight); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | if (__rh_recovery_prepare(rh) <= 0) { |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 549 | atomic_dec(&rh->recovery_in_flight); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | up(&rh->recovery_count); |
| 551 | break; |
| 552 | } |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 553 | } |
| 554 | |
| 555 | /* Drop the extra reference */ |
| 556 | if (atomic_dec_and_test(&rh->recovery_in_flight)) |
| 557 | wake_up_all(&_kmirrord_recovery_stopped); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | } |
| 559 | |
| 560 | /* |
| 561 | * Returns any quiesced regions. |
| 562 | */ |
| 563 | static struct region *rh_recovery_start(struct region_hash *rh) |
| 564 | { |
| 565 | struct region *reg = NULL; |
| 566 | |
| 567 | spin_lock_irq(&rh->region_lock); |
| 568 | if (!list_empty(&rh->quiesced_regions)) { |
| 569 | reg = list_entry(rh->quiesced_regions.next, |
| 570 | struct region, list); |
| 571 | list_del_init(®->list); /* remove from the quiesced list */ |
| 572 | } |
| 573 | spin_unlock_irq(&rh->region_lock); |
| 574 | |
| 575 | return reg; |
| 576 | } |
| 577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | static void rh_recovery_end(struct region *reg, int success) |
| 579 | { |
| 580 | struct region_hash *rh = reg->rh; |
| 581 | |
| 582 | spin_lock_irq(&rh->region_lock); |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 583 | if (success) |
| 584 | list_add(®->list, ®->rh->recovered_regions); |
| 585 | else { |
| 586 | reg->state = RH_NOSYNC; |
| 587 | list_add(®->list, ®->rh->failed_recovered_regions); |
| 588 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | spin_unlock_irq(&rh->region_lock); |
| 590 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 591 | wake(rh->ms); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | } |
| 593 | |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 594 | static int rh_flush(struct region_hash *rh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | { |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 596 | return rh->log->type->flush(rh->log); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 597 | } |
| 598 | |
| 599 | static void rh_delay(struct region_hash *rh, struct bio *bio) |
| 600 | { |
| 601 | struct region *reg; |
| 602 | |
| 603 | read_lock(&rh->hash_lock); |
| 604 | reg = __rh_find(rh, bio_to_region(rh, bio)); |
| 605 | bio_list_add(®->delayed_bios, bio); |
| 606 | read_unlock(&rh->hash_lock); |
| 607 | } |
| 608 | |
| 609 | static void rh_stop_recovery(struct region_hash *rh) |
| 610 | { |
| 611 | int i; |
| 612 | |
| 613 | /* wait for any recovering regions */ |
| 614 | for (i = 0; i < MAX_RECOVERY; i++) |
| 615 | down(&rh->recovery_count); |
| 616 | } |
| 617 | |
| 618 | static void rh_start_recovery(struct region_hash *rh) |
| 619 | { |
| 620 | int i; |
| 621 | |
| 622 | for (i = 0; i < MAX_RECOVERY; i++) |
| 623 | up(&rh->recovery_count); |
| 624 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 625 | wake(rh->ms); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | } |
| 627 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | /* |
| 629 | * Every mirror should look like this one. |
| 630 | */ |
| 631 | #define DEFAULT_MIRROR 0 |
| 632 | |
| 633 | /* |
| 634 | * This is yucky. We squirrel the mirror_set struct away inside |
| 635 | * bi_next for write buffers. This is safe since the bh |
| 636 | * doesn't get submitted to the lower levels of block layer. |
| 637 | */ |
| 638 | static struct mirror_set *bio_get_ms(struct bio *bio) |
| 639 | { |
| 640 | return (struct mirror_set *) bio->bi_next; |
| 641 | } |
| 642 | |
| 643 | static void bio_set_ms(struct bio *bio, struct mirror_set *ms) |
| 644 | { |
| 645 | bio->bi_next = (struct bio *) ms; |
| 646 | } |
| 647 | |
| 648 | /*----------------------------------------------------------------- |
| 649 | * Recovery. |
| 650 | * |
| 651 | * When a mirror is first activated we may find that some regions |
| 652 | * are in the no-sync state. We have to recover these by |
| 653 | * recopying from the default mirror to all the others. |
| 654 | *---------------------------------------------------------------*/ |
| 655 | static void recovery_complete(int read_err, unsigned int write_err, |
| 656 | void *context) |
| 657 | { |
| 658 | struct region *reg = (struct region *) context; |
| 659 | |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 660 | if (read_err) |
| 661 | /* Read error means the failure of default mirror. */ |
| 662 | DMERR_LIMIT("Unable to read primary mirror during recovery"); |
| 663 | |
| 664 | if (write_err) |
| 665 | DMERR_LIMIT("Write error during recovery (error = 0x%x)", |
| 666 | write_err); |
| 667 | |
Jonathan Brassow | ce503f5 | 2006-06-26 00:27:30 -0700 | [diff] [blame] | 668 | rh_recovery_end(reg, !(read_err || write_err)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | } |
| 670 | |
| 671 | static int recover(struct mirror_set *ms, struct region *reg) |
| 672 | { |
| 673 | int r; |
| 674 | unsigned int i; |
| 675 | struct io_region from, to[KCOPYD_MAX_REGIONS], *dest; |
| 676 | struct mirror *m; |
| 677 | unsigned long flags = 0; |
| 678 | |
| 679 | /* fill in the source */ |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 680 | m = ms->default_mirror; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | from.bdev = m->dev->bdev; |
| 682 | from.sector = m->offset + region_to_sector(reg->rh, reg->key); |
| 683 | if (reg->key == (ms->nr_regions - 1)) { |
| 684 | /* |
| 685 | * The final region may be smaller than |
| 686 | * region_size. |
| 687 | */ |
| 688 | from.count = ms->ti->len & (reg->rh->region_size - 1); |
| 689 | if (!from.count) |
| 690 | from.count = reg->rh->region_size; |
| 691 | } else |
| 692 | from.count = reg->rh->region_size; |
| 693 | |
| 694 | /* fill in the destinations */ |
| 695 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 696 | if (&ms->mirror[i] == ms->default_mirror) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | continue; |
| 698 | |
| 699 | m = ms->mirror + i; |
| 700 | dest->bdev = m->dev->bdev; |
| 701 | dest->sector = m->offset + region_to_sector(reg->rh, reg->key); |
| 702 | dest->count = from.count; |
| 703 | dest++; |
| 704 | } |
| 705 | |
| 706 | /* hand to kcopyd */ |
| 707 | set_bit(KCOPYD_IGNORE_ERROR, &flags); |
| 708 | r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags, |
| 709 | recovery_complete, reg); |
| 710 | |
| 711 | return r; |
| 712 | } |
| 713 | |
| 714 | static void do_recovery(struct mirror_set *ms) |
| 715 | { |
| 716 | int r; |
| 717 | struct region *reg; |
| 718 | struct dirty_log *log = ms->rh.log; |
| 719 | |
| 720 | /* |
| 721 | * Start quiescing some regions. |
| 722 | */ |
| 723 | rh_recovery_prepare(&ms->rh); |
| 724 | |
| 725 | /* |
| 726 | * Copy any already quiesced regions. |
| 727 | */ |
| 728 | while ((reg = rh_recovery_start(&ms->rh))) { |
| 729 | r = recover(ms, reg); |
| 730 | if (r) |
| 731 | rh_recovery_end(reg, 0); |
| 732 | } |
| 733 | |
| 734 | /* |
| 735 | * Update the in sync flag. |
| 736 | */ |
| 737 | if (!ms->in_sync && |
| 738 | (log->type->get_sync_count(log) == ms->nr_regions)) { |
| 739 | /* the sync is complete */ |
| 740 | dm_table_event(ms->ti->table); |
| 741 | ms->in_sync = 1; |
| 742 | } |
| 743 | } |
| 744 | |
| 745 | /*----------------------------------------------------------------- |
| 746 | * Reads |
| 747 | *---------------------------------------------------------------*/ |
| 748 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) |
| 749 | { |
| 750 | /* FIXME: add read balancing */ |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 751 | return ms->default_mirror; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | } |
| 753 | |
| 754 | /* |
| 755 | * remap a buffer to a particular mirror. |
| 756 | */ |
| 757 | static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio) |
| 758 | { |
| 759 | bio->bi_bdev = m->dev->bdev; |
| 760 | bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin); |
| 761 | } |
| 762 | |
| 763 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) |
| 764 | { |
| 765 | region_t region; |
| 766 | struct bio *bio; |
| 767 | struct mirror *m; |
| 768 | |
| 769 | while ((bio = bio_list_pop(reads))) { |
| 770 | region = bio_to_region(&ms->rh, bio); |
| 771 | |
| 772 | /* |
| 773 | * We can only read balance if the region is in sync. |
| 774 | */ |
Jonathan Brassow | b997b82 | 2007-05-09 02:33:08 -0700 | [diff] [blame] | 775 | if (rh_in_sync(&ms->rh, region, 1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | m = choose_mirror(ms, bio->bi_sector); |
| 777 | else |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 778 | m = ms->default_mirror; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | |
| 780 | map_bio(ms, m, bio); |
| 781 | generic_make_request(bio); |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | /*----------------------------------------------------------------- |
| 786 | * Writes. |
| 787 | * |
| 788 | * We do different things with the write io depending on the |
| 789 | * state of the region that it's in: |
| 790 | * |
| 791 | * SYNC: increment pending, use kcopyd to write to *all* mirrors |
| 792 | * RECOVERING: delay the io until recovery completes |
| 793 | * NOSYNC: increment pending, just write to the default mirror |
| 794 | *---------------------------------------------------------------*/ |
| 795 | static void write_callback(unsigned long error, void *context) |
| 796 | { |
| 797 | unsigned int i; |
| 798 | int uptodate = 1; |
| 799 | struct bio *bio = (struct bio *) context; |
| 800 | struct mirror_set *ms; |
| 801 | |
| 802 | ms = bio_get_ms(bio); |
| 803 | bio_set_ms(bio, NULL); |
| 804 | |
| 805 | /* |
| 806 | * NOTE: We don't decrement the pending count here, |
| 807 | * instead it is done by the targets endio function. |
| 808 | * This way we handle both writes to SYNC and NOSYNC |
| 809 | * regions with the same code. |
| 810 | */ |
| 811 | |
| 812 | if (error) { |
| 813 | /* |
| 814 | * only error the io if all mirrors failed. |
| 815 | * FIXME: bogus |
| 816 | */ |
| 817 | uptodate = 0; |
| 818 | for (i = 0; i < ms->nr_mirrors; i++) |
| 819 | if (!test_bit(i, &error)) { |
| 820 | uptodate = 1; |
| 821 | break; |
| 822 | } |
| 823 | } |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 824 | bio_endio(bio, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | } |
| 826 | |
| 827 | static void do_write(struct mirror_set *ms, struct bio *bio) |
| 828 | { |
| 829 | unsigned int i; |
| 830 | struct io_region io[KCOPYD_MAX_REGIONS+1]; |
| 831 | struct mirror *m; |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 832 | struct dm_io_request io_req = { |
| 833 | .bi_rw = WRITE, |
| 834 | .mem.type = DM_IO_BVEC, |
| 835 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, |
| 836 | .notify.fn = write_callback, |
| 837 | .notify.context = bio, |
| 838 | .client = ms->io_client, |
| 839 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | |
| 841 | for (i = 0; i < ms->nr_mirrors; i++) { |
| 842 | m = ms->mirror + i; |
| 843 | |
| 844 | io[i].bdev = m->dev->bdev; |
| 845 | io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin); |
| 846 | io[i].count = bio->bi_size >> 9; |
| 847 | } |
| 848 | |
| 849 | bio_set_ms(bio, ms); |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 850 | |
| 851 | (void) dm_io(&io_req, ms->nr_mirrors, io, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | } |
| 853 | |
| 854 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) |
| 855 | { |
| 856 | int state; |
| 857 | struct bio *bio; |
| 858 | struct bio_list sync, nosync, recover, *this_list = NULL; |
| 859 | |
| 860 | if (!writes->head) |
| 861 | return; |
| 862 | |
| 863 | /* |
| 864 | * Classify each write. |
| 865 | */ |
| 866 | bio_list_init(&sync); |
| 867 | bio_list_init(&nosync); |
| 868 | bio_list_init(&recover); |
| 869 | |
| 870 | while ((bio = bio_list_pop(writes))) { |
| 871 | state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1); |
| 872 | switch (state) { |
| 873 | case RH_CLEAN: |
| 874 | case RH_DIRTY: |
| 875 | this_list = &sync; |
| 876 | break; |
| 877 | |
| 878 | case RH_NOSYNC: |
| 879 | this_list = &nosync; |
| 880 | break; |
| 881 | |
| 882 | case RH_RECOVERING: |
| 883 | this_list = &recover; |
| 884 | break; |
| 885 | } |
| 886 | |
| 887 | bio_list_add(this_list, bio); |
| 888 | } |
| 889 | |
| 890 | /* |
| 891 | * Increment the pending counts for any regions that will |
| 892 | * be written to (writes to recover regions are going to |
| 893 | * be delayed). |
| 894 | */ |
| 895 | rh_inc_pending(&ms->rh, &sync); |
| 896 | rh_inc_pending(&ms->rh, &nosync); |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 897 | ms->log_failure = rh_flush(&ms->rh) ? 1 : 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | |
| 899 | /* |
| 900 | * Dispatch io. |
| 901 | */ |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 902 | if (unlikely(ms->log_failure)) |
| 903 | while ((bio = bio_list_pop(&sync))) |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 904 | bio_endio(bio, -EIO); |
Jonathan Brassow | fc1ff95 | 2007-07-12 17:29:15 +0100 | [diff] [blame] | 905 | else while ((bio = bio_list_pop(&sync))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 906 | do_write(ms, bio); |
| 907 | |
| 908 | while ((bio = bio_list_pop(&recover))) |
| 909 | rh_delay(&ms->rh, bio); |
| 910 | |
| 911 | while ((bio = bio_list_pop(&nosync))) { |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 912 | map_bio(ms, ms->default_mirror, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 913 | generic_make_request(bio); |
| 914 | } |
| 915 | } |
| 916 | |
| 917 | /*----------------------------------------------------------------- |
| 918 | * kmirrord |
| 919 | *---------------------------------------------------------------*/ |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 920 | static void do_mirror(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | { |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 922 | struct mirror_set *ms =container_of(work, struct mirror_set, |
| 923 | kmirrord_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | struct bio_list reads, writes; |
| 925 | |
| 926 | spin_lock(&ms->lock); |
| 927 | reads = ms->reads; |
| 928 | writes = ms->writes; |
| 929 | bio_list_init(&ms->reads); |
| 930 | bio_list_init(&ms->writes); |
| 931 | spin_unlock(&ms->lock); |
| 932 | |
| 933 | rh_update_states(&ms->rh); |
| 934 | do_recovery(ms); |
| 935 | do_reads(ms, &reads); |
| 936 | do_writes(ms, &writes); |
| 937 | } |
| 938 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | /*----------------------------------------------------------------- |
| 940 | * Target functions |
| 941 | *---------------------------------------------------------------*/ |
| 942 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, |
| 943 | uint32_t region_size, |
| 944 | struct dm_target *ti, |
| 945 | struct dirty_log *dl) |
| 946 | { |
| 947 | size_t len; |
| 948 | struct mirror_set *ms = NULL; |
| 949 | |
| 950 | if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors)) |
| 951 | return NULL; |
| 952 | |
| 953 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
| 954 | |
Yoann Padioleau | dd00cc4 | 2007-07-19 01:49:03 -0700 | [diff] [blame] | 955 | ms = kzalloc(len, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | if (!ms) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 957 | ti->error = "Cannot allocate mirror context"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 958 | return NULL; |
| 959 | } |
| 960 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | spin_lock_init(&ms->lock); |
| 962 | |
| 963 | ms->ti = ti; |
| 964 | ms->nr_mirrors = nr_mirrors; |
| 965 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); |
| 966 | ms->in_sync = 0; |
Jonathan E Brassow | a1a1908 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 967 | ms->default_mirror = &ms->mirror[DEFAULT_MIRROR]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 969 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
| 970 | if (IS_ERR(ms->io_client)) { |
| 971 | ti->error = "Error creating dm_io client"; |
| 972 | kfree(ms); |
| 973 | return NULL; |
| 974 | } |
| 975 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 977 | ti->error = "Error creating dirty region hash"; |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 978 | dm_io_client_destroy(ms->io_client); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | kfree(ms); |
| 980 | return NULL; |
| 981 | } |
| 982 | |
| 983 | return ms; |
| 984 | } |
| 985 | |
| 986 | static void free_context(struct mirror_set *ms, struct dm_target *ti, |
| 987 | unsigned int m) |
| 988 | { |
| 989 | while (m--) |
| 990 | dm_put_device(ti, ms->mirror[m].dev); |
| 991 | |
Milan Broz | 88be163 | 2007-05-09 02:33:04 -0700 | [diff] [blame] | 992 | dm_io_client_destroy(ms->io_client); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | rh_exit(&ms->rh); |
| 994 | kfree(ms); |
| 995 | } |
| 996 | |
| 997 | static inline int _check_region_size(struct dm_target *ti, uint32_t size) |
| 998 | { |
vignesh babu | 6f3c3f0 | 2007-10-19 22:38:44 +0100 | [diff] [blame^] | 999 | return !(size % (PAGE_SIZE >> 9) || !is_power_of_2(size) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | size > ti->len); |
| 1001 | } |
| 1002 | |
| 1003 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
| 1004 | unsigned int mirror, char **argv) |
| 1005 | { |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 1006 | unsigned long long offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 1008 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1009 | ti->error = "Invalid offset"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1010 | return -EINVAL; |
| 1011 | } |
| 1012 | |
| 1013 | if (dm_get_device(ti, argv[0], offset, ti->len, |
| 1014 | dm_table_get_mode(ti->table), |
| 1015 | &ms->mirror[mirror].dev)) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1016 | ti->error = "Device lookup failure"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | return -ENXIO; |
| 1018 | } |
| 1019 | |
| 1020 | ms->mirror[mirror].offset = offset; |
| 1021 | |
| 1022 | return 0; |
| 1023 | } |
| 1024 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | /* |
| 1026 | * Create dirty log: log_type #log_params <log_params> |
| 1027 | */ |
| 1028 | static struct dirty_log *create_dirty_log(struct dm_target *ti, |
| 1029 | unsigned int argc, char **argv, |
| 1030 | unsigned int *args_used) |
| 1031 | { |
| 1032 | unsigned int param_count; |
| 1033 | struct dirty_log *dl; |
| 1034 | |
| 1035 | if (argc < 2) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1036 | ti->error = "Insufficient mirror log arguments"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1037 | return NULL; |
| 1038 | } |
| 1039 | |
| 1040 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1041 | ti->error = "Invalid mirror log argument count"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1042 | return NULL; |
| 1043 | } |
| 1044 | |
| 1045 | *args_used = 2 + param_count; |
| 1046 | |
| 1047 | if (argc < *args_used) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1048 | ti->error = "Insufficient mirror log arguments"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | return NULL; |
| 1050 | } |
| 1051 | |
| 1052 | dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2); |
| 1053 | if (!dl) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1054 | ti->error = "Error creating mirror dirty log"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | return NULL; |
| 1056 | } |
| 1057 | |
| 1058 | if (!_check_region_size(ti, dl->type->get_region_size(dl))) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1059 | ti->error = "Invalid region size"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | dm_destroy_dirty_log(dl); |
| 1061 | return NULL; |
| 1062 | } |
| 1063 | |
| 1064 | return dl; |
| 1065 | } |
| 1066 | |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1067 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
| 1068 | unsigned *args_used) |
| 1069 | { |
| 1070 | unsigned num_features; |
| 1071 | struct dm_target *ti = ms->ti; |
| 1072 | |
| 1073 | *args_used = 0; |
| 1074 | |
| 1075 | if (!argc) |
| 1076 | return 0; |
| 1077 | |
| 1078 | if (sscanf(argv[0], "%u", &num_features) != 1) { |
| 1079 | ti->error = "Invalid number of features"; |
| 1080 | return -EINVAL; |
| 1081 | } |
| 1082 | |
| 1083 | argc--; |
| 1084 | argv++; |
| 1085 | (*args_used)++; |
| 1086 | |
| 1087 | if (num_features > argc) { |
| 1088 | ti->error = "Not enough arguments to support feature count"; |
| 1089 | return -EINVAL; |
| 1090 | } |
| 1091 | |
| 1092 | if (!strcmp("handle_errors", argv[0])) |
| 1093 | ms->features |= DM_RAID1_HANDLE_ERRORS; |
| 1094 | else { |
| 1095 | ti->error = "Unrecognised feature requested"; |
| 1096 | return -EINVAL; |
| 1097 | } |
| 1098 | |
| 1099 | (*args_used)++; |
| 1100 | |
| 1101 | return 0; |
| 1102 | } |
| 1103 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | /* |
| 1105 | * Construct a mirror mapping: |
| 1106 | * |
| 1107 | * log_type #log_params <log_params> |
| 1108 | * #mirrors [mirror_path offset]{2,} |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1109 | * [#features <features>] |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | * |
| 1111 | * log_type is "core" or "disk" |
| 1112 | * #log_params is between 1 and 3 |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1113 | * |
| 1114 | * If present, features must be "handle_errors". |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1115 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
| 1117 | { |
| 1118 | int r; |
| 1119 | unsigned int nr_mirrors, m, args_used; |
| 1120 | struct mirror_set *ms; |
| 1121 | struct dirty_log *dl; |
| 1122 | |
| 1123 | dl = create_dirty_log(ti, argc, argv, &args_used); |
| 1124 | if (!dl) |
| 1125 | return -EINVAL; |
| 1126 | |
| 1127 | argv += args_used; |
| 1128 | argc -= args_used; |
| 1129 | |
| 1130 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || |
| 1131 | nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1132 | ti->error = "Invalid number of mirrors"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | dm_destroy_dirty_log(dl); |
| 1134 | return -EINVAL; |
| 1135 | } |
| 1136 | |
| 1137 | argv++, argc--; |
| 1138 | |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1139 | if (argc < nr_mirrors * 2) { |
| 1140 | ti->error = "Too few mirror arguments"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | dm_destroy_dirty_log(dl); |
| 1142 | return -EINVAL; |
| 1143 | } |
| 1144 | |
| 1145 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); |
| 1146 | if (!ms) { |
| 1147 | dm_destroy_dirty_log(dl); |
| 1148 | return -ENOMEM; |
| 1149 | } |
| 1150 | |
| 1151 | /* Get the mirror parameter sets */ |
| 1152 | for (m = 0; m < nr_mirrors; m++) { |
| 1153 | r = get_mirror(ms, ti, m, argv); |
| 1154 | if (r) { |
| 1155 | free_context(ms, ti, m); |
| 1156 | return r; |
| 1157 | } |
| 1158 | argv += 2; |
| 1159 | argc -= 2; |
| 1160 | } |
| 1161 | |
| 1162 | ti->private = ms; |
Alasdair G Kergon | d88854f | 2005-07-07 17:59:34 -0700 | [diff] [blame] | 1163 | ti->split_io = ms->rh.region_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1164 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1165 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
| 1166 | if (!ms->kmirrord_wq) { |
| 1167 | DMERR("couldn't start kmirrord"); |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 1168 | r = -ENOMEM; |
| 1169 | goto err_free_context; |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1170 | } |
| 1171 | INIT_WORK(&ms->kmirrord_work, do_mirror); |
| 1172 | |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1173 | r = parse_features(ms, argc, argv, &args_used); |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 1174 | if (r) |
| 1175 | goto err_destroy_wq; |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1176 | |
| 1177 | argv += args_used; |
| 1178 | argc -= args_used; |
| 1179 | |
Jonathan Brassow | f44db67 | 2007-07-12 17:29:04 +0100 | [diff] [blame] | 1180 | /* |
| 1181 | * Any read-balancing addition depends on the |
| 1182 | * DM_RAID1_HANDLE_ERRORS flag being present. |
| 1183 | * This is because the decision to balance depends |
| 1184 | * on the sync state of a region. If the above |
| 1185 | * flag is not present, we ignore errors; and |
| 1186 | * the sync state may be inaccurate. |
| 1187 | */ |
| 1188 | |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1189 | if (argc) { |
| 1190 | ti->error = "Too many mirror arguments"; |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 1191 | r = -EINVAL; |
| 1192 | goto err_destroy_wq; |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1193 | } |
| 1194 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client); |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 1196 | if (r) |
| 1197 | goto err_destroy_wq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1199 | wake(ms); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1200 | return 0; |
Dmitry Monakhov | a72cf73 | 2007-10-19 22:38:39 +0100 | [diff] [blame] | 1201 | |
| 1202 | err_destroy_wq: |
| 1203 | destroy_workqueue(ms->kmirrord_wq); |
| 1204 | err_free_context: |
| 1205 | free_context(ms, ti, ms->nr_mirrors); |
| 1206 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | } |
| 1208 | |
| 1209 | static void mirror_dtr(struct dm_target *ti) |
| 1210 | { |
| 1211 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
| 1212 | |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1213 | flush_workqueue(ms->kmirrord_wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1214 | kcopyd_client_destroy(ms->kcopyd_client); |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1215 | destroy_workqueue(ms->kmirrord_wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | free_context(ms, ti, ms->nr_mirrors); |
| 1217 | } |
| 1218 | |
| 1219 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) |
| 1220 | { |
| 1221 | int should_wake = 0; |
| 1222 | struct bio_list *bl; |
| 1223 | |
| 1224 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; |
| 1225 | spin_lock(&ms->lock); |
| 1226 | should_wake = !(bl->head); |
| 1227 | bio_list_add(bl, bio); |
| 1228 | spin_unlock(&ms->lock); |
| 1229 | |
| 1230 | if (should_wake) |
Holger Smolinski | 6ad36fe | 2007-05-09 02:32:50 -0700 | [diff] [blame] | 1231 | wake(ms); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | } |
| 1233 | |
| 1234 | /* |
| 1235 | * Mirror mapping function |
| 1236 | */ |
| 1237 | static int mirror_map(struct dm_target *ti, struct bio *bio, |
| 1238 | union map_info *map_context) |
| 1239 | { |
| 1240 | int r, rw = bio_rw(bio); |
| 1241 | struct mirror *m; |
| 1242 | struct mirror_set *ms = ti->private; |
| 1243 | |
Neil Brown | e4c8b3b | 2006-06-26 00:27:26 -0700 | [diff] [blame] | 1244 | map_context->ll = bio_to_region(&ms->rh, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | |
| 1246 | if (rw == WRITE) { |
| 1247 | queue_bio(ms, bio, rw); |
Kiyoshi Ueda | d2a7ad2 | 2006-12-08 02:41:06 -0800 | [diff] [blame] | 1248 | return DM_MAPIO_SUBMITTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | } |
| 1250 | |
| 1251 | r = ms->rh.log->type->in_sync(ms->rh.log, |
| 1252 | bio_to_region(&ms->rh, bio), 0); |
| 1253 | if (r < 0 && r != -EWOULDBLOCK) |
| 1254 | return r; |
| 1255 | |
| 1256 | if (r == -EWOULDBLOCK) /* FIXME: ugly */ |
Kiyoshi Ueda | d2a7ad2 | 2006-12-08 02:41:06 -0800 | [diff] [blame] | 1257 | r = DM_MAPIO_SUBMITTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1258 | |
| 1259 | /* |
| 1260 | * We don't want to fast track a recovery just for a read |
| 1261 | * ahead. So we just let it silently fail. |
| 1262 | * FIXME: get rid of this. |
| 1263 | */ |
| 1264 | if (!r && rw == READA) |
| 1265 | return -EIO; |
| 1266 | |
| 1267 | if (!r) { |
| 1268 | /* Pass this io over to the daemon */ |
| 1269 | queue_bio(ms, bio, rw); |
Kiyoshi Ueda | d2a7ad2 | 2006-12-08 02:41:06 -0800 | [diff] [blame] | 1270 | return DM_MAPIO_SUBMITTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1271 | } |
| 1272 | |
| 1273 | m = choose_mirror(ms, bio->bi_sector); |
| 1274 | if (!m) |
| 1275 | return -EIO; |
| 1276 | |
| 1277 | map_bio(ms, m, bio); |
Kiyoshi Ueda | d2a7ad2 | 2006-12-08 02:41:06 -0800 | [diff] [blame] | 1278 | return DM_MAPIO_REMAPPED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | } |
| 1280 | |
| 1281 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, |
| 1282 | int error, union map_info *map_context) |
| 1283 | { |
| 1284 | int rw = bio_rw(bio); |
| 1285 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
| 1286 | region_t region = map_context->ll; |
| 1287 | |
| 1288 | /* |
| 1289 | * We need to dec pending if this was a write. |
| 1290 | */ |
| 1291 | if (rw == WRITE) |
| 1292 | rh_dec(&ms->rh, region); |
| 1293 | |
| 1294 | return 0; |
| 1295 | } |
| 1296 | |
| 1297 | static void mirror_postsuspend(struct dm_target *ti) |
| 1298 | { |
| 1299 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
| 1300 | struct dirty_log *log = ms->rh.log; |
| 1301 | |
| 1302 | rh_stop_recovery(&ms->rh); |
Jonathan E Brassow | 3318404 | 2006-11-08 17:44:44 -0800 | [diff] [blame] | 1303 | |
| 1304 | /* Wait for all I/O we generated to complete */ |
| 1305 | wait_event(_kmirrord_recovery_stopped, |
| 1306 | !atomic_read(&ms->rh.recovery_in_flight)); |
| 1307 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1308 | if (log->type->suspend && log->type->suspend(log)) |
| 1309 | /* FIXME: need better error handling */ |
| 1310 | DMWARN("log suspend failed"); |
| 1311 | } |
| 1312 | |
| 1313 | static void mirror_resume(struct dm_target *ti) |
| 1314 | { |
| 1315 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
| 1316 | struct dirty_log *log = ms->rh.log; |
| 1317 | if (log->type->resume && log->type->resume(log)) |
| 1318 | /* FIXME: need better error handling */ |
| 1319 | DMWARN("log resume failed"); |
| 1320 | rh_start_recovery(&ms->rh); |
| 1321 | } |
| 1322 | |
| 1323 | static int mirror_status(struct dm_target *ti, status_type_t type, |
| 1324 | char *result, unsigned int maxlen) |
| 1325 | { |
Jonathan E Brassow | 315dcc2 | 2007-05-09 02:32:58 -0700 | [diff] [blame] | 1326 | unsigned int m, sz = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
| 1328 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | switch (type) { |
| 1330 | case STATUSTYPE_INFO: |
| 1331 | DMEMIT("%d ", ms->nr_mirrors); |
| 1332 | for (m = 0; m < ms->nr_mirrors; m++) |
| 1333 | DMEMIT("%s ", ms->mirror[m].dev->name); |
| 1334 | |
Milan Broz | c95bc20 | 2007-07-12 17:27:24 +0100 | [diff] [blame] | 1335 | DMEMIT("%llu/%llu 0 ", |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 1336 | (unsigned long long)ms->rh.log->type-> |
| 1337 | get_sync_count(ms->rh.log), |
| 1338 | (unsigned long long)ms->nr_regions); |
Jonathan E Brassow | 315dcc2 | 2007-05-09 02:32:58 -0700 | [diff] [blame] | 1339 | |
Milan Broz | c95bc20 | 2007-07-12 17:27:24 +0100 | [diff] [blame] | 1340 | sz += ms->rh.log->type->status(ms->rh.log, type, result+sz, maxlen-sz); |
Jonathan E Brassow | 315dcc2 | 2007-05-09 02:32:58 -0700 | [diff] [blame] | 1341 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1342 | break; |
| 1343 | |
| 1344 | case STATUSTYPE_TABLE: |
Jonathan E Brassow | 315dcc2 | 2007-05-09 02:32:58 -0700 | [diff] [blame] | 1345 | sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen); |
| 1346 | |
Jonathan Brassow | e52b8f6 | 2006-10-03 01:15:32 -0700 | [diff] [blame] | 1347 | DMEMIT("%d", ms->nr_mirrors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | for (m = 0; m < ms->nr_mirrors; m++) |
Jonathan Brassow | e52b8f6 | 2006-10-03 01:15:32 -0700 | [diff] [blame] | 1349 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 1350 | (unsigned long long)ms->mirror[m].offset); |
Jonathan E Brassow | a8e6afa | 2007-05-09 02:32:59 -0700 | [diff] [blame] | 1351 | |
| 1352 | if (ms->features & DM_RAID1_HANDLE_ERRORS) |
| 1353 | DMEMIT(" 1 handle_errors"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | } |
| 1355 | |
| 1356 | return 0; |
| 1357 | } |
| 1358 | |
| 1359 | static struct target_type mirror_target = { |
| 1360 | .name = "mirror", |
Jonathan E Brassow | 315dcc2 | 2007-05-09 02:32:58 -0700 | [diff] [blame] | 1361 | .version = {1, 0, 3}, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | .module = THIS_MODULE, |
| 1363 | .ctr = mirror_ctr, |
| 1364 | .dtr = mirror_dtr, |
| 1365 | .map = mirror_map, |
| 1366 | .end_io = mirror_end_io, |
| 1367 | .postsuspend = mirror_postsuspend, |
| 1368 | .resume = mirror_resume, |
| 1369 | .status = mirror_status, |
| 1370 | }; |
| 1371 | |
| 1372 | static int __init dm_mirror_init(void) |
| 1373 | { |
| 1374 | int r; |
| 1375 | |
| 1376 | r = dm_dirty_log_init(); |
| 1377 | if (r) |
| 1378 | return r; |
| 1379 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | r = dm_register_target(&mirror_target); |
| 1381 | if (r < 0) { |
Alasdair G Kergon | 0cd3312 | 2007-07-12 17:27:01 +0100 | [diff] [blame] | 1382 | DMERR("Failed to register mirror target"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | dm_dirty_log_exit(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | } |
| 1385 | |
| 1386 | return r; |
| 1387 | } |
| 1388 | |
| 1389 | static void __exit dm_mirror_exit(void) |
| 1390 | { |
| 1391 | int r; |
| 1392 | |
| 1393 | r = dm_unregister_target(&mirror_target); |
| 1394 | if (r < 0) |
Alasdair G Kergon | 0cd3312 | 2007-07-12 17:27:01 +0100 | [diff] [blame] | 1395 | DMERR("unregister failed %d", r); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | dm_dirty_log_exit(); |
| 1398 | } |
| 1399 | |
| 1400 | /* Module hooks */ |
| 1401 | module_init(dm_mirror_init); |
| 1402 | module_exit(dm_mirror_exit); |
| 1403 | |
| 1404 | MODULE_DESCRIPTION(DM_NAME " mirror target"); |
| 1405 | MODULE_AUTHOR("Joe Thornber"); |
| 1406 | MODULE_LICENSE("GPL"); |