Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * dm-snapshot.c |
| 3 | * |
| 4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. |
| 5 | * |
| 6 | * This file is released under the GPL. |
| 7 | */ |
| 8 | |
| 9 | #include <linux/blkdev.h> |
| 10 | #include <linux/config.h> |
| 11 | #include <linux/ctype.h> |
| 12 | #include <linux/device-mapper.h> |
| 13 | #include <linux/fs.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/kdev_t.h> |
| 16 | #include <linux/list.h> |
| 17 | #include <linux/mempool.h> |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/slab.h> |
| 20 | #include <linux/vmalloc.h> |
| 21 | |
| 22 | #include "dm-snap.h" |
| 23 | #include "dm-bio-list.h" |
| 24 | #include "kcopyd.h" |
| 25 | |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 26 | #define DM_MSG_PREFIX "snapshots" |
| 27 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | /* |
| 29 | * The percentage increment we will wake up users at |
| 30 | */ |
| 31 | #define WAKE_UP_PERCENT 5 |
| 32 | |
| 33 | /* |
| 34 | * kcopyd priority of snapshot operations |
| 35 | */ |
| 36 | #define SNAPSHOT_COPY_PRIORITY 2 |
| 37 | |
| 38 | /* |
| 39 | * Each snapshot reserves this many pages for io |
| 40 | */ |
| 41 | #define SNAPSHOT_PAGES 256 |
| 42 | |
| 43 | struct pending_exception { |
| 44 | struct exception e; |
| 45 | |
| 46 | /* |
| 47 | * Origin buffers waiting for this to complete are held |
| 48 | * in a bio list |
| 49 | */ |
| 50 | struct bio_list origin_bios; |
| 51 | struct bio_list snapshot_bios; |
| 52 | |
| 53 | /* |
Alasdair G Kergon | eccf081 | 2006-03-27 01:17:42 -0800 | [diff] [blame] | 54 | * Short-term queue of pending exceptions prior to submission. |
| 55 | */ |
| 56 | struct list_head list; |
| 57 | |
| 58 | /* |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 59 | * The primary pending_exception is the one that holds |
| 60 | * the sibling_count and the list of origin_bios for a |
| 61 | * group of pending_exceptions. It is always last to get freed. |
| 62 | * These fields get set up when writing to the origin. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | */ |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 64 | struct pending_exception *primary_pe; |
| 65 | |
| 66 | /* |
| 67 | * Number of pending_exceptions processing this chunk. |
| 68 | * When this drops to zero we must complete the origin bios. |
| 69 | * If incrementing or decrementing this, hold pe->snap->lock for |
| 70 | * the sibling concerned and not pe->primary_pe->snap->lock unless |
| 71 | * they are the same. |
| 72 | */ |
| 73 | atomic_t sibling_count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | |
| 75 | /* Pointer back to snapshot context */ |
| 76 | struct dm_snapshot *snap; |
| 77 | |
| 78 | /* |
| 79 | * 1 indicates the exception has already been sent to |
| 80 | * kcopyd. |
| 81 | */ |
| 82 | int started; |
| 83 | }; |
| 84 | |
| 85 | /* |
| 86 | * Hash table mapping origin volumes to lists of snapshots and |
| 87 | * a lock to protect it |
| 88 | */ |
| 89 | static kmem_cache_t *exception_cache; |
| 90 | static kmem_cache_t *pending_cache; |
| 91 | static mempool_t *pending_pool; |
| 92 | |
| 93 | /* |
| 94 | * One of these per registered origin, held in the snapshot_origins hash |
| 95 | */ |
| 96 | struct origin { |
| 97 | /* The origin device */ |
| 98 | struct block_device *bdev; |
| 99 | |
| 100 | struct list_head hash_list; |
| 101 | |
| 102 | /* List of snapshots for this origin */ |
| 103 | struct list_head snapshots; |
| 104 | }; |
| 105 | |
| 106 | /* |
| 107 | * Size of the hash table for origin volumes. If we make this |
| 108 | * the size of the minors list then it should be nearly perfect |
| 109 | */ |
| 110 | #define ORIGIN_HASH_SIZE 256 |
| 111 | #define ORIGIN_MASK 0xFF |
| 112 | static struct list_head *_origins; |
| 113 | static struct rw_semaphore _origins_lock; |
| 114 | |
| 115 | static int init_origin_hash(void) |
| 116 | { |
| 117 | int i; |
| 118 | |
| 119 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), |
| 120 | GFP_KERNEL); |
| 121 | if (!_origins) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 122 | DMERR("unable to allocate memory"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | return -ENOMEM; |
| 124 | } |
| 125 | |
| 126 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
| 127 | INIT_LIST_HEAD(_origins + i); |
| 128 | init_rwsem(&_origins_lock); |
| 129 | |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | static void exit_origin_hash(void) |
| 134 | { |
| 135 | kfree(_origins); |
| 136 | } |
| 137 | |
| 138 | static inline unsigned int origin_hash(struct block_device *bdev) |
| 139 | { |
| 140 | return bdev->bd_dev & ORIGIN_MASK; |
| 141 | } |
| 142 | |
| 143 | static struct origin *__lookup_origin(struct block_device *origin) |
| 144 | { |
| 145 | struct list_head *ol; |
| 146 | struct origin *o; |
| 147 | |
| 148 | ol = &_origins[origin_hash(origin)]; |
| 149 | list_for_each_entry (o, ol, hash_list) |
| 150 | if (bdev_equal(o->bdev, origin)) |
| 151 | return o; |
| 152 | |
| 153 | return NULL; |
| 154 | } |
| 155 | |
| 156 | static void __insert_origin(struct origin *o) |
| 157 | { |
| 158 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; |
| 159 | list_add_tail(&o->hash_list, sl); |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * Make a note of the snapshot and its origin so we can look it |
| 164 | * up when the origin has a write on it. |
| 165 | */ |
| 166 | static int register_snapshot(struct dm_snapshot *snap) |
| 167 | { |
| 168 | struct origin *o; |
| 169 | struct block_device *bdev = snap->origin->bdev; |
| 170 | |
| 171 | down_write(&_origins_lock); |
| 172 | o = __lookup_origin(bdev); |
| 173 | |
| 174 | if (!o) { |
| 175 | /* New origin */ |
| 176 | o = kmalloc(sizeof(*o), GFP_KERNEL); |
| 177 | if (!o) { |
| 178 | up_write(&_origins_lock); |
| 179 | return -ENOMEM; |
| 180 | } |
| 181 | |
| 182 | /* Initialise the struct */ |
| 183 | INIT_LIST_HEAD(&o->snapshots); |
| 184 | o->bdev = bdev; |
| 185 | |
| 186 | __insert_origin(o); |
| 187 | } |
| 188 | |
| 189 | list_add_tail(&snap->list, &o->snapshots); |
| 190 | |
| 191 | up_write(&_origins_lock); |
| 192 | return 0; |
| 193 | } |
| 194 | |
| 195 | static void unregister_snapshot(struct dm_snapshot *s) |
| 196 | { |
| 197 | struct origin *o; |
| 198 | |
| 199 | down_write(&_origins_lock); |
| 200 | o = __lookup_origin(s->origin->bdev); |
| 201 | |
| 202 | list_del(&s->list); |
| 203 | if (list_empty(&o->snapshots)) { |
| 204 | list_del(&o->hash_list); |
| 205 | kfree(o); |
| 206 | } |
| 207 | |
| 208 | up_write(&_origins_lock); |
| 209 | } |
| 210 | |
| 211 | /* |
| 212 | * Implementation of the exception hash tables. |
| 213 | */ |
| 214 | static int init_exception_table(struct exception_table *et, uint32_t size) |
| 215 | { |
| 216 | unsigned int i; |
| 217 | |
| 218 | et->hash_mask = size - 1; |
| 219 | et->table = dm_vcalloc(size, sizeof(struct list_head)); |
| 220 | if (!et->table) |
| 221 | return -ENOMEM; |
| 222 | |
| 223 | for (i = 0; i < size; i++) |
| 224 | INIT_LIST_HEAD(et->table + i); |
| 225 | |
| 226 | return 0; |
| 227 | } |
| 228 | |
| 229 | static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem) |
| 230 | { |
| 231 | struct list_head *slot; |
| 232 | struct exception *ex, *next; |
| 233 | int i, size; |
| 234 | |
| 235 | size = et->hash_mask + 1; |
| 236 | for (i = 0; i < size; i++) { |
| 237 | slot = et->table + i; |
| 238 | |
| 239 | list_for_each_entry_safe (ex, next, slot, hash_list) |
| 240 | kmem_cache_free(mem, ex); |
| 241 | } |
| 242 | |
| 243 | vfree(et->table); |
| 244 | } |
| 245 | |
| 246 | static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
| 247 | { |
| 248 | return chunk & et->hash_mask; |
| 249 | } |
| 250 | |
| 251 | static void insert_exception(struct exception_table *eh, struct exception *e) |
| 252 | { |
| 253 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; |
| 254 | list_add(&e->hash_list, l); |
| 255 | } |
| 256 | |
| 257 | static inline void remove_exception(struct exception *e) |
| 258 | { |
| 259 | list_del(&e->hash_list); |
| 260 | } |
| 261 | |
| 262 | /* |
| 263 | * Return the exception data for a sector, or NULL if not |
| 264 | * remapped. |
| 265 | */ |
| 266 | static struct exception *lookup_exception(struct exception_table *et, |
| 267 | chunk_t chunk) |
| 268 | { |
| 269 | struct list_head *slot; |
| 270 | struct exception *e; |
| 271 | |
| 272 | slot = &et->table[exception_hash(et, chunk)]; |
| 273 | list_for_each_entry (e, slot, hash_list) |
| 274 | if (e->old_chunk == chunk) |
| 275 | return e; |
| 276 | |
| 277 | return NULL; |
| 278 | } |
| 279 | |
| 280 | static inline struct exception *alloc_exception(void) |
| 281 | { |
| 282 | struct exception *e; |
| 283 | |
| 284 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); |
| 285 | if (!e) |
| 286 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
| 287 | |
| 288 | return e; |
| 289 | } |
| 290 | |
| 291 | static inline void free_exception(struct exception *e) |
| 292 | { |
| 293 | kmem_cache_free(exception_cache, e); |
| 294 | } |
| 295 | |
| 296 | static inline struct pending_exception *alloc_pending_exception(void) |
| 297 | { |
| 298 | return mempool_alloc(pending_pool, GFP_NOIO); |
| 299 | } |
| 300 | |
| 301 | static inline void free_pending_exception(struct pending_exception *pe) |
| 302 | { |
| 303 | mempool_free(pe, pending_pool); |
| 304 | } |
| 305 | |
| 306 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) |
| 307 | { |
| 308 | struct exception *e; |
| 309 | |
| 310 | e = alloc_exception(); |
| 311 | if (!e) |
| 312 | return -ENOMEM; |
| 313 | |
| 314 | e->old_chunk = old; |
| 315 | e->new_chunk = new; |
| 316 | insert_exception(&s->complete, e); |
| 317 | return 0; |
| 318 | } |
| 319 | |
| 320 | /* |
| 321 | * Hard coded magic. |
| 322 | */ |
| 323 | static int calc_max_buckets(void) |
| 324 | { |
| 325 | /* use a fixed size of 2MB */ |
| 326 | unsigned long mem = 2 * 1024 * 1024; |
| 327 | mem /= sizeof(struct list_head); |
| 328 | |
| 329 | return mem; |
| 330 | } |
| 331 | |
| 332 | /* |
| 333 | * Rounds a number down to a power of 2. |
| 334 | */ |
| 335 | static inline uint32_t round_down(uint32_t n) |
| 336 | { |
| 337 | while (n & (n - 1)) |
| 338 | n &= (n - 1); |
| 339 | return n; |
| 340 | } |
| 341 | |
| 342 | /* |
| 343 | * Allocate room for a suitable hash table. |
| 344 | */ |
| 345 | static int init_hash_tables(struct dm_snapshot *s) |
| 346 | { |
| 347 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; |
| 348 | |
| 349 | /* |
| 350 | * Calculate based on the size of the original volume or |
| 351 | * the COW volume... |
| 352 | */ |
| 353 | cow_dev_size = get_dev_size(s->cow->bdev); |
| 354 | origin_dev_size = get_dev_size(s->origin->bdev); |
| 355 | max_buckets = calc_max_buckets(); |
| 356 | |
| 357 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; |
| 358 | hash_size = min(hash_size, max_buckets); |
| 359 | |
| 360 | /* Round it down to a power of 2 */ |
| 361 | hash_size = round_down(hash_size); |
| 362 | if (init_exception_table(&s->complete, hash_size)) |
| 363 | return -ENOMEM; |
| 364 | |
| 365 | /* |
| 366 | * Allocate hash table for in-flight exceptions |
| 367 | * Make this smaller than the real hash table |
| 368 | */ |
| 369 | hash_size >>= 3; |
| 370 | if (hash_size < 64) |
| 371 | hash_size = 64; |
| 372 | |
| 373 | if (init_exception_table(&s->pending, hash_size)) { |
| 374 | exit_exception_table(&s->complete, exception_cache); |
| 375 | return -ENOMEM; |
| 376 | } |
| 377 | |
| 378 | return 0; |
| 379 | } |
| 380 | |
| 381 | /* |
| 382 | * Round a number up to the nearest 'size' boundary. size must |
| 383 | * be a power of 2. |
| 384 | */ |
| 385 | static inline ulong round_up(ulong n, ulong size) |
| 386 | { |
| 387 | size--; |
| 388 | return (n + size) & ~size; |
| 389 | } |
| 390 | |
Alasdair G Kergon | 2d38fe2 | 2006-01-06 00:20:02 -0800 | [diff] [blame] | 391 | static void read_snapshot_metadata(struct dm_snapshot *s) |
| 392 | { |
Alasdair G Kergon | 2d38fe2 | 2006-01-06 00:20:02 -0800 | [diff] [blame] | 393 | if (s->store.read_metadata(&s->store)) { |
| 394 | down_write(&s->lock); |
| 395 | s->valid = 0; |
| 396 | up_write(&s->lock); |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 397 | |
| 398 | dm_table_event(s->table); |
Alasdair G Kergon | 2d38fe2 | 2006-01-06 00:20:02 -0800 | [diff] [blame] | 399 | } |
Alasdair G Kergon | 2d38fe2 | 2006-01-06 00:20:02 -0800 | [diff] [blame] | 400 | } |
| 401 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | /* |
| 403 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> |
| 404 | */ |
| 405 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
| 406 | { |
| 407 | struct dm_snapshot *s; |
| 408 | unsigned long chunk_size; |
| 409 | int r = -EINVAL; |
| 410 | char persistent; |
| 411 | char *origin_path; |
| 412 | char *cow_path; |
| 413 | char *value; |
| 414 | int blocksize; |
| 415 | |
| 416 | if (argc < 4) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 417 | ti->error = "requires exactly 4 arguments"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | r = -EINVAL; |
| 419 | goto bad1; |
| 420 | } |
| 421 | |
| 422 | origin_path = argv[0]; |
| 423 | cow_path = argv[1]; |
| 424 | persistent = toupper(*argv[2]); |
| 425 | |
| 426 | if (persistent != 'P' && persistent != 'N') { |
| 427 | ti->error = "Persistent flag is not P or N"; |
| 428 | r = -EINVAL; |
| 429 | goto bad1; |
| 430 | } |
| 431 | |
| 432 | chunk_size = simple_strtoul(argv[3], &value, 10); |
| 433 | if (chunk_size == 0 || value == NULL) { |
| 434 | ti->error = "Invalid chunk size"; |
| 435 | r = -EINVAL; |
| 436 | goto bad1; |
| 437 | } |
| 438 | |
| 439 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
| 440 | if (s == NULL) { |
| 441 | ti->error = "Cannot allocate snapshot context private " |
| 442 | "structure"; |
| 443 | r = -ENOMEM; |
| 444 | goto bad1; |
| 445 | } |
| 446 | |
| 447 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); |
| 448 | if (r) { |
| 449 | ti->error = "Cannot get origin device"; |
| 450 | goto bad2; |
| 451 | } |
| 452 | |
| 453 | r = dm_get_device(ti, cow_path, 0, 0, |
| 454 | FMODE_READ | FMODE_WRITE, &s->cow); |
| 455 | if (r) { |
| 456 | dm_put_device(ti, s->origin); |
| 457 | ti->error = "Cannot get COW device"; |
| 458 | goto bad2; |
| 459 | } |
| 460 | |
| 461 | /* |
| 462 | * Chunk size must be multiple of page size. Silently |
| 463 | * round up if it's not. |
| 464 | */ |
| 465 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); |
| 466 | |
| 467 | /* Validate the chunk size against the device block size */ |
| 468 | blocksize = s->cow->bdev->bd_disk->queue->hardsect_size; |
| 469 | if (chunk_size % (blocksize >> 9)) { |
| 470 | ti->error = "Chunk size is not a multiple of device blocksize"; |
| 471 | r = -EINVAL; |
| 472 | goto bad3; |
| 473 | } |
| 474 | |
| 475 | /* Check chunk_size is a power of 2 */ |
| 476 | if (chunk_size & (chunk_size - 1)) { |
| 477 | ti->error = "Chunk size is not a power of 2"; |
| 478 | r = -EINVAL; |
| 479 | goto bad3; |
| 480 | } |
| 481 | |
| 482 | s->chunk_size = chunk_size; |
| 483 | s->chunk_mask = chunk_size - 1; |
| 484 | s->type = persistent; |
| 485 | s->chunk_shift = ffs(chunk_size) - 1; |
| 486 | |
| 487 | s->valid = 1; |
Alasdair G Kergon | aa14ede | 2006-02-01 03:04:50 -0800 | [diff] [blame] | 488 | s->active = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | s->last_percent = 0; |
| 490 | init_rwsem(&s->lock); |
| 491 | s->table = ti->table; |
| 492 | |
| 493 | /* Allocate hash table for COW data */ |
| 494 | if (init_hash_tables(s)) { |
| 495 | ti->error = "Unable to allocate hash table space"; |
| 496 | r = -ENOMEM; |
| 497 | goto bad3; |
| 498 | } |
| 499 | |
| 500 | /* |
| 501 | * Check the persistent flag - done here because we need the iobuf |
| 502 | * to check the LV header |
| 503 | */ |
| 504 | s->store.snap = s; |
| 505 | |
| 506 | if (persistent == 'P') |
| 507 | r = dm_create_persistent(&s->store, chunk_size); |
| 508 | else |
| 509 | r = dm_create_transient(&s->store, s, blocksize); |
| 510 | |
| 511 | if (r) { |
| 512 | ti->error = "Couldn't create exception store"; |
| 513 | r = -EINVAL; |
| 514 | goto bad4; |
| 515 | } |
| 516 | |
| 517 | r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
| 518 | if (r) { |
| 519 | ti->error = "Could not create kcopyd client"; |
| 520 | goto bad5; |
| 521 | } |
| 522 | |
Alasdair G Kergon | aa14ede | 2006-02-01 03:04:50 -0800 | [diff] [blame] | 523 | /* Metadata must only be loaded into one table at once */ |
| 524 | read_snapshot_metadata(s); |
| 525 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | /* Add snapshot to the list of snapshots for this origin */ |
Alasdair G Kergon | aa14ede | 2006-02-01 03:04:50 -0800 | [diff] [blame] | 527 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | if (register_snapshot(s)) { |
| 529 | r = -EINVAL; |
| 530 | ti->error = "Cannot register snapshot origin"; |
| 531 | goto bad6; |
| 532 | } |
| 533 | |
| 534 | ti->private = s; |
Alasdair G Kergon | c51c275 | 2006-06-26 00:27:18 -0700 | [diff] [blame] | 535 | ti->split_io = s->chunk_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | |
| 537 | return 0; |
| 538 | |
| 539 | bad6: |
| 540 | kcopyd_client_destroy(s->kcopyd_client); |
| 541 | |
| 542 | bad5: |
| 543 | s->store.destroy(&s->store); |
| 544 | |
| 545 | bad4: |
| 546 | exit_exception_table(&s->pending, pending_cache); |
| 547 | exit_exception_table(&s->complete, exception_cache); |
| 548 | |
| 549 | bad3: |
| 550 | dm_put_device(ti, s->cow); |
| 551 | dm_put_device(ti, s->origin); |
| 552 | |
| 553 | bad2: |
| 554 | kfree(s); |
| 555 | |
| 556 | bad1: |
| 557 | return r; |
| 558 | } |
| 559 | |
| 560 | static void snapshot_dtr(struct dm_target *ti) |
| 561 | { |
| 562 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; |
| 563 | |
Alasdair G Kergon | 138728d | 2006-03-27 01:17:50 -0800 | [diff] [blame] | 564 | /* Prevent further origin writes from using this snapshot. */ |
| 565 | /* After this returns there can be no new kcopyd jobs. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | unregister_snapshot(s); |
| 567 | |
Alasdair G Kergon | 138728d | 2006-03-27 01:17:50 -0800 | [diff] [blame] | 568 | kcopyd_client_destroy(s->kcopyd_client); |
| 569 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | exit_exception_table(&s->pending, pending_cache); |
| 571 | exit_exception_table(&s->complete, exception_cache); |
| 572 | |
| 573 | /* Deallocate memory used */ |
| 574 | s->store.destroy(&s->store); |
| 575 | |
| 576 | dm_put_device(ti, s->origin); |
| 577 | dm_put_device(ti, s->cow); |
Alasdair G Kergon | 138728d | 2006-03-27 01:17:50 -0800 | [diff] [blame] | 578 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | kfree(s); |
| 580 | } |
| 581 | |
| 582 | /* |
| 583 | * Flush a list of buffers. |
| 584 | */ |
| 585 | static void flush_bios(struct bio *bio) |
| 586 | { |
| 587 | struct bio *n; |
| 588 | |
| 589 | while (bio) { |
| 590 | n = bio->bi_next; |
| 591 | bio->bi_next = NULL; |
| 592 | generic_make_request(bio); |
| 593 | bio = n; |
| 594 | } |
| 595 | } |
| 596 | |
| 597 | /* |
| 598 | * Error a list of buffers. |
| 599 | */ |
| 600 | static void error_bios(struct bio *bio) |
| 601 | { |
| 602 | struct bio *n; |
| 603 | |
| 604 | while (bio) { |
| 605 | n = bio->bi_next; |
| 606 | bio->bi_next = NULL; |
| 607 | bio_io_error(bio, bio->bi_size); |
| 608 | bio = n; |
| 609 | } |
| 610 | } |
| 611 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 612 | static inline void error_snapshot_bios(struct pending_exception *pe) |
| 613 | { |
| 614 | error_bios(bio_list_get(&pe->snapshot_bios)); |
| 615 | } |
| 616 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | static struct bio *__flush_bios(struct pending_exception *pe) |
| 618 | { |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 619 | /* |
| 620 | * If this pe is involved in a write to the origin and |
| 621 | * it is the last sibling to complete then release |
| 622 | * the bios for the original write to the origin. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | */ |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 624 | |
| 625 | if (pe->primary_pe && |
| 626 | atomic_dec_and_test(&pe->primary_pe->sibling_count)) |
| 627 | return bio_list_get(&pe->primary_pe->origin_bios); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | |
| 629 | return NULL; |
| 630 | } |
| 631 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 632 | static void __invalidate_snapshot(struct dm_snapshot *s, |
| 633 | struct pending_exception *pe, int err) |
| 634 | { |
| 635 | if (!s->valid) |
| 636 | return; |
| 637 | |
| 638 | if (err == -EIO) |
| 639 | DMERR("Invalidating snapshot: Error reading/writing."); |
| 640 | else if (err == -ENOMEM) |
| 641 | DMERR("Invalidating snapshot: Unable to allocate exception."); |
| 642 | |
| 643 | if (pe) |
| 644 | remove_exception(&pe->e); |
| 645 | |
| 646 | if (s->store.drop_snapshot) |
| 647 | s->store.drop_snapshot(&s->store); |
| 648 | |
| 649 | s->valid = 0; |
| 650 | |
| 651 | dm_table_event(s->table); |
| 652 | } |
| 653 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | static void pending_complete(struct pending_exception *pe, int success) |
| 655 | { |
| 656 | struct exception *e; |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 657 | struct pending_exception *primary_pe; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | struct dm_snapshot *s = pe->snap; |
| 659 | struct bio *flush = NULL; |
| 660 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 661 | if (!success) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | /* Read/write error - snapshot is unusable */ |
| 663 | down_write(&s->lock); |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 664 | __invalidate_snapshot(s, pe, -EIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 665 | flush = __flush_bios(pe); |
| 666 | up_write(&s->lock); |
| 667 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 668 | error_snapshot_bios(pe); |
| 669 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | } |
| 671 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 672 | e = alloc_exception(); |
| 673 | if (!e) { |
| 674 | down_write(&s->lock); |
| 675 | __invalidate_snapshot(s, pe, -ENOMEM); |
| 676 | flush = __flush_bios(pe); |
| 677 | up_write(&s->lock); |
| 678 | |
| 679 | error_snapshot_bios(pe); |
| 680 | goto out; |
| 681 | } |
| 682 | *e = pe->e; |
| 683 | |
| 684 | /* |
| 685 | * Add a proper exception, and remove the |
| 686 | * in-flight exception from the list. |
| 687 | */ |
| 688 | down_write(&s->lock); |
| 689 | if (!s->valid) { |
| 690 | flush = __flush_bios(pe); |
| 691 | up_write(&s->lock); |
| 692 | |
| 693 | free_exception(e); |
| 694 | |
| 695 | error_snapshot_bios(pe); |
| 696 | goto out; |
| 697 | } |
| 698 | |
| 699 | insert_exception(&s->complete, e); |
| 700 | remove_exception(&pe->e); |
| 701 | flush = __flush_bios(pe); |
| 702 | |
| 703 | up_write(&s->lock); |
| 704 | |
| 705 | /* Submit any pending write bios */ |
| 706 | flush_bios(bio_list_get(&pe->snapshot_bios)); |
| 707 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | out: |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 709 | primary_pe = pe->primary_pe; |
| 710 | |
| 711 | /* |
| 712 | * Free the pe if it's not linked to an origin write or if |
| 713 | * it's not itself a primary pe. |
| 714 | */ |
| 715 | if (!primary_pe || primary_pe != pe) |
| 716 | free_pending_exception(pe); |
| 717 | |
| 718 | /* |
| 719 | * Free the primary pe if nothing references it. |
| 720 | */ |
| 721 | if (primary_pe && !atomic_read(&primary_pe->sibling_count)) |
| 722 | free_pending_exception(primary_pe); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | |
| 724 | if (flush) |
| 725 | flush_bios(flush); |
| 726 | } |
| 727 | |
| 728 | static void commit_callback(void *context, int success) |
| 729 | { |
| 730 | struct pending_exception *pe = (struct pending_exception *) context; |
| 731 | pending_complete(pe, success); |
| 732 | } |
| 733 | |
| 734 | /* |
| 735 | * Called when the copy I/O has finished. kcopyd actually runs |
| 736 | * this code so don't block. |
| 737 | */ |
| 738 | static void copy_callback(int read_err, unsigned int write_err, void *context) |
| 739 | { |
| 740 | struct pending_exception *pe = (struct pending_exception *) context; |
| 741 | struct dm_snapshot *s = pe->snap; |
| 742 | |
| 743 | if (read_err || write_err) |
| 744 | pending_complete(pe, 0); |
| 745 | |
| 746 | else |
| 747 | /* Update the metadata if we are persistent */ |
| 748 | s->store.commit_exception(&s->store, &pe->e, commit_callback, |
| 749 | pe); |
| 750 | } |
| 751 | |
| 752 | /* |
| 753 | * Dispatches the copy operation to kcopyd. |
| 754 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 755 | static void start_copy(struct pending_exception *pe) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | { |
| 757 | struct dm_snapshot *s = pe->snap; |
| 758 | struct io_region src, dest; |
| 759 | struct block_device *bdev = s->origin->bdev; |
| 760 | sector_t dev_size; |
| 761 | |
| 762 | dev_size = get_dev_size(bdev); |
| 763 | |
| 764 | src.bdev = bdev; |
| 765 | src.sector = chunk_to_sector(s, pe->e.old_chunk); |
| 766 | src.count = min(s->chunk_size, dev_size - src.sector); |
| 767 | |
| 768 | dest.bdev = s->cow->bdev; |
| 769 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); |
| 770 | dest.count = src.count; |
| 771 | |
| 772 | /* Hand over to kcopyd */ |
| 773 | kcopyd_copy(s->kcopyd_client, |
| 774 | &src, 1, &dest, 0, copy_callback, pe); |
| 775 | } |
| 776 | |
| 777 | /* |
| 778 | * Looks to see if this snapshot already has a pending exception |
| 779 | * for this chunk, otherwise it allocates a new one and inserts |
| 780 | * it into the pending table. |
| 781 | * |
| 782 | * NOTE: a write lock must be held on snap->lock before calling |
| 783 | * this. |
| 784 | */ |
| 785 | static struct pending_exception * |
| 786 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
| 787 | { |
| 788 | struct exception *e; |
| 789 | struct pending_exception *pe; |
| 790 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
| 791 | |
| 792 | /* |
| 793 | * Is there a pending exception for this already ? |
| 794 | */ |
| 795 | e = lookup_exception(&s->pending, chunk); |
| 796 | if (e) { |
| 797 | /* cast the exception to a pending exception */ |
| 798 | pe = container_of(e, struct pending_exception, e); |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 799 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | } |
| 801 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 802 | /* |
| 803 | * Create a new pending exception, we don't want |
| 804 | * to hold the lock while we do this. |
| 805 | */ |
| 806 | up_write(&s->lock); |
| 807 | pe = alloc_pending_exception(); |
| 808 | down_write(&s->lock); |
| 809 | |
| 810 | if (!s->valid) { |
| 811 | free_pending_exception(pe); |
| 812 | return NULL; |
| 813 | } |
| 814 | |
| 815 | e = lookup_exception(&s->pending, chunk); |
| 816 | if (e) { |
| 817 | free_pending_exception(pe); |
| 818 | pe = container_of(e, struct pending_exception, e); |
| 819 | goto out; |
| 820 | } |
| 821 | |
| 822 | pe->e.old_chunk = chunk; |
| 823 | bio_list_init(&pe->origin_bios); |
| 824 | bio_list_init(&pe->snapshot_bios); |
| 825 | pe->primary_pe = NULL; |
| 826 | atomic_set(&pe->sibling_count, 1); |
| 827 | pe->snap = s; |
| 828 | pe->started = 0; |
| 829 | |
| 830 | if (s->store.prepare_exception(&s->store, &pe->e)) { |
| 831 | free_pending_exception(pe); |
| 832 | return NULL; |
| 833 | } |
| 834 | |
| 835 | insert_exception(&s->pending, &pe->e); |
| 836 | |
| 837 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | return pe; |
| 839 | } |
| 840 | |
| 841 | static inline void remap_exception(struct dm_snapshot *s, struct exception *e, |
| 842 | struct bio *bio) |
| 843 | { |
| 844 | bio->bi_bdev = s->cow->bdev; |
| 845 | bio->bi_sector = chunk_to_sector(s, e->new_chunk) + |
| 846 | (bio->bi_sector & s->chunk_mask); |
| 847 | } |
| 848 | |
| 849 | static int snapshot_map(struct dm_target *ti, struct bio *bio, |
| 850 | union map_info *map_context) |
| 851 | { |
| 852 | struct exception *e; |
| 853 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 854 | int copy_needed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | int r = 1; |
| 856 | chunk_t chunk; |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 857 | struct pending_exception *pe = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | |
| 859 | chunk = sector_to_chunk(s, bio->bi_sector); |
| 860 | |
| 861 | /* Full snapshots are not usable */ |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 862 | /* To get here the table must be live so s->active is always set. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | if (!s->valid) |
Alasdair G Kergon | f6a80ea | 2005-07-12 15:53:01 -0700 | [diff] [blame] | 864 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | |
Alasdair G Kergon | 4aac0a6 | 2006-02-01 03:04:55 -0800 | [diff] [blame] | 866 | if (unlikely(bio_barrier(bio))) |
| 867 | return -EOPNOTSUPP; |
| 868 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | /* |
| 870 | * Write to snapshot - higher level takes care of RW/RO |
| 871 | * flags so we should only get this if we are |
| 872 | * writeable. |
| 873 | */ |
| 874 | if (bio_rw(bio) == WRITE) { |
| 875 | |
| 876 | /* FIXME: should only take write lock if we need |
| 877 | * to copy an exception */ |
| 878 | down_write(&s->lock); |
| 879 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 880 | if (!s->valid) { |
| 881 | r = -EIO; |
| 882 | goto out_unlock; |
| 883 | } |
| 884 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | /* If the block is already remapped - use that, else remap it */ |
| 886 | e = lookup_exception(&s->complete, chunk); |
| 887 | if (e) { |
| 888 | remap_exception(s, e, bio); |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 889 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | } |
| 891 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 892 | pe = __find_pending_exception(s, bio); |
| 893 | if (!pe) { |
| 894 | __invalidate_snapshot(s, pe, -ENOMEM); |
| 895 | r = -EIO; |
| 896 | goto out_unlock; |
| 897 | } |
| 898 | |
| 899 | remap_exception(s, &pe->e, bio); |
| 900 | bio_list_add(&pe->snapshot_bios, bio); |
| 901 | |
| 902 | if (!pe->started) { |
| 903 | /* this is protected by snap->lock */ |
| 904 | pe->started = 1; |
| 905 | copy_needed = 1; |
| 906 | } |
| 907 | |
| 908 | r = 0; |
| 909 | |
| 910 | out_unlock: |
| 911 | up_write(&s->lock); |
| 912 | |
| 913 | if (copy_needed) |
| 914 | start_copy(pe); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 915 | } else { |
| 916 | /* |
| 917 | * FIXME: this read path scares me because we |
| 918 | * always use the origin when we have a pending |
| 919 | * exception. However I can't think of a |
| 920 | * situation where this is wrong - ejt. |
| 921 | */ |
| 922 | |
| 923 | /* Do reads */ |
| 924 | down_read(&s->lock); |
| 925 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 926 | if (!s->valid) { |
| 927 | up_read(&s->lock); |
| 928 | return -EIO; |
| 929 | } |
| 930 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 931 | /* See if it it has been remapped */ |
| 932 | e = lookup_exception(&s->complete, chunk); |
| 933 | if (e) |
| 934 | remap_exception(s, e, bio); |
| 935 | else |
| 936 | bio->bi_bdev = s->origin->bdev; |
| 937 | |
| 938 | up_read(&s->lock); |
| 939 | } |
| 940 | |
| 941 | return r; |
| 942 | } |
| 943 | |
| 944 | static void snapshot_resume(struct dm_target *ti) |
| 945 | { |
| 946 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; |
| 947 | |
Alasdair G Kergon | aa14ede | 2006-02-01 03:04:50 -0800 | [diff] [blame] | 948 | down_write(&s->lock); |
| 949 | s->active = 1; |
| 950 | up_write(&s->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | } |
| 952 | |
| 953 | static int snapshot_status(struct dm_target *ti, status_type_t type, |
| 954 | char *result, unsigned int maxlen) |
| 955 | { |
| 956 | struct dm_snapshot *snap = (struct dm_snapshot *) ti->private; |
| 957 | |
| 958 | switch (type) { |
| 959 | case STATUSTYPE_INFO: |
| 960 | if (!snap->valid) |
| 961 | snprintf(result, maxlen, "Invalid"); |
| 962 | else { |
| 963 | if (snap->store.fraction_full) { |
| 964 | sector_t numerator, denominator; |
| 965 | snap->store.fraction_full(&snap->store, |
| 966 | &numerator, |
| 967 | &denominator); |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 968 | snprintf(result, maxlen, "%llu/%llu", |
| 969 | (unsigned long long)numerator, |
| 970 | (unsigned long long)denominator); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | } |
| 972 | else |
| 973 | snprintf(result, maxlen, "Unknown"); |
| 974 | } |
| 975 | break; |
| 976 | |
| 977 | case STATUSTYPE_TABLE: |
| 978 | /* |
| 979 | * kdevname returns a static pointer so we need |
| 980 | * to make private copies if the output is to |
| 981 | * make sense. |
| 982 | */ |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 983 | snprintf(result, maxlen, "%s %s %c %llu", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | snap->origin->name, snap->cow->name, |
Andrew Morton | 4ee218c | 2006-03-27 01:17:48 -0800 | [diff] [blame] | 985 | snap->type, |
| 986 | (unsigned long long)snap->chunk_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | break; |
| 988 | } |
| 989 | |
| 990 | return 0; |
| 991 | } |
| 992 | |
| 993 | /*----------------------------------------------------------------- |
| 994 | * Origin methods |
| 995 | *---------------------------------------------------------------*/ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
| 997 | { |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 998 | int r = 1, first = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | struct dm_snapshot *snap; |
| 1000 | struct exception *e; |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 1001 | struct pending_exception *pe, *next_pe, *primary_pe = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | chunk_t chunk; |
Alasdair G Kergon | eccf081 | 2006-03-27 01:17:42 -0800 | [diff] [blame] | 1003 | LIST_HEAD(pe_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1004 | |
| 1005 | /* Do all the snapshots on this origin */ |
| 1006 | list_for_each_entry (snap, snapshots, list) { |
| 1007 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1008 | down_write(&snap->lock); |
| 1009 | |
Alasdair G Kergon | aa14ede | 2006-02-01 03:04:50 -0800 | [diff] [blame] | 1010 | /* Only deal with valid and active snapshots */ |
| 1011 | if (!snap->valid || !snap->active) |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1012 | goto next_snapshot; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1013 | |
Alasdair G Kergon | d5e404c | 2005-07-12 15:53:05 -0700 | [diff] [blame] | 1014 | /* Nothing to do if writing beyond end of snapshot */ |
| 1015 | if (bio->bi_sector >= dm_table_get_size(snap->table)) |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1016 | goto next_snapshot; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | |
| 1018 | /* |
| 1019 | * Remember, different snapshots can have |
| 1020 | * different chunk sizes. |
| 1021 | */ |
| 1022 | chunk = sector_to_chunk(snap, bio->bi_sector); |
| 1023 | |
| 1024 | /* |
| 1025 | * Check exception table to see if block |
| 1026 | * is already remapped in this snapshot |
| 1027 | * and trigger an exception if not. |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 1028 | * |
| 1029 | * sibling_count is initialised to 1 so pending_complete() |
| 1030 | * won't destroy the primary_pe while we're inside this loop. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | */ |
| 1032 | e = lookup_exception(&snap->complete, chunk); |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1033 | if (e) |
| 1034 | goto next_snapshot; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1036 | pe = __find_pending_exception(snap, bio); |
| 1037 | if (!pe) { |
| 1038 | __invalidate_snapshot(snap, pe, ENOMEM); |
| 1039 | goto next_snapshot; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | } |
| 1041 | |
Alasdair G Kergon | 76df1c6 | 2006-03-27 01:17:45 -0800 | [diff] [blame] | 1042 | if (!primary_pe) { |
| 1043 | /* |
| 1044 | * Either every pe here has same |
| 1045 | * primary_pe or none has one yet. |
| 1046 | */ |
| 1047 | if (pe->primary_pe) |
| 1048 | primary_pe = pe->primary_pe; |
| 1049 | else { |
| 1050 | primary_pe = pe; |
| 1051 | first = 1; |
| 1052 | } |
| 1053 | |
| 1054 | bio_list_add(&primary_pe->origin_bios, bio); |
| 1055 | |
| 1056 | r = 0; |
| 1057 | } |
| 1058 | |
| 1059 | if (!pe->primary_pe) { |
| 1060 | atomic_inc(&primary_pe->sibling_count); |
| 1061 | pe->primary_pe = primary_pe; |
| 1062 | } |
| 1063 | |
| 1064 | if (!pe->started) { |
| 1065 | pe->started = 1; |
| 1066 | list_add_tail(&pe->list, &pe_queue); |
| 1067 | } |
| 1068 | |
| 1069 | next_snapshot: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | up_write(&snap->lock); |
| 1071 | } |
| 1072 | |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 1073 | if (!primary_pe) |
| 1074 | goto out; |
| 1075 | |
| 1076 | /* |
| 1077 | * If this is the first time we're processing this chunk and |
| 1078 | * sibling_count is now 1 it means all the pending exceptions |
| 1079 | * got completed while we were in the loop above, so it falls to |
| 1080 | * us here to remove the primary_pe and submit any origin_bios. |
| 1081 | */ |
| 1082 | |
| 1083 | if (first && atomic_dec_and_test(&primary_pe->sibling_count)) { |
| 1084 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
| 1085 | free_pending_exception(primary_pe); |
| 1086 | /* If we got here, pe_queue is necessarily empty. */ |
| 1087 | goto out; |
| 1088 | } |
| 1089 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | /* |
| 1091 | * Now that we have a complete pe list we can start the copying. |
| 1092 | */ |
Alasdair G Kergon | eccf081 | 2006-03-27 01:17:42 -0800 | [diff] [blame] | 1093 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
| 1094 | start_copy(pe); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | |
Alasdair G Kergon | b4b610f | 2006-03-27 01:17:44 -0800 | [diff] [blame] | 1096 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | return r; |
| 1098 | } |
| 1099 | |
| 1100 | /* |
| 1101 | * Called on a write from the origin driver. |
| 1102 | */ |
| 1103 | static int do_origin(struct dm_dev *origin, struct bio *bio) |
| 1104 | { |
| 1105 | struct origin *o; |
| 1106 | int r = 1; |
| 1107 | |
| 1108 | down_read(&_origins_lock); |
| 1109 | o = __lookup_origin(origin->bdev); |
| 1110 | if (o) |
| 1111 | r = __origin_write(&o->snapshots, bio); |
| 1112 | up_read(&_origins_lock); |
| 1113 | |
| 1114 | return r; |
| 1115 | } |
| 1116 | |
| 1117 | /* |
| 1118 | * Origin: maps a linear range of a device, with hooks for snapshotting. |
| 1119 | */ |
| 1120 | |
| 1121 | /* |
| 1122 | * Construct an origin mapping: <dev_path> |
| 1123 | * The context for an origin is merely a 'struct dm_dev *' |
| 1124 | * pointing to the real device. |
| 1125 | */ |
| 1126 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
| 1127 | { |
| 1128 | int r; |
| 1129 | struct dm_dev *dev; |
| 1130 | |
| 1131 | if (argc != 1) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1132 | ti->error = "origin: incorrect number of arguments"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | return -EINVAL; |
| 1134 | } |
| 1135 | |
| 1136 | r = dm_get_device(ti, argv[0], 0, ti->len, |
| 1137 | dm_table_get_mode(ti->table), &dev); |
| 1138 | if (r) { |
| 1139 | ti->error = "Cannot get target device"; |
| 1140 | return r; |
| 1141 | } |
| 1142 | |
| 1143 | ti->private = dev; |
| 1144 | return 0; |
| 1145 | } |
| 1146 | |
| 1147 | static void origin_dtr(struct dm_target *ti) |
| 1148 | { |
| 1149 | struct dm_dev *dev = (struct dm_dev *) ti->private; |
| 1150 | dm_put_device(ti, dev); |
| 1151 | } |
| 1152 | |
| 1153 | static int origin_map(struct dm_target *ti, struct bio *bio, |
| 1154 | union map_info *map_context) |
| 1155 | { |
| 1156 | struct dm_dev *dev = (struct dm_dev *) ti->private; |
| 1157 | bio->bi_bdev = dev->bdev; |
| 1158 | |
Alasdair G Kergon | 4aac0a6 | 2006-02-01 03:04:55 -0800 | [diff] [blame] | 1159 | if (unlikely(bio_barrier(bio))) |
| 1160 | return -EOPNOTSUPP; |
| 1161 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | /* Only tell snapshots if this is a write */ |
| 1163 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1; |
| 1164 | } |
| 1165 | |
| 1166 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
| 1167 | |
| 1168 | /* |
| 1169 | * Set the target "split_io" field to the minimum of all the snapshots' |
| 1170 | * chunk sizes. |
| 1171 | */ |
| 1172 | static void origin_resume(struct dm_target *ti) |
| 1173 | { |
| 1174 | struct dm_dev *dev = (struct dm_dev *) ti->private; |
| 1175 | struct dm_snapshot *snap; |
| 1176 | struct origin *o; |
| 1177 | chunk_t chunk_size = 0; |
| 1178 | |
| 1179 | down_read(&_origins_lock); |
| 1180 | o = __lookup_origin(dev->bdev); |
| 1181 | if (o) |
| 1182 | list_for_each_entry (snap, &o->snapshots, list) |
| 1183 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); |
| 1184 | up_read(&_origins_lock); |
| 1185 | |
| 1186 | ti->split_io = chunk_size; |
| 1187 | } |
| 1188 | |
| 1189 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, |
| 1190 | unsigned int maxlen) |
| 1191 | { |
| 1192 | struct dm_dev *dev = (struct dm_dev *) ti->private; |
| 1193 | |
| 1194 | switch (type) { |
| 1195 | case STATUSTYPE_INFO: |
| 1196 | result[0] = '\0'; |
| 1197 | break; |
| 1198 | |
| 1199 | case STATUSTYPE_TABLE: |
| 1200 | snprintf(result, maxlen, "%s", dev->name); |
| 1201 | break; |
| 1202 | } |
| 1203 | |
| 1204 | return 0; |
| 1205 | } |
| 1206 | |
| 1207 | static struct target_type origin_target = { |
| 1208 | .name = "snapshot-origin", |
Alasdair G Kergon | c51c275 | 2006-06-26 00:27:18 -0700 | [diff] [blame] | 1209 | .version = {1, 4, 0}, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1210 | .module = THIS_MODULE, |
| 1211 | .ctr = origin_ctr, |
| 1212 | .dtr = origin_dtr, |
| 1213 | .map = origin_map, |
| 1214 | .resume = origin_resume, |
| 1215 | .status = origin_status, |
| 1216 | }; |
| 1217 | |
| 1218 | static struct target_type snapshot_target = { |
| 1219 | .name = "snapshot", |
Alasdair G Kergon | c51c275 | 2006-06-26 00:27:18 -0700 | [diff] [blame] | 1220 | .version = {1, 4, 0}, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1221 | .module = THIS_MODULE, |
| 1222 | .ctr = snapshot_ctr, |
| 1223 | .dtr = snapshot_dtr, |
| 1224 | .map = snapshot_map, |
| 1225 | .resume = snapshot_resume, |
| 1226 | .status = snapshot_status, |
| 1227 | }; |
| 1228 | |
| 1229 | static int __init dm_snapshot_init(void) |
| 1230 | { |
| 1231 | int r; |
| 1232 | |
| 1233 | r = dm_register_target(&snapshot_target); |
| 1234 | if (r) { |
| 1235 | DMERR("snapshot target register failed %d", r); |
| 1236 | return r; |
| 1237 | } |
| 1238 | |
| 1239 | r = dm_register_target(&origin_target); |
| 1240 | if (r < 0) { |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 1241 | DMERR("Origin target register failed %d", r); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | goto bad1; |
| 1243 | } |
| 1244 | |
| 1245 | r = init_origin_hash(); |
| 1246 | if (r) { |
| 1247 | DMERR("init_origin_hash failed."); |
| 1248 | goto bad2; |
| 1249 | } |
| 1250 | |
| 1251 | exception_cache = kmem_cache_create("dm-snapshot-ex", |
| 1252 | sizeof(struct exception), |
| 1253 | __alignof__(struct exception), |
| 1254 | 0, NULL, NULL); |
| 1255 | if (!exception_cache) { |
| 1256 | DMERR("Couldn't create exception cache."); |
| 1257 | r = -ENOMEM; |
| 1258 | goto bad3; |
| 1259 | } |
| 1260 | |
| 1261 | pending_cache = |
| 1262 | kmem_cache_create("dm-snapshot-in", |
| 1263 | sizeof(struct pending_exception), |
| 1264 | __alignof__(struct pending_exception), |
| 1265 | 0, NULL, NULL); |
| 1266 | if (!pending_cache) { |
| 1267 | DMERR("Couldn't create pending cache."); |
| 1268 | r = -ENOMEM; |
| 1269 | goto bad4; |
| 1270 | } |
| 1271 | |
Matthew Dobson | 93d2341 | 2006-03-26 01:37:50 -0800 | [diff] [blame] | 1272 | pending_pool = mempool_create_slab_pool(128, pending_cache); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | if (!pending_pool) { |
| 1274 | DMERR("Couldn't create pending pool."); |
| 1275 | r = -ENOMEM; |
| 1276 | goto bad5; |
| 1277 | } |
| 1278 | |
| 1279 | return 0; |
| 1280 | |
| 1281 | bad5: |
| 1282 | kmem_cache_destroy(pending_cache); |
| 1283 | bad4: |
| 1284 | kmem_cache_destroy(exception_cache); |
| 1285 | bad3: |
| 1286 | exit_origin_hash(); |
| 1287 | bad2: |
| 1288 | dm_unregister_target(&origin_target); |
| 1289 | bad1: |
| 1290 | dm_unregister_target(&snapshot_target); |
| 1291 | return r; |
| 1292 | } |
| 1293 | |
| 1294 | static void __exit dm_snapshot_exit(void) |
| 1295 | { |
| 1296 | int r; |
| 1297 | |
| 1298 | r = dm_unregister_target(&snapshot_target); |
| 1299 | if (r) |
| 1300 | DMERR("snapshot unregister failed %d", r); |
| 1301 | |
| 1302 | r = dm_unregister_target(&origin_target); |
| 1303 | if (r) |
| 1304 | DMERR("origin unregister failed %d", r); |
| 1305 | |
| 1306 | exit_origin_hash(); |
| 1307 | mempool_destroy(pending_pool); |
| 1308 | kmem_cache_destroy(pending_cache); |
| 1309 | kmem_cache_destroy(exception_cache); |
| 1310 | } |
| 1311 | |
| 1312 | /* Module hooks */ |
| 1313 | module_init(dm_snapshot_init); |
| 1314 | module_exit(dm_snapshot_exit); |
| 1315 | |
| 1316 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); |
| 1317 | MODULE_AUTHOR("Joe Thornber"); |
| 1318 | MODULE_LICENSE("GPL"); |