Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1 | /* |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2 | * Copyright (C) 2011-2012 Red Hat UK. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3 | * |
| 4 | * This file is released under the GPL. |
| 5 | */ |
| 6 | |
| 7 | #include "dm-thin-metadata.h" |
Mike Snitzer | 4f81a41 | 2012-10-12 21:02:13 +0100 | [diff] [blame] | 8 | #include "dm-bio-prison.h" |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 9 | #include "dm.h" |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 10 | |
| 11 | #include <linux/device-mapper.h> |
| 12 | #include <linux/dm-io.h> |
| 13 | #include <linux/dm-kcopyd.h> |
| 14 | #include <linux/list.h> |
| 15 | #include <linux/init.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/slab.h> |
| 18 | |
| 19 | #define DM_MSG_PREFIX "thin" |
| 20 | |
| 21 | /* |
| 22 | * Tunable constants |
| 23 | */ |
Alasdair G Kergon | 7768ed3 | 2012-07-27 15:07:57 +0100 | [diff] [blame] | 24 | #define ENDIO_HOOK_POOL_SIZE 1024 |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 25 | #define MAPPING_POOL_SIZE 1024 |
| 26 | #define PRISON_CELLS 1024 |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 27 | #define COMMIT_PERIOD HZ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 28 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 29 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
| 30 | "A percentage of time allocated for copy on write"); |
| 31 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 32 | /* |
| 33 | * The block size of the device holding pool data must be |
| 34 | * between 64KB and 1GB. |
| 35 | */ |
| 36 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) |
| 37 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
| 38 | |
| 39 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 40 | * Device id is restricted to 24 bits. |
| 41 | */ |
| 42 | #define MAX_DEV_ID ((1 << 24) - 1) |
| 43 | |
| 44 | /* |
| 45 | * How do we handle breaking sharing of data blocks? |
| 46 | * ================================================= |
| 47 | * |
| 48 | * We use a standard copy-on-write btree to store the mappings for the |
| 49 | * devices (note I'm talking about copy-on-write of the metadata here, not |
| 50 | * the data). When you take an internal snapshot you clone the root node |
| 51 | * of the origin btree. After this there is no concept of an origin or a |
| 52 | * snapshot. They are just two device trees that happen to point to the |
| 53 | * same data blocks. |
| 54 | * |
| 55 | * When we get a write in we decide if it's to a shared data block using |
| 56 | * some timestamp magic. If it is, we have to break sharing. |
| 57 | * |
| 58 | * Let's say we write to a shared block in what was the origin. The |
| 59 | * steps are: |
| 60 | * |
| 61 | * i) plug io further to this physical block. (see bio_prison code). |
| 62 | * |
| 63 | * ii) quiesce any read io to that shared data block. Obviously |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 64 | * including all devices that share this block. (see dm_deferred_set code) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 65 | * |
| 66 | * iii) copy the data block to a newly allocate block. This step can be |
| 67 | * missed out if the io covers the block. (schedule_copy). |
| 68 | * |
| 69 | * iv) insert the new mapping into the origin's btree |
Joe Thornber | fe878f3 | 2012-03-28 18:41:24 +0100 | [diff] [blame] | 70 | * (process_prepared_mapping). This act of inserting breaks some |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 71 | * sharing of btree nodes between the two devices. Breaking sharing only |
| 72 | * effects the btree of that specific device. Btrees for the other |
| 73 | * devices that share the block never change. The btree for the origin |
| 74 | * device as it was after the last commit is untouched, ie. we're using |
| 75 | * persistent data structures in the functional programming sense. |
| 76 | * |
| 77 | * v) unplug io to this physical block, including the io that triggered |
| 78 | * the breaking of sharing. |
| 79 | * |
| 80 | * Steps (ii) and (iii) occur in parallel. |
| 81 | * |
| 82 | * The metadata _doesn't_ need to be committed before the io continues. We |
| 83 | * get away with this because the io is always written to a _new_ block. |
| 84 | * If there's a crash, then: |
| 85 | * |
| 86 | * - The origin mapping will point to the old origin block (the shared |
| 87 | * one). This will contain the data as it was before the io that triggered |
| 88 | * the breaking of sharing came in. |
| 89 | * |
| 90 | * - The snap mapping still points to the old block. As it would after |
| 91 | * the commit. |
| 92 | * |
| 93 | * The downside of this scheme is the timestamp magic isn't perfect, and |
| 94 | * will continue to think that data block in the snapshot device is shared |
| 95 | * even after the write to the origin has broken sharing. I suspect data |
| 96 | * blocks will typically be shared by many different devices, so we're |
| 97 | * breaking sharing n + 1 times, rather than n, where n is the number of |
| 98 | * devices that reference this data block. At the moment I think the |
| 99 | * benefits far, far outweigh the disadvantages. |
| 100 | */ |
| 101 | |
| 102 | /*----------------------------------------------------------------*/ |
| 103 | |
| 104 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 105 | * Key building. |
| 106 | */ |
| 107 | static void build_data_key(struct dm_thin_device *td, |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 108 | dm_block_t b, struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 109 | { |
| 110 | key->virtual = 0; |
| 111 | key->dev = dm_thin_dev_id(td); |
| 112 | key->block = b; |
| 113 | } |
| 114 | |
| 115 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 116 | struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 117 | { |
| 118 | key->virtual = 1; |
| 119 | key->dev = dm_thin_dev_id(td); |
| 120 | key->block = b; |
| 121 | } |
| 122 | |
| 123 | /*----------------------------------------------------------------*/ |
| 124 | |
| 125 | /* |
| 126 | * A pool device ties together a metadata device and a data device. It |
| 127 | * also provides the interface for creating and destroying internal |
| 128 | * devices. |
| 129 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 130 | struct dm_thin_new_mapping; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 131 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 132 | /* |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 133 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 134 | */ |
| 135 | enum pool_mode { |
| 136 | PM_WRITE, /* metadata may be changed */ |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 137 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 138 | PM_READ_ONLY, /* metadata may not be changed */ |
| 139 | PM_FAIL, /* all I/O fails */ |
| 140 | }; |
| 141 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 142 | struct pool_features { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 143 | enum pool_mode mode; |
| 144 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 145 | bool zero_new_blocks:1; |
| 146 | bool discard_enabled:1; |
| 147 | bool discard_passdown:1; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 148 | bool error_if_no_space:1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 149 | }; |
| 150 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 151 | struct thin_c; |
| 152 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); |
| 153 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
| 154 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 155 | struct pool { |
| 156 | struct list_head list; |
| 157 | struct dm_target *ti; /* Only set if a pool target is bound */ |
| 158 | |
| 159 | struct mapped_device *pool_md; |
| 160 | struct block_device *md_dev; |
| 161 | struct dm_pool_metadata *pmd; |
| 162 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 163 | dm_block_t low_water_blocks; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 164 | uint32_t sectors_per_block; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 165 | int sectors_per_block_shift; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 166 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 167 | struct pool_features pf; |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 168 | bool low_water_triggered:1; /* A dm event has been sent */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 169 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 170 | struct dm_bio_prison *prison; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 171 | struct dm_kcopyd_client *copier; |
| 172 | |
| 173 | struct workqueue_struct *wq; |
| 174 | struct work_struct worker; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 175 | struct delayed_work waker; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 176 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 177 | unsigned long last_commit_jiffies; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 178 | unsigned ref_count; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 179 | |
| 180 | spinlock_t lock; |
| 181 | struct bio_list deferred_bios; |
| 182 | struct bio_list deferred_flush_bios; |
| 183 | struct list_head prepared_mappings; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 184 | struct list_head prepared_discards; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 185 | |
| 186 | struct bio_list retry_on_resume_list; |
| 187 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 188 | struct dm_deferred_set *shared_read_ds; |
| 189 | struct dm_deferred_set *all_io_ds; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 190 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 191 | struct dm_thin_new_mapping *next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 192 | mempool_t *mapping_pool; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 193 | |
| 194 | process_bio_fn process_bio; |
| 195 | process_bio_fn process_discard; |
| 196 | |
| 197 | process_mapping_fn process_prepared_mapping; |
| 198 | process_mapping_fn process_prepared_discard; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 199 | }; |
| 200 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 201 | static enum pool_mode get_pool_mode(struct pool *pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 202 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 203 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 204 | /* |
| 205 | * Target context for a pool. |
| 206 | */ |
| 207 | struct pool_c { |
| 208 | struct dm_target *ti; |
| 209 | struct pool *pool; |
| 210 | struct dm_dev *data_dev; |
| 211 | struct dm_dev *metadata_dev; |
| 212 | struct dm_target_callbacks callbacks; |
| 213 | |
| 214 | dm_block_t low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 215 | struct pool_features requested_pf; /* Features requested during table load */ |
| 216 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 217 | }; |
| 218 | |
| 219 | /* |
| 220 | * Target context for a thin. |
| 221 | */ |
| 222 | struct thin_c { |
| 223 | struct dm_dev *pool_dev; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 224 | struct dm_dev *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 225 | dm_thin_id dev_id; |
| 226 | |
| 227 | struct pool *pool; |
| 228 | struct dm_thin_device *td; |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 229 | bool requeue_mode:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 230 | }; |
| 231 | |
| 232 | /*----------------------------------------------------------------*/ |
| 233 | |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 234 | /* |
| 235 | * wake_worker() is used when new work is queued and when pool_resume is |
| 236 | * ready to continue deferred IO processing. |
| 237 | */ |
| 238 | static void wake_worker(struct pool *pool) |
| 239 | { |
| 240 | queue_work(pool->wq, &pool->worker); |
| 241 | } |
| 242 | |
| 243 | /*----------------------------------------------------------------*/ |
| 244 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 245 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
| 246 | struct dm_bio_prison_cell **cell_result) |
| 247 | { |
| 248 | int r; |
| 249 | struct dm_bio_prison_cell *cell_prealloc; |
| 250 | |
| 251 | /* |
| 252 | * Allocate a cell from the prison's mempool. |
| 253 | * This might block but it can't fail. |
| 254 | */ |
| 255 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); |
| 256 | |
| 257 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); |
| 258 | if (r) |
| 259 | /* |
| 260 | * We reused an old cell; we can get rid of |
| 261 | * the new one. |
| 262 | */ |
| 263 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); |
| 264 | |
| 265 | return r; |
| 266 | } |
| 267 | |
| 268 | static void cell_release(struct pool *pool, |
| 269 | struct dm_bio_prison_cell *cell, |
| 270 | struct bio_list *bios) |
| 271 | { |
| 272 | dm_cell_release(pool->prison, cell, bios); |
| 273 | dm_bio_prison_free_cell(pool->prison, cell); |
| 274 | } |
| 275 | |
| 276 | static void cell_release_no_holder(struct pool *pool, |
| 277 | struct dm_bio_prison_cell *cell, |
| 278 | struct bio_list *bios) |
| 279 | { |
| 280 | dm_cell_release_no_holder(pool->prison, cell, bios); |
| 281 | dm_bio_prison_free_cell(pool->prison, cell); |
| 282 | } |
| 283 | |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 284 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
| 285 | struct dm_bio_prison_cell *cell) |
| 286 | { |
| 287 | struct pool *pool = tc->pool; |
| 288 | unsigned long flags; |
| 289 | |
| 290 | spin_lock_irqsave(&pool->lock, flags); |
| 291 | dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios); |
| 292 | spin_unlock_irqrestore(&pool->lock, flags); |
| 293 | |
| 294 | wake_worker(pool); |
| 295 | } |
| 296 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 297 | static void cell_error(struct pool *pool, |
| 298 | struct dm_bio_prison_cell *cell) |
| 299 | { |
| 300 | dm_cell_error(pool->prison, cell); |
| 301 | dm_bio_prison_free_cell(pool->prison, cell); |
| 302 | } |
| 303 | |
| 304 | /*----------------------------------------------------------------*/ |
| 305 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 306 | /* |
| 307 | * A global list of pools that uses a struct mapped_device as a key. |
| 308 | */ |
| 309 | static struct dm_thin_pool_table { |
| 310 | struct mutex mutex; |
| 311 | struct list_head pools; |
| 312 | } dm_thin_pool_table; |
| 313 | |
| 314 | static void pool_table_init(void) |
| 315 | { |
| 316 | mutex_init(&dm_thin_pool_table.mutex); |
| 317 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); |
| 318 | } |
| 319 | |
| 320 | static void __pool_table_insert(struct pool *pool) |
| 321 | { |
| 322 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 323 | list_add(&pool->list, &dm_thin_pool_table.pools); |
| 324 | } |
| 325 | |
| 326 | static void __pool_table_remove(struct pool *pool) |
| 327 | { |
| 328 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 329 | list_del(&pool->list); |
| 330 | } |
| 331 | |
| 332 | static struct pool *__pool_table_lookup(struct mapped_device *md) |
| 333 | { |
| 334 | struct pool *pool = NULL, *tmp; |
| 335 | |
| 336 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 337 | |
| 338 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 339 | if (tmp->pool_md == md) { |
| 340 | pool = tmp; |
| 341 | break; |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | return pool; |
| 346 | } |
| 347 | |
| 348 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) |
| 349 | { |
| 350 | struct pool *pool = NULL, *tmp; |
| 351 | |
| 352 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 353 | |
| 354 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 355 | if (tmp->md_dev == md_dev) { |
| 356 | pool = tmp; |
| 357 | break; |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | return pool; |
| 362 | } |
| 363 | |
| 364 | /*----------------------------------------------------------------*/ |
| 365 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 366 | struct dm_thin_endio_hook { |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 367 | struct thin_c *tc; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 368 | struct dm_deferred_entry *shared_read_entry; |
| 369 | struct dm_deferred_entry *all_io_entry; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 370 | struct dm_thin_new_mapping *overwrite_mapping; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 371 | }; |
| 372 | |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 373 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 374 | { |
| 375 | struct bio *bio; |
| 376 | struct bio_list bios; |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 377 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 378 | |
| 379 | bio_list_init(&bios); |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 380 | |
| 381 | spin_lock_irqsave(&tc->pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 382 | bio_list_merge(&bios, master); |
| 383 | bio_list_init(master); |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 384 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 385 | |
| 386 | while ((bio = bio_list_pop(&bios))) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 387 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 388 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 389 | if (h->tc == tc) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 390 | bio_endio(bio, DM_ENDIO_REQUEUE); |
| 391 | else |
| 392 | bio_list_add(master, bio); |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | static void requeue_io(struct thin_c *tc) |
| 397 | { |
| 398 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 399 | |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 400 | requeue_bio_list(tc, &pool->deferred_bios); |
| 401 | requeue_bio_list(tc, &pool->retry_on_resume_list); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 402 | } |
| 403 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 404 | static void error_retry_list(struct pool *pool) |
| 405 | { |
| 406 | struct bio *bio; |
| 407 | unsigned long flags; |
| 408 | struct bio_list bios; |
| 409 | |
| 410 | bio_list_init(&bios); |
| 411 | |
| 412 | spin_lock_irqsave(&pool->lock, flags); |
| 413 | bio_list_merge(&bios, &pool->retry_on_resume_list); |
| 414 | bio_list_init(&pool->retry_on_resume_list); |
| 415 | spin_unlock_irqrestore(&pool->lock, flags); |
| 416 | |
| 417 | while ((bio = bio_list_pop(&bios))) |
| 418 | bio_io_error(bio); |
| 419 | } |
| 420 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 421 | /* |
| 422 | * This section of code contains the logic for processing a thin device's IO. |
| 423 | * Much of the code depends on pool object resources (lists, workqueues, etc) |
| 424 | * but most is exclusively called from the thin target rather than the thin-pool |
| 425 | * target. |
| 426 | */ |
| 427 | |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 428 | static bool block_size_is_power_of_two(struct pool *pool) |
| 429 | { |
| 430 | return pool->sectors_per_block_shift >= 0; |
| 431 | } |
| 432 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 433 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
| 434 | { |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 435 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 436 | sector_t block_nr = bio->bi_iter.bi_sector; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 437 | |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 438 | if (block_size_is_power_of_two(pool)) |
| 439 | block_nr >>= pool->sectors_per_block_shift; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 440 | else |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 441 | (void) sector_div(block_nr, pool->sectors_per_block); |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 442 | |
| 443 | return block_nr; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 444 | } |
| 445 | |
| 446 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
| 447 | { |
| 448 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 449 | sector_t bi_sector = bio->bi_iter.bi_sector; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 450 | |
| 451 | bio->bi_bdev = tc->pool_dev->bdev; |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 452 | if (block_size_is_power_of_two(pool)) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 453 | bio->bi_iter.bi_sector = |
| 454 | (block << pool->sectors_per_block_shift) | |
| 455 | (bi_sector & (pool->sectors_per_block - 1)); |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 456 | else |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 457 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 458 | sector_div(bi_sector, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 459 | } |
| 460 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 461 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
| 462 | { |
| 463 | bio->bi_bdev = tc->origin_dev->bdev; |
| 464 | } |
| 465 | |
Joe Thornber | 4afdd68 | 2012-07-27 15:08:14 +0100 | [diff] [blame] | 466 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
| 467 | { |
| 468 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && |
| 469 | dm_thin_changed_this_transaction(tc->td); |
| 470 | } |
| 471 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 472 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
| 473 | { |
| 474 | struct dm_thin_endio_hook *h; |
| 475 | |
| 476 | if (bio->bi_rw & REQ_DISCARD) |
| 477 | return; |
| 478 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 479 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 480 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
| 481 | } |
| 482 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 483 | static void issue(struct thin_c *tc, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 484 | { |
| 485 | struct pool *pool = tc->pool; |
| 486 | unsigned long flags; |
| 487 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 488 | if (!bio_triggers_commit(tc, bio)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 489 | generic_make_request(bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 490 | return; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * Complete bio with an error if earlier I/O caused changes to |
| 495 | * the metadata that can't be committed e.g, due to I/O errors |
| 496 | * on the metadata device. |
| 497 | */ |
| 498 | if (dm_thin_aborted_changes(tc->td)) { |
| 499 | bio_io_error(bio); |
| 500 | return; |
| 501 | } |
| 502 | |
| 503 | /* |
| 504 | * Batch together any bios that trigger commits and then issue a |
| 505 | * single commit for them in process_deferred_bios(). |
| 506 | */ |
| 507 | spin_lock_irqsave(&pool->lock, flags); |
| 508 | bio_list_add(&pool->deferred_flush_bios, bio); |
| 509 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 510 | } |
| 511 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 512 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
| 513 | { |
| 514 | remap_to_origin(tc, bio); |
| 515 | issue(tc, bio); |
| 516 | } |
| 517 | |
| 518 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, |
| 519 | dm_block_t block) |
| 520 | { |
| 521 | remap(tc, bio, block); |
| 522 | issue(tc, bio); |
| 523 | } |
| 524 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 525 | /*----------------------------------------------------------------*/ |
| 526 | |
| 527 | /* |
| 528 | * Bio endio functions. |
| 529 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 530 | struct dm_thin_new_mapping { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 531 | struct list_head list; |
| 532 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 533 | bool quiesced:1; |
| 534 | bool prepared:1; |
| 535 | bool pass_discard:1; |
| 536 | bool definitely_not_shared:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 537 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 538 | int err; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 539 | struct thin_c *tc; |
| 540 | dm_block_t virt_block; |
| 541 | dm_block_t data_block; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 542 | struct dm_bio_prison_cell *cell, *cell2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 543 | |
| 544 | /* |
| 545 | * If the bio covers the whole area of a block then we can avoid |
| 546 | * zeroing or copying. Instead this bio is hooked. The bio will |
| 547 | * still be in the cell, so care has to be taken to avoid issuing |
| 548 | * the bio twice. |
| 549 | */ |
| 550 | struct bio *bio; |
| 551 | bio_end_io_t *saved_bi_end_io; |
| 552 | }; |
| 553 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 554 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 555 | { |
| 556 | struct pool *pool = m->tc->pool; |
| 557 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 558 | if (m->quiesced && m->prepared) { |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 559 | list_add_tail(&m->list, &pool->prepared_mappings); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 560 | wake_worker(pool); |
| 561 | } |
| 562 | } |
| 563 | |
| 564 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
| 565 | { |
| 566 | unsigned long flags; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 567 | struct dm_thin_new_mapping *m = context; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 568 | struct pool *pool = m->tc->pool; |
| 569 | |
| 570 | m->err = read_err || write_err ? -EIO : 0; |
| 571 | |
| 572 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 573 | m->prepared = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 574 | __maybe_add_mapping(m); |
| 575 | spin_unlock_irqrestore(&pool->lock, flags); |
| 576 | } |
| 577 | |
| 578 | static void overwrite_endio(struct bio *bio, int err) |
| 579 | { |
| 580 | unsigned long flags; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 581 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 582 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 583 | struct pool *pool = m->tc->pool; |
| 584 | |
| 585 | m->err = err; |
| 586 | |
| 587 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 588 | m->prepared = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 589 | __maybe_add_mapping(m); |
| 590 | spin_unlock_irqrestore(&pool->lock, flags); |
| 591 | } |
| 592 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 593 | /*----------------------------------------------------------------*/ |
| 594 | |
| 595 | /* |
| 596 | * Workqueue. |
| 597 | */ |
| 598 | |
| 599 | /* |
| 600 | * Prepared mapping jobs. |
| 601 | */ |
| 602 | |
| 603 | /* |
| 604 | * This sends the bios in the cell back to the deferred_bios list. |
| 605 | */ |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 606 | 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] | 607 | { |
| 608 | struct pool *pool = tc->pool; |
| 609 | unsigned long flags; |
| 610 | |
| 611 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 612 | cell_release(pool, cell, &pool->deferred_bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 613 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
| 614 | |
| 615 | wake_worker(pool); |
| 616 | } |
| 617 | |
| 618 | /* |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 619 | * Same as cell_defer above, except it omits the original holder of the cell. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 620 | */ |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 621 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 622 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 623 | struct pool *pool = tc->pool; |
| 624 | unsigned long flags; |
| 625 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 626 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 627 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 628 | spin_unlock_irqrestore(&pool->lock, flags); |
| 629 | |
| 630 | wake_worker(pool); |
| 631 | } |
| 632 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 633 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
| 634 | { |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 635 | if (m->bio) { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 636 | m->bio->bi_end_io = m->saved_bi_end_io; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 637 | atomic_inc(&m->bio->bi_remaining); |
| 638 | } |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 639 | cell_error(m->tc->pool, m->cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 640 | list_del(&m->list); |
| 641 | mempool_free(m, m->tc->pool->mapping_pool); |
| 642 | } |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 643 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 644 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 645 | { |
| 646 | struct thin_c *tc = m->tc; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 647 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 648 | struct bio *bio; |
| 649 | int r; |
| 650 | |
| 651 | bio = m->bio; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 652 | if (bio) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 653 | bio->bi_end_io = m->saved_bi_end_io; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 654 | atomic_inc(&bio->bi_remaining); |
| 655 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 656 | |
| 657 | if (m->err) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 658 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 659 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 660 | } |
| 661 | |
| 662 | /* |
| 663 | * Commit the prepared block into the mapping btree. |
| 664 | * Any I/O for this block arriving after this point will get |
| 665 | * remapped to it directly. |
| 666 | */ |
| 667 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); |
| 668 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 669 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 670 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 671 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 672 | } |
| 673 | |
| 674 | /* |
| 675 | * Release any bios held while the block was being provisioned. |
| 676 | * If we are processing a write bio that completely covers the block, |
| 677 | * we already processed it so can ignore it now when processing |
| 678 | * the bios in the cell. |
| 679 | */ |
| 680 | if (bio) { |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 681 | cell_defer_no_holder(tc, m->cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 682 | bio_endio(bio, 0); |
| 683 | } else |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 684 | cell_defer(tc, m->cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 685 | |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 686 | out: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 687 | list_del(&m->list); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 688 | mempool_free(m, pool->mapping_pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 689 | } |
| 690 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 691 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 692 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 693 | struct thin_c *tc = m->tc; |
| 694 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 695 | bio_io_error(m->bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 696 | cell_defer_no_holder(tc, m->cell); |
| 697 | cell_defer_no_holder(tc, m->cell2); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 698 | mempool_free(m, tc->pool->mapping_pool); |
| 699 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 700 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 701 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) |
| 702 | { |
| 703 | struct thin_c *tc = m->tc; |
| 704 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 705 | inc_all_io_entry(tc->pool, m->bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 706 | cell_defer_no_holder(tc, m->cell); |
| 707 | cell_defer_no_holder(tc, m->cell2); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 708 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 709 | if (m->pass_discard) |
Joe Thornber | 19fa1a6 | 2013-12-17 12:09:40 -0500 | [diff] [blame] | 710 | if (m->definitely_not_shared) |
| 711 | remap_and_issue(tc, m->bio, m->data_block); |
| 712 | else { |
| 713 | bool used = false; |
| 714 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) |
| 715 | bio_endio(m->bio, 0); |
| 716 | else |
| 717 | remap_and_issue(tc, m->bio, m->data_block); |
| 718 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 719 | else |
| 720 | bio_endio(m->bio, 0); |
| 721 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 722 | mempool_free(m, tc->pool->mapping_pool); |
| 723 | } |
| 724 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 725 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
| 726 | { |
| 727 | int r; |
| 728 | struct thin_c *tc = m->tc; |
| 729 | |
| 730 | r = dm_thin_remove_block(tc->td, m->virt_block); |
| 731 | if (r) |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 732 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 733 | |
| 734 | process_prepared_discard_passdown(m); |
| 735 | } |
| 736 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 737 | static void process_prepared(struct pool *pool, struct list_head *head, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 738 | process_mapping_fn *fn) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 739 | { |
| 740 | unsigned long flags; |
| 741 | struct list_head maps; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 742 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 743 | |
| 744 | INIT_LIST_HEAD(&maps); |
| 745 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 746 | list_splice_init(head, &maps); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 747 | spin_unlock_irqrestore(&pool->lock, flags); |
| 748 | |
| 749 | list_for_each_entry_safe(m, tmp, &maps, list) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 750 | (*fn)(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 751 | } |
| 752 | |
| 753 | /* |
| 754 | * Deferred bio jobs. |
| 755 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 756 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
| 757 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 758 | return bio->bi_iter.bi_size == |
| 759 | (pool->sectors_per_block << SECTOR_SHIFT); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 760 | } |
| 761 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 762 | static int io_overwrites_block(struct pool *pool, struct bio *bio) |
| 763 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 764 | return (bio_data_dir(bio) == WRITE) && |
| 765 | io_overlaps_block(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 766 | } |
| 767 | |
| 768 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, |
| 769 | bio_end_io_t *fn) |
| 770 | { |
| 771 | *save = bio->bi_end_io; |
| 772 | bio->bi_end_io = fn; |
| 773 | } |
| 774 | |
| 775 | static int ensure_next_mapping(struct pool *pool) |
| 776 | { |
| 777 | if (pool->next_mapping) |
| 778 | return 0; |
| 779 | |
| 780 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); |
| 781 | |
| 782 | return pool->next_mapping ? 0 : -ENOMEM; |
| 783 | } |
| 784 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 785 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 786 | { |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 787 | struct dm_thin_new_mapping *m = pool->next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 788 | |
| 789 | BUG_ON(!pool->next_mapping); |
| 790 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 791 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
| 792 | INIT_LIST_HEAD(&m->list); |
| 793 | m->bio = NULL; |
| 794 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 795 | pool->next_mapping = NULL; |
| 796 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 797 | return m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 798 | } |
| 799 | |
| 800 | 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] | 801 | struct dm_dev *origin, dm_block_t data_origin, |
| 802 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 803 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 804 | { |
| 805 | int r; |
| 806 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 807 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 808 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 809 | m->tc = tc; |
| 810 | m->virt_block = virt_block; |
| 811 | m->data_block = data_dest; |
| 812 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 813 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 814 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 815 | m->quiesced = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 816 | |
| 817 | /* |
| 818 | * IO to pool_dev remaps to the pool target's data_dev. |
| 819 | * |
| 820 | * If the whole block of data is being overwritten, we can issue the |
| 821 | * bio immediately. Otherwise we use kcopyd to clone the data first. |
| 822 | */ |
| 823 | if (io_overwrites_block(pool, bio)) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 824 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 825 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 826 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 827 | m->bio = bio; |
| 828 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 829 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 830 | remap_and_issue(tc, bio, data_dest); |
| 831 | } else { |
| 832 | struct dm_io_region from, to; |
| 833 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 834 | from.bdev = origin->bdev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 835 | from.sector = data_origin * pool->sectors_per_block; |
| 836 | from.count = pool->sectors_per_block; |
| 837 | |
| 838 | to.bdev = tc->pool_dev->bdev; |
| 839 | to.sector = data_dest * pool->sectors_per_block; |
| 840 | to.count = pool->sectors_per_block; |
| 841 | |
| 842 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, |
| 843 | 0, copy_complete, m); |
| 844 | if (r < 0) { |
| 845 | mempool_free(m, pool->mapping_pool); |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 846 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 847 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 848 | } |
| 849 | } |
| 850 | } |
| 851 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 852 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
| 853 | dm_block_t data_origin, dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 854 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 855 | { |
| 856 | schedule_copy(tc, virt_block, tc->pool_dev, |
| 857 | data_origin, data_dest, cell, bio); |
| 858 | } |
| 859 | |
| 860 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
| 861 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 862 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 863 | { |
| 864 | schedule_copy(tc, virt_block, tc->origin_dev, |
| 865 | virt_block, data_dest, cell, bio); |
| 866 | } |
| 867 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 868 | 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] | 869 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 870 | struct bio *bio) |
| 871 | { |
| 872 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 873 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 874 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 875 | m->quiesced = true; |
| 876 | m->prepared = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 877 | m->tc = tc; |
| 878 | m->virt_block = virt_block; |
| 879 | m->data_block = data_block; |
| 880 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 881 | |
| 882 | /* |
| 883 | * If the whole block of data is being overwritten or we are not |
| 884 | * zeroing pre-existing data, we can issue the bio immediately. |
| 885 | * Otherwise we use kcopyd to zero the data first. |
| 886 | */ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 887 | if (!pool->pf.zero_new_blocks) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 888 | process_prepared_mapping(m); |
| 889 | |
| 890 | else if (io_overwrites_block(pool, bio)) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 891 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 892 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 893 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 894 | m->bio = bio; |
| 895 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 896 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 897 | remap_and_issue(tc, bio, data_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 898 | } else { |
| 899 | int r; |
| 900 | struct dm_io_region to; |
| 901 | |
| 902 | to.bdev = tc->pool_dev->bdev; |
| 903 | to.sector = data_block * pool->sectors_per_block; |
| 904 | to.count = pool->sectors_per_block; |
| 905 | |
| 906 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); |
| 907 | if (r < 0) { |
| 908 | mempool_free(m, pool->mapping_pool); |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 909 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 910 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 911 | } |
| 912 | } |
| 913 | } |
| 914 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 915 | /* |
| 916 | * A non-zero return indicates read_only or fail_io mode. |
| 917 | * Many callers don't care about the return value. |
| 918 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 919 | static int commit(struct pool *pool) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 920 | { |
| 921 | int r; |
| 922 | |
| 923 | if (get_pool_mode(pool) != PM_WRITE) |
| 924 | return -EINVAL; |
| 925 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 926 | r = dm_pool_commit_metadata(pool->pmd); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 927 | if (r) |
| 928 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 929 | |
| 930 | return r; |
| 931 | } |
| 932 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 933 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
| 934 | { |
| 935 | unsigned long flags; |
| 936 | |
| 937 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
| 938 | DMWARN("%s: reached low water mark for data device: sending event.", |
| 939 | dm_device_name(pool->pool_md)); |
| 940 | spin_lock_irqsave(&pool->lock, flags); |
| 941 | pool->low_water_triggered = true; |
| 942 | spin_unlock_irqrestore(&pool->lock, flags); |
| 943 | dm_table_event(pool->ti->table); |
| 944 | } |
| 945 | } |
| 946 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 947 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
| 948 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 949 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
| 950 | { |
| 951 | int r; |
| 952 | dm_block_t free_blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 953 | struct pool *pool = tc->pool; |
| 954 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 955 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
Joe Thornber | 8d30abf | 2013-12-04 19:16:11 -0500 | [diff] [blame] | 956 | return -EINVAL; |
| 957 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 958 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 959 | if (r) { |
| 960 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 961 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 962 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 963 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 964 | check_low_water_mark(pool, free_blocks); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 965 | |
| 966 | if (!free_blocks) { |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 967 | /* |
| 968 | * Try to commit to see if that will free up some |
| 969 | * more space. |
| 970 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 971 | r = commit(pool); |
| 972 | if (r) |
| 973 | return r; |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 974 | |
| 975 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 976 | if (r) { |
| 977 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 978 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 979 | } |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 980 | |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 981 | if (!free_blocks) { |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 982 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 983 | return -ENOSPC; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 984 | } |
| 985 | } |
| 986 | |
| 987 | r = dm_pool_alloc_data_block(pool->pmd, result); |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 988 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 989 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 990 | return r; |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 991 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 992 | |
| 993 | return 0; |
| 994 | } |
| 995 | |
| 996 | /* |
| 997 | * If we have run out of space, queue bios until the device is |
| 998 | * resumed, presumably after having been reloaded with more space. |
| 999 | */ |
| 1000 | static void retry_on_resume(struct bio *bio) |
| 1001 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1002 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1003 | struct thin_c *tc = h->tc; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1004 | struct pool *pool = tc->pool; |
| 1005 | unsigned long flags; |
| 1006 | |
| 1007 | spin_lock_irqsave(&pool->lock, flags); |
| 1008 | bio_list_add(&pool->retry_on_resume_list, bio); |
| 1009 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1010 | } |
| 1011 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1012 | static bool should_error_unserviceable_bio(struct pool *pool) |
| 1013 | { |
| 1014 | enum pool_mode m = get_pool_mode(pool); |
| 1015 | |
| 1016 | switch (m) { |
| 1017 | case PM_WRITE: |
| 1018 | /* Shouldn't get here */ |
| 1019 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); |
| 1020 | return true; |
| 1021 | |
| 1022 | case PM_OUT_OF_DATA_SPACE: |
| 1023 | return pool->pf.error_if_no_space; |
| 1024 | |
| 1025 | case PM_READ_ONLY: |
| 1026 | case PM_FAIL: |
| 1027 | return true; |
| 1028 | default: |
| 1029 | /* Shouldn't get here */ |
| 1030 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); |
| 1031 | return true; |
| 1032 | } |
| 1033 | } |
| 1034 | |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1035 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
| 1036 | { |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1037 | if (should_error_unserviceable_bio(pool)) |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1038 | bio_io_error(bio); |
Mike Snitzer | 6d16202 | 2013-12-20 18:09:02 -0500 | [diff] [blame] | 1039 | else |
| 1040 | retry_on_resume(bio); |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1041 | } |
| 1042 | |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1043 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1044 | { |
| 1045 | struct bio *bio; |
| 1046 | struct bio_list bios; |
| 1047 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1048 | if (should_error_unserviceable_bio(pool)) { |
| 1049 | cell_error(pool, cell); |
| 1050 | return; |
| 1051 | } |
| 1052 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1053 | bio_list_init(&bios); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1054 | cell_release(pool, cell, &bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1055 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1056 | if (should_error_unserviceable_bio(pool)) |
| 1057 | while ((bio = bio_list_pop(&bios))) |
| 1058 | bio_io_error(bio); |
| 1059 | else |
| 1060 | while ((bio = bio_list_pop(&bios))) |
| 1061 | retry_on_resume(bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1062 | } |
| 1063 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1064 | static void process_discard(struct thin_c *tc, struct bio *bio) |
| 1065 | { |
| 1066 | int r; |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1067 | unsigned long flags; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1068 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1069 | struct dm_bio_prison_cell *cell, *cell2; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1070 | struct dm_cell_key key, key2; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1071 | dm_block_t block = get_bio_block(tc, bio); |
| 1072 | struct dm_thin_lookup_result lookup_result; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1073 | struct dm_thin_new_mapping *m; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1074 | |
| 1075 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1076 | if (bio_detain(tc->pool, &key, bio, &cell)) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1077 | return; |
| 1078 | |
| 1079 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1080 | switch (r) { |
| 1081 | case 0: |
| 1082 | /* |
| 1083 | * Check nobody is fiddling with this pool block. This can |
| 1084 | * happen if someone's in the process of breaking sharing |
| 1085 | * on this block. |
| 1086 | */ |
| 1087 | build_data_key(tc->td, lookup_result.block, &key2); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1088 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1089 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1090 | break; |
| 1091 | } |
| 1092 | |
| 1093 | if (io_overlaps_block(pool, bio)) { |
| 1094 | /* |
| 1095 | * IO may still be going to the destination block. We must |
| 1096 | * quiesce before we can do the removal. |
| 1097 | */ |
| 1098 | m = get_next_mapping(pool); |
| 1099 | m->tc = tc; |
Joe Thornber | 19fa1a6 | 2013-12-17 12:09:40 -0500 | [diff] [blame] | 1100 | m->pass_discard = pool->pf.discard_passdown; |
| 1101 | m->definitely_not_shared = !lookup_result.shared; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1102 | m->virt_block = block; |
| 1103 | m->data_block = lookup_result.block; |
| 1104 | m->cell = cell; |
| 1105 | m->cell2 = cell2; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1106 | m->bio = bio; |
| 1107 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1108 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1109 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 1110 | list_add_tail(&m->list, &pool->prepared_discards); |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1111 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1112 | wake_worker(pool); |
| 1113 | } |
| 1114 | } else { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1115 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1116 | cell_defer_no_holder(tc, cell); |
| 1117 | cell_defer_no_holder(tc, cell2); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1118 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1119 | /* |
Mikulas Patocka | 4929630 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1120 | * The DM core makes sure that the discard doesn't span |
| 1121 | * a block boundary. So we submit the discard of a |
| 1122 | * partial block appropriately. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1123 | */ |
Mikulas Patocka | 650d2a0 | 2012-07-20 14:25:05 +0100 | [diff] [blame] | 1124 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
| 1125 | remap_and_issue(tc, bio, lookup_result.block); |
| 1126 | else |
| 1127 | bio_endio(bio, 0); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1128 | } |
| 1129 | break; |
| 1130 | |
| 1131 | case -ENODATA: |
| 1132 | /* |
| 1133 | * It isn't provisioned, just forget it. |
| 1134 | */ |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1135 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1136 | bio_endio(bio, 0); |
| 1137 | break; |
| 1138 | |
| 1139 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1140 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1141 | __func__, r); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1142 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1143 | bio_io_error(bio); |
| 1144 | break; |
| 1145 | } |
| 1146 | } |
| 1147 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1148 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1149 | struct dm_cell_key *key, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1150 | struct dm_thin_lookup_result *lookup_result, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1151 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1152 | { |
| 1153 | int r; |
| 1154 | dm_block_t data_block; |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1155 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1156 | |
| 1157 | r = alloc_data_block(tc, &data_block); |
| 1158 | switch (r) { |
| 1159 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1160 | schedule_internal_copy(tc, block, lookup_result->block, |
| 1161 | data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1162 | break; |
| 1163 | |
| 1164 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1165 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1166 | break; |
| 1167 | |
| 1168 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1169 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1170 | __func__, r); |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1171 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1172 | break; |
| 1173 | } |
| 1174 | } |
| 1175 | |
| 1176 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
| 1177 | dm_block_t block, |
| 1178 | struct dm_thin_lookup_result *lookup_result) |
| 1179 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1180 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1181 | struct pool *pool = tc->pool; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1182 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1183 | |
| 1184 | /* |
| 1185 | * If cell is already occupied, then sharing is already in the process |
| 1186 | * of being broken so we have nothing further to do here. |
| 1187 | */ |
| 1188 | build_data_key(tc->td, lookup_result->block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1189 | if (bio_detain(pool, &key, bio, &cell)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1190 | return; |
| 1191 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1192 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1193 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
| 1194 | else { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1195 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1196 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1197 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1198 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1199 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1200 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1201 | remap_and_issue(tc, bio, lookup_result->block); |
| 1202 | } |
| 1203 | } |
| 1204 | |
| 1205 | 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] | 1206 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1207 | { |
| 1208 | int r; |
| 1209 | dm_block_t data_block; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1210 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1211 | |
| 1212 | /* |
| 1213 | * Remap empty bios (flushes) immediately, without provisioning. |
| 1214 | */ |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1215 | if (!bio->bi_iter.bi_size) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1216 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1217 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1218 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1219 | remap_and_issue(tc, bio, 0); |
| 1220 | return; |
| 1221 | } |
| 1222 | |
| 1223 | /* |
| 1224 | * Fill read bios with zeroes and complete them immediately. |
| 1225 | */ |
| 1226 | if (bio_data_dir(bio) == READ) { |
| 1227 | zero_fill_bio(bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1228 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1229 | bio_endio(bio, 0); |
| 1230 | return; |
| 1231 | } |
| 1232 | |
| 1233 | r = alloc_data_block(tc, &data_block); |
| 1234 | switch (r) { |
| 1235 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1236 | if (tc->origin_dev) |
| 1237 | schedule_external_copy(tc, block, data_block, cell, bio); |
| 1238 | else |
| 1239 | schedule_zero(tc, block, data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1240 | break; |
| 1241 | |
| 1242 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1243 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1244 | break; |
| 1245 | |
| 1246 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1247 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1248 | __func__, r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1249 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1250 | break; |
| 1251 | } |
| 1252 | } |
| 1253 | |
| 1254 | static void process_bio(struct thin_c *tc, struct bio *bio) |
| 1255 | { |
| 1256 | int r; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1257 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1258 | dm_block_t block = get_bio_block(tc, bio); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1259 | struct dm_bio_prison_cell *cell; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1260 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1261 | struct dm_thin_lookup_result lookup_result; |
| 1262 | |
| 1263 | /* |
| 1264 | * If cell is already occupied, then the block is already |
| 1265 | * being provisioned so we have nothing further to do here. |
| 1266 | */ |
| 1267 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1268 | if (bio_detain(pool, &key, bio, &cell)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1269 | return; |
| 1270 | |
| 1271 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1272 | switch (r) { |
| 1273 | case 0: |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1274 | if (lookup_result.shared) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1275 | process_shared_bio(tc, bio, block, &lookup_result); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1276 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1277 | } else { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1278 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1279 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1280 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1281 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1282 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1283 | break; |
| 1284 | |
| 1285 | case -ENODATA: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1286 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1287 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1288 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1289 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1290 | remap_to_origin_and_issue(tc, bio); |
| 1291 | } else |
| 1292 | provision_block(tc, bio, block, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1293 | break; |
| 1294 | |
| 1295 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1296 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1297 | __func__, r); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1298 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1299 | bio_io_error(bio); |
| 1300 | break; |
| 1301 | } |
| 1302 | } |
| 1303 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1304 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
| 1305 | { |
| 1306 | int r; |
| 1307 | int rw = bio_data_dir(bio); |
| 1308 | dm_block_t block = get_bio_block(tc, bio); |
| 1309 | struct dm_thin_lookup_result lookup_result; |
| 1310 | |
| 1311 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1312 | switch (r) { |
| 1313 | case 0: |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1314 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1315 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1316 | else { |
| 1317 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1318 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1319 | } |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1320 | break; |
| 1321 | |
| 1322 | case -ENODATA: |
| 1323 | if (rw != READ) { |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1324 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1325 | break; |
| 1326 | } |
| 1327 | |
| 1328 | if (tc->origin_dev) { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1329 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1330 | remap_to_origin_and_issue(tc, bio); |
| 1331 | break; |
| 1332 | } |
| 1333 | |
| 1334 | zero_fill_bio(bio); |
| 1335 | bio_endio(bio, 0); |
| 1336 | break; |
| 1337 | |
| 1338 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1339 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1340 | __func__, r); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1341 | bio_io_error(bio); |
| 1342 | break; |
| 1343 | } |
| 1344 | } |
| 1345 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1346 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
| 1347 | { |
| 1348 | bio_endio(bio, 0); |
| 1349 | } |
| 1350 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1351 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
| 1352 | { |
| 1353 | bio_io_error(bio); |
| 1354 | } |
| 1355 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 1356 | /* |
| 1357 | * FIXME: should we also commit due to size of transaction, measured in |
| 1358 | * metadata blocks? |
| 1359 | */ |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1360 | static int need_commit_due_to_time(struct pool *pool) |
| 1361 | { |
| 1362 | return jiffies < pool->last_commit_jiffies || |
| 1363 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; |
| 1364 | } |
| 1365 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1366 | static void process_deferred_bios(struct pool *pool) |
| 1367 | { |
| 1368 | unsigned long flags; |
| 1369 | struct bio *bio; |
| 1370 | struct bio_list bios; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1371 | |
| 1372 | bio_list_init(&bios); |
| 1373 | |
| 1374 | spin_lock_irqsave(&pool->lock, flags); |
| 1375 | bio_list_merge(&bios, &pool->deferred_bios); |
| 1376 | bio_list_init(&pool->deferred_bios); |
| 1377 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1378 | |
| 1379 | while ((bio = bio_list_pop(&bios))) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1380 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1381 | struct thin_c *tc = h->tc; |
| 1382 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1383 | if (tc->requeue_mode) { |
| 1384 | bio_endio(bio, DM_ENDIO_REQUEUE); |
| 1385 | continue; |
| 1386 | } |
| 1387 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1388 | /* |
| 1389 | * If we've got no free new_mapping structs, and processing |
| 1390 | * this bio might require one, we pause until there are some |
| 1391 | * prepared mappings to process. |
| 1392 | */ |
| 1393 | if (ensure_next_mapping(pool)) { |
| 1394 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | fe76cd8 | 2014-03-28 02:15:02 -0400 | [diff] [blame^] | 1395 | bio_list_add(&pool->deferred_bios, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1396 | bio_list_merge(&pool->deferred_bios, &bios); |
| 1397 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1398 | break; |
| 1399 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1400 | |
| 1401 | if (bio->bi_rw & REQ_DISCARD) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1402 | pool->process_discard(tc, bio); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1403 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1404 | pool->process_bio(tc, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1405 | } |
| 1406 | |
| 1407 | /* |
| 1408 | * If there are any deferred flush bios, we must commit |
| 1409 | * the metadata before issuing them. |
| 1410 | */ |
| 1411 | bio_list_init(&bios); |
| 1412 | spin_lock_irqsave(&pool->lock, flags); |
| 1413 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
| 1414 | bio_list_init(&pool->deferred_flush_bios); |
| 1415 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1416 | |
Mike Snitzer | 4d1662a | 2014-02-06 06:08:56 -0500 | [diff] [blame] | 1417 | if (bio_list_empty(&bios) && |
| 1418 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1419 | return; |
| 1420 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 1421 | if (commit(pool)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1422 | while ((bio = bio_list_pop(&bios))) |
| 1423 | bio_io_error(bio); |
| 1424 | return; |
| 1425 | } |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1426 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1427 | |
| 1428 | while ((bio = bio_list_pop(&bios))) |
| 1429 | generic_make_request(bio); |
| 1430 | } |
| 1431 | |
| 1432 | static void do_worker(struct work_struct *ws) |
| 1433 | { |
| 1434 | struct pool *pool = container_of(ws, struct pool, worker); |
| 1435 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1436 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
| 1437 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1438 | process_deferred_bios(pool); |
| 1439 | } |
| 1440 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1441 | /* |
| 1442 | * We want to commit periodically so that not too much |
| 1443 | * unwritten data builds up. |
| 1444 | */ |
| 1445 | static void do_waker(struct work_struct *ws) |
| 1446 | { |
| 1447 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); |
| 1448 | wake_worker(pool); |
| 1449 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); |
| 1450 | } |
| 1451 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1452 | /*----------------------------------------------------------------*/ |
| 1453 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1454 | struct noflush_work { |
| 1455 | struct work_struct worker; |
| 1456 | struct thin_c *tc; |
| 1457 | |
| 1458 | atomic_t complete; |
| 1459 | wait_queue_head_t wait; |
| 1460 | }; |
| 1461 | |
| 1462 | static void complete_noflush_work(struct noflush_work *w) |
| 1463 | { |
| 1464 | atomic_set(&w->complete, 1); |
| 1465 | wake_up(&w->wait); |
| 1466 | } |
| 1467 | |
| 1468 | static void do_noflush_start(struct work_struct *ws) |
| 1469 | { |
| 1470 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); |
| 1471 | w->tc->requeue_mode = true; |
| 1472 | requeue_io(w->tc); |
| 1473 | complete_noflush_work(w); |
| 1474 | } |
| 1475 | |
| 1476 | static void do_noflush_stop(struct work_struct *ws) |
| 1477 | { |
| 1478 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); |
| 1479 | w->tc->requeue_mode = false; |
| 1480 | complete_noflush_work(w); |
| 1481 | } |
| 1482 | |
| 1483 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) |
| 1484 | { |
| 1485 | struct noflush_work w; |
| 1486 | |
| 1487 | INIT_WORK(&w.worker, fn); |
| 1488 | w.tc = tc; |
| 1489 | atomic_set(&w.complete, 0); |
| 1490 | init_waitqueue_head(&w.wait); |
| 1491 | |
| 1492 | queue_work(tc->pool->wq, &w.worker); |
| 1493 | |
| 1494 | wait_event(w.wait, atomic_read(&w.complete)); |
| 1495 | } |
| 1496 | |
| 1497 | /*----------------------------------------------------------------*/ |
| 1498 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1499 | static enum pool_mode get_pool_mode(struct pool *pool) |
| 1500 | { |
| 1501 | return pool->pf.mode; |
| 1502 | } |
| 1503 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1504 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
| 1505 | { |
| 1506 | dm_table_event(pool->ti->table); |
| 1507 | DMINFO("%s: switching pool to %s mode", |
| 1508 | dm_device_name(pool->pool_md), new_mode); |
| 1509 | } |
| 1510 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1511 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1512 | { |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 1513 | struct pool_c *pt = pool->ti->private; |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1514 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
| 1515 | enum pool_mode old_mode = get_pool_mode(pool); |
| 1516 | |
| 1517 | /* |
| 1518 | * Never allow the pool to transition to PM_WRITE mode if user |
| 1519 | * intervention is required to verify metadata and data consistency. |
| 1520 | */ |
| 1521 | if (new_mode == PM_WRITE && needs_check) { |
| 1522 | DMERR("%s: unable to switch pool to write mode until repaired.", |
| 1523 | dm_device_name(pool->pool_md)); |
| 1524 | if (old_mode != new_mode) |
| 1525 | new_mode = old_mode; |
| 1526 | else |
| 1527 | new_mode = PM_READ_ONLY; |
| 1528 | } |
| 1529 | /* |
| 1530 | * If we were in PM_FAIL mode, rollback of metadata failed. We're |
| 1531 | * not going to recover without a thin_repair. So we never let the |
| 1532 | * pool move out of the old mode. |
| 1533 | */ |
| 1534 | if (old_mode == PM_FAIL) |
| 1535 | new_mode = old_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1536 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1537 | switch (new_mode) { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1538 | case PM_FAIL: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1539 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1540 | notify_of_pool_mode_change(pool, "failure"); |
Joe Thornber | 5383ef3 | 2013-12-04 16:30:01 -0500 | [diff] [blame] | 1541 | dm_pool_metadata_read_only(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1542 | pool->process_bio = process_bio_fail; |
| 1543 | pool->process_discard = process_bio_fail; |
| 1544 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
| 1545 | pool->process_prepared_discard = process_prepared_discard_fail; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1546 | |
| 1547 | error_retry_list(pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1548 | break; |
| 1549 | |
| 1550 | case PM_READ_ONLY: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1551 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1552 | notify_of_pool_mode_change(pool, "read-only"); |
| 1553 | dm_pool_metadata_read_only(pool->pmd); |
| 1554 | pool->process_bio = process_bio_read_only; |
| 1555 | pool->process_discard = process_bio_success; |
| 1556 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
| 1557 | pool->process_prepared_discard = process_prepared_discard_passdown; |
| 1558 | |
| 1559 | error_retry_list(pool); |
| 1560 | break; |
| 1561 | |
| 1562 | case PM_OUT_OF_DATA_SPACE: |
| 1563 | /* |
| 1564 | * Ideally we'd never hit this state; the low water mark |
| 1565 | * would trigger userland to extend the pool before we |
| 1566 | * completely run out of data space. However, many small |
| 1567 | * IOs to unprovisioned space can consume data space at an |
| 1568 | * alarming rate. Adjust your low water mark if you're |
| 1569 | * frequently seeing this mode. |
| 1570 | */ |
| 1571 | if (old_mode != new_mode) |
| 1572 | notify_of_pool_mode_change(pool, "out-of-data-space"); |
| 1573 | pool->process_bio = process_bio_read_only; |
| 1574 | pool->process_discard = process_discard; |
| 1575 | pool->process_prepared_mapping = process_prepared_mapping; |
| 1576 | pool->process_prepared_discard = process_prepared_discard_passdown; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1577 | break; |
| 1578 | |
| 1579 | case PM_WRITE: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1580 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1581 | notify_of_pool_mode_change(pool, "write"); |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 1582 | dm_pool_metadata_read_write(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1583 | pool->process_bio = process_bio; |
| 1584 | pool->process_discard = process_discard; |
| 1585 | pool->process_prepared_mapping = process_prepared_mapping; |
| 1586 | pool->process_prepared_discard = process_prepared_discard; |
| 1587 | break; |
| 1588 | } |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1589 | |
| 1590 | pool->pf.mode = new_mode; |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 1591 | /* |
| 1592 | * The pool mode may have changed, sync it so bind_control_target() |
| 1593 | * doesn't cause an unexpected mode transition on resume. |
| 1594 | */ |
| 1595 | pt->adjusted_pf.mode = new_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1596 | } |
| 1597 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1598 | static void abort_transaction(struct pool *pool) |
| 1599 | { |
| 1600 | const char *dev_name = dm_device_name(pool->pool_md); |
| 1601 | |
| 1602 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
| 1603 | if (dm_pool_abort_metadata(pool->pmd)) { |
| 1604 | DMERR("%s: failed to abort metadata transaction", dev_name); |
| 1605 | set_pool_mode(pool, PM_FAIL); |
| 1606 | } |
| 1607 | |
| 1608 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { |
| 1609 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
| 1610 | set_pool_mode(pool, PM_FAIL); |
| 1611 | } |
| 1612 | } |
| 1613 | |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1614 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
| 1615 | { |
| 1616 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
| 1617 | dm_device_name(pool->pool_md), op, r); |
| 1618 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1619 | abort_transaction(pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1620 | set_pool_mode(pool, PM_READ_ONLY); |
| 1621 | } |
| 1622 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1623 | /*----------------------------------------------------------------*/ |
| 1624 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1625 | /* |
| 1626 | * Mapping functions. |
| 1627 | */ |
| 1628 | |
| 1629 | /* |
| 1630 | * Called only while mapping a thin bio to hand it over to the workqueue. |
| 1631 | */ |
| 1632 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) |
| 1633 | { |
| 1634 | unsigned long flags; |
| 1635 | struct pool *pool = tc->pool; |
| 1636 | |
| 1637 | spin_lock_irqsave(&pool->lock, flags); |
| 1638 | bio_list_add(&pool->deferred_bios, bio); |
| 1639 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1640 | |
| 1641 | wake_worker(pool); |
| 1642 | } |
| 1643 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1644 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1645 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1646 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1647 | |
| 1648 | h->tc = tc; |
| 1649 | h->shared_read_entry = NULL; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1650 | h->all_io_entry = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1651 | h->overwrite_mapping = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1652 | } |
| 1653 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1654 | /* |
| 1655 | * Non-blocking function called from the thin target's map function. |
| 1656 | */ |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 1657 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1658 | { |
| 1659 | int r; |
| 1660 | struct thin_c *tc = ti->private; |
| 1661 | dm_block_t block = get_bio_block(tc, bio); |
| 1662 | struct dm_thin_device *td = tc->td; |
| 1663 | struct dm_thin_lookup_result result; |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1664 | struct dm_bio_prison_cell cell1, cell2; |
| 1665 | struct dm_bio_prison_cell *cell_result; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1666 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1667 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1668 | thin_hook_bio(tc, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1669 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1670 | if (tc->requeue_mode) { |
| 1671 | bio_endio(bio, DM_ENDIO_REQUEUE); |
| 1672 | return DM_MAPIO_SUBMITTED; |
| 1673 | } |
| 1674 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1675 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 1676 | bio_io_error(bio); |
| 1677 | return DM_MAPIO_SUBMITTED; |
| 1678 | } |
| 1679 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1680 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1681 | thin_defer_bio(tc, bio); |
| 1682 | return DM_MAPIO_SUBMITTED; |
| 1683 | } |
| 1684 | |
| 1685 | r = dm_thin_find_block(td, block, 0, &result); |
| 1686 | |
| 1687 | /* |
| 1688 | * Note that we defer readahead too. |
| 1689 | */ |
| 1690 | switch (r) { |
| 1691 | case 0: |
| 1692 | if (unlikely(result.shared)) { |
| 1693 | /* |
| 1694 | * We have a race condition here between the |
| 1695 | * result.shared value returned by the lookup and |
| 1696 | * snapshot creation, which may cause new |
| 1697 | * sharing. |
| 1698 | * |
| 1699 | * To avoid this always quiesce the origin before |
| 1700 | * taking the snap. You want to do this anyway to |
| 1701 | * ensure a consistent application view |
| 1702 | * (i.e. lockfs). |
| 1703 | * |
| 1704 | * More distant ancestors are irrelevant. The |
| 1705 | * shared flag will be set in their case. |
| 1706 | */ |
| 1707 | thin_defer_bio(tc, bio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1708 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1709 | } |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1710 | |
| 1711 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1712 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1713 | return DM_MAPIO_SUBMITTED; |
| 1714 | |
| 1715 | build_data_key(tc->td, result.block, &key); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1716 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
| 1717 | cell_defer_no_holder_no_free(tc, &cell1); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1718 | return DM_MAPIO_SUBMITTED; |
| 1719 | } |
| 1720 | |
| 1721 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1722 | cell_defer_no_holder_no_free(tc, &cell2); |
| 1723 | cell_defer_no_holder_no_free(tc, &cell1); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1724 | |
| 1725 | remap(tc, bio, result.block); |
| 1726 | return DM_MAPIO_REMAPPED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1727 | |
| 1728 | case -ENODATA: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1729 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
| 1730 | /* |
| 1731 | * This block isn't provisioned, and we have no way |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1732 | * of doing so. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1733 | */ |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1734 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1735 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1736 | } |
| 1737 | /* fall through */ |
| 1738 | |
| 1739 | case -EWOULDBLOCK: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1740 | /* |
| 1741 | * In future, the failed dm_thin_find_block above could |
| 1742 | * provide the hint to load the metadata into cache. |
| 1743 | */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1744 | thin_defer_bio(tc, bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1745 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1746 | |
| 1747 | default: |
| 1748 | /* |
| 1749 | * Must always call bio_io_error on failure. |
| 1750 | * dm_thin_find_block can fail with -EINVAL if the |
| 1751 | * pool is switched to fail-io mode. |
| 1752 | */ |
| 1753 | bio_io_error(bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1754 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1755 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1756 | } |
| 1757 | |
| 1758 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) |
| 1759 | { |
| 1760 | int r; |
| 1761 | unsigned long flags; |
| 1762 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
| 1763 | |
| 1764 | spin_lock_irqsave(&pt->pool->lock, flags); |
| 1765 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); |
| 1766 | spin_unlock_irqrestore(&pt->pool->lock, flags); |
| 1767 | |
| 1768 | if (!r) { |
| 1769 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 1770 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
| 1771 | } |
| 1772 | |
| 1773 | return r; |
| 1774 | } |
| 1775 | |
| 1776 | static void __requeue_bios(struct pool *pool) |
| 1777 | { |
| 1778 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); |
| 1779 | bio_list_init(&pool->retry_on_resume_list); |
| 1780 | } |
| 1781 | |
| 1782 | /*---------------------------------------------------------------- |
| 1783 | * Binding of control targets to a pool object |
| 1784 | *--------------------------------------------------------------*/ |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1785 | static bool data_dev_supports_discard(struct pool_c *pt) |
| 1786 | { |
| 1787 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 1788 | |
| 1789 | return q && blk_queue_discard(q); |
| 1790 | } |
| 1791 | |
Joe Thornber | 58051b9 | 2013-03-20 17:21:25 +0000 | [diff] [blame] | 1792 | static bool is_factor(sector_t block_size, uint32_t n) |
| 1793 | { |
| 1794 | return !sector_div(block_size, n); |
| 1795 | } |
| 1796 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1797 | /* |
| 1798 | * If discard_passdown was enabled verify that the data device |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1799 | * supports discards. Disable discard_passdown if not. |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1800 | */ |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1801 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1802 | { |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1803 | struct pool *pool = pt->pool; |
| 1804 | struct block_device *data_bdev = pt->data_dev->bdev; |
| 1805 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; |
| 1806 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; |
| 1807 | const char *reason = NULL; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1808 | char buf[BDEVNAME_SIZE]; |
| 1809 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1810 | if (!pt->adjusted_pf.discard_passdown) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1811 | return; |
| 1812 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1813 | if (!data_dev_supports_discard(pt)) |
| 1814 | reason = "discard unsupported"; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1815 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1816 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) |
| 1817 | reason = "max discard sectors smaller than a block"; |
| 1818 | |
| 1819 | else if (data_limits->discard_granularity > block_size) |
| 1820 | reason = "discard granularity larger than a block"; |
| 1821 | |
Joe Thornber | 58051b9 | 2013-03-20 17:21:25 +0000 | [diff] [blame] | 1822 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1823 | reason = "discard granularity not a factor of block size"; |
| 1824 | |
| 1825 | if (reason) { |
| 1826 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); |
| 1827 | pt->adjusted_pf.discard_passdown = false; |
| 1828 | } |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1829 | } |
| 1830 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1831 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
| 1832 | { |
| 1833 | struct pool_c *pt = ti->private; |
| 1834 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1835 | /* |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 1836 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1837 | */ |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1838 | enum pool_mode old_mode = get_pool_mode(pool); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1839 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1840 | |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 1841 | /* |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1842 | * Don't change the pool's mode until set_pool_mode() below. |
| 1843 | * Otherwise the pool's process_* function pointers may |
| 1844 | * not match the desired pool mode. |
| 1845 | */ |
| 1846 | pt->adjusted_pf.mode = old_mode; |
| 1847 | |
| 1848 | pool->ti = ti; |
| 1849 | pool->pf = pt->adjusted_pf; |
| 1850 | pool->low_water_blocks = pt->low_water_blocks; |
| 1851 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1852 | set_pool_mode(pool, new_mode); |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 1853 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1854 | return 0; |
| 1855 | } |
| 1856 | |
| 1857 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) |
| 1858 | { |
| 1859 | if (pool->ti == ti) |
| 1860 | pool->ti = NULL; |
| 1861 | } |
| 1862 | |
| 1863 | /*---------------------------------------------------------------- |
| 1864 | * Pool creation |
| 1865 | *--------------------------------------------------------------*/ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1866 | /* Initialize pool features. */ |
| 1867 | static void pool_features_init(struct pool_features *pf) |
| 1868 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1869 | pf->mode = PM_WRITE; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1870 | pf->zero_new_blocks = true; |
| 1871 | pf->discard_enabled = true; |
| 1872 | pf->discard_passdown = true; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 1873 | pf->error_if_no_space = false; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1874 | } |
| 1875 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1876 | static void __pool_destroy(struct pool *pool) |
| 1877 | { |
| 1878 | __pool_table_remove(pool); |
| 1879 | |
| 1880 | if (dm_pool_metadata_close(pool->pmd) < 0) |
| 1881 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 1882 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1883 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1884 | dm_kcopyd_client_destroy(pool->copier); |
| 1885 | |
| 1886 | if (pool->wq) |
| 1887 | destroy_workqueue(pool->wq); |
| 1888 | |
| 1889 | if (pool->next_mapping) |
| 1890 | mempool_free(pool->next_mapping, pool->mapping_pool); |
| 1891 | mempool_destroy(pool->mapping_pool); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1892 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 1893 | dm_deferred_set_destroy(pool->all_io_ds); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1894 | kfree(pool); |
| 1895 | } |
| 1896 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1897 | static struct kmem_cache *_new_mapping_cache; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1898 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1899 | static struct pool *pool_create(struct mapped_device *pool_md, |
| 1900 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1901 | unsigned long block_size, |
| 1902 | int read_only, char **error) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1903 | { |
| 1904 | int r; |
| 1905 | void *err_p; |
| 1906 | struct pool *pool; |
| 1907 | struct dm_pool_metadata *pmd; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1908 | bool format_device = read_only ? false : true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1909 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1910 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1911 | if (IS_ERR(pmd)) { |
| 1912 | *error = "Error creating metadata object"; |
| 1913 | return (struct pool *)pmd; |
| 1914 | } |
| 1915 | |
| 1916 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); |
| 1917 | if (!pool) { |
| 1918 | *error = "Error allocating memory for pool"; |
| 1919 | err_p = ERR_PTR(-ENOMEM); |
| 1920 | goto bad_pool; |
| 1921 | } |
| 1922 | |
| 1923 | pool->pmd = pmd; |
| 1924 | pool->sectors_per_block = block_size; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1925 | if (block_size & (block_size - 1)) |
| 1926 | pool->sectors_per_block_shift = -1; |
| 1927 | else |
| 1928 | pool->sectors_per_block_shift = __ffs(block_size); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1929 | pool->low_water_blocks = 0; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 1930 | pool_features_init(&pool->pf); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1931 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1932 | if (!pool->prison) { |
| 1933 | *error = "Error creating pool's bio prison"; |
| 1934 | err_p = ERR_PTR(-ENOMEM); |
| 1935 | goto bad_prison; |
| 1936 | } |
| 1937 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 1938 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1939 | if (IS_ERR(pool->copier)) { |
| 1940 | r = PTR_ERR(pool->copier); |
| 1941 | *error = "Error creating pool's kcopyd client"; |
| 1942 | err_p = ERR_PTR(r); |
| 1943 | goto bad_kcopyd_client; |
| 1944 | } |
| 1945 | |
| 1946 | /* |
| 1947 | * Create singlethreaded workqueue that will service all devices |
| 1948 | * that use this metadata. |
| 1949 | */ |
| 1950 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); |
| 1951 | if (!pool->wq) { |
| 1952 | *error = "Error creating pool's workqueue"; |
| 1953 | err_p = ERR_PTR(-ENOMEM); |
| 1954 | goto bad_wq; |
| 1955 | } |
| 1956 | |
| 1957 | INIT_WORK(&pool->worker, do_worker); |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1958 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1959 | spin_lock_init(&pool->lock); |
| 1960 | bio_list_init(&pool->deferred_bios); |
| 1961 | bio_list_init(&pool->deferred_flush_bios); |
| 1962 | INIT_LIST_HEAD(&pool->prepared_mappings); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1963 | INIT_LIST_HEAD(&pool->prepared_discards); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 1964 | pool->low_water_triggered = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1965 | bio_list_init(&pool->retry_on_resume_list); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1966 | |
| 1967 | pool->shared_read_ds = dm_deferred_set_create(); |
| 1968 | if (!pool->shared_read_ds) { |
| 1969 | *error = "Error creating pool's shared read deferred set"; |
| 1970 | err_p = ERR_PTR(-ENOMEM); |
| 1971 | goto bad_shared_read_ds; |
| 1972 | } |
| 1973 | |
| 1974 | pool->all_io_ds = dm_deferred_set_create(); |
| 1975 | if (!pool->all_io_ds) { |
| 1976 | *error = "Error creating pool's all io deferred set"; |
| 1977 | err_p = ERR_PTR(-ENOMEM); |
| 1978 | goto bad_all_io_ds; |
| 1979 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1980 | |
| 1981 | pool->next_mapping = NULL; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1982 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
| 1983 | _new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1984 | if (!pool->mapping_pool) { |
| 1985 | *error = "Error creating pool's mapping mempool"; |
| 1986 | err_p = ERR_PTR(-ENOMEM); |
| 1987 | goto bad_mapping_pool; |
| 1988 | } |
| 1989 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1990 | pool->ref_count = 1; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1991 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1992 | pool->pool_md = pool_md; |
| 1993 | pool->md_dev = metadata_dev; |
| 1994 | __pool_table_insert(pool); |
| 1995 | |
| 1996 | return pool; |
| 1997 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1998 | bad_mapping_pool: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1999 | dm_deferred_set_destroy(pool->all_io_ds); |
| 2000 | bad_all_io_ds: |
| 2001 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 2002 | bad_shared_read_ds: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2003 | destroy_workqueue(pool->wq); |
| 2004 | bad_wq: |
| 2005 | dm_kcopyd_client_destroy(pool->copier); |
| 2006 | bad_kcopyd_client: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2007 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2008 | bad_prison: |
| 2009 | kfree(pool); |
| 2010 | bad_pool: |
| 2011 | if (dm_pool_metadata_close(pmd)) |
| 2012 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 2013 | |
| 2014 | return err_p; |
| 2015 | } |
| 2016 | |
| 2017 | static void __pool_inc(struct pool *pool) |
| 2018 | { |
| 2019 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2020 | pool->ref_count++; |
| 2021 | } |
| 2022 | |
| 2023 | static void __pool_dec(struct pool *pool) |
| 2024 | { |
| 2025 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2026 | BUG_ON(!pool->ref_count); |
| 2027 | if (!--pool->ref_count) |
| 2028 | __pool_destroy(pool); |
| 2029 | } |
| 2030 | |
| 2031 | static struct pool *__pool_find(struct mapped_device *pool_md, |
| 2032 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2033 | unsigned long block_size, int read_only, |
| 2034 | char **error, int *created) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2035 | { |
| 2036 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); |
| 2037 | |
| 2038 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2039 | if (pool->pool_md != pool_md) { |
| 2040 | *error = "metadata device already in use by a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2041 | return ERR_PTR(-EBUSY); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2042 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2043 | __pool_inc(pool); |
| 2044 | |
| 2045 | } else { |
| 2046 | pool = __pool_table_lookup(pool_md); |
| 2047 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2048 | if (pool->md_dev != metadata_dev) { |
| 2049 | *error = "different pool cannot replace a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2050 | return ERR_PTR(-EINVAL); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2051 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2052 | __pool_inc(pool); |
| 2053 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2054 | } else { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2055 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2056 | *created = 1; |
| 2057 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2058 | } |
| 2059 | |
| 2060 | return pool; |
| 2061 | } |
| 2062 | |
| 2063 | /*---------------------------------------------------------------- |
| 2064 | * Pool target methods |
| 2065 | *--------------------------------------------------------------*/ |
| 2066 | static void pool_dtr(struct dm_target *ti) |
| 2067 | { |
| 2068 | struct pool_c *pt = ti->private; |
| 2069 | |
| 2070 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2071 | |
| 2072 | unbind_control_target(pt->pool, ti); |
| 2073 | __pool_dec(pt->pool); |
| 2074 | dm_put_device(ti, pt->metadata_dev); |
| 2075 | dm_put_device(ti, pt->data_dev); |
| 2076 | kfree(pt); |
| 2077 | |
| 2078 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2079 | } |
| 2080 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2081 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
| 2082 | struct dm_target *ti) |
| 2083 | { |
| 2084 | int r; |
| 2085 | unsigned argc; |
| 2086 | const char *arg_name; |
| 2087 | |
| 2088 | static struct dm_arg _args[] = { |
Mike Snitzer | 74aa45c | 2014-01-15 19:07:58 -0500 | [diff] [blame] | 2089 | {0, 4, "Invalid number of pool feature arguments"}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2090 | }; |
| 2091 | |
| 2092 | /* |
| 2093 | * No feature arguments supplied. |
| 2094 | */ |
| 2095 | if (!as->argc) |
| 2096 | return 0; |
| 2097 | |
| 2098 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
| 2099 | if (r) |
| 2100 | return -EINVAL; |
| 2101 | |
| 2102 | while (argc && !r) { |
| 2103 | arg_name = dm_shift_arg(as); |
| 2104 | argc--; |
| 2105 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2106 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2107 | pf->zero_new_blocks = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2108 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2109 | else if (!strcasecmp(arg_name, "ignore_discard")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2110 | pf->discard_enabled = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2111 | |
| 2112 | else if (!strcasecmp(arg_name, "no_discard_passdown")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2113 | pf->discard_passdown = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2114 | |
| 2115 | else if (!strcasecmp(arg_name, "read_only")) |
| 2116 | pf->mode = PM_READ_ONLY; |
| 2117 | |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2118 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
| 2119 | pf->error_if_no_space = true; |
| 2120 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2121 | else { |
| 2122 | ti->error = "Unrecognised pool feature requested"; |
| 2123 | r = -EINVAL; |
| 2124 | break; |
| 2125 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2126 | } |
| 2127 | |
| 2128 | return r; |
| 2129 | } |
| 2130 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2131 | static void metadata_low_callback(void *context) |
| 2132 | { |
| 2133 | struct pool *pool = context; |
| 2134 | |
| 2135 | DMWARN("%s: reached low water mark for metadata device: sending event.", |
| 2136 | dm_device_name(pool->pool_md)); |
| 2137 | |
| 2138 | dm_table_event(pool->ti->table); |
| 2139 | } |
| 2140 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2141 | static sector_t get_dev_size(struct block_device *bdev) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2142 | { |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2143 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; |
| 2144 | } |
| 2145 | |
| 2146 | static void warn_if_metadata_device_too_big(struct block_device *bdev) |
| 2147 | { |
| 2148 | sector_t metadata_dev_size = get_dev_size(bdev); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2149 | char buffer[BDEVNAME_SIZE]; |
| 2150 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2151 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2152 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
| 2153 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2154 | } |
| 2155 | |
| 2156 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
| 2157 | { |
| 2158 | sector_t metadata_dev_size = get_dev_size(bdev); |
| 2159 | |
| 2160 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) |
| 2161 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2162 | |
| 2163 | return metadata_dev_size; |
| 2164 | } |
| 2165 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2166 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
| 2167 | { |
| 2168 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); |
| 2169 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2170 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2171 | |
| 2172 | return metadata_dev_size; |
| 2173 | } |
| 2174 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2175 | /* |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2176 | * When a metadata threshold is crossed a dm event is triggered, and |
| 2177 | * userland should respond by growing the metadata device. We could let |
| 2178 | * userland set the threshold, like we do with the data threshold, but I'm |
| 2179 | * not sure they know enough to do this well. |
| 2180 | */ |
| 2181 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) |
| 2182 | { |
| 2183 | /* |
| 2184 | * 4M is ample for all ops with the possible exception of thin |
| 2185 | * device deletion which is harmless if it fails (just retry the |
| 2186 | * delete after you've grown the device). |
| 2187 | */ |
| 2188 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; |
| 2189 | return min((dm_block_t)1024ULL /* 4M */, quarter); |
| 2190 | } |
| 2191 | |
| 2192 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2193 | * thin-pool <metadata dev> <data dev> |
| 2194 | * <data block size (sectors)> |
| 2195 | * <low water mark (blocks)> |
| 2196 | * [<#feature args> [<arg>]*] |
| 2197 | * |
| 2198 | * Optional feature arguments are: |
| 2199 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2200 | * ignore_discard: disable discard |
| 2201 | * no_discard_passdown: don't pass discards down to the data device |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2202 | * read_only: Don't allow any changes to be made to the pool metadata. |
| 2203 | * error_if_no_space: error IOs, instead of queueing, if no space. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2204 | */ |
| 2205 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 2206 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2207 | int r, pool_created = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2208 | struct pool_c *pt; |
| 2209 | struct pool *pool; |
| 2210 | struct pool_features pf; |
| 2211 | struct dm_arg_set as; |
| 2212 | struct dm_dev *data_dev; |
| 2213 | unsigned long block_size; |
| 2214 | dm_block_t low_water_blocks; |
| 2215 | struct dm_dev *metadata_dev; |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2216 | fmode_t metadata_mode; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2217 | |
| 2218 | /* |
| 2219 | * FIXME Remove validation from scope of lock. |
| 2220 | */ |
| 2221 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2222 | |
| 2223 | if (argc < 4) { |
| 2224 | ti->error = "Invalid argument count"; |
| 2225 | r = -EINVAL; |
| 2226 | goto out_unlock; |
| 2227 | } |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2228 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2229 | as.argc = argc; |
| 2230 | as.argv = argv; |
| 2231 | |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2232 | /* |
| 2233 | * Set default pool features. |
| 2234 | */ |
| 2235 | pool_features_init(&pf); |
| 2236 | |
| 2237 | dm_consume_args(&as, 4); |
| 2238 | r = parse_pool_features(&as, &pf, ti); |
| 2239 | if (r) |
| 2240 | goto out_unlock; |
| 2241 | |
| 2242 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); |
| 2243 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2244 | if (r) { |
| 2245 | ti->error = "Error opening metadata block device"; |
| 2246 | goto out_unlock; |
| 2247 | } |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2248 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2249 | |
| 2250 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); |
| 2251 | if (r) { |
| 2252 | ti->error = "Error getting data device"; |
| 2253 | goto out_metadata; |
| 2254 | } |
| 2255 | |
| 2256 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || |
| 2257 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || |
| 2258 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2259 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2260 | ti->error = "Invalid block size"; |
| 2261 | r = -EINVAL; |
| 2262 | goto out; |
| 2263 | } |
| 2264 | |
| 2265 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { |
| 2266 | ti->error = "Invalid low water mark"; |
| 2267 | r = -EINVAL; |
| 2268 | goto out; |
| 2269 | } |
| 2270 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2271 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
| 2272 | if (!pt) { |
| 2273 | r = -ENOMEM; |
| 2274 | goto out; |
| 2275 | } |
| 2276 | |
| 2277 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2278 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2279 | if (IS_ERR(pool)) { |
| 2280 | r = PTR_ERR(pool); |
| 2281 | goto out_free_pt; |
| 2282 | } |
| 2283 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2284 | /* |
| 2285 | * 'pool_created' reflects whether this is the first table load. |
| 2286 | * Top level discard support is not allowed to be changed after |
| 2287 | * initial load. This would require a pool reload to trigger thin |
| 2288 | * device changes. |
| 2289 | */ |
| 2290 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { |
| 2291 | ti->error = "Discard support cannot be disabled once enabled"; |
| 2292 | r = -EINVAL; |
| 2293 | goto out_flags_changed; |
| 2294 | } |
| 2295 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2296 | pt->pool = pool; |
| 2297 | pt->ti = ti; |
| 2298 | pt->metadata_dev = metadata_dev; |
| 2299 | pt->data_dev = data_dev; |
| 2300 | pt->low_water_blocks = low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2301 | pt->adjusted_pf = pt->requested_pf = pf; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 2302 | ti->num_flush_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2303 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2304 | /* |
| 2305 | * Only need to enable discards if the pool should pass |
| 2306 | * them down to the data device. The thin device's discard |
| 2307 | * processing will cause mappings to be removed from the btree. |
| 2308 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 2309 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2310 | if (pf.discard_enabled && pf.discard_passdown) { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 2311 | ti->num_discard_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2312 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2313 | /* |
| 2314 | * Setting 'discards_supported' circumvents the normal |
| 2315 | * stacking of discard limits (this keeps the pool and |
| 2316 | * thin devices' discard limits consistent). |
| 2317 | */ |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 2318 | ti->discards_supported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2319 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2320 | ti->private = pt; |
| 2321 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2322 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
| 2323 | calc_metadata_threshold(pt), |
| 2324 | metadata_low_callback, |
| 2325 | pool); |
| 2326 | if (r) |
| 2327 | goto out_free_pt; |
| 2328 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2329 | pt->callbacks.congested_fn = pool_is_congested; |
| 2330 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); |
| 2331 | |
| 2332 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2333 | |
| 2334 | return 0; |
| 2335 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2336 | out_flags_changed: |
| 2337 | __pool_dec(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2338 | out_free_pt: |
| 2339 | kfree(pt); |
| 2340 | out: |
| 2341 | dm_put_device(ti, data_dev); |
| 2342 | out_metadata: |
| 2343 | dm_put_device(ti, metadata_dev); |
| 2344 | out_unlock: |
| 2345 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2346 | |
| 2347 | return r; |
| 2348 | } |
| 2349 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 2350 | static int pool_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2351 | { |
| 2352 | int r; |
| 2353 | struct pool_c *pt = ti->private; |
| 2354 | struct pool *pool = pt->pool; |
| 2355 | unsigned long flags; |
| 2356 | |
| 2357 | /* |
| 2358 | * As this is a singleton target, ti->begin is always zero. |
| 2359 | */ |
| 2360 | spin_lock_irqsave(&pool->lock, flags); |
| 2361 | bio->bi_bdev = pt->data_dev->bdev; |
| 2362 | r = DM_MAPIO_REMAPPED; |
| 2363 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2364 | |
| 2365 | return r; |
| 2366 | } |
| 2367 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2368 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
| 2369 | { |
| 2370 | int r; |
| 2371 | struct pool_c *pt = ti->private; |
| 2372 | struct pool *pool = pt->pool; |
| 2373 | sector_t data_size = ti->len; |
| 2374 | dm_block_t sb_data_size; |
| 2375 | |
| 2376 | *need_commit = false; |
| 2377 | |
| 2378 | (void) sector_div(data_size, pool->sectors_per_block); |
| 2379 | |
| 2380 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
| 2381 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2382 | DMERR("%s: failed to retrieve data device size", |
| 2383 | dm_device_name(pool->pool_md)); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2384 | return r; |
| 2385 | } |
| 2386 | |
| 2387 | if (data_size < sb_data_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2388 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
| 2389 | dm_device_name(pool->pool_md), |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2390 | (unsigned long long)data_size, sb_data_size); |
| 2391 | return -EINVAL; |
| 2392 | |
| 2393 | } else if (data_size > sb_data_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2394 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 2395 | DMERR("%s: unable to grow the data device until repaired.", |
| 2396 | dm_device_name(pool->pool_md)); |
| 2397 | return 0; |
| 2398 | } |
| 2399 | |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 2400 | if (sb_data_size) |
| 2401 | DMINFO("%s: growing the data device from %llu to %llu blocks", |
| 2402 | dm_device_name(pool->pool_md), |
| 2403 | sb_data_size, (unsigned long long)data_size); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2404 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
| 2405 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2406 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2407 | return r; |
| 2408 | } |
| 2409 | |
| 2410 | *need_commit = true; |
| 2411 | } |
| 2412 | |
| 2413 | return 0; |
| 2414 | } |
| 2415 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2416 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
| 2417 | { |
| 2418 | int r; |
| 2419 | struct pool_c *pt = ti->private; |
| 2420 | struct pool *pool = pt->pool; |
| 2421 | dm_block_t metadata_dev_size, sb_metadata_dev_size; |
| 2422 | |
| 2423 | *need_commit = false; |
| 2424 | |
Alasdair G Kergon | 610bba8 | 2013-05-19 18:57:50 +0100 | [diff] [blame] | 2425 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2426 | |
| 2427 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); |
| 2428 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2429 | DMERR("%s: failed to retrieve metadata device size", |
| 2430 | dm_device_name(pool->pool_md)); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2431 | return r; |
| 2432 | } |
| 2433 | |
| 2434 | if (metadata_dev_size < sb_metadata_dev_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2435 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
| 2436 | dm_device_name(pool->pool_md), |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2437 | metadata_dev_size, sb_metadata_dev_size); |
| 2438 | return -EINVAL; |
| 2439 | |
| 2440 | } else if (metadata_dev_size > sb_metadata_dev_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2441 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 2442 | DMERR("%s: unable to grow the metadata device until repaired.", |
| 2443 | dm_device_name(pool->pool_md)); |
| 2444 | return 0; |
| 2445 | } |
| 2446 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2447 | warn_if_metadata_device_too_big(pool->md_dev); |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 2448 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
| 2449 | dm_device_name(pool->pool_md), |
| 2450 | sb_metadata_dev_size, metadata_dev_size); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2451 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
| 2452 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2453 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2454 | return r; |
| 2455 | } |
| 2456 | |
| 2457 | *need_commit = true; |
| 2458 | } |
| 2459 | |
| 2460 | return 0; |
| 2461 | } |
| 2462 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2463 | /* |
| 2464 | * Retrieves the number of blocks of the data device from |
| 2465 | * the superblock and compares it to the actual device size, |
| 2466 | * thus resizing the data device in case it has grown. |
| 2467 | * |
| 2468 | * This both copes with opening preallocated data devices in the ctr |
| 2469 | * being followed by a resume |
| 2470 | * -and- |
| 2471 | * calling the resume method individually after userspace has |
| 2472 | * grown the data device in reaction to a table event. |
| 2473 | */ |
| 2474 | static int pool_preresume(struct dm_target *ti) |
| 2475 | { |
| 2476 | int r; |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2477 | bool need_commit1, need_commit2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2478 | struct pool_c *pt = ti->private; |
| 2479 | struct pool *pool = pt->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2480 | |
| 2481 | /* |
| 2482 | * Take control of the pool object. |
| 2483 | */ |
| 2484 | r = bind_control_target(pool, ti); |
| 2485 | if (r) |
| 2486 | return r; |
| 2487 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2488 | r = maybe_resize_data_dev(ti, &need_commit1); |
| 2489 | if (r) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2490 | return r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2491 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2492 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
| 2493 | if (r) |
| 2494 | return r; |
| 2495 | |
| 2496 | if (need_commit1 || need_commit2) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2497 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2498 | |
| 2499 | return 0; |
| 2500 | } |
| 2501 | |
| 2502 | static void pool_resume(struct dm_target *ti) |
| 2503 | { |
| 2504 | struct pool_c *pt = ti->private; |
| 2505 | struct pool *pool = pt->pool; |
| 2506 | unsigned long flags; |
| 2507 | |
| 2508 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 2509 | pool->low_water_triggered = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2510 | __requeue_bios(pool); |
| 2511 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2512 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2513 | do_waker(&pool->waker.work); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2514 | } |
| 2515 | |
| 2516 | static void pool_postsuspend(struct dm_target *ti) |
| 2517 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2518 | struct pool_c *pt = ti->private; |
| 2519 | struct pool *pool = pt->pool; |
| 2520 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2521 | cancel_delayed_work(&pool->waker); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2522 | flush_workqueue(pool->wq); |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2523 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2524 | } |
| 2525 | |
| 2526 | static int check_arg_count(unsigned argc, unsigned args_required) |
| 2527 | { |
| 2528 | if (argc != args_required) { |
| 2529 | DMWARN("Message received with %u arguments instead of %u.", |
| 2530 | argc, args_required); |
| 2531 | return -EINVAL; |
| 2532 | } |
| 2533 | |
| 2534 | return 0; |
| 2535 | } |
| 2536 | |
| 2537 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) |
| 2538 | { |
| 2539 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && |
| 2540 | *dev_id <= MAX_DEV_ID) |
| 2541 | return 0; |
| 2542 | |
| 2543 | if (warning) |
| 2544 | DMWARN("Message received with invalid device id: %s", arg); |
| 2545 | |
| 2546 | return -EINVAL; |
| 2547 | } |
| 2548 | |
| 2549 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2550 | { |
| 2551 | dm_thin_id dev_id; |
| 2552 | int r; |
| 2553 | |
| 2554 | r = check_arg_count(argc, 2); |
| 2555 | if (r) |
| 2556 | return r; |
| 2557 | |
| 2558 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2559 | if (r) |
| 2560 | return r; |
| 2561 | |
| 2562 | r = dm_pool_create_thin(pool->pmd, dev_id); |
| 2563 | if (r) { |
| 2564 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", |
| 2565 | argv[1]); |
| 2566 | return r; |
| 2567 | } |
| 2568 | |
| 2569 | return 0; |
| 2570 | } |
| 2571 | |
| 2572 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2573 | { |
| 2574 | dm_thin_id dev_id; |
| 2575 | dm_thin_id origin_dev_id; |
| 2576 | int r; |
| 2577 | |
| 2578 | r = check_arg_count(argc, 3); |
| 2579 | if (r) |
| 2580 | return r; |
| 2581 | |
| 2582 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2583 | if (r) |
| 2584 | return r; |
| 2585 | |
| 2586 | r = read_dev_id(argv[2], &origin_dev_id, 1); |
| 2587 | if (r) |
| 2588 | return r; |
| 2589 | |
| 2590 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); |
| 2591 | if (r) { |
| 2592 | DMWARN("Creation of new snapshot %s of device %s failed.", |
| 2593 | argv[1], argv[2]); |
| 2594 | return r; |
| 2595 | } |
| 2596 | |
| 2597 | return 0; |
| 2598 | } |
| 2599 | |
| 2600 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2601 | { |
| 2602 | dm_thin_id dev_id; |
| 2603 | int r; |
| 2604 | |
| 2605 | r = check_arg_count(argc, 2); |
| 2606 | if (r) |
| 2607 | return r; |
| 2608 | |
| 2609 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2610 | if (r) |
| 2611 | return r; |
| 2612 | |
| 2613 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); |
| 2614 | if (r) |
| 2615 | DMWARN("Deletion of thin device %s failed.", argv[1]); |
| 2616 | |
| 2617 | return r; |
| 2618 | } |
| 2619 | |
| 2620 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2621 | { |
| 2622 | dm_thin_id old_id, new_id; |
| 2623 | int r; |
| 2624 | |
| 2625 | r = check_arg_count(argc, 3); |
| 2626 | if (r) |
| 2627 | return r; |
| 2628 | |
| 2629 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { |
| 2630 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); |
| 2631 | return -EINVAL; |
| 2632 | } |
| 2633 | |
| 2634 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { |
| 2635 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); |
| 2636 | return -EINVAL; |
| 2637 | } |
| 2638 | |
| 2639 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); |
| 2640 | if (r) { |
| 2641 | DMWARN("Failed to change transaction id from %s to %s.", |
| 2642 | argv[1], argv[2]); |
| 2643 | return r; |
| 2644 | } |
| 2645 | |
| 2646 | return 0; |
| 2647 | } |
| 2648 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2649 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2650 | { |
| 2651 | int r; |
| 2652 | |
| 2653 | r = check_arg_count(argc, 1); |
| 2654 | if (r) |
| 2655 | return r; |
| 2656 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2657 | (void) commit(pool); |
Joe Thornber | 0d200ae | 2012-07-03 12:55:31 +0100 | [diff] [blame] | 2658 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2659 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
| 2660 | if (r) |
| 2661 | DMWARN("reserve_metadata_snap message failed."); |
| 2662 | |
| 2663 | return r; |
| 2664 | } |
| 2665 | |
| 2666 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2667 | { |
| 2668 | int r; |
| 2669 | |
| 2670 | r = check_arg_count(argc, 1); |
| 2671 | if (r) |
| 2672 | return r; |
| 2673 | |
| 2674 | r = dm_pool_release_metadata_snap(pool->pmd); |
| 2675 | if (r) |
| 2676 | DMWARN("release_metadata_snap message failed."); |
| 2677 | |
| 2678 | return r; |
| 2679 | } |
| 2680 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2681 | /* |
| 2682 | * Messages supported: |
| 2683 | * create_thin <dev_id> |
| 2684 | * create_snap <dev_id> <origin_id> |
| 2685 | * delete <dev_id> |
| 2686 | * trim <dev_id> <new_size_in_sectors> |
| 2687 | * set_transaction_id <current_trans_id> <new_trans_id> |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2688 | * reserve_metadata_snap |
| 2689 | * release_metadata_snap |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2690 | */ |
| 2691 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) |
| 2692 | { |
| 2693 | int r = -EINVAL; |
| 2694 | struct pool_c *pt = ti->private; |
| 2695 | struct pool *pool = pt->pool; |
| 2696 | |
| 2697 | if (!strcasecmp(argv[0], "create_thin")) |
| 2698 | r = process_create_thin_mesg(argc, argv, pool); |
| 2699 | |
| 2700 | else if (!strcasecmp(argv[0], "create_snap")) |
| 2701 | r = process_create_snap_mesg(argc, argv, pool); |
| 2702 | |
| 2703 | else if (!strcasecmp(argv[0], "delete")) |
| 2704 | r = process_delete_mesg(argc, argv, pool); |
| 2705 | |
| 2706 | else if (!strcasecmp(argv[0], "set_transaction_id")) |
| 2707 | r = process_set_transaction_id_mesg(argc, argv, pool); |
| 2708 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2709 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
| 2710 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); |
| 2711 | |
| 2712 | else if (!strcasecmp(argv[0], "release_metadata_snap")) |
| 2713 | r = process_release_metadata_snap_mesg(argc, argv, pool); |
| 2714 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2715 | else |
| 2716 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); |
| 2717 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2718 | if (!r) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2719 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2720 | |
| 2721 | return r; |
| 2722 | } |
| 2723 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2724 | static void emit_flags(struct pool_features *pf, char *result, |
| 2725 | unsigned sz, unsigned maxlen) |
| 2726 | { |
| 2727 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2728 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
| 2729 | pf->error_if_no_space; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2730 | DMEMIT("%u ", count); |
| 2731 | |
| 2732 | if (!pf->zero_new_blocks) |
| 2733 | DMEMIT("skip_block_zeroing "); |
| 2734 | |
| 2735 | if (!pf->discard_enabled) |
| 2736 | DMEMIT("ignore_discard "); |
| 2737 | |
| 2738 | if (!pf->discard_passdown) |
| 2739 | DMEMIT("no_discard_passdown "); |
| 2740 | |
| 2741 | if (pf->mode == PM_READ_ONLY) |
| 2742 | DMEMIT("read_only "); |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2743 | |
| 2744 | if (pf->error_if_no_space) |
| 2745 | DMEMIT("error_if_no_space "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2746 | } |
| 2747 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2748 | /* |
| 2749 | * Status line is: |
| 2750 | * <transaction id> <used metadata sectors>/<total metadata sectors> |
| 2751 | * <used data sectors>/<total data sectors> <held metadata root> |
| 2752 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2753 | static void pool_status(struct dm_target *ti, status_type_t type, |
| 2754 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2755 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2756 | int r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2757 | unsigned sz = 0; |
| 2758 | uint64_t transaction_id; |
| 2759 | dm_block_t nr_free_blocks_data; |
| 2760 | dm_block_t nr_free_blocks_metadata; |
| 2761 | dm_block_t nr_blocks_data; |
| 2762 | dm_block_t nr_blocks_metadata; |
| 2763 | dm_block_t held_root; |
| 2764 | char buf[BDEVNAME_SIZE]; |
| 2765 | char buf2[BDEVNAME_SIZE]; |
| 2766 | struct pool_c *pt = ti->private; |
| 2767 | struct pool *pool = pt->pool; |
| 2768 | |
| 2769 | switch (type) { |
| 2770 | case STATUSTYPE_INFO: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2771 | if (get_pool_mode(pool) == PM_FAIL) { |
| 2772 | DMEMIT("Fail"); |
| 2773 | break; |
| 2774 | } |
| 2775 | |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2776 | /* Commit to ensure statistics aren't out-of-date */ |
| 2777 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2778 | (void) commit(pool); |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2779 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2780 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
| 2781 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2782 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
| 2783 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2784 | goto err; |
| 2785 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2786 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2787 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
| 2788 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2789 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
| 2790 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2791 | goto err; |
| 2792 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2793 | |
| 2794 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2795 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2796 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
| 2797 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2798 | goto err; |
| 2799 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2800 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2801 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
| 2802 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2803 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
| 2804 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2805 | goto err; |
| 2806 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2807 | |
| 2808 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2809 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2810 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
| 2811 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2812 | goto err; |
| 2813 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2814 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2815 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2816 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2817 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
| 2818 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2819 | goto err; |
| 2820 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2821 | |
| 2822 | DMEMIT("%llu %llu/%llu %llu/%llu ", |
| 2823 | (unsigned long long)transaction_id, |
| 2824 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
| 2825 | (unsigned long long)nr_blocks_metadata, |
| 2826 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), |
| 2827 | (unsigned long long)nr_blocks_data); |
| 2828 | |
| 2829 | if (held_root) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2830 | DMEMIT("%llu ", held_root); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2831 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2832 | DMEMIT("- "); |
| 2833 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2834 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
| 2835 | DMEMIT("out_of_data_space "); |
| 2836 | else if (pool->pf.mode == PM_READ_ONLY) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2837 | DMEMIT("ro "); |
| 2838 | else |
| 2839 | DMEMIT("rw "); |
| 2840 | |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 2841 | if (!pool->pf.discard_enabled) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2842 | DMEMIT("ignore_discard "); |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 2843 | else if (pool->pf.discard_passdown) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2844 | DMEMIT("discard_passdown "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2845 | else |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2846 | DMEMIT("no_discard_passdown "); |
| 2847 | |
| 2848 | if (pool->pf.error_if_no_space) |
| 2849 | DMEMIT("error_if_no_space "); |
| 2850 | else |
| 2851 | DMEMIT("queue_if_no_space "); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2852 | |
| 2853 | break; |
| 2854 | |
| 2855 | case STATUSTYPE_TABLE: |
| 2856 | DMEMIT("%s %s %lu %llu ", |
| 2857 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), |
| 2858 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), |
| 2859 | (unsigned long)pool->sectors_per_block, |
| 2860 | (unsigned long long)pt->low_water_blocks); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2861 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2862 | break; |
| 2863 | } |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2864 | return; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2865 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2866 | err: |
| 2867 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2868 | } |
| 2869 | |
| 2870 | static int pool_iterate_devices(struct dm_target *ti, |
| 2871 | iterate_devices_callout_fn fn, void *data) |
| 2872 | { |
| 2873 | struct pool_c *pt = ti->private; |
| 2874 | |
| 2875 | return fn(ti, pt->data_dev, 0, ti->len, data); |
| 2876 | } |
| 2877 | |
| 2878 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
| 2879 | struct bio_vec *biovec, int max_size) |
| 2880 | { |
| 2881 | struct pool_c *pt = ti->private; |
| 2882 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 2883 | |
| 2884 | if (!q->merge_bvec_fn) |
| 2885 | return max_size; |
| 2886 | |
| 2887 | bvm->bi_bdev = pt->data_dev->bdev; |
| 2888 | |
| 2889 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); |
| 2890 | } |
| 2891 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2892 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2893 | { |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2894 | struct pool *pool = pt->pool; |
| 2895 | struct queue_limits *data_limits; |
| 2896 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2897 | limits->max_discard_sectors = pool->sectors_per_block; |
| 2898 | |
| 2899 | /* |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2900 | * discard_granularity is just a hint, and not enforced. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2901 | */ |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2902 | if (pt->adjusted_pf.discard_passdown) { |
| 2903 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; |
| 2904 | limits->discard_granularity = data_limits->discard_granularity; |
Mike Snitzer | f13945d | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2905 | } else |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2906 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2907 | } |
| 2908 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2909 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 2910 | { |
| 2911 | struct pool_c *pt = ti->private; |
| 2912 | struct pool *pool = pt->pool; |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 2913 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2914 | |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 2915 | /* |
| 2916 | * If the system-determined stacked limits are compatible with the |
| 2917 | * pool's blocksize (io_opt is a factor) do not override them. |
| 2918 | */ |
| 2919 | if (io_opt_sectors < pool->sectors_per_block || |
| 2920 | do_div(io_opt_sectors, pool->sectors_per_block)) { |
| 2921 | blk_limits_io_min(limits, 0); |
| 2922 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
| 2923 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2924 | |
| 2925 | /* |
| 2926 | * pt->adjusted_pf is a staging area for the actual features to use. |
| 2927 | * They get transferred to the live pool in bind_control_target() |
| 2928 | * called from pool_preresume(). |
| 2929 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 2930 | if (!pt->adjusted_pf.discard_enabled) { |
| 2931 | /* |
| 2932 | * Must explicitly disallow stacking discard limits otherwise the |
| 2933 | * block layer will stack them if pool's data device has support. |
| 2934 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the |
| 2935 | * user to see that, so make sure to set all discard limits to 0. |
| 2936 | */ |
| 2937 | limits->discard_granularity = 0; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2938 | return; |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 2939 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2940 | |
| 2941 | disable_passdown_if_not_supported(pt); |
| 2942 | |
| 2943 | set_discard_limits(pt, limits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2944 | } |
| 2945 | |
| 2946 | static struct target_type pool_target = { |
| 2947 | .name = "thin-pool", |
| 2948 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | |
| 2949 | DM_TARGET_IMMUTABLE, |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2950 | .version = {1, 11, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2951 | .module = THIS_MODULE, |
| 2952 | .ctr = pool_ctr, |
| 2953 | .dtr = pool_dtr, |
| 2954 | .map = pool_map, |
| 2955 | .postsuspend = pool_postsuspend, |
| 2956 | .preresume = pool_preresume, |
| 2957 | .resume = pool_resume, |
| 2958 | .message = pool_message, |
| 2959 | .status = pool_status, |
| 2960 | .merge = pool_merge, |
| 2961 | .iterate_devices = pool_iterate_devices, |
| 2962 | .io_hints = pool_io_hints, |
| 2963 | }; |
| 2964 | |
| 2965 | /*---------------------------------------------------------------- |
| 2966 | * Thin target methods |
| 2967 | *--------------------------------------------------------------*/ |
| 2968 | static void thin_dtr(struct dm_target *ti) |
| 2969 | { |
| 2970 | struct thin_c *tc = ti->private; |
| 2971 | |
| 2972 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2973 | |
| 2974 | __pool_dec(tc->pool); |
| 2975 | dm_pool_close_thin_device(tc->td); |
| 2976 | dm_put_device(ti, tc->pool_dev); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2977 | if (tc->origin_dev) |
| 2978 | dm_put_device(ti, tc->origin_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2979 | kfree(tc); |
| 2980 | |
| 2981 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2982 | } |
| 2983 | |
| 2984 | /* |
| 2985 | * Thin target parameters: |
| 2986 | * |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2987 | * <pool_dev> <dev_id> [origin_dev] |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2988 | * |
| 2989 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) |
| 2990 | * dev_id: the internal device identifier |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2991 | * 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] | 2992 | * |
| 2993 | * If the pool device has discards disabled, they get disabled for the thin |
| 2994 | * device as well. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2995 | */ |
| 2996 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 2997 | { |
| 2998 | int r; |
| 2999 | struct thin_c *tc; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3000 | struct dm_dev *pool_dev, *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3001 | struct mapped_device *pool_md; |
| 3002 | |
| 3003 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3004 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3005 | if (argc != 2 && argc != 3) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3006 | ti->error = "Invalid argument count"; |
| 3007 | r = -EINVAL; |
| 3008 | goto out_unlock; |
| 3009 | } |
| 3010 | |
| 3011 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); |
| 3012 | if (!tc) { |
| 3013 | ti->error = "Out of memory"; |
| 3014 | r = -ENOMEM; |
| 3015 | goto out_unlock; |
| 3016 | } |
| 3017 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3018 | if (argc == 3) { |
| 3019 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
| 3020 | if (r) { |
| 3021 | ti->error = "Error opening origin device"; |
| 3022 | goto bad_origin_dev; |
| 3023 | } |
| 3024 | tc->origin_dev = origin_dev; |
| 3025 | } |
| 3026 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3027 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
| 3028 | if (r) { |
| 3029 | ti->error = "Error opening pool device"; |
| 3030 | goto bad_pool_dev; |
| 3031 | } |
| 3032 | tc->pool_dev = pool_dev; |
| 3033 | |
| 3034 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { |
| 3035 | ti->error = "Invalid device id"; |
| 3036 | r = -EINVAL; |
| 3037 | goto bad_common; |
| 3038 | } |
| 3039 | |
| 3040 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); |
| 3041 | if (!pool_md) { |
| 3042 | ti->error = "Couldn't get pool mapped device"; |
| 3043 | r = -EINVAL; |
| 3044 | goto bad_common; |
| 3045 | } |
| 3046 | |
| 3047 | tc->pool = __pool_table_lookup(pool_md); |
| 3048 | if (!tc->pool) { |
| 3049 | ti->error = "Couldn't find pool object"; |
| 3050 | r = -EINVAL; |
| 3051 | goto bad_pool_lookup; |
| 3052 | } |
| 3053 | __pool_inc(tc->pool); |
| 3054 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3055 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 3056 | ti->error = "Couldn't open thin device, Pool is in fail mode"; |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3057 | r = -EINVAL; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3058 | goto bad_thin_open; |
| 3059 | } |
| 3060 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3061 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
| 3062 | if (r) { |
| 3063 | ti->error = "Couldn't open thin internal device"; |
| 3064 | goto bad_thin_open; |
| 3065 | } |
| 3066 | |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 3067 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
| 3068 | if (r) |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3069 | goto bad_target_max_io_len; |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 3070 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3071 | ti->num_flush_bios = 1; |
Joe Thornber | 16ad3d1 | 2012-07-27 15:08:07 +0100 | [diff] [blame] | 3072 | ti->flush_supported = true; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 3073 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3074 | |
| 3075 | /* In case the pool supports discards, pass them on. */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3076 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3077 | if (tc->pool->pf.discard_enabled) { |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 3078 | ti->discards_supported = true; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3079 | ti->num_discard_bios = 1; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3080 | /* Discard bios must be split on a block boundary */ |
| 3081 | ti->split_discard_bios = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3082 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3083 | |
| 3084 | dm_put(pool_md); |
| 3085 | |
| 3086 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3087 | |
| 3088 | return 0; |
| 3089 | |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3090 | bad_target_max_io_len: |
| 3091 | dm_pool_close_thin_device(tc->td); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3092 | bad_thin_open: |
| 3093 | __pool_dec(tc->pool); |
| 3094 | bad_pool_lookup: |
| 3095 | dm_put(pool_md); |
| 3096 | bad_common: |
| 3097 | dm_put_device(ti, tc->pool_dev); |
| 3098 | bad_pool_dev: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3099 | if (tc->origin_dev) |
| 3100 | dm_put_device(ti, tc->origin_dev); |
| 3101 | bad_origin_dev: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3102 | kfree(tc); |
| 3103 | out_unlock: |
| 3104 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3105 | |
| 3106 | return r; |
| 3107 | } |
| 3108 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3109 | static int thin_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3110 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 3111 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3112 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3113 | return thin_bio_map(ti, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3114 | } |
| 3115 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3116 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3117 | { |
| 3118 | unsigned long flags; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 3119 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3120 | struct list_head work; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3121 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3122 | struct pool *pool = h->tc->pool; |
| 3123 | |
| 3124 | if (h->shared_read_entry) { |
| 3125 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 3126 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3127 | |
| 3128 | spin_lock_irqsave(&pool->lock, flags); |
| 3129 | list_for_each_entry_safe(m, tmp, &work, list) { |
| 3130 | list_del(&m->list); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 3131 | m->quiesced = true; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3132 | __maybe_add_mapping(m); |
| 3133 | } |
| 3134 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3135 | } |
| 3136 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3137 | if (h->all_io_entry) { |
| 3138 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 3139 | dm_deferred_entry_dec(h->all_io_entry, &work); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 3140 | if (!list_empty(&work)) { |
| 3141 | spin_lock_irqsave(&pool->lock, flags); |
| 3142 | list_for_each_entry_safe(m, tmp, &work, list) |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 3143 | list_add_tail(&m->list, &pool->prepared_discards); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 3144 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3145 | wake_worker(pool); |
| 3146 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3147 | } |
| 3148 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3149 | return 0; |
| 3150 | } |
| 3151 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3152 | static void thin_presuspend(struct dm_target *ti) |
| 3153 | { |
| 3154 | struct thin_c *tc = ti->private; |
| 3155 | |
| 3156 | if (dm_noflush_suspending(ti)) |
| 3157 | noflush_work(tc, do_noflush_start); |
| 3158 | } |
| 3159 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3160 | static void thin_postsuspend(struct dm_target *ti) |
| 3161 | { |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3162 | struct thin_c *tc = ti->private; |
| 3163 | |
| 3164 | /* |
| 3165 | * The dm_noflush_suspending flag has been cleared by now, so |
| 3166 | * unfortunately we must always run this. |
| 3167 | */ |
| 3168 | noflush_work(tc, do_noflush_stop); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3169 | } |
| 3170 | |
| 3171 | /* |
| 3172 | * <nr mapped sectors> <highest mapped sector> |
| 3173 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3174 | static void thin_status(struct dm_target *ti, status_type_t type, |
| 3175 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3176 | { |
| 3177 | int r; |
| 3178 | ssize_t sz = 0; |
| 3179 | dm_block_t mapped, highest; |
| 3180 | char buf[BDEVNAME_SIZE]; |
| 3181 | struct thin_c *tc = ti->private; |
| 3182 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3183 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 3184 | DMEMIT("Fail"); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3185 | return; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3186 | } |
| 3187 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3188 | if (!tc->td) |
| 3189 | DMEMIT("-"); |
| 3190 | else { |
| 3191 | switch (type) { |
| 3192 | case STATUSTYPE_INFO: |
| 3193 | r = dm_thin_get_mapped_count(tc->td, &mapped); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3194 | if (r) { |
| 3195 | DMERR("dm_thin_get_mapped_count returned %d", r); |
| 3196 | goto err; |
| 3197 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3198 | |
| 3199 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3200 | if (r < 0) { |
| 3201 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); |
| 3202 | goto err; |
| 3203 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3204 | |
| 3205 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); |
| 3206 | if (r) |
| 3207 | DMEMIT("%llu", ((highest + 1) * |
| 3208 | tc->pool->sectors_per_block) - 1); |
| 3209 | else |
| 3210 | DMEMIT("-"); |
| 3211 | break; |
| 3212 | |
| 3213 | case STATUSTYPE_TABLE: |
| 3214 | DMEMIT("%s %lu", |
| 3215 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), |
| 3216 | (unsigned long) tc->dev_id); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3217 | if (tc->origin_dev) |
| 3218 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3219 | break; |
| 3220 | } |
| 3221 | } |
| 3222 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3223 | return; |
| 3224 | |
| 3225 | err: |
| 3226 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3227 | } |
| 3228 | |
| 3229 | static int thin_iterate_devices(struct dm_target *ti, |
| 3230 | iterate_devices_callout_fn fn, void *data) |
| 3231 | { |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3232 | sector_t blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3233 | struct thin_c *tc = ti->private; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3234 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3235 | |
| 3236 | /* |
| 3237 | * We can't call dm_pool_get_data_dev_size() since that blocks. So |
| 3238 | * we follow a more convoluted path through to the pool's target. |
| 3239 | */ |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3240 | if (!pool->ti) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3241 | return 0; /* nothing is bound */ |
| 3242 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3243 | blocks = pool->ti->len; |
| 3244 | (void) sector_div(blocks, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3245 | if (blocks) |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3246 | 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] | 3247 | |
| 3248 | return 0; |
| 3249 | } |
| 3250 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3251 | static struct target_type thin_target = { |
| 3252 | .name = "thin", |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 3253 | .version = {1, 11, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3254 | .module = THIS_MODULE, |
| 3255 | .ctr = thin_ctr, |
| 3256 | .dtr = thin_dtr, |
| 3257 | .map = thin_map, |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3258 | .end_io = thin_endio, |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3259 | .presuspend = thin_presuspend, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3260 | .postsuspend = thin_postsuspend, |
| 3261 | .status = thin_status, |
| 3262 | .iterate_devices = thin_iterate_devices, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3263 | }; |
| 3264 | |
| 3265 | /*----------------------------------------------------------------*/ |
| 3266 | |
| 3267 | static int __init dm_thin_init(void) |
| 3268 | { |
| 3269 | int r; |
| 3270 | |
| 3271 | pool_table_init(); |
| 3272 | |
| 3273 | r = dm_register_target(&thin_target); |
| 3274 | if (r) |
| 3275 | return r; |
| 3276 | |
| 3277 | r = dm_register_target(&pool_target); |
| 3278 | if (r) |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3279 | goto bad_pool_target; |
| 3280 | |
| 3281 | r = -ENOMEM; |
| 3282 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3283 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
| 3284 | if (!_new_mapping_cache) |
| 3285 | goto bad_new_mapping_cache; |
| 3286 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3287 | return 0; |
| 3288 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3289 | bad_new_mapping_cache: |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3290 | dm_unregister_target(&pool_target); |
| 3291 | bad_pool_target: |
| 3292 | dm_unregister_target(&thin_target); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3293 | |
| 3294 | return r; |
| 3295 | } |
| 3296 | |
| 3297 | static void dm_thin_exit(void) |
| 3298 | { |
| 3299 | dm_unregister_target(&thin_target); |
| 3300 | dm_unregister_target(&pool_target); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3301 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3302 | kmem_cache_destroy(_new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3303 | } |
| 3304 | |
| 3305 | module_init(dm_thin_init); |
| 3306 | module_exit(dm_thin_exit); |
| 3307 | |
Alasdair G Kergon | 7cab8bf | 2012-05-12 01:43:19 +0100 | [diff] [blame] | 3308 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3309 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 3310 | MODULE_LICENSE("GPL"); |