Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2011 Red Hat UK. |
| 3 | * |
| 4 | * This file is released under the GPL. |
| 5 | */ |
| 6 | |
| 7 | #include "dm-thin-metadata.h" |
| 8 | |
| 9 | #include <linux/device-mapper.h> |
| 10 | #include <linux/dm-io.h> |
| 11 | #include <linux/dm-kcopyd.h> |
| 12 | #include <linux/list.h> |
| 13 | #include <linux/init.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/slab.h> |
| 16 | |
| 17 | #define DM_MSG_PREFIX "thin" |
| 18 | |
| 19 | /* |
| 20 | * Tunable constants |
| 21 | */ |
Alasdair G Kergon | 7768ed3 | 2012-07-27 15:07:57 +0100 | [diff] [blame] | 22 | #define ENDIO_HOOK_POOL_SIZE 1024 |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 23 | #define DEFERRED_SET_SIZE 64 |
| 24 | #define MAPPING_POOL_SIZE 1024 |
| 25 | #define PRISON_CELLS 1024 |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 26 | #define COMMIT_PERIOD HZ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 27 | |
| 28 | /* |
| 29 | * The block size of the device holding pool data must be |
| 30 | * between 64KB and 1GB. |
| 31 | */ |
| 32 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) |
| 33 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
| 34 | |
| 35 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 36 | * Device id is restricted to 24 bits. |
| 37 | */ |
| 38 | #define MAX_DEV_ID ((1 << 24) - 1) |
| 39 | |
| 40 | /* |
| 41 | * How do we handle breaking sharing of data blocks? |
| 42 | * ================================================= |
| 43 | * |
| 44 | * We use a standard copy-on-write btree to store the mappings for the |
| 45 | * devices (note I'm talking about copy-on-write of the metadata here, not |
| 46 | * the data). When you take an internal snapshot you clone the root node |
| 47 | * of the origin btree. After this there is no concept of an origin or a |
| 48 | * snapshot. They are just two device trees that happen to point to the |
| 49 | * same data blocks. |
| 50 | * |
| 51 | * When we get a write in we decide if it's to a shared data block using |
| 52 | * some timestamp magic. If it is, we have to break sharing. |
| 53 | * |
| 54 | * Let's say we write to a shared block in what was the origin. The |
| 55 | * steps are: |
| 56 | * |
| 57 | * i) plug io further to this physical block. (see bio_prison code). |
| 58 | * |
| 59 | * ii) quiesce any read io to that shared data block. Obviously |
| 60 | * including all devices that share this block. (see deferred_set code) |
| 61 | * |
| 62 | * iii) copy the data block to a newly allocate block. This step can be |
| 63 | * missed out if the io covers the block. (schedule_copy). |
| 64 | * |
| 65 | * iv) insert the new mapping into the origin's btree |
Joe Thornber | fe878f3 | 2012-03-28 18:41:24 +0100 | [diff] [blame] | 66 | * (process_prepared_mapping). This act of inserting breaks some |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 67 | * sharing of btree nodes between the two devices. Breaking sharing only |
| 68 | * effects the btree of that specific device. Btrees for the other |
| 69 | * devices that share the block never change. The btree for the origin |
| 70 | * device as it was after the last commit is untouched, ie. we're using |
| 71 | * persistent data structures in the functional programming sense. |
| 72 | * |
| 73 | * v) unplug io to this physical block, including the io that triggered |
| 74 | * the breaking of sharing. |
| 75 | * |
| 76 | * Steps (ii) and (iii) occur in parallel. |
| 77 | * |
| 78 | * The metadata _doesn't_ need to be committed before the io continues. We |
| 79 | * get away with this because the io is always written to a _new_ block. |
| 80 | * If there's a crash, then: |
| 81 | * |
| 82 | * - The origin mapping will point to the old origin block (the shared |
| 83 | * one). This will contain the data as it was before the io that triggered |
| 84 | * the breaking of sharing came in. |
| 85 | * |
| 86 | * - The snap mapping still points to the old block. As it would after |
| 87 | * the commit. |
| 88 | * |
| 89 | * The downside of this scheme is the timestamp magic isn't perfect, and |
| 90 | * will continue to think that data block in the snapshot device is shared |
| 91 | * even after the write to the origin has broken sharing. I suspect data |
| 92 | * blocks will typically be shared by many different devices, so we're |
| 93 | * breaking sharing n + 1 times, rather than n, where n is the number of |
| 94 | * devices that reference this data block. At the moment I think the |
| 95 | * benefits far, far outweigh the disadvantages. |
| 96 | */ |
| 97 | |
| 98 | /*----------------------------------------------------------------*/ |
| 99 | |
| 100 | /* |
| 101 | * Sometimes we can't deal with a bio straight away. We put them in prison |
| 102 | * where they can't cause any mischief. Bios are put in a cell identified |
| 103 | * by a key, multiple bios can be in the same cell. When the cell is |
| 104 | * subsequently unlocked the bios become available. |
| 105 | */ |
| 106 | struct bio_prison; |
| 107 | |
| 108 | struct cell_key { |
| 109 | int virtual; |
| 110 | dm_thin_id dev; |
| 111 | dm_block_t block; |
| 112 | }; |
| 113 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 114 | struct dm_bio_prison_cell { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 115 | struct hlist_node list; |
| 116 | struct bio_prison *prison; |
| 117 | struct cell_key key; |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 118 | struct bio *holder; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 119 | struct bio_list bios; |
| 120 | }; |
| 121 | |
| 122 | struct bio_prison { |
| 123 | spinlock_t lock; |
| 124 | mempool_t *cell_pool; |
| 125 | |
| 126 | unsigned nr_buckets; |
| 127 | unsigned hash_mask; |
| 128 | struct hlist_head *cells; |
| 129 | }; |
| 130 | |
| 131 | static uint32_t calc_nr_buckets(unsigned nr_cells) |
| 132 | { |
| 133 | uint32_t n = 128; |
| 134 | |
| 135 | nr_cells /= 4; |
| 136 | nr_cells = min(nr_cells, 8192u); |
| 137 | |
| 138 | while (n < nr_cells) |
| 139 | n <<= 1; |
| 140 | |
| 141 | return n; |
| 142 | } |
| 143 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 144 | static struct kmem_cache *_cell_cache; |
| 145 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 146 | /* |
| 147 | * @nr_cells should be the number of cells you want in use _concurrently_. |
| 148 | * Don't confuse it with the number of distinct keys. |
| 149 | */ |
| 150 | static struct bio_prison *prison_create(unsigned nr_cells) |
| 151 | { |
| 152 | unsigned i; |
| 153 | uint32_t nr_buckets = calc_nr_buckets(nr_cells); |
| 154 | size_t len = sizeof(struct bio_prison) + |
| 155 | (sizeof(struct hlist_head) * nr_buckets); |
| 156 | struct bio_prison *prison = kmalloc(len, GFP_KERNEL); |
| 157 | |
| 158 | if (!prison) |
| 159 | return NULL; |
| 160 | |
| 161 | spin_lock_init(&prison->lock); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 162 | prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 163 | if (!prison->cell_pool) { |
| 164 | kfree(prison); |
| 165 | return NULL; |
| 166 | } |
| 167 | |
| 168 | prison->nr_buckets = nr_buckets; |
| 169 | prison->hash_mask = nr_buckets - 1; |
| 170 | prison->cells = (struct hlist_head *) (prison + 1); |
| 171 | for (i = 0; i < nr_buckets; i++) |
| 172 | INIT_HLIST_HEAD(prison->cells + i); |
| 173 | |
| 174 | return prison; |
| 175 | } |
| 176 | |
| 177 | static void prison_destroy(struct bio_prison *prison) |
| 178 | { |
| 179 | mempool_destroy(prison->cell_pool); |
| 180 | kfree(prison); |
| 181 | } |
| 182 | |
| 183 | static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key) |
| 184 | { |
| 185 | const unsigned long BIG_PRIME = 4294967291UL; |
| 186 | uint64_t hash = key->block * BIG_PRIME; |
| 187 | |
| 188 | return (uint32_t) (hash & prison->hash_mask); |
| 189 | } |
| 190 | |
| 191 | static int keys_equal(struct cell_key *lhs, struct cell_key *rhs) |
| 192 | { |
| 193 | return (lhs->virtual == rhs->virtual) && |
| 194 | (lhs->dev == rhs->dev) && |
| 195 | (lhs->block == rhs->block); |
| 196 | } |
| 197 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 198 | static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket, |
| 199 | struct cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 200 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 201 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 202 | struct hlist_node *tmp; |
| 203 | |
| 204 | hlist_for_each_entry(cell, tmp, bucket, list) |
| 205 | if (keys_equal(&cell->key, key)) |
| 206 | return cell; |
| 207 | |
| 208 | return NULL; |
| 209 | } |
| 210 | |
| 211 | /* |
| 212 | * This may block if a new cell needs allocating. You must ensure that |
| 213 | * cells will be unlocked even if the calling thread is blocked. |
| 214 | * |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 215 | * Returns 1 if the cell was already held, 0 if @inmate is the new holder. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 216 | */ |
| 217 | static int bio_detain(struct bio_prison *prison, struct cell_key *key, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 218 | struct bio *inmate, struct dm_bio_prison_cell **ref) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 219 | { |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 220 | int r = 1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 221 | unsigned long flags; |
| 222 | uint32_t hash = hash_key(prison, key); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 223 | struct dm_bio_prison_cell *cell, *cell2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 224 | |
| 225 | BUG_ON(hash > prison->nr_buckets); |
| 226 | |
| 227 | spin_lock_irqsave(&prison->lock, flags); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 228 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 229 | cell = __search_bucket(prison->cells + hash, key); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 230 | if (cell) { |
| 231 | bio_list_add(&cell->bios, inmate); |
| 232 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 233 | } |
| 234 | |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 235 | /* |
| 236 | * Allocate a new cell |
| 237 | */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 238 | spin_unlock_irqrestore(&prison->lock, flags); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 239 | cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO); |
| 240 | spin_lock_irqsave(&prison->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 241 | |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 242 | /* |
| 243 | * We've been unlocked, so we have to double check that |
| 244 | * nobody else has inserted this cell in the meantime. |
| 245 | */ |
| 246 | cell = __search_bucket(prison->cells + hash, key); |
| 247 | if (cell) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 248 | mempool_free(cell2, prison->cell_pool); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 249 | bio_list_add(&cell->bios, inmate); |
| 250 | goto out; |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * Use new cell. |
| 255 | */ |
| 256 | cell = cell2; |
| 257 | |
| 258 | cell->prison = prison; |
| 259 | memcpy(&cell->key, key, sizeof(cell->key)); |
| 260 | cell->holder = inmate; |
| 261 | bio_list_init(&cell->bios); |
| 262 | hlist_add_head(&cell->list, prison->cells + hash); |
| 263 | |
| 264 | r = 0; |
| 265 | |
| 266 | out: |
| 267 | spin_unlock_irqrestore(&prison->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 268 | |
| 269 | *ref = cell; |
| 270 | |
| 271 | return r; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * @inmates must have been initialised prior to this call |
| 276 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 277 | static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 278 | { |
| 279 | struct bio_prison *prison = cell->prison; |
| 280 | |
| 281 | hlist_del(&cell->list); |
| 282 | |
Mike Snitzer | 03aaae7 | 2012-05-12 01:43:12 +0100 | [diff] [blame] | 283 | if (inmates) { |
| 284 | bio_list_add(inmates, cell->holder); |
| 285 | bio_list_merge(inmates, &cell->bios); |
| 286 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 287 | |
| 288 | mempool_free(cell, prison->cell_pool); |
| 289 | } |
| 290 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 291 | static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 292 | { |
| 293 | unsigned long flags; |
| 294 | struct bio_prison *prison = cell->prison; |
| 295 | |
| 296 | spin_lock_irqsave(&prison->lock, flags); |
| 297 | __cell_release(cell, bios); |
| 298 | spin_unlock_irqrestore(&prison->lock, flags); |
| 299 | } |
| 300 | |
| 301 | /* |
| 302 | * There are a couple of places where we put a bio into a cell briefly |
| 303 | * before taking it out again. In these situations we know that no other |
| 304 | * bio may be in the cell. This function releases the cell, and also does |
| 305 | * a sanity check. |
| 306 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 307 | static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 308 | { |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 309 | BUG_ON(cell->holder != bio); |
| 310 | BUG_ON(!bio_list_empty(&cell->bios)); |
Mike Snitzer | 03aaae7 | 2012-05-12 01:43:12 +0100 | [diff] [blame] | 311 | |
| 312 | __cell_release(cell, NULL); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 313 | } |
| 314 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 315 | static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 316 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 317 | unsigned long flags; |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 318 | struct bio_prison *prison = cell->prison; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 319 | |
| 320 | spin_lock_irqsave(&prison->lock, flags); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 321 | __cell_release_singleton(cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 322 | spin_unlock_irqrestore(&prison->lock, flags); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 323 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 324 | |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 325 | /* |
| 326 | * Sometimes we don't want the holder, just the additional bios. |
| 327 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 328 | static void __cell_release_no_holder(struct dm_bio_prison_cell *cell, |
| 329 | struct bio_list *inmates) |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 330 | { |
| 331 | struct bio_prison *prison = cell->prison; |
| 332 | |
| 333 | hlist_del(&cell->list); |
| 334 | bio_list_merge(inmates, &cell->bios); |
| 335 | |
| 336 | mempool_free(cell, prison->cell_pool); |
| 337 | } |
| 338 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 339 | static void cell_release_no_holder(struct dm_bio_prison_cell *cell, |
| 340 | struct bio_list *inmates) |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 341 | { |
| 342 | unsigned long flags; |
| 343 | struct bio_prison *prison = cell->prison; |
| 344 | |
| 345 | spin_lock_irqsave(&prison->lock, flags); |
| 346 | __cell_release_no_holder(cell, inmates); |
| 347 | spin_unlock_irqrestore(&prison->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 348 | } |
| 349 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 350 | static void cell_error(struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 351 | { |
| 352 | struct bio_prison *prison = cell->prison; |
| 353 | struct bio_list bios; |
| 354 | struct bio *bio; |
| 355 | unsigned long flags; |
| 356 | |
| 357 | bio_list_init(&bios); |
| 358 | |
| 359 | spin_lock_irqsave(&prison->lock, flags); |
| 360 | __cell_release(cell, &bios); |
| 361 | spin_unlock_irqrestore(&prison->lock, flags); |
| 362 | |
| 363 | while ((bio = bio_list_pop(&bios))) |
| 364 | bio_io_error(bio); |
| 365 | } |
| 366 | |
| 367 | /*----------------------------------------------------------------*/ |
| 368 | |
| 369 | /* |
| 370 | * We use the deferred set to keep track of pending reads to shared blocks. |
| 371 | * We do this to ensure the new mapping caused by a write isn't performed |
| 372 | * until these prior reads have completed. Otherwise the insertion of the |
| 373 | * new mapping could free the old block that the read bios are mapped to. |
| 374 | */ |
| 375 | |
| 376 | struct deferred_set; |
| 377 | struct deferred_entry { |
| 378 | struct deferred_set *ds; |
| 379 | unsigned count; |
| 380 | struct list_head work_items; |
| 381 | }; |
| 382 | |
| 383 | struct deferred_set { |
| 384 | spinlock_t lock; |
| 385 | unsigned current_entry; |
| 386 | unsigned sweeper; |
| 387 | struct deferred_entry entries[DEFERRED_SET_SIZE]; |
| 388 | }; |
| 389 | |
| 390 | static void ds_init(struct deferred_set *ds) |
| 391 | { |
| 392 | int i; |
| 393 | |
| 394 | spin_lock_init(&ds->lock); |
| 395 | ds->current_entry = 0; |
| 396 | ds->sweeper = 0; |
| 397 | for (i = 0; i < DEFERRED_SET_SIZE; i++) { |
| 398 | ds->entries[i].ds = ds; |
| 399 | ds->entries[i].count = 0; |
| 400 | INIT_LIST_HEAD(&ds->entries[i].work_items); |
| 401 | } |
| 402 | } |
| 403 | |
| 404 | static struct deferred_entry *ds_inc(struct deferred_set *ds) |
| 405 | { |
| 406 | unsigned long flags; |
| 407 | struct deferred_entry *entry; |
| 408 | |
| 409 | spin_lock_irqsave(&ds->lock, flags); |
| 410 | entry = ds->entries + ds->current_entry; |
| 411 | entry->count++; |
| 412 | spin_unlock_irqrestore(&ds->lock, flags); |
| 413 | |
| 414 | return entry; |
| 415 | } |
| 416 | |
| 417 | static unsigned ds_next(unsigned index) |
| 418 | { |
| 419 | return (index + 1) % DEFERRED_SET_SIZE; |
| 420 | } |
| 421 | |
| 422 | static void __sweep(struct deferred_set *ds, struct list_head *head) |
| 423 | { |
| 424 | while ((ds->sweeper != ds->current_entry) && |
| 425 | !ds->entries[ds->sweeper].count) { |
| 426 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); |
| 427 | ds->sweeper = ds_next(ds->sweeper); |
| 428 | } |
| 429 | |
| 430 | if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count) |
| 431 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); |
| 432 | } |
| 433 | |
| 434 | static void ds_dec(struct deferred_entry *entry, struct list_head *head) |
| 435 | { |
| 436 | unsigned long flags; |
| 437 | |
| 438 | spin_lock_irqsave(&entry->ds->lock, flags); |
| 439 | BUG_ON(!entry->count); |
| 440 | --entry->count; |
| 441 | __sweep(entry->ds, head); |
| 442 | spin_unlock_irqrestore(&entry->ds->lock, flags); |
| 443 | } |
| 444 | |
| 445 | /* |
| 446 | * Returns 1 if deferred or 0 if no pending items to delay job. |
| 447 | */ |
| 448 | static int ds_add_work(struct deferred_set *ds, struct list_head *work) |
| 449 | { |
| 450 | int r = 1; |
| 451 | unsigned long flags; |
| 452 | unsigned next_entry; |
| 453 | |
| 454 | spin_lock_irqsave(&ds->lock, flags); |
| 455 | if ((ds->sweeper == ds->current_entry) && |
| 456 | !ds->entries[ds->current_entry].count) |
| 457 | r = 0; |
| 458 | else { |
| 459 | list_add(work, &ds->entries[ds->current_entry].work_items); |
| 460 | next_entry = ds_next(ds->current_entry); |
| 461 | if (!ds->entries[next_entry].count) |
| 462 | ds->current_entry = next_entry; |
| 463 | } |
| 464 | spin_unlock_irqrestore(&ds->lock, flags); |
| 465 | |
| 466 | return r; |
| 467 | } |
| 468 | |
| 469 | /*----------------------------------------------------------------*/ |
| 470 | |
| 471 | /* |
| 472 | * Key building. |
| 473 | */ |
| 474 | static void build_data_key(struct dm_thin_device *td, |
| 475 | dm_block_t b, struct cell_key *key) |
| 476 | { |
| 477 | key->virtual = 0; |
| 478 | key->dev = dm_thin_dev_id(td); |
| 479 | key->block = b; |
| 480 | } |
| 481 | |
| 482 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, |
| 483 | struct cell_key *key) |
| 484 | { |
| 485 | key->virtual = 1; |
| 486 | key->dev = dm_thin_dev_id(td); |
| 487 | key->block = b; |
| 488 | } |
| 489 | |
| 490 | /*----------------------------------------------------------------*/ |
| 491 | |
| 492 | /* |
| 493 | * A pool device ties together a metadata device and a data device. It |
| 494 | * also provides the interface for creating and destroying internal |
| 495 | * devices. |
| 496 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 497 | struct dm_thin_new_mapping; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 498 | |
| 499 | struct pool_features { |
| 500 | unsigned zero_new_blocks:1; |
| 501 | unsigned discard_enabled:1; |
| 502 | unsigned discard_passdown:1; |
| 503 | }; |
| 504 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 505 | struct pool { |
| 506 | struct list_head list; |
| 507 | struct dm_target *ti; /* Only set if a pool target is bound */ |
| 508 | |
| 509 | struct mapped_device *pool_md; |
| 510 | struct block_device *md_dev; |
| 511 | struct dm_pool_metadata *pmd; |
| 512 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 513 | dm_block_t low_water_blocks; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 514 | uint32_t sectors_per_block; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 515 | int sectors_per_block_shift; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 516 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 517 | struct pool_features pf; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 518 | unsigned low_water_triggered:1; /* A dm event has been sent */ |
| 519 | unsigned no_free_space:1; /* A -ENOSPC warning has been issued */ |
| 520 | |
| 521 | struct bio_prison *prison; |
| 522 | struct dm_kcopyd_client *copier; |
| 523 | |
| 524 | struct workqueue_struct *wq; |
| 525 | struct work_struct worker; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 526 | struct delayed_work waker; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 527 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 528 | unsigned long last_commit_jiffies; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 529 | unsigned ref_count; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 530 | |
| 531 | spinlock_t lock; |
| 532 | struct bio_list deferred_bios; |
| 533 | struct bio_list deferred_flush_bios; |
| 534 | struct list_head prepared_mappings; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 535 | struct list_head prepared_discards; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 536 | |
| 537 | struct bio_list retry_on_resume_list; |
| 538 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 539 | struct deferred_set shared_read_ds; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 540 | struct deferred_set all_io_ds; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 541 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 542 | struct dm_thin_new_mapping *next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 543 | mempool_t *mapping_pool; |
| 544 | mempool_t *endio_hook_pool; |
| 545 | }; |
| 546 | |
| 547 | /* |
| 548 | * Target context for a pool. |
| 549 | */ |
| 550 | struct pool_c { |
| 551 | struct dm_target *ti; |
| 552 | struct pool *pool; |
| 553 | struct dm_dev *data_dev; |
| 554 | struct dm_dev *metadata_dev; |
| 555 | struct dm_target_callbacks callbacks; |
| 556 | |
| 557 | dm_block_t low_water_blocks; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 558 | struct pool_features pf; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 559 | }; |
| 560 | |
| 561 | /* |
| 562 | * Target context for a thin. |
| 563 | */ |
| 564 | struct thin_c { |
| 565 | struct dm_dev *pool_dev; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 566 | struct dm_dev *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 567 | dm_thin_id dev_id; |
| 568 | |
| 569 | struct pool *pool; |
| 570 | struct dm_thin_device *td; |
| 571 | }; |
| 572 | |
| 573 | /*----------------------------------------------------------------*/ |
| 574 | |
| 575 | /* |
| 576 | * A global list of pools that uses a struct mapped_device as a key. |
| 577 | */ |
| 578 | static struct dm_thin_pool_table { |
| 579 | struct mutex mutex; |
| 580 | struct list_head pools; |
| 581 | } dm_thin_pool_table; |
| 582 | |
| 583 | static void pool_table_init(void) |
| 584 | { |
| 585 | mutex_init(&dm_thin_pool_table.mutex); |
| 586 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); |
| 587 | } |
| 588 | |
| 589 | static void __pool_table_insert(struct pool *pool) |
| 590 | { |
| 591 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 592 | list_add(&pool->list, &dm_thin_pool_table.pools); |
| 593 | } |
| 594 | |
| 595 | static void __pool_table_remove(struct pool *pool) |
| 596 | { |
| 597 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 598 | list_del(&pool->list); |
| 599 | } |
| 600 | |
| 601 | static struct pool *__pool_table_lookup(struct mapped_device *md) |
| 602 | { |
| 603 | struct pool *pool = NULL, *tmp; |
| 604 | |
| 605 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 606 | |
| 607 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 608 | if (tmp->pool_md == md) { |
| 609 | pool = tmp; |
| 610 | break; |
| 611 | } |
| 612 | } |
| 613 | |
| 614 | return pool; |
| 615 | } |
| 616 | |
| 617 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) |
| 618 | { |
| 619 | struct pool *pool = NULL, *tmp; |
| 620 | |
| 621 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 622 | |
| 623 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 624 | if (tmp->md_dev == md_dev) { |
| 625 | pool = tmp; |
| 626 | break; |
| 627 | } |
| 628 | } |
| 629 | |
| 630 | return pool; |
| 631 | } |
| 632 | |
| 633 | /*----------------------------------------------------------------*/ |
| 634 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 635 | struct dm_thin_endio_hook { |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 636 | struct thin_c *tc; |
| 637 | struct deferred_entry *shared_read_entry; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 638 | struct deferred_entry *all_io_entry; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 639 | struct dm_thin_new_mapping *overwrite_mapping; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 640 | }; |
| 641 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 642 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
| 643 | { |
| 644 | struct bio *bio; |
| 645 | struct bio_list bios; |
| 646 | |
| 647 | bio_list_init(&bios); |
| 648 | bio_list_merge(&bios, master); |
| 649 | bio_list_init(master); |
| 650 | |
| 651 | while ((bio = bio_list_pop(&bios))) { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 652 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
| 653 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 654 | if (h->tc == tc) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 655 | bio_endio(bio, DM_ENDIO_REQUEUE); |
| 656 | else |
| 657 | bio_list_add(master, bio); |
| 658 | } |
| 659 | } |
| 660 | |
| 661 | static void requeue_io(struct thin_c *tc) |
| 662 | { |
| 663 | struct pool *pool = tc->pool; |
| 664 | unsigned long flags; |
| 665 | |
| 666 | spin_lock_irqsave(&pool->lock, flags); |
| 667 | __requeue_bio_list(tc, &pool->deferred_bios); |
| 668 | __requeue_bio_list(tc, &pool->retry_on_resume_list); |
| 669 | spin_unlock_irqrestore(&pool->lock, flags); |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * This section of code contains the logic for processing a thin device's IO. |
| 674 | * Much of the code depends on pool object resources (lists, workqueues, etc) |
| 675 | * but most is exclusively called from the thin target rather than the thin-pool |
| 676 | * target. |
| 677 | */ |
| 678 | |
| 679 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
| 680 | { |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 681 | sector_t block_nr = bio->bi_sector; |
| 682 | |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 683 | if (tc->pool->sectors_per_block_shift < 0) |
| 684 | (void) sector_div(block_nr, tc->pool->sectors_per_block); |
| 685 | else |
| 686 | block_nr >>= tc->pool->sectors_per_block_shift; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 687 | |
| 688 | return block_nr; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 689 | } |
| 690 | |
| 691 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
| 692 | { |
| 693 | struct pool *pool = tc->pool; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 694 | sector_t bi_sector = bio->bi_sector; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 695 | |
| 696 | bio->bi_bdev = tc->pool_dev->bdev; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 697 | if (tc->pool->sectors_per_block_shift < 0) |
| 698 | bio->bi_sector = (block * pool->sectors_per_block) + |
| 699 | sector_div(bi_sector, pool->sectors_per_block); |
| 700 | else |
| 701 | bio->bi_sector = (block << pool->sectors_per_block_shift) | |
| 702 | (bi_sector & (pool->sectors_per_block - 1)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 703 | } |
| 704 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 705 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
| 706 | { |
| 707 | bio->bi_bdev = tc->origin_dev->bdev; |
| 708 | } |
| 709 | |
| 710 | static void issue(struct thin_c *tc, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 711 | { |
| 712 | struct pool *pool = tc->pool; |
| 713 | unsigned long flags; |
| 714 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 715 | /* |
| 716 | * Batch together any FUA/FLUSH bios we find and then issue |
| 717 | * a single commit for them in process_deferred_bios(). |
| 718 | */ |
| 719 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
| 720 | spin_lock_irqsave(&pool->lock, flags); |
| 721 | bio_list_add(&pool->deferred_flush_bios, bio); |
| 722 | spin_unlock_irqrestore(&pool->lock, flags); |
| 723 | } else |
| 724 | generic_make_request(bio); |
| 725 | } |
| 726 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 727 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
| 728 | { |
| 729 | remap_to_origin(tc, bio); |
| 730 | issue(tc, bio); |
| 731 | } |
| 732 | |
| 733 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, |
| 734 | dm_block_t block) |
| 735 | { |
| 736 | remap(tc, bio, block); |
| 737 | issue(tc, bio); |
| 738 | } |
| 739 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 740 | /* |
| 741 | * wake_worker() is used when new work is queued and when pool_resume is |
| 742 | * ready to continue deferred IO processing. |
| 743 | */ |
| 744 | static void wake_worker(struct pool *pool) |
| 745 | { |
| 746 | queue_work(pool->wq, &pool->worker); |
| 747 | } |
| 748 | |
| 749 | /*----------------------------------------------------------------*/ |
| 750 | |
| 751 | /* |
| 752 | * Bio endio functions. |
| 753 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 754 | struct dm_thin_new_mapping { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 755 | struct list_head list; |
| 756 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 757 | unsigned quiesced:1; |
| 758 | unsigned prepared:1; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 759 | unsigned pass_discard:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 760 | |
| 761 | struct thin_c *tc; |
| 762 | dm_block_t virt_block; |
| 763 | dm_block_t data_block; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 764 | struct dm_bio_prison_cell *cell, *cell2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 765 | int err; |
| 766 | |
| 767 | /* |
| 768 | * If the bio covers the whole area of a block then we can avoid |
| 769 | * zeroing or copying. Instead this bio is hooked. The bio will |
| 770 | * still be in the cell, so care has to be taken to avoid issuing |
| 771 | * the bio twice. |
| 772 | */ |
| 773 | struct bio *bio; |
| 774 | bio_end_io_t *saved_bi_end_io; |
| 775 | }; |
| 776 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 777 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 778 | { |
| 779 | struct pool *pool = m->tc->pool; |
| 780 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 781 | if (m->quiesced && m->prepared) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 782 | list_add(&m->list, &pool->prepared_mappings); |
| 783 | wake_worker(pool); |
| 784 | } |
| 785 | } |
| 786 | |
| 787 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
| 788 | { |
| 789 | unsigned long flags; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 790 | struct dm_thin_new_mapping *m = context; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 791 | struct pool *pool = m->tc->pool; |
| 792 | |
| 793 | m->err = read_err || write_err ? -EIO : 0; |
| 794 | |
| 795 | spin_lock_irqsave(&pool->lock, flags); |
| 796 | m->prepared = 1; |
| 797 | __maybe_add_mapping(m); |
| 798 | spin_unlock_irqrestore(&pool->lock, flags); |
| 799 | } |
| 800 | |
| 801 | static void overwrite_endio(struct bio *bio, int err) |
| 802 | { |
| 803 | unsigned long flags; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 804 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
| 805 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 806 | struct pool *pool = m->tc->pool; |
| 807 | |
| 808 | m->err = err; |
| 809 | |
| 810 | spin_lock_irqsave(&pool->lock, flags); |
| 811 | m->prepared = 1; |
| 812 | __maybe_add_mapping(m); |
| 813 | spin_unlock_irqrestore(&pool->lock, flags); |
| 814 | } |
| 815 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 816 | /*----------------------------------------------------------------*/ |
| 817 | |
| 818 | /* |
| 819 | * Workqueue. |
| 820 | */ |
| 821 | |
| 822 | /* |
| 823 | * Prepared mapping jobs. |
| 824 | */ |
| 825 | |
| 826 | /* |
| 827 | * This sends the bios in the cell back to the deferred_bios list. |
| 828 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 829 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 830 | dm_block_t data_block) |
| 831 | { |
| 832 | struct pool *pool = tc->pool; |
| 833 | unsigned long flags; |
| 834 | |
| 835 | spin_lock_irqsave(&pool->lock, flags); |
| 836 | cell_release(cell, &pool->deferred_bios); |
| 837 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
| 838 | |
| 839 | wake_worker(pool); |
| 840 | } |
| 841 | |
| 842 | /* |
| 843 | * Same as cell_defer above, except it omits one particular detainee, |
| 844 | * a write bio that covers the block and has already been processed. |
| 845 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 846 | static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 847 | { |
| 848 | struct bio_list bios; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 849 | struct pool *pool = tc->pool; |
| 850 | unsigned long flags; |
| 851 | |
| 852 | bio_list_init(&bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 853 | |
| 854 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 855 | cell_release_no_holder(cell, &pool->deferred_bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 856 | spin_unlock_irqrestore(&pool->lock, flags); |
| 857 | |
| 858 | wake_worker(pool); |
| 859 | } |
| 860 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 861 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 862 | { |
| 863 | struct thin_c *tc = m->tc; |
| 864 | struct bio *bio; |
| 865 | int r; |
| 866 | |
| 867 | bio = m->bio; |
| 868 | if (bio) |
| 869 | bio->bi_end_io = m->saved_bi_end_io; |
| 870 | |
| 871 | if (m->err) { |
| 872 | cell_error(m->cell); |
| 873 | return; |
| 874 | } |
| 875 | |
| 876 | /* |
| 877 | * Commit the prepared block into the mapping btree. |
| 878 | * Any I/O for this block arriving after this point will get |
| 879 | * remapped to it directly. |
| 880 | */ |
| 881 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); |
| 882 | if (r) { |
| 883 | DMERR("dm_thin_insert_block() failed"); |
| 884 | cell_error(m->cell); |
| 885 | return; |
| 886 | } |
| 887 | |
| 888 | /* |
| 889 | * Release any bios held while the block was being provisioned. |
| 890 | * If we are processing a write bio that completely covers the block, |
| 891 | * we already processed it so can ignore it now when processing |
| 892 | * the bios in the cell. |
| 893 | */ |
| 894 | if (bio) { |
Joe Thornber | 6f94a4c | 2012-03-28 18:41:23 +0100 | [diff] [blame] | 895 | cell_defer_except(tc, m->cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 896 | bio_endio(bio, 0); |
| 897 | } else |
| 898 | cell_defer(tc, m->cell, m->data_block); |
| 899 | |
| 900 | list_del(&m->list); |
| 901 | mempool_free(m, tc->pool->mapping_pool); |
| 902 | } |
| 903 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 904 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 905 | { |
| 906 | int r; |
| 907 | struct thin_c *tc = m->tc; |
| 908 | |
| 909 | r = dm_thin_remove_block(tc->td, m->virt_block); |
| 910 | if (r) |
| 911 | DMERR("dm_thin_remove_block() failed"); |
| 912 | |
| 913 | /* |
| 914 | * Pass the discard down to the underlying device? |
| 915 | */ |
| 916 | if (m->pass_discard) |
| 917 | remap_and_issue(tc, m->bio, m->data_block); |
| 918 | else |
| 919 | bio_endio(m->bio, 0); |
| 920 | |
| 921 | cell_defer_except(tc, m->cell); |
| 922 | cell_defer_except(tc, m->cell2); |
| 923 | mempool_free(m, tc->pool->mapping_pool); |
| 924 | } |
| 925 | |
| 926 | static void process_prepared(struct pool *pool, struct list_head *head, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 927 | void (*fn)(struct dm_thin_new_mapping *)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 928 | { |
| 929 | unsigned long flags; |
| 930 | struct list_head maps; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 931 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 932 | |
| 933 | INIT_LIST_HEAD(&maps); |
| 934 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 935 | list_splice_init(head, &maps); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 936 | spin_unlock_irqrestore(&pool->lock, flags); |
| 937 | |
| 938 | list_for_each_entry_safe(m, tmp, &maps, list) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 939 | fn(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 940 | } |
| 941 | |
| 942 | /* |
| 943 | * Deferred bio jobs. |
| 944 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 945 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
| 946 | { |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 947 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 948 | } |
| 949 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 950 | static int io_overwrites_block(struct pool *pool, struct bio *bio) |
| 951 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 952 | return (bio_data_dir(bio) == WRITE) && |
| 953 | io_overlaps_block(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 954 | } |
| 955 | |
| 956 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, |
| 957 | bio_end_io_t *fn) |
| 958 | { |
| 959 | *save = bio->bi_end_io; |
| 960 | bio->bi_end_io = fn; |
| 961 | } |
| 962 | |
| 963 | static int ensure_next_mapping(struct pool *pool) |
| 964 | { |
| 965 | if (pool->next_mapping) |
| 966 | return 0; |
| 967 | |
| 968 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); |
| 969 | |
| 970 | return pool->next_mapping ? 0 : -ENOMEM; |
| 971 | } |
| 972 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 973 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 974 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 975 | struct dm_thin_new_mapping *r = pool->next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 976 | |
| 977 | BUG_ON(!pool->next_mapping); |
| 978 | |
| 979 | pool->next_mapping = NULL; |
| 980 | |
| 981 | return r; |
| 982 | } |
| 983 | |
| 984 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 985 | struct dm_dev *origin, dm_block_t data_origin, |
| 986 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 987 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 988 | { |
| 989 | int r; |
| 990 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 991 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 992 | |
| 993 | INIT_LIST_HEAD(&m->list); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 994 | m->quiesced = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 995 | m->prepared = 0; |
| 996 | m->tc = tc; |
| 997 | m->virt_block = virt_block; |
| 998 | m->data_block = data_dest; |
| 999 | m->cell = cell; |
| 1000 | m->err = 0; |
| 1001 | m->bio = NULL; |
| 1002 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1003 | if (!ds_add_work(&pool->shared_read_ds, &m->list)) |
| 1004 | m->quiesced = 1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1005 | |
| 1006 | /* |
| 1007 | * IO to pool_dev remaps to the pool target's data_dev. |
| 1008 | * |
| 1009 | * If the whole block of data is being overwritten, we can issue the |
| 1010 | * bio immediately. Otherwise we use kcopyd to clone the data first. |
| 1011 | */ |
| 1012 | if (io_overwrites_block(pool, bio)) { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1013 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
| 1014 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1015 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1016 | m->bio = bio; |
| 1017 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1018 | remap_and_issue(tc, bio, data_dest); |
| 1019 | } else { |
| 1020 | struct dm_io_region from, to; |
| 1021 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1022 | from.bdev = origin->bdev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1023 | from.sector = data_origin * pool->sectors_per_block; |
| 1024 | from.count = pool->sectors_per_block; |
| 1025 | |
| 1026 | to.bdev = tc->pool_dev->bdev; |
| 1027 | to.sector = data_dest * pool->sectors_per_block; |
| 1028 | to.count = pool->sectors_per_block; |
| 1029 | |
| 1030 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, |
| 1031 | 0, copy_complete, m); |
| 1032 | if (r < 0) { |
| 1033 | mempool_free(m, pool->mapping_pool); |
| 1034 | DMERR("dm_kcopyd_copy() failed"); |
| 1035 | cell_error(cell); |
| 1036 | } |
| 1037 | } |
| 1038 | } |
| 1039 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1040 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
| 1041 | dm_block_t data_origin, dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1042 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1043 | { |
| 1044 | schedule_copy(tc, virt_block, tc->pool_dev, |
| 1045 | data_origin, data_dest, cell, bio); |
| 1046 | } |
| 1047 | |
| 1048 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
| 1049 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1050 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1051 | { |
| 1052 | schedule_copy(tc, virt_block, tc->origin_dev, |
| 1053 | virt_block, data_dest, cell, bio); |
| 1054 | } |
| 1055 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1056 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1057 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1058 | struct bio *bio) |
| 1059 | { |
| 1060 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1061 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1062 | |
| 1063 | INIT_LIST_HEAD(&m->list); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1064 | m->quiesced = 1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1065 | m->prepared = 0; |
| 1066 | m->tc = tc; |
| 1067 | m->virt_block = virt_block; |
| 1068 | m->data_block = data_block; |
| 1069 | m->cell = cell; |
| 1070 | m->err = 0; |
| 1071 | m->bio = NULL; |
| 1072 | |
| 1073 | /* |
| 1074 | * If the whole block of data is being overwritten or we are not |
| 1075 | * zeroing pre-existing data, we can issue the bio immediately. |
| 1076 | * Otherwise we use kcopyd to zero the data first. |
| 1077 | */ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1078 | if (!pool->pf.zero_new_blocks) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1079 | process_prepared_mapping(m); |
| 1080 | |
| 1081 | else if (io_overwrites_block(pool, bio)) { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1082 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
| 1083 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1084 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1085 | m->bio = bio; |
| 1086 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1087 | remap_and_issue(tc, bio, data_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1088 | } else { |
| 1089 | int r; |
| 1090 | struct dm_io_region to; |
| 1091 | |
| 1092 | to.bdev = tc->pool_dev->bdev; |
| 1093 | to.sector = data_block * pool->sectors_per_block; |
| 1094 | to.count = pool->sectors_per_block; |
| 1095 | |
| 1096 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); |
| 1097 | if (r < 0) { |
| 1098 | mempool_free(m, pool->mapping_pool); |
| 1099 | DMERR("dm_kcopyd_zero() failed"); |
| 1100 | cell_error(cell); |
| 1101 | } |
| 1102 | } |
| 1103 | } |
| 1104 | |
| 1105 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
| 1106 | { |
| 1107 | int r; |
| 1108 | dm_block_t free_blocks; |
| 1109 | unsigned long flags; |
| 1110 | struct pool *pool = tc->pool; |
| 1111 | |
| 1112 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
| 1113 | if (r) |
| 1114 | return r; |
| 1115 | |
| 1116 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
| 1117 | DMWARN("%s: reached low water mark, sending event.", |
| 1118 | dm_device_name(pool->pool_md)); |
| 1119 | spin_lock_irqsave(&pool->lock, flags); |
| 1120 | pool->low_water_triggered = 1; |
| 1121 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1122 | dm_table_event(pool->ti->table); |
| 1123 | } |
| 1124 | |
| 1125 | if (!free_blocks) { |
| 1126 | if (pool->no_free_space) |
| 1127 | return -ENOSPC; |
| 1128 | else { |
| 1129 | /* |
| 1130 | * Try to commit to see if that will free up some |
| 1131 | * more space. |
| 1132 | */ |
| 1133 | r = dm_pool_commit_metadata(pool->pmd); |
| 1134 | if (r) { |
| 1135 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", |
| 1136 | __func__, r); |
| 1137 | return r; |
| 1138 | } |
| 1139 | |
| 1140 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
| 1141 | if (r) |
| 1142 | return r; |
| 1143 | |
| 1144 | /* |
| 1145 | * If we still have no space we set a flag to avoid |
| 1146 | * doing all this checking and return -ENOSPC. |
| 1147 | */ |
| 1148 | if (!free_blocks) { |
| 1149 | DMWARN("%s: no free space available.", |
| 1150 | dm_device_name(pool->pool_md)); |
| 1151 | spin_lock_irqsave(&pool->lock, flags); |
| 1152 | pool->no_free_space = 1; |
| 1153 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1154 | return -ENOSPC; |
| 1155 | } |
| 1156 | } |
| 1157 | } |
| 1158 | |
| 1159 | r = dm_pool_alloc_data_block(pool->pmd, result); |
| 1160 | if (r) |
| 1161 | return r; |
| 1162 | |
| 1163 | return 0; |
| 1164 | } |
| 1165 | |
| 1166 | /* |
| 1167 | * If we have run out of space, queue bios until the device is |
| 1168 | * resumed, presumably after having been reloaded with more space. |
| 1169 | */ |
| 1170 | static void retry_on_resume(struct bio *bio) |
| 1171 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1172 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1173 | struct thin_c *tc = h->tc; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1174 | struct pool *pool = tc->pool; |
| 1175 | unsigned long flags; |
| 1176 | |
| 1177 | spin_lock_irqsave(&pool->lock, flags); |
| 1178 | bio_list_add(&pool->retry_on_resume_list, bio); |
| 1179 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1180 | } |
| 1181 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1182 | static void no_space(struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1183 | { |
| 1184 | struct bio *bio; |
| 1185 | struct bio_list bios; |
| 1186 | |
| 1187 | bio_list_init(&bios); |
| 1188 | cell_release(cell, &bios); |
| 1189 | |
| 1190 | while ((bio = bio_list_pop(&bios))) |
| 1191 | retry_on_resume(bio); |
| 1192 | } |
| 1193 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1194 | static void process_discard(struct thin_c *tc, struct bio *bio) |
| 1195 | { |
| 1196 | int r; |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1197 | unsigned long flags; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1198 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1199 | struct dm_bio_prison_cell *cell, *cell2; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1200 | struct cell_key key, key2; |
| 1201 | dm_block_t block = get_bio_block(tc, bio); |
| 1202 | struct dm_thin_lookup_result lookup_result; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1203 | struct dm_thin_new_mapping *m; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1204 | |
| 1205 | build_virtual_key(tc->td, block, &key); |
| 1206 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) |
| 1207 | return; |
| 1208 | |
| 1209 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1210 | switch (r) { |
| 1211 | case 0: |
| 1212 | /* |
| 1213 | * Check nobody is fiddling with this pool block. This can |
| 1214 | * happen if someone's in the process of breaking sharing |
| 1215 | * on this block. |
| 1216 | */ |
| 1217 | build_data_key(tc->td, lookup_result.block, &key2); |
| 1218 | if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) { |
| 1219 | cell_release_singleton(cell, bio); |
| 1220 | break; |
| 1221 | } |
| 1222 | |
| 1223 | if (io_overlaps_block(pool, bio)) { |
| 1224 | /* |
| 1225 | * IO may still be going to the destination block. We must |
| 1226 | * quiesce before we can do the removal. |
| 1227 | */ |
| 1228 | m = get_next_mapping(pool); |
| 1229 | m->tc = tc; |
Mike Snitzer | 17b7d63 | 2012-07-27 15:07:57 +0100 | [diff] [blame] | 1230 | m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1231 | m->virt_block = block; |
| 1232 | m->data_block = lookup_result.block; |
| 1233 | m->cell = cell; |
| 1234 | m->cell2 = cell2; |
| 1235 | m->err = 0; |
| 1236 | m->bio = bio; |
| 1237 | |
| 1238 | if (!ds_add_work(&pool->all_io_ds, &m->list)) { |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1239 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1240 | list_add(&m->list, &pool->prepared_discards); |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1241 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1242 | wake_worker(pool); |
| 1243 | } |
| 1244 | } else { |
| 1245 | /* |
Mikulas Patocka | 4929630 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1246 | * The DM core makes sure that the discard doesn't span |
| 1247 | * a block boundary. So we submit the discard of a |
| 1248 | * partial block appropriately. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1249 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1250 | cell_release_singleton(cell, bio); |
| 1251 | cell_release_singleton(cell2, bio); |
Mikulas Patocka | 650d2a0 | 2012-07-20 14:25:05 +0100 | [diff] [blame] | 1252 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
| 1253 | remap_and_issue(tc, bio, lookup_result.block); |
| 1254 | else |
| 1255 | bio_endio(bio, 0); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1256 | } |
| 1257 | break; |
| 1258 | |
| 1259 | case -ENODATA: |
| 1260 | /* |
| 1261 | * It isn't provisioned, just forget it. |
| 1262 | */ |
| 1263 | cell_release_singleton(cell, bio); |
| 1264 | bio_endio(bio, 0); |
| 1265 | break; |
| 1266 | |
| 1267 | default: |
| 1268 | DMERR("discard: find block unexpectedly returned %d", r); |
| 1269 | cell_release_singleton(cell, bio); |
| 1270 | bio_io_error(bio); |
| 1271 | break; |
| 1272 | } |
| 1273 | } |
| 1274 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1275 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
| 1276 | struct cell_key *key, |
| 1277 | struct dm_thin_lookup_result *lookup_result, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1278 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1279 | { |
| 1280 | int r; |
| 1281 | dm_block_t data_block; |
| 1282 | |
| 1283 | r = alloc_data_block(tc, &data_block); |
| 1284 | switch (r) { |
| 1285 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1286 | schedule_internal_copy(tc, block, lookup_result->block, |
| 1287 | data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1288 | break; |
| 1289 | |
| 1290 | case -ENOSPC: |
| 1291 | no_space(cell); |
| 1292 | break; |
| 1293 | |
| 1294 | default: |
| 1295 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); |
| 1296 | cell_error(cell); |
| 1297 | break; |
| 1298 | } |
| 1299 | } |
| 1300 | |
| 1301 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
| 1302 | dm_block_t block, |
| 1303 | struct dm_thin_lookup_result *lookup_result) |
| 1304 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1305 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1306 | struct pool *pool = tc->pool; |
| 1307 | struct cell_key key; |
| 1308 | |
| 1309 | /* |
| 1310 | * If cell is already occupied, then sharing is already in the process |
| 1311 | * of being broken so we have nothing further to do here. |
| 1312 | */ |
| 1313 | build_data_key(tc->td, lookup_result->block, &key); |
| 1314 | if (bio_detain(pool->prison, &key, bio, &cell)) |
| 1315 | return; |
| 1316 | |
| 1317 | if (bio_data_dir(bio) == WRITE) |
| 1318 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
| 1319 | else { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1320 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1321 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1322 | h->shared_read_entry = ds_inc(&pool->shared_read_ds); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1323 | |
| 1324 | cell_release_singleton(cell, bio); |
| 1325 | remap_and_issue(tc, bio, lookup_result->block); |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1330 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1331 | { |
| 1332 | int r; |
| 1333 | dm_block_t data_block; |
| 1334 | |
| 1335 | /* |
| 1336 | * Remap empty bios (flushes) immediately, without provisioning. |
| 1337 | */ |
| 1338 | if (!bio->bi_size) { |
| 1339 | cell_release_singleton(cell, bio); |
| 1340 | remap_and_issue(tc, bio, 0); |
| 1341 | return; |
| 1342 | } |
| 1343 | |
| 1344 | /* |
| 1345 | * Fill read bios with zeroes and complete them immediately. |
| 1346 | */ |
| 1347 | if (bio_data_dir(bio) == READ) { |
| 1348 | zero_fill_bio(bio); |
| 1349 | cell_release_singleton(cell, bio); |
| 1350 | bio_endio(bio, 0); |
| 1351 | return; |
| 1352 | } |
| 1353 | |
| 1354 | r = alloc_data_block(tc, &data_block); |
| 1355 | switch (r) { |
| 1356 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1357 | if (tc->origin_dev) |
| 1358 | schedule_external_copy(tc, block, data_block, cell, bio); |
| 1359 | else |
| 1360 | schedule_zero(tc, block, data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1361 | break; |
| 1362 | |
| 1363 | case -ENOSPC: |
| 1364 | no_space(cell); |
| 1365 | break; |
| 1366 | |
| 1367 | default: |
| 1368 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); |
| 1369 | cell_error(cell); |
| 1370 | break; |
| 1371 | } |
| 1372 | } |
| 1373 | |
| 1374 | static void process_bio(struct thin_c *tc, struct bio *bio) |
| 1375 | { |
| 1376 | int r; |
| 1377 | dm_block_t block = get_bio_block(tc, bio); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1378 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1379 | struct cell_key key; |
| 1380 | struct dm_thin_lookup_result lookup_result; |
| 1381 | |
| 1382 | /* |
| 1383 | * If cell is already occupied, then the block is already |
| 1384 | * being provisioned so we have nothing further to do here. |
| 1385 | */ |
| 1386 | build_virtual_key(tc->td, block, &key); |
| 1387 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) |
| 1388 | return; |
| 1389 | |
| 1390 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1391 | switch (r) { |
| 1392 | case 0: |
| 1393 | /* |
| 1394 | * We can release this cell now. This thread is the only |
| 1395 | * one that puts bios into a cell, and we know there were |
| 1396 | * no preceding bios. |
| 1397 | */ |
| 1398 | /* |
| 1399 | * TODO: this will probably have to change when discard goes |
| 1400 | * back in. |
| 1401 | */ |
| 1402 | cell_release_singleton(cell, bio); |
| 1403 | |
| 1404 | if (lookup_result.shared) |
| 1405 | process_shared_bio(tc, bio, block, &lookup_result); |
| 1406 | else |
| 1407 | remap_and_issue(tc, bio, lookup_result.block); |
| 1408 | break; |
| 1409 | |
| 1410 | case -ENODATA: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1411 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
| 1412 | cell_release_singleton(cell, bio); |
| 1413 | remap_to_origin_and_issue(tc, bio); |
| 1414 | } else |
| 1415 | provision_block(tc, bio, block, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1416 | break; |
| 1417 | |
| 1418 | default: |
| 1419 | DMERR("dm_thin_find_block() failed, error = %d", r); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1420 | cell_release_singleton(cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1421 | bio_io_error(bio); |
| 1422 | break; |
| 1423 | } |
| 1424 | } |
| 1425 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1426 | static int need_commit_due_to_time(struct pool *pool) |
| 1427 | { |
| 1428 | return jiffies < pool->last_commit_jiffies || |
| 1429 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; |
| 1430 | } |
| 1431 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1432 | static void process_deferred_bios(struct pool *pool) |
| 1433 | { |
| 1434 | unsigned long flags; |
| 1435 | struct bio *bio; |
| 1436 | struct bio_list bios; |
| 1437 | int r; |
| 1438 | |
| 1439 | bio_list_init(&bios); |
| 1440 | |
| 1441 | spin_lock_irqsave(&pool->lock, flags); |
| 1442 | bio_list_merge(&bios, &pool->deferred_bios); |
| 1443 | bio_list_init(&pool->deferred_bios); |
| 1444 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1445 | |
| 1446 | while ((bio = bio_list_pop(&bios))) { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1447 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1448 | struct thin_c *tc = h->tc; |
| 1449 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1450 | /* |
| 1451 | * If we've got no free new_mapping structs, and processing |
| 1452 | * this bio might require one, we pause until there are some |
| 1453 | * prepared mappings to process. |
| 1454 | */ |
| 1455 | if (ensure_next_mapping(pool)) { |
| 1456 | spin_lock_irqsave(&pool->lock, flags); |
| 1457 | bio_list_merge(&pool->deferred_bios, &bios); |
| 1458 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1459 | |
| 1460 | break; |
| 1461 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1462 | |
| 1463 | if (bio->bi_rw & REQ_DISCARD) |
| 1464 | process_discard(tc, bio); |
| 1465 | else |
| 1466 | process_bio(tc, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1467 | } |
| 1468 | |
| 1469 | /* |
| 1470 | * If there are any deferred flush bios, we must commit |
| 1471 | * the metadata before issuing them. |
| 1472 | */ |
| 1473 | bio_list_init(&bios); |
| 1474 | spin_lock_irqsave(&pool->lock, flags); |
| 1475 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
| 1476 | bio_list_init(&pool->deferred_flush_bios); |
| 1477 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1478 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1479 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1480 | return; |
| 1481 | |
| 1482 | r = dm_pool_commit_metadata(pool->pmd); |
| 1483 | if (r) { |
| 1484 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", |
| 1485 | __func__, r); |
| 1486 | while ((bio = bio_list_pop(&bios))) |
| 1487 | bio_io_error(bio); |
| 1488 | return; |
| 1489 | } |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1490 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1491 | |
| 1492 | while ((bio = bio_list_pop(&bios))) |
| 1493 | generic_make_request(bio); |
| 1494 | } |
| 1495 | |
| 1496 | static void do_worker(struct work_struct *ws) |
| 1497 | { |
| 1498 | struct pool *pool = container_of(ws, struct pool, worker); |
| 1499 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1500 | process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping); |
| 1501 | process_prepared(pool, &pool->prepared_discards, process_prepared_discard); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1502 | process_deferred_bios(pool); |
| 1503 | } |
| 1504 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1505 | /* |
| 1506 | * We want to commit periodically so that not too much |
| 1507 | * unwritten data builds up. |
| 1508 | */ |
| 1509 | static void do_waker(struct work_struct *ws) |
| 1510 | { |
| 1511 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); |
| 1512 | wake_worker(pool); |
| 1513 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); |
| 1514 | } |
| 1515 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1516 | /*----------------------------------------------------------------*/ |
| 1517 | |
| 1518 | /* |
| 1519 | * Mapping functions. |
| 1520 | */ |
| 1521 | |
| 1522 | /* |
| 1523 | * Called only while mapping a thin bio to hand it over to the workqueue. |
| 1524 | */ |
| 1525 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) |
| 1526 | { |
| 1527 | unsigned long flags; |
| 1528 | struct pool *pool = tc->pool; |
| 1529 | |
| 1530 | spin_lock_irqsave(&pool->lock, flags); |
| 1531 | bio_list_add(&pool->deferred_bios, bio); |
| 1532 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1533 | |
| 1534 | wake_worker(pool); |
| 1535 | } |
| 1536 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1537 | static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio) |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1538 | { |
| 1539 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1540 | struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1541 | |
| 1542 | h->tc = tc; |
| 1543 | h->shared_read_entry = NULL; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1544 | h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1545 | h->overwrite_mapping = NULL; |
| 1546 | |
| 1547 | return h; |
| 1548 | } |
| 1549 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1550 | /* |
| 1551 | * Non-blocking function called from the thin target's map function. |
| 1552 | */ |
| 1553 | static int thin_bio_map(struct dm_target *ti, struct bio *bio, |
| 1554 | union map_info *map_context) |
| 1555 | { |
| 1556 | int r; |
| 1557 | struct thin_c *tc = ti->private; |
| 1558 | dm_block_t block = get_bio_block(tc, bio); |
| 1559 | struct dm_thin_device *td = tc->td; |
| 1560 | struct dm_thin_lookup_result result; |
| 1561 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1562 | map_context->ptr = thin_hook_bio(tc, bio); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1563 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1564 | thin_defer_bio(tc, bio); |
| 1565 | return DM_MAPIO_SUBMITTED; |
| 1566 | } |
| 1567 | |
| 1568 | r = dm_thin_find_block(td, block, 0, &result); |
| 1569 | |
| 1570 | /* |
| 1571 | * Note that we defer readahead too. |
| 1572 | */ |
| 1573 | switch (r) { |
| 1574 | case 0: |
| 1575 | if (unlikely(result.shared)) { |
| 1576 | /* |
| 1577 | * We have a race condition here between the |
| 1578 | * result.shared value returned by the lookup and |
| 1579 | * snapshot creation, which may cause new |
| 1580 | * sharing. |
| 1581 | * |
| 1582 | * To avoid this always quiesce the origin before |
| 1583 | * taking the snap. You want to do this anyway to |
| 1584 | * ensure a consistent application view |
| 1585 | * (i.e. lockfs). |
| 1586 | * |
| 1587 | * More distant ancestors are irrelevant. The |
| 1588 | * shared flag will be set in their case. |
| 1589 | */ |
| 1590 | thin_defer_bio(tc, bio); |
| 1591 | r = DM_MAPIO_SUBMITTED; |
| 1592 | } else { |
| 1593 | remap(tc, bio, result.block); |
| 1594 | r = DM_MAPIO_REMAPPED; |
| 1595 | } |
| 1596 | break; |
| 1597 | |
| 1598 | case -ENODATA: |
| 1599 | /* |
| 1600 | * In future, the failed dm_thin_find_block above could |
| 1601 | * provide the hint to load the metadata into cache. |
| 1602 | */ |
| 1603 | case -EWOULDBLOCK: |
| 1604 | thin_defer_bio(tc, bio); |
| 1605 | r = DM_MAPIO_SUBMITTED; |
| 1606 | break; |
| 1607 | } |
| 1608 | |
| 1609 | return r; |
| 1610 | } |
| 1611 | |
| 1612 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) |
| 1613 | { |
| 1614 | int r; |
| 1615 | unsigned long flags; |
| 1616 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
| 1617 | |
| 1618 | spin_lock_irqsave(&pt->pool->lock, flags); |
| 1619 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); |
| 1620 | spin_unlock_irqrestore(&pt->pool->lock, flags); |
| 1621 | |
| 1622 | if (!r) { |
| 1623 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 1624 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
| 1625 | } |
| 1626 | |
| 1627 | return r; |
| 1628 | } |
| 1629 | |
| 1630 | static void __requeue_bios(struct pool *pool) |
| 1631 | { |
| 1632 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); |
| 1633 | bio_list_init(&pool->retry_on_resume_list); |
| 1634 | } |
| 1635 | |
| 1636 | /*---------------------------------------------------------------- |
| 1637 | * Binding of control targets to a pool object |
| 1638 | *--------------------------------------------------------------*/ |
| 1639 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
| 1640 | { |
| 1641 | struct pool_c *pt = ti->private; |
| 1642 | |
| 1643 | pool->ti = ti; |
| 1644 | pool->low_water_blocks = pt->low_water_blocks; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1645 | pool->pf = pt->pf; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1646 | |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 1647 | /* |
| 1648 | * If discard_passdown was enabled verify that the data device |
| 1649 | * supports discards. Disable discard_passdown if not; otherwise |
| 1650 | * -EOPNOTSUPP will be returned. |
| 1651 | */ |
| 1652 | if (pt->pf.discard_passdown) { |
| 1653 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 1654 | if (!q || !blk_queue_discard(q)) { |
| 1655 | char buf[BDEVNAME_SIZE]; |
| 1656 | DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.", |
| 1657 | bdevname(pt->data_dev->bdev, buf)); |
| 1658 | pool->pf.discard_passdown = 0; |
| 1659 | } |
| 1660 | } |
| 1661 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1662 | return 0; |
| 1663 | } |
| 1664 | |
| 1665 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) |
| 1666 | { |
| 1667 | if (pool->ti == ti) |
| 1668 | pool->ti = NULL; |
| 1669 | } |
| 1670 | |
| 1671 | /*---------------------------------------------------------------- |
| 1672 | * Pool creation |
| 1673 | *--------------------------------------------------------------*/ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1674 | /* Initialize pool features. */ |
| 1675 | static void pool_features_init(struct pool_features *pf) |
| 1676 | { |
| 1677 | pf->zero_new_blocks = 1; |
| 1678 | pf->discard_enabled = 1; |
| 1679 | pf->discard_passdown = 1; |
| 1680 | } |
| 1681 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1682 | static void __pool_destroy(struct pool *pool) |
| 1683 | { |
| 1684 | __pool_table_remove(pool); |
| 1685 | |
| 1686 | if (dm_pool_metadata_close(pool->pmd) < 0) |
| 1687 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 1688 | |
| 1689 | prison_destroy(pool->prison); |
| 1690 | dm_kcopyd_client_destroy(pool->copier); |
| 1691 | |
| 1692 | if (pool->wq) |
| 1693 | destroy_workqueue(pool->wq); |
| 1694 | |
| 1695 | if (pool->next_mapping) |
| 1696 | mempool_free(pool->next_mapping, pool->mapping_pool); |
| 1697 | mempool_destroy(pool->mapping_pool); |
| 1698 | mempool_destroy(pool->endio_hook_pool); |
| 1699 | kfree(pool); |
| 1700 | } |
| 1701 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1702 | static struct kmem_cache *_new_mapping_cache; |
| 1703 | static struct kmem_cache *_endio_hook_cache; |
| 1704 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1705 | static struct pool *pool_create(struct mapped_device *pool_md, |
| 1706 | struct block_device *metadata_dev, |
| 1707 | unsigned long block_size, char **error) |
| 1708 | { |
| 1709 | int r; |
| 1710 | void *err_p; |
| 1711 | struct pool *pool; |
| 1712 | struct dm_pool_metadata *pmd; |
| 1713 | |
| 1714 | pmd = dm_pool_metadata_open(metadata_dev, block_size); |
| 1715 | if (IS_ERR(pmd)) { |
| 1716 | *error = "Error creating metadata object"; |
| 1717 | return (struct pool *)pmd; |
| 1718 | } |
| 1719 | |
| 1720 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); |
| 1721 | if (!pool) { |
| 1722 | *error = "Error allocating memory for pool"; |
| 1723 | err_p = ERR_PTR(-ENOMEM); |
| 1724 | goto bad_pool; |
| 1725 | } |
| 1726 | |
| 1727 | pool->pmd = pmd; |
| 1728 | pool->sectors_per_block = block_size; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1729 | if (block_size & (block_size - 1)) |
| 1730 | pool->sectors_per_block_shift = -1; |
| 1731 | else |
| 1732 | pool->sectors_per_block_shift = __ffs(block_size); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1733 | pool->low_water_blocks = 0; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1734 | pool_features_init(&pool->pf); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1735 | pool->prison = prison_create(PRISON_CELLS); |
| 1736 | if (!pool->prison) { |
| 1737 | *error = "Error creating pool's bio prison"; |
| 1738 | err_p = ERR_PTR(-ENOMEM); |
| 1739 | goto bad_prison; |
| 1740 | } |
| 1741 | |
| 1742 | pool->copier = dm_kcopyd_client_create(); |
| 1743 | if (IS_ERR(pool->copier)) { |
| 1744 | r = PTR_ERR(pool->copier); |
| 1745 | *error = "Error creating pool's kcopyd client"; |
| 1746 | err_p = ERR_PTR(r); |
| 1747 | goto bad_kcopyd_client; |
| 1748 | } |
| 1749 | |
| 1750 | /* |
| 1751 | * Create singlethreaded workqueue that will service all devices |
| 1752 | * that use this metadata. |
| 1753 | */ |
| 1754 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); |
| 1755 | if (!pool->wq) { |
| 1756 | *error = "Error creating pool's workqueue"; |
| 1757 | err_p = ERR_PTR(-ENOMEM); |
| 1758 | goto bad_wq; |
| 1759 | } |
| 1760 | |
| 1761 | INIT_WORK(&pool->worker, do_worker); |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1762 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1763 | spin_lock_init(&pool->lock); |
| 1764 | bio_list_init(&pool->deferred_bios); |
| 1765 | bio_list_init(&pool->deferred_flush_bios); |
| 1766 | INIT_LIST_HEAD(&pool->prepared_mappings); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1767 | INIT_LIST_HEAD(&pool->prepared_discards); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1768 | pool->low_water_triggered = 0; |
| 1769 | pool->no_free_space = 0; |
| 1770 | bio_list_init(&pool->retry_on_resume_list); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1771 | ds_init(&pool->shared_read_ds); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1772 | ds_init(&pool->all_io_ds); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1773 | |
| 1774 | pool->next_mapping = NULL; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1775 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
| 1776 | _new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1777 | if (!pool->mapping_pool) { |
| 1778 | *error = "Error creating pool's mapping mempool"; |
| 1779 | err_p = ERR_PTR(-ENOMEM); |
| 1780 | goto bad_mapping_pool; |
| 1781 | } |
| 1782 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1783 | pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE, |
| 1784 | _endio_hook_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1785 | if (!pool->endio_hook_pool) { |
| 1786 | *error = "Error creating pool's endio_hook mempool"; |
| 1787 | err_p = ERR_PTR(-ENOMEM); |
| 1788 | goto bad_endio_hook_pool; |
| 1789 | } |
| 1790 | pool->ref_count = 1; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1791 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1792 | pool->pool_md = pool_md; |
| 1793 | pool->md_dev = metadata_dev; |
| 1794 | __pool_table_insert(pool); |
| 1795 | |
| 1796 | return pool; |
| 1797 | |
| 1798 | bad_endio_hook_pool: |
| 1799 | mempool_destroy(pool->mapping_pool); |
| 1800 | bad_mapping_pool: |
| 1801 | destroy_workqueue(pool->wq); |
| 1802 | bad_wq: |
| 1803 | dm_kcopyd_client_destroy(pool->copier); |
| 1804 | bad_kcopyd_client: |
| 1805 | prison_destroy(pool->prison); |
| 1806 | bad_prison: |
| 1807 | kfree(pool); |
| 1808 | bad_pool: |
| 1809 | if (dm_pool_metadata_close(pmd)) |
| 1810 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 1811 | |
| 1812 | return err_p; |
| 1813 | } |
| 1814 | |
| 1815 | static void __pool_inc(struct pool *pool) |
| 1816 | { |
| 1817 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 1818 | pool->ref_count++; |
| 1819 | } |
| 1820 | |
| 1821 | static void __pool_dec(struct pool *pool) |
| 1822 | { |
| 1823 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 1824 | BUG_ON(!pool->ref_count); |
| 1825 | if (!--pool->ref_count) |
| 1826 | __pool_destroy(pool); |
| 1827 | } |
| 1828 | |
| 1829 | static struct pool *__pool_find(struct mapped_device *pool_md, |
| 1830 | struct block_device *metadata_dev, |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1831 | unsigned long block_size, char **error, |
| 1832 | int *created) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1833 | { |
| 1834 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); |
| 1835 | |
| 1836 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 1837 | if (pool->pool_md != pool_md) { |
| 1838 | *error = "metadata device already in use by a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1839 | return ERR_PTR(-EBUSY); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 1840 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1841 | __pool_inc(pool); |
| 1842 | |
| 1843 | } else { |
| 1844 | pool = __pool_table_lookup(pool_md); |
| 1845 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 1846 | if (pool->md_dev != metadata_dev) { |
| 1847 | *error = "different pool cannot replace a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1848 | return ERR_PTR(-EINVAL); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 1849 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1850 | __pool_inc(pool); |
| 1851 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1852 | } else { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1853 | pool = pool_create(pool_md, metadata_dev, block_size, error); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1854 | *created = 1; |
| 1855 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1856 | } |
| 1857 | |
| 1858 | return pool; |
| 1859 | } |
| 1860 | |
| 1861 | /*---------------------------------------------------------------- |
| 1862 | * Pool target methods |
| 1863 | *--------------------------------------------------------------*/ |
| 1864 | static void pool_dtr(struct dm_target *ti) |
| 1865 | { |
| 1866 | struct pool_c *pt = ti->private; |
| 1867 | |
| 1868 | mutex_lock(&dm_thin_pool_table.mutex); |
| 1869 | |
| 1870 | unbind_control_target(pt->pool, ti); |
| 1871 | __pool_dec(pt->pool); |
| 1872 | dm_put_device(ti, pt->metadata_dev); |
| 1873 | dm_put_device(ti, pt->data_dev); |
| 1874 | kfree(pt); |
| 1875 | |
| 1876 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 1877 | } |
| 1878 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1879 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
| 1880 | struct dm_target *ti) |
| 1881 | { |
| 1882 | int r; |
| 1883 | unsigned argc; |
| 1884 | const char *arg_name; |
| 1885 | |
| 1886 | static struct dm_arg _args[] = { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1887 | {0, 3, "Invalid number of pool feature arguments"}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1888 | }; |
| 1889 | |
| 1890 | /* |
| 1891 | * No feature arguments supplied. |
| 1892 | */ |
| 1893 | if (!as->argc) |
| 1894 | return 0; |
| 1895 | |
| 1896 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
| 1897 | if (r) |
| 1898 | return -EINVAL; |
| 1899 | |
| 1900 | while (argc && !r) { |
| 1901 | arg_name = dm_shift_arg(as); |
| 1902 | argc--; |
| 1903 | |
| 1904 | if (!strcasecmp(arg_name, "skip_block_zeroing")) { |
| 1905 | pf->zero_new_blocks = 0; |
| 1906 | continue; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1907 | } else if (!strcasecmp(arg_name, "ignore_discard")) { |
| 1908 | pf->discard_enabled = 0; |
| 1909 | continue; |
| 1910 | } else if (!strcasecmp(arg_name, "no_discard_passdown")) { |
| 1911 | pf->discard_passdown = 0; |
| 1912 | continue; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1913 | } |
| 1914 | |
| 1915 | ti->error = "Unrecognised pool feature requested"; |
| 1916 | r = -EINVAL; |
| 1917 | } |
| 1918 | |
| 1919 | return r; |
| 1920 | } |
| 1921 | |
| 1922 | /* |
| 1923 | * thin-pool <metadata dev> <data dev> |
| 1924 | * <data block size (sectors)> |
| 1925 | * <low water mark (blocks)> |
| 1926 | * [<#feature args> [<arg>]*] |
| 1927 | * |
| 1928 | * Optional feature arguments are: |
| 1929 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1930 | * ignore_discard: disable discard |
| 1931 | * no_discard_passdown: don't pass discards down to the data device |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1932 | */ |
| 1933 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 1934 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1935 | int r, pool_created = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1936 | struct pool_c *pt; |
| 1937 | struct pool *pool; |
| 1938 | struct pool_features pf; |
| 1939 | struct dm_arg_set as; |
| 1940 | struct dm_dev *data_dev; |
| 1941 | unsigned long block_size; |
| 1942 | dm_block_t low_water_blocks; |
| 1943 | struct dm_dev *metadata_dev; |
| 1944 | sector_t metadata_dev_size; |
Mike Snitzer | c4a69ec | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1945 | char b[BDEVNAME_SIZE]; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1946 | |
| 1947 | /* |
| 1948 | * FIXME Remove validation from scope of lock. |
| 1949 | */ |
| 1950 | mutex_lock(&dm_thin_pool_table.mutex); |
| 1951 | |
| 1952 | if (argc < 4) { |
| 1953 | ti->error = "Invalid argument count"; |
| 1954 | r = -EINVAL; |
| 1955 | goto out_unlock; |
| 1956 | } |
| 1957 | as.argc = argc; |
| 1958 | as.argv = argv; |
| 1959 | |
| 1960 | r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev); |
| 1961 | if (r) { |
| 1962 | ti->error = "Error opening metadata block device"; |
| 1963 | goto out_unlock; |
| 1964 | } |
| 1965 | |
| 1966 | metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT; |
Mike Snitzer | c4a69ec | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1967 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
| 1968 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
| 1969 | bdevname(metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1970 | |
| 1971 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); |
| 1972 | if (r) { |
| 1973 | ti->error = "Error getting data device"; |
| 1974 | goto out_metadata; |
| 1975 | } |
| 1976 | |
| 1977 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || |
| 1978 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || |
| 1979 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 1980 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1981 | ti->error = "Invalid block size"; |
| 1982 | r = -EINVAL; |
| 1983 | goto out; |
| 1984 | } |
| 1985 | |
| 1986 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { |
| 1987 | ti->error = "Invalid low water mark"; |
| 1988 | r = -EINVAL; |
| 1989 | goto out; |
| 1990 | } |
| 1991 | |
| 1992 | /* |
| 1993 | * Set default pool features. |
| 1994 | */ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1995 | pool_features_init(&pf); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1996 | |
| 1997 | dm_consume_args(&as, 4); |
| 1998 | r = parse_pool_features(&as, &pf, ti); |
| 1999 | if (r) |
| 2000 | goto out; |
| 2001 | |
| 2002 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
| 2003 | if (!pt) { |
| 2004 | r = -ENOMEM; |
| 2005 | goto out; |
| 2006 | } |
| 2007 | |
| 2008 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2009 | block_size, &ti->error, &pool_created); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2010 | if (IS_ERR(pool)) { |
| 2011 | r = PTR_ERR(pool); |
| 2012 | goto out_free_pt; |
| 2013 | } |
| 2014 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2015 | /* |
| 2016 | * 'pool_created' reflects whether this is the first table load. |
| 2017 | * Top level discard support is not allowed to be changed after |
| 2018 | * initial load. This would require a pool reload to trigger thin |
| 2019 | * device changes. |
| 2020 | */ |
| 2021 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { |
| 2022 | ti->error = "Discard support cannot be disabled once enabled"; |
| 2023 | r = -EINVAL; |
| 2024 | goto out_flags_changed; |
| 2025 | } |
| 2026 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2027 | /* |
| 2028 | * The block layer requires discard_granularity to be a power of 2. |
| 2029 | */ |
| 2030 | if (pf.discard_enabled && !is_power_of_2(block_size)) { |
| 2031 | ti->error = "Discard support must be disabled when the block size is not a power of 2"; |
| 2032 | r = -EINVAL; |
| 2033 | goto out_flags_changed; |
| 2034 | } |
| 2035 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2036 | pt->pool = pool; |
| 2037 | pt->ti = ti; |
| 2038 | pt->metadata_dev = metadata_dev; |
| 2039 | pt->data_dev = data_dev; |
| 2040 | pt->low_water_blocks = low_water_blocks; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2041 | pt->pf = pf; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2042 | ti->num_flush_requests = 1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2043 | /* |
| 2044 | * Only need to enable discards if the pool should pass |
| 2045 | * them down to the data device. The thin device's discard |
| 2046 | * processing will cause mappings to be removed from the btree. |
| 2047 | */ |
| 2048 | if (pf.discard_enabled && pf.discard_passdown) { |
| 2049 | ti->num_discard_requests = 1; |
| 2050 | /* |
| 2051 | * Setting 'discards_supported' circumvents the normal |
| 2052 | * stacking of discard limits (this keeps the pool and |
| 2053 | * thin devices' discard limits consistent). |
| 2054 | */ |
| 2055 | ti->discards_supported = 1; |
| 2056 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2057 | ti->private = pt; |
| 2058 | |
| 2059 | pt->callbacks.congested_fn = pool_is_congested; |
| 2060 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); |
| 2061 | |
| 2062 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2063 | |
| 2064 | return 0; |
| 2065 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2066 | out_flags_changed: |
| 2067 | __pool_dec(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2068 | out_free_pt: |
| 2069 | kfree(pt); |
| 2070 | out: |
| 2071 | dm_put_device(ti, data_dev); |
| 2072 | out_metadata: |
| 2073 | dm_put_device(ti, metadata_dev); |
| 2074 | out_unlock: |
| 2075 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2076 | |
| 2077 | return r; |
| 2078 | } |
| 2079 | |
| 2080 | static int pool_map(struct dm_target *ti, struct bio *bio, |
| 2081 | union map_info *map_context) |
| 2082 | { |
| 2083 | int r; |
| 2084 | struct pool_c *pt = ti->private; |
| 2085 | struct pool *pool = pt->pool; |
| 2086 | unsigned long flags; |
| 2087 | |
| 2088 | /* |
| 2089 | * As this is a singleton target, ti->begin is always zero. |
| 2090 | */ |
| 2091 | spin_lock_irqsave(&pool->lock, flags); |
| 2092 | bio->bi_bdev = pt->data_dev->bdev; |
| 2093 | r = DM_MAPIO_REMAPPED; |
| 2094 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2095 | |
| 2096 | return r; |
| 2097 | } |
| 2098 | |
| 2099 | /* |
| 2100 | * Retrieves the number of blocks of the data device from |
| 2101 | * the superblock and compares it to the actual device size, |
| 2102 | * thus resizing the data device in case it has grown. |
| 2103 | * |
| 2104 | * This both copes with opening preallocated data devices in the ctr |
| 2105 | * being followed by a resume |
| 2106 | * -and- |
| 2107 | * calling the resume method individually after userspace has |
| 2108 | * grown the data device in reaction to a table event. |
| 2109 | */ |
| 2110 | static int pool_preresume(struct dm_target *ti) |
| 2111 | { |
| 2112 | int r; |
| 2113 | struct pool_c *pt = ti->private; |
| 2114 | struct pool *pool = pt->pool; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2115 | sector_t data_size = ti->len; |
| 2116 | dm_block_t sb_data_size; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2117 | |
| 2118 | /* |
| 2119 | * Take control of the pool object. |
| 2120 | */ |
| 2121 | r = bind_control_target(pool, ti); |
| 2122 | if (r) |
| 2123 | return r; |
| 2124 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2125 | (void) sector_div(data_size, pool->sectors_per_block); |
| 2126 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2127 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
| 2128 | if (r) { |
| 2129 | DMERR("failed to retrieve data device size"); |
| 2130 | return r; |
| 2131 | } |
| 2132 | |
| 2133 | if (data_size < sb_data_size) { |
| 2134 | DMERR("pool target too small, is %llu blocks (expected %llu)", |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2135 | (unsigned long long)data_size, sb_data_size); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2136 | return -EINVAL; |
| 2137 | |
| 2138 | } else if (data_size > sb_data_size) { |
| 2139 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
| 2140 | if (r) { |
| 2141 | DMERR("failed to resize data device"); |
| 2142 | return r; |
| 2143 | } |
| 2144 | |
| 2145 | r = dm_pool_commit_metadata(pool->pmd); |
| 2146 | if (r) { |
| 2147 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", |
| 2148 | __func__, r); |
| 2149 | return r; |
| 2150 | } |
| 2151 | } |
| 2152 | |
| 2153 | return 0; |
| 2154 | } |
| 2155 | |
| 2156 | static void pool_resume(struct dm_target *ti) |
| 2157 | { |
| 2158 | struct pool_c *pt = ti->private; |
| 2159 | struct pool *pool = pt->pool; |
| 2160 | unsigned long flags; |
| 2161 | |
| 2162 | spin_lock_irqsave(&pool->lock, flags); |
| 2163 | pool->low_water_triggered = 0; |
| 2164 | pool->no_free_space = 0; |
| 2165 | __requeue_bios(pool); |
| 2166 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2167 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2168 | do_waker(&pool->waker.work); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2169 | } |
| 2170 | |
| 2171 | static void pool_postsuspend(struct dm_target *ti) |
| 2172 | { |
| 2173 | int r; |
| 2174 | struct pool_c *pt = ti->private; |
| 2175 | struct pool *pool = pt->pool; |
| 2176 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2177 | cancel_delayed_work(&pool->waker); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2178 | flush_workqueue(pool->wq); |
| 2179 | |
| 2180 | r = dm_pool_commit_metadata(pool->pmd); |
| 2181 | if (r < 0) { |
| 2182 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", |
| 2183 | __func__, r); |
| 2184 | /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/ |
| 2185 | } |
| 2186 | } |
| 2187 | |
| 2188 | static int check_arg_count(unsigned argc, unsigned args_required) |
| 2189 | { |
| 2190 | if (argc != args_required) { |
| 2191 | DMWARN("Message received with %u arguments instead of %u.", |
| 2192 | argc, args_required); |
| 2193 | return -EINVAL; |
| 2194 | } |
| 2195 | |
| 2196 | return 0; |
| 2197 | } |
| 2198 | |
| 2199 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) |
| 2200 | { |
| 2201 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && |
| 2202 | *dev_id <= MAX_DEV_ID) |
| 2203 | return 0; |
| 2204 | |
| 2205 | if (warning) |
| 2206 | DMWARN("Message received with invalid device id: %s", arg); |
| 2207 | |
| 2208 | return -EINVAL; |
| 2209 | } |
| 2210 | |
| 2211 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2212 | { |
| 2213 | dm_thin_id dev_id; |
| 2214 | int r; |
| 2215 | |
| 2216 | r = check_arg_count(argc, 2); |
| 2217 | if (r) |
| 2218 | return r; |
| 2219 | |
| 2220 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2221 | if (r) |
| 2222 | return r; |
| 2223 | |
| 2224 | r = dm_pool_create_thin(pool->pmd, dev_id); |
| 2225 | if (r) { |
| 2226 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", |
| 2227 | argv[1]); |
| 2228 | return r; |
| 2229 | } |
| 2230 | |
| 2231 | return 0; |
| 2232 | } |
| 2233 | |
| 2234 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2235 | { |
| 2236 | dm_thin_id dev_id; |
| 2237 | dm_thin_id origin_dev_id; |
| 2238 | int r; |
| 2239 | |
| 2240 | r = check_arg_count(argc, 3); |
| 2241 | if (r) |
| 2242 | return r; |
| 2243 | |
| 2244 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2245 | if (r) |
| 2246 | return r; |
| 2247 | |
| 2248 | r = read_dev_id(argv[2], &origin_dev_id, 1); |
| 2249 | if (r) |
| 2250 | return r; |
| 2251 | |
| 2252 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); |
| 2253 | if (r) { |
| 2254 | DMWARN("Creation of new snapshot %s of device %s failed.", |
| 2255 | argv[1], argv[2]); |
| 2256 | return r; |
| 2257 | } |
| 2258 | |
| 2259 | return 0; |
| 2260 | } |
| 2261 | |
| 2262 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2263 | { |
| 2264 | dm_thin_id dev_id; |
| 2265 | int r; |
| 2266 | |
| 2267 | r = check_arg_count(argc, 2); |
| 2268 | if (r) |
| 2269 | return r; |
| 2270 | |
| 2271 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2272 | if (r) |
| 2273 | return r; |
| 2274 | |
| 2275 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); |
| 2276 | if (r) |
| 2277 | DMWARN("Deletion of thin device %s failed.", argv[1]); |
| 2278 | |
| 2279 | return r; |
| 2280 | } |
| 2281 | |
| 2282 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2283 | { |
| 2284 | dm_thin_id old_id, new_id; |
| 2285 | int r; |
| 2286 | |
| 2287 | r = check_arg_count(argc, 3); |
| 2288 | if (r) |
| 2289 | return r; |
| 2290 | |
| 2291 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { |
| 2292 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); |
| 2293 | return -EINVAL; |
| 2294 | } |
| 2295 | |
| 2296 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { |
| 2297 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); |
| 2298 | return -EINVAL; |
| 2299 | } |
| 2300 | |
| 2301 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); |
| 2302 | if (r) { |
| 2303 | DMWARN("Failed to change transaction id from %s to %s.", |
| 2304 | argv[1], argv[2]); |
| 2305 | return r; |
| 2306 | } |
| 2307 | |
| 2308 | return 0; |
| 2309 | } |
| 2310 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2311 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2312 | { |
| 2313 | int r; |
| 2314 | |
| 2315 | r = check_arg_count(argc, 1); |
| 2316 | if (r) |
| 2317 | return r; |
| 2318 | |
Joe Thornber | 0d200ae | 2012-07-03 12:55:31 +0100 | [diff] [blame] | 2319 | r = dm_pool_commit_metadata(pool->pmd); |
| 2320 | if (r) { |
| 2321 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", |
| 2322 | __func__, r); |
| 2323 | return r; |
| 2324 | } |
| 2325 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2326 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
| 2327 | if (r) |
| 2328 | DMWARN("reserve_metadata_snap message failed."); |
| 2329 | |
| 2330 | return r; |
| 2331 | } |
| 2332 | |
| 2333 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2334 | { |
| 2335 | int r; |
| 2336 | |
| 2337 | r = check_arg_count(argc, 1); |
| 2338 | if (r) |
| 2339 | return r; |
| 2340 | |
| 2341 | r = dm_pool_release_metadata_snap(pool->pmd); |
| 2342 | if (r) |
| 2343 | DMWARN("release_metadata_snap message failed."); |
| 2344 | |
| 2345 | return r; |
| 2346 | } |
| 2347 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2348 | /* |
| 2349 | * Messages supported: |
| 2350 | * create_thin <dev_id> |
| 2351 | * create_snap <dev_id> <origin_id> |
| 2352 | * delete <dev_id> |
| 2353 | * trim <dev_id> <new_size_in_sectors> |
| 2354 | * set_transaction_id <current_trans_id> <new_trans_id> |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2355 | * reserve_metadata_snap |
| 2356 | * release_metadata_snap |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2357 | */ |
| 2358 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) |
| 2359 | { |
| 2360 | int r = -EINVAL; |
| 2361 | struct pool_c *pt = ti->private; |
| 2362 | struct pool *pool = pt->pool; |
| 2363 | |
| 2364 | if (!strcasecmp(argv[0], "create_thin")) |
| 2365 | r = process_create_thin_mesg(argc, argv, pool); |
| 2366 | |
| 2367 | else if (!strcasecmp(argv[0], "create_snap")) |
| 2368 | r = process_create_snap_mesg(argc, argv, pool); |
| 2369 | |
| 2370 | else if (!strcasecmp(argv[0], "delete")) |
| 2371 | r = process_delete_mesg(argc, argv, pool); |
| 2372 | |
| 2373 | else if (!strcasecmp(argv[0], "set_transaction_id")) |
| 2374 | r = process_set_transaction_id_mesg(argc, argv, pool); |
| 2375 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2376 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
| 2377 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); |
| 2378 | |
| 2379 | else if (!strcasecmp(argv[0], "release_metadata_snap")) |
| 2380 | r = process_release_metadata_snap_mesg(argc, argv, pool); |
| 2381 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2382 | else |
| 2383 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); |
| 2384 | |
| 2385 | if (!r) { |
| 2386 | r = dm_pool_commit_metadata(pool->pmd); |
| 2387 | if (r) |
| 2388 | DMERR("%s message: dm_pool_commit_metadata() failed, error = %d", |
| 2389 | argv[0], r); |
| 2390 | } |
| 2391 | |
| 2392 | return r; |
| 2393 | } |
| 2394 | |
| 2395 | /* |
| 2396 | * Status line is: |
| 2397 | * <transaction id> <used metadata sectors>/<total metadata sectors> |
| 2398 | * <used data sectors>/<total data sectors> <held metadata root> |
| 2399 | */ |
| 2400 | static int pool_status(struct dm_target *ti, status_type_t type, |
| 2401 | char *result, unsigned maxlen) |
| 2402 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2403 | int r, count; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2404 | unsigned sz = 0; |
| 2405 | uint64_t transaction_id; |
| 2406 | dm_block_t nr_free_blocks_data; |
| 2407 | dm_block_t nr_free_blocks_metadata; |
| 2408 | dm_block_t nr_blocks_data; |
| 2409 | dm_block_t nr_blocks_metadata; |
| 2410 | dm_block_t held_root; |
| 2411 | char buf[BDEVNAME_SIZE]; |
| 2412 | char buf2[BDEVNAME_SIZE]; |
| 2413 | struct pool_c *pt = ti->private; |
| 2414 | struct pool *pool = pt->pool; |
| 2415 | |
| 2416 | switch (type) { |
| 2417 | case STATUSTYPE_INFO: |
| 2418 | r = dm_pool_get_metadata_transaction_id(pool->pmd, |
| 2419 | &transaction_id); |
| 2420 | if (r) |
| 2421 | return r; |
| 2422 | |
| 2423 | r = dm_pool_get_free_metadata_block_count(pool->pmd, |
| 2424 | &nr_free_blocks_metadata); |
| 2425 | if (r) |
| 2426 | return r; |
| 2427 | |
| 2428 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); |
| 2429 | if (r) |
| 2430 | return r; |
| 2431 | |
| 2432 | r = dm_pool_get_free_block_count(pool->pmd, |
| 2433 | &nr_free_blocks_data); |
| 2434 | if (r) |
| 2435 | return r; |
| 2436 | |
| 2437 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); |
| 2438 | if (r) |
| 2439 | return r; |
| 2440 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2441 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2442 | if (r) |
| 2443 | return r; |
| 2444 | |
| 2445 | DMEMIT("%llu %llu/%llu %llu/%llu ", |
| 2446 | (unsigned long long)transaction_id, |
| 2447 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
| 2448 | (unsigned long long)nr_blocks_metadata, |
| 2449 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), |
| 2450 | (unsigned long long)nr_blocks_data); |
| 2451 | |
| 2452 | if (held_root) |
| 2453 | DMEMIT("%llu", held_root); |
| 2454 | else |
| 2455 | DMEMIT("-"); |
| 2456 | |
| 2457 | break; |
| 2458 | |
| 2459 | case STATUSTYPE_TABLE: |
| 2460 | DMEMIT("%s %s %lu %llu ", |
| 2461 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), |
| 2462 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), |
| 2463 | (unsigned long)pool->sectors_per_block, |
| 2464 | (unsigned long long)pt->low_water_blocks); |
| 2465 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2466 | count = !pool->pf.zero_new_blocks + !pool->pf.discard_enabled + |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 2467 | !pt->pf.discard_passdown; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2468 | DMEMIT("%u ", count); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2469 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2470 | if (!pool->pf.zero_new_blocks) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2471 | DMEMIT("skip_block_zeroing "); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2472 | |
| 2473 | if (!pool->pf.discard_enabled) |
| 2474 | DMEMIT("ignore_discard "); |
| 2475 | |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 2476 | if (!pt->pf.discard_passdown) |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2477 | DMEMIT("no_discard_passdown "); |
| 2478 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2479 | break; |
| 2480 | } |
| 2481 | |
| 2482 | return 0; |
| 2483 | } |
| 2484 | |
| 2485 | static int pool_iterate_devices(struct dm_target *ti, |
| 2486 | iterate_devices_callout_fn fn, void *data) |
| 2487 | { |
| 2488 | struct pool_c *pt = ti->private; |
| 2489 | |
| 2490 | return fn(ti, pt->data_dev, 0, ti->len, data); |
| 2491 | } |
| 2492 | |
| 2493 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
| 2494 | struct bio_vec *biovec, int max_size) |
| 2495 | { |
| 2496 | struct pool_c *pt = ti->private; |
| 2497 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 2498 | |
| 2499 | if (!q->merge_bvec_fn) |
| 2500 | return max_size; |
| 2501 | |
| 2502 | bvm->bi_bdev = pt->data_dev->bdev; |
| 2503 | |
| 2504 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); |
| 2505 | } |
| 2506 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2507 | static void set_discard_limits(struct pool *pool, struct queue_limits *limits) |
| 2508 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2509 | /* |
| 2510 | * FIXME: these limits may be incompatible with the pool's data device |
| 2511 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2512 | limits->max_discard_sectors = pool->sectors_per_block; |
| 2513 | |
| 2514 | /* |
| 2515 | * This is just a hint, and not enforced. We have to cope with |
Mikulas Patocka | 4929630 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 2516 | * bios that cover a block partially. A discard that spans a block |
| 2517 | * boundary is not sent to this target. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2518 | */ |
| 2519 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2520 | limits->discard_zeroes_data = pool->pf.zero_new_blocks; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2521 | } |
| 2522 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2523 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 2524 | { |
| 2525 | struct pool_c *pt = ti->private; |
| 2526 | struct pool *pool = pt->pool; |
| 2527 | |
| 2528 | blk_limits_io_min(limits, 0); |
| 2529 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2530 | if (pool->pf.discard_enabled) |
| 2531 | set_discard_limits(pool, limits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2532 | } |
| 2533 | |
| 2534 | static struct target_type pool_target = { |
| 2535 | .name = "thin-pool", |
| 2536 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | |
| 2537 | DM_TARGET_IMMUTABLE, |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2538 | .version = {1, 2, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2539 | .module = THIS_MODULE, |
| 2540 | .ctr = pool_ctr, |
| 2541 | .dtr = pool_dtr, |
| 2542 | .map = pool_map, |
| 2543 | .postsuspend = pool_postsuspend, |
| 2544 | .preresume = pool_preresume, |
| 2545 | .resume = pool_resume, |
| 2546 | .message = pool_message, |
| 2547 | .status = pool_status, |
| 2548 | .merge = pool_merge, |
| 2549 | .iterate_devices = pool_iterate_devices, |
| 2550 | .io_hints = pool_io_hints, |
| 2551 | }; |
| 2552 | |
| 2553 | /*---------------------------------------------------------------- |
| 2554 | * Thin target methods |
| 2555 | *--------------------------------------------------------------*/ |
| 2556 | static void thin_dtr(struct dm_target *ti) |
| 2557 | { |
| 2558 | struct thin_c *tc = ti->private; |
| 2559 | |
| 2560 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2561 | |
| 2562 | __pool_dec(tc->pool); |
| 2563 | dm_pool_close_thin_device(tc->td); |
| 2564 | dm_put_device(ti, tc->pool_dev); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2565 | if (tc->origin_dev) |
| 2566 | dm_put_device(ti, tc->origin_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2567 | kfree(tc); |
| 2568 | |
| 2569 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2570 | } |
| 2571 | |
| 2572 | /* |
| 2573 | * Thin target parameters: |
| 2574 | * |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2575 | * <pool_dev> <dev_id> [origin_dev] |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2576 | * |
| 2577 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) |
| 2578 | * dev_id: the internal device identifier |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2579 | * origin_dev: a device external to the pool that should act as the origin |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2580 | * |
| 2581 | * If the pool device has discards disabled, they get disabled for the thin |
| 2582 | * device as well. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2583 | */ |
| 2584 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 2585 | { |
| 2586 | int r; |
| 2587 | struct thin_c *tc; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2588 | struct dm_dev *pool_dev, *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2589 | struct mapped_device *pool_md; |
| 2590 | |
| 2591 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2592 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2593 | if (argc != 2 && argc != 3) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2594 | ti->error = "Invalid argument count"; |
| 2595 | r = -EINVAL; |
| 2596 | goto out_unlock; |
| 2597 | } |
| 2598 | |
| 2599 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); |
| 2600 | if (!tc) { |
| 2601 | ti->error = "Out of memory"; |
| 2602 | r = -ENOMEM; |
| 2603 | goto out_unlock; |
| 2604 | } |
| 2605 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2606 | if (argc == 3) { |
| 2607 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
| 2608 | if (r) { |
| 2609 | ti->error = "Error opening origin device"; |
| 2610 | goto bad_origin_dev; |
| 2611 | } |
| 2612 | tc->origin_dev = origin_dev; |
| 2613 | } |
| 2614 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2615 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
| 2616 | if (r) { |
| 2617 | ti->error = "Error opening pool device"; |
| 2618 | goto bad_pool_dev; |
| 2619 | } |
| 2620 | tc->pool_dev = pool_dev; |
| 2621 | |
| 2622 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { |
| 2623 | ti->error = "Invalid device id"; |
| 2624 | r = -EINVAL; |
| 2625 | goto bad_common; |
| 2626 | } |
| 2627 | |
| 2628 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); |
| 2629 | if (!pool_md) { |
| 2630 | ti->error = "Couldn't get pool mapped device"; |
| 2631 | r = -EINVAL; |
| 2632 | goto bad_common; |
| 2633 | } |
| 2634 | |
| 2635 | tc->pool = __pool_table_lookup(pool_md); |
| 2636 | if (!tc->pool) { |
| 2637 | ti->error = "Couldn't find pool object"; |
| 2638 | r = -EINVAL; |
| 2639 | goto bad_pool_lookup; |
| 2640 | } |
| 2641 | __pool_inc(tc->pool); |
| 2642 | |
| 2643 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
| 2644 | if (r) { |
| 2645 | ti->error = "Couldn't open thin internal device"; |
| 2646 | goto bad_thin_open; |
| 2647 | } |
| 2648 | |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 2649 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
| 2650 | if (r) |
| 2651 | goto bad_thin_open; |
| 2652 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2653 | ti->num_flush_requests = 1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2654 | |
| 2655 | /* In case the pool supports discards, pass them on. */ |
| 2656 | if (tc->pool->pf.discard_enabled) { |
| 2657 | ti->discards_supported = 1; |
| 2658 | ti->num_discard_requests = 1; |
Mikulas Patocka | 650d2a0 | 2012-07-20 14:25:05 +0100 | [diff] [blame] | 2659 | ti->discard_zeroes_data_unsupported = 1; |
Mikulas Patocka | 4929630 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 2660 | /* Discard requests must be split on a block boundary */ |
| 2661 | ti->split_discard_requests = 1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2662 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2663 | |
| 2664 | dm_put(pool_md); |
| 2665 | |
| 2666 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2667 | |
| 2668 | return 0; |
| 2669 | |
| 2670 | bad_thin_open: |
| 2671 | __pool_dec(tc->pool); |
| 2672 | bad_pool_lookup: |
| 2673 | dm_put(pool_md); |
| 2674 | bad_common: |
| 2675 | dm_put_device(ti, tc->pool_dev); |
| 2676 | bad_pool_dev: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2677 | if (tc->origin_dev) |
| 2678 | dm_put_device(ti, tc->origin_dev); |
| 2679 | bad_origin_dev: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2680 | kfree(tc); |
| 2681 | out_unlock: |
| 2682 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2683 | |
| 2684 | return r; |
| 2685 | } |
| 2686 | |
| 2687 | static int thin_map(struct dm_target *ti, struct bio *bio, |
| 2688 | union map_info *map_context) |
| 2689 | { |
Alasdair G Kergon | 6efd6e8 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2690 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2691 | |
| 2692 | return thin_bio_map(ti, bio, map_context); |
| 2693 | } |
| 2694 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2695 | static int thin_endio(struct dm_target *ti, |
| 2696 | struct bio *bio, int err, |
| 2697 | union map_info *map_context) |
| 2698 | { |
| 2699 | unsigned long flags; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2700 | struct dm_thin_endio_hook *h = map_context->ptr; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2701 | struct list_head work; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2702 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2703 | struct pool *pool = h->tc->pool; |
| 2704 | |
| 2705 | if (h->shared_read_entry) { |
| 2706 | INIT_LIST_HEAD(&work); |
| 2707 | ds_dec(h->shared_read_entry, &work); |
| 2708 | |
| 2709 | spin_lock_irqsave(&pool->lock, flags); |
| 2710 | list_for_each_entry_safe(m, tmp, &work, list) { |
| 2711 | list_del(&m->list); |
| 2712 | m->quiesced = 1; |
| 2713 | __maybe_add_mapping(m); |
| 2714 | } |
| 2715 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2716 | } |
| 2717 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2718 | if (h->all_io_entry) { |
| 2719 | INIT_LIST_HEAD(&work); |
| 2720 | ds_dec(h->all_io_entry, &work); |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 2721 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2722 | list_for_each_entry_safe(m, tmp, &work, list) |
| 2723 | list_add(&m->list, &pool->prepared_discards); |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 2724 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2725 | } |
| 2726 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2727 | mempool_free(h, pool->endio_hook_pool); |
| 2728 | |
| 2729 | return 0; |
| 2730 | } |
| 2731 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2732 | static void thin_postsuspend(struct dm_target *ti) |
| 2733 | { |
| 2734 | if (dm_noflush_suspending(ti)) |
| 2735 | requeue_io((struct thin_c *)ti->private); |
| 2736 | } |
| 2737 | |
| 2738 | /* |
| 2739 | * <nr mapped sectors> <highest mapped sector> |
| 2740 | */ |
| 2741 | static int thin_status(struct dm_target *ti, status_type_t type, |
| 2742 | char *result, unsigned maxlen) |
| 2743 | { |
| 2744 | int r; |
| 2745 | ssize_t sz = 0; |
| 2746 | dm_block_t mapped, highest; |
| 2747 | char buf[BDEVNAME_SIZE]; |
| 2748 | struct thin_c *tc = ti->private; |
| 2749 | |
| 2750 | if (!tc->td) |
| 2751 | DMEMIT("-"); |
| 2752 | else { |
| 2753 | switch (type) { |
| 2754 | case STATUSTYPE_INFO: |
| 2755 | r = dm_thin_get_mapped_count(tc->td, &mapped); |
| 2756 | if (r) |
| 2757 | return r; |
| 2758 | |
| 2759 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); |
| 2760 | if (r < 0) |
| 2761 | return r; |
| 2762 | |
| 2763 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); |
| 2764 | if (r) |
| 2765 | DMEMIT("%llu", ((highest + 1) * |
| 2766 | tc->pool->sectors_per_block) - 1); |
| 2767 | else |
| 2768 | DMEMIT("-"); |
| 2769 | break; |
| 2770 | |
| 2771 | case STATUSTYPE_TABLE: |
| 2772 | DMEMIT("%s %lu", |
| 2773 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), |
| 2774 | (unsigned long) tc->dev_id); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2775 | if (tc->origin_dev) |
| 2776 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2777 | break; |
| 2778 | } |
| 2779 | } |
| 2780 | |
| 2781 | return 0; |
| 2782 | } |
| 2783 | |
| 2784 | static int thin_iterate_devices(struct dm_target *ti, |
| 2785 | iterate_devices_callout_fn fn, void *data) |
| 2786 | { |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2787 | sector_t blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2788 | struct thin_c *tc = ti->private; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2789 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2790 | |
| 2791 | /* |
| 2792 | * We can't call dm_pool_get_data_dev_size() since that blocks. So |
| 2793 | * we follow a more convoluted path through to the pool's target. |
| 2794 | */ |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2795 | if (!pool->ti) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2796 | return 0; /* nothing is bound */ |
| 2797 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2798 | blocks = pool->ti->len; |
| 2799 | (void) sector_div(blocks, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2800 | if (blocks) |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2801 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2802 | |
| 2803 | return 0; |
| 2804 | } |
| 2805 | |
| 2806 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 2807 | { |
| 2808 | struct thin_c *tc = ti->private; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2809 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2810 | |
| 2811 | blk_limits_io_min(limits, 0); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2812 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
| 2813 | set_discard_limits(pool, limits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2814 | } |
| 2815 | |
| 2816 | static struct target_type thin_target = { |
| 2817 | .name = "thin", |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2818 | .version = {1, 2, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2819 | .module = THIS_MODULE, |
| 2820 | .ctr = thin_ctr, |
| 2821 | .dtr = thin_dtr, |
| 2822 | .map = thin_map, |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2823 | .end_io = thin_endio, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2824 | .postsuspend = thin_postsuspend, |
| 2825 | .status = thin_status, |
| 2826 | .iterate_devices = thin_iterate_devices, |
| 2827 | .io_hints = thin_io_hints, |
| 2828 | }; |
| 2829 | |
| 2830 | /*----------------------------------------------------------------*/ |
| 2831 | |
| 2832 | static int __init dm_thin_init(void) |
| 2833 | { |
| 2834 | int r; |
| 2835 | |
| 2836 | pool_table_init(); |
| 2837 | |
| 2838 | r = dm_register_target(&thin_target); |
| 2839 | if (r) |
| 2840 | return r; |
| 2841 | |
| 2842 | r = dm_register_target(&pool_target); |
| 2843 | if (r) |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2844 | goto bad_pool_target; |
| 2845 | |
| 2846 | r = -ENOMEM; |
| 2847 | |
| 2848 | _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0); |
| 2849 | if (!_cell_cache) |
| 2850 | goto bad_cell_cache; |
| 2851 | |
| 2852 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
| 2853 | if (!_new_mapping_cache) |
| 2854 | goto bad_new_mapping_cache; |
| 2855 | |
| 2856 | _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0); |
| 2857 | if (!_endio_hook_cache) |
| 2858 | goto bad_endio_hook_cache; |
| 2859 | |
| 2860 | return 0; |
| 2861 | |
| 2862 | bad_endio_hook_cache: |
| 2863 | kmem_cache_destroy(_new_mapping_cache); |
| 2864 | bad_new_mapping_cache: |
| 2865 | kmem_cache_destroy(_cell_cache); |
| 2866 | bad_cell_cache: |
| 2867 | dm_unregister_target(&pool_target); |
| 2868 | bad_pool_target: |
| 2869 | dm_unregister_target(&thin_target); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2870 | |
| 2871 | return r; |
| 2872 | } |
| 2873 | |
| 2874 | static void dm_thin_exit(void) |
| 2875 | { |
| 2876 | dm_unregister_target(&thin_target); |
| 2877 | dm_unregister_target(&pool_target); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2878 | |
| 2879 | kmem_cache_destroy(_cell_cache); |
| 2880 | kmem_cache_destroy(_new_mapping_cache); |
| 2881 | kmem_cache_destroy(_endio_hook_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2882 | } |
| 2883 | |
| 2884 | module_init(dm_thin_init); |
| 2885 | module_exit(dm_thin_exit); |
| 2886 | |
Alasdair G Kergon | 7cab8bf | 2012-05-12 01:43:19 +0100 | [diff] [blame] | 2887 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2888 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 2889 | MODULE_LICENSE("GPL"); |