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> |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 15 | #include <linux/rculist.h> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 16 | #include <linux/init.h> |
| 17 | #include <linux/module.h> |
| 18 | #include <linux/slab.h> |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 19 | #include <linux/rbtree.h> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 20 | |
| 21 | #define DM_MSG_PREFIX "thin" |
| 22 | |
| 23 | /* |
| 24 | * Tunable constants |
| 25 | */ |
Alasdair G Kergon | 7768ed3 | 2012-07-27 15:07:57 +0100 | [diff] [blame] | 26 | #define ENDIO_HOOK_POOL_SIZE 1024 |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 27 | #define MAPPING_POOL_SIZE 1024 |
| 28 | #define PRISON_CELLS 1024 |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 29 | #define COMMIT_PERIOD HZ |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 30 | #define NO_SPACE_TIMEOUT_SECS 60 |
| 31 | |
| 32 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 33 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 34 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
| 35 | "A percentage of time allocated for copy on write"); |
| 36 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 37 | /* |
| 38 | * The block size of the device holding pool data must be |
| 39 | * between 64KB and 1GB. |
| 40 | */ |
| 41 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) |
| 42 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
| 43 | |
| 44 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 45 | * Device id is restricted to 24 bits. |
| 46 | */ |
| 47 | #define MAX_DEV_ID ((1 << 24) - 1) |
| 48 | |
| 49 | /* |
| 50 | * How do we handle breaking sharing of data blocks? |
| 51 | * ================================================= |
| 52 | * |
| 53 | * We use a standard copy-on-write btree to store the mappings for the |
| 54 | * devices (note I'm talking about copy-on-write of the metadata here, not |
| 55 | * the data). When you take an internal snapshot you clone the root node |
| 56 | * of the origin btree. After this there is no concept of an origin or a |
| 57 | * snapshot. They are just two device trees that happen to point to the |
| 58 | * same data blocks. |
| 59 | * |
| 60 | * When we get a write in we decide if it's to a shared data block using |
| 61 | * some timestamp magic. If it is, we have to break sharing. |
| 62 | * |
| 63 | * Let's say we write to a shared block in what was the origin. The |
| 64 | * steps are: |
| 65 | * |
| 66 | * i) plug io further to this physical block. (see bio_prison code). |
| 67 | * |
| 68 | * ii) quiesce any read io to that shared data block. Obviously |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 69 | * including all devices that share this block. (see dm_deferred_set code) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 70 | * |
| 71 | * iii) copy the data block to a newly allocate block. This step can be |
| 72 | * missed out if the io covers the block. (schedule_copy). |
| 73 | * |
| 74 | * iv) insert the new mapping into the origin's btree |
Joe Thornber | fe878f3 | 2012-03-28 18:41:24 +0100 | [diff] [blame] | 75 | * (process_prepared_mapping). This act of inserting breaks some |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 76 | * sharing of btree nodes between the two devices. Breaking sharing only |
| 77 | * effects the btree of that specific device. Btrees for the other |
| 78 | * devices that share the block never change. The btree for the origin |
| 79 | * device as it was after the last commit is untouched, ie. we're using |
| 80 | * persistent data structures in the functional programming sense. |
| 81 | * |
| 82 | * v) unplug io to this physical block, including the io that triggered |
| 83 | * the breaking of sharing. |
| 84 | * |
| 85 | * Steps (ii) and (iii) occur in parallel. |
| 86 | * |
| 87 | * The metadata _doesn't_ need to be committed before the io continues. We |
| 88 | * get away with this because the io is always written to a _new_ block. |
| 89 | * If there's a crash, then: |
| 90 | * |
| 91 | * - The origin mapping will point to the old origin block (the shared |
| 92 | * one). This will contain the data as it was before the io that triggered |
| 93 | * the breaking of sharing came in. |
| 94 | * |
| 95 | * - The snap mapping still points to the old block. As it would after |
| 96 | * the commit. |
| 97 | * |
| 98 | * The downside of this scheme is the timestamp magic isn't perfect, and |
| 99 | * will continue to think that data block in the snapshot device is shared |
| 100 | * even after the write to the origin has broken sharing. I suspect data |
| 101 | * blocks will typically be shared by many different devices, so we're |
| 102 | * breaking sharing n + 1 times, rather than n, where n is the number of |
| 103 | * devices that reference this data block. At the moment I think the |
| 104 | * benefits far, far outweigh the disadvantages. |
| 105 | */ |
| 106 | |
| 107 | /*----------------------------------------------------------------*/ |
| 108 | |
| 109 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 110 | * Key building. |
| 111 | */ |
| 112 | static void build_data_key(struct dm_thin_device *td, |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 113 | dm_block_t b, struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 114 | { |
| 115 | key->virtual = 0; |
| 116 | key->dev = dm_thin_dev_id(td); |
| 117 | key->block = b; |
| 118 | } |
| 119 | |
| 120 | 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] | 121 | struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 122 | { |
| 123 | key->virtual = 1; |
| 124 | key->dev = dm_thin_dev_id(td); |
| 125 | key->block = b; |
| 126 | } |
| 127 | |
| 128 | /*----------------------------------------------------------------*/ |
| 129 | |
| 130 | /* |
| 131 | * A pool device ties together a metadata device and a data device. It |
| 132 | * also provides the interface for creating and destroying internal |
| 133 | * devices. |
| 134 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 135 | struct dm_thin_new_mapping; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 136 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 137 | /* |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 138 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 139 | */ |
| 140 | enum pool_mode { |
| 141 | PM_WRITE, /* metadata may be changed */ |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 142 | 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] | 143 | PM_READ_ONLY, /* metadata may not be changed */ |
| 144 | PM_FAIL, /* all I/O fails */ |
| 145 | }; |
| 146 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 147 | struct pool_features { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 148 | enum pool_mode mode; |
| 149 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 150 | bool zero_new_blocks:1; |
| 151 | bool discard_enabled:1; |
| 152 | bool discard_passdown:1; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 153 | bool error_if_no_space:1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 154 | }; |
| 155 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 156 | struct thin_c; |
| 157 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); |
| 158 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
| 159 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 160 | struct pool { |
| 161 | struct list_head list; |
| 162 | struct dm_target *ti; /* Only set if a pool target is bound */ |
| 163 | |
| 164 | struct mapped_device *pool_md; |
| 165 | struct block_device *md_dev; |
| 166 | struct dm_pool_metadata *pmd; |
| 167 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 168 | dm_block_t low_water_blocks; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 169 | uint32_t sectors_per_block; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 170 | int sectors_per_block_shift; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 171 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 172 | struct pool_features pf; |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 173 | bool low_water_triggered:1; /* A dm event has been sent */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 174 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 175 | struct dm_bio_prison *prison; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 176 | struct dm_kcopyd_client *copier; |
| 177 | |
| 178 | struct workqueue_struct *wq; |
| 179 | struct work_struct worker; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 180 | struct delayed_work waker; |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 181 | struct delayed_work no_space_timeout; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 182 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 183 | unsigned long last_commit_jiffies; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 184 | unsigned ref_count; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 185 | |
| 186 | spinlock_t lock; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 187 | struct bio_list deferred_flush_bios; |
| 188 | struct list_head prepared_mappings; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 189 | struct list_head prepared_discards; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 190 | struct list_head active_thins; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 191 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 192 | struct dm_deferred_set *shared_read_ds; |
| 193 | struct dm_deferred_set *all_io_ds; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 194 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 195 | struct dm_thin_new_mapping *next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 196 | mempool_t *mapping_pool; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 197 | |
| 198 | process_bio_fn process_bio; |
| 199 | process_bio_fn process_discard; |
| 200 | |
| 201 | process_mapping_fn process_prepared_mapping; |
| 202 | process_mapping_fn process_prepared_discard; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 203 | }; |
| 204 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 205 | static enum pool_mode get_pool_mode(struct pool *pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 206 | 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] | 207 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 208 | /* |
| 209 | * Target context for a pool. |
| 210 | */ |
| 211 | struct pool_c { |
| 212 | struct dm_target *ti; |
| 213 | struct pool *pool; |
| 214 | struct dm_dev *data_dev; |
| 215 | struct dm_dev *metadata_dev; |
| 216 | struct dm_target_callbacks callbacks; |
| 217 | |
| 218 | dm_block_t low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 219 | struct pool_features requested_pf; /* Features requested during table load */ |
| 220 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 221 | }; |
| 222 | |
| 223 | /* |
| 224 | * Target context for a thin. |
| 225 | */ |
| 226 | struct thin_c { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 227 | struct list_head list; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 228 | struct dm_dev *pool_dev; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 229 | struct dm_dev *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 230 | dm_thin_id dev_id; |
| 231 | |
| 232 | struct pool *pool; |
| 233 | struct dm_thin_device *td; |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 234 | bool requeue_mode:1; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 235 | spinlock_t lock; |
| 236 | struct bio_list deferred_bio_list; |
| 237 | struct bio_list retry_on_resume_list; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 238 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 239 | |
| 240 | /* |
| 241 | * Ensures the thin is not destroyed until the worker has finished |
| 242 | * iterating the active_thins list. |
| 243 | */ |
| 244 | atomic_t refcount; |
| 245 | struct completion can_destroy; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 246 | }; |
| 247 | |
| 248 | /*----------------------------------------------------------------*/ |
| 249 | |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 250 | /* |
| 251 | * wake_worker() is used when new work is queued and when pool_resume is |
| 252 | * ready to continue deferred IO processing. |
| 253 | */ |
| 254 | static void wake_worker(struct pool *pool) |
| 255 | { |
| 256 | queue_work(pool->wq, &pool->worker); |
| 257 | } |
| 258 | |
| 259 | /*----------------------------------------------------------------*/ |
| 260 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 261 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
| 262 | struct dm_bio_prison_cell **cell_result) |
| 263 | { |
| 264 | int r; |
| 265 | struct dm_bio_prison_cell *cell_prealloc; |
| 266 | |
| 267 | /* |
| 268 | * Allocate a cell from the prison's mempool. |
| 269 | * This might block but it can't fail. |
| 270 | */ |
| 271 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); |
| 272 | |
| 273 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); |
| 274 | if (r) |
| 275 | /* |
| 276 | * We reused an old cell; we can get rid of |
| 277 | * the new one. |
| 278 | */ |
| 279 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); |
| 280 | |
| 281 | return r; |
| 282 | } |
| 283 | |
| 284 | static void cell_release(struct pool *pool, |
| 285 | struct dm_bio_prison_cell *cell, |
| 286 | struct bio_list *bios) |
| 287 | { |
| 288 | dm_cell_release(pool->prison, cell, bios); |
| 289 | dm_bio_prison_free_cell(pool->prison, cell); |
| 290 | } |
| 291 | |
| 292 | static void cell_release_no_holder(struct pool *pool, |
| 293 | struct dm_bio_prison_cell *cell, |
| 294 | struct bio_list *bios) |
| 295 | { |
| 296 | dm_cell_release_no_holder(pool->prison, cell, bios); |
| 297 | dm_bio_prison_free_cell(pool->prison, cell); |
| 298 | } |
| 299 | |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 300 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
| 301 | struct dm_bio_prison_cell *cell) |
| 302 | { |
| 303 | struct pool *pool = tc->pool; |
| 304 | unsigned long flags; |
| 305 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 306 | spin_lock_irqsave(&tc->lock, flags); |
| 307 | dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list); |
| 308 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 309 | |
| 310 | wake_worker(pool); |
| 311 | } |
| 312 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 313 | static void cell_error_with_code(struct pool *pool, |
| 314 | struct dm_bio_prison_cell *cell, int error_code) |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 315 | { |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 316 | dm_cell_error(pool->prison, cell, error_code); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 317 | dm_bio_prison_free_cell(pool->prison, cell); |
| 318 | } |
| 319 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 320 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
| 321 | { |
| 322 | cell_error_with_code(pool, cell, -EIO); |
| 323 | } |
| 324 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 325 | /*----------------------------------------------------------------*/ |
| 326 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 327 | /* |
| 328 | * A global list of pools that uses a struct mapped_device as a key. |
| 329 | */ |
| 330 | static struct dm_thin_pool_table { |
| 331 | struct mutex mutex; |
| 332 | struct list_head pools; |
| 333 | } dm_thin_pool_table; |
| 334 | |
| 335 | static void pool_table_init(void) |
| 336 | { |
| 337 | mutex_init(&dm_thin_pool_table.mutex); |
| 338 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); |
| 339 | } |
| 340 | |
| 341 | static void __pool_table_insert(struct pool *pool) |
| 342 | { |
| 343 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 344 | list_add(&pool->list, &dm_thin_pool_table.pools); |
| 345 | } |
| 346 | |
| 347 | static void __pool_table_remove(struct pool *pool) |
| 348 | { |
| 349 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 350 | list_del(&pool->list); |
| 351 | } |
| 352 | |
| 353 | static struct pool *__pool_table_lookup(struct mapped_device *md) |
| 354 | { |
| 355 | struct pool *pool = NULL, *tmp; |
| 356 | |
| 357 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 358 | |
| 359 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 360 | if (tmp->pool_md == md) { |
| 361 | pool = tmp; |
| 362 | break; |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | return pool; |
| 367 | } |
| 368 | |
| 369 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) |
| 370 | { |
| 371 | struct pool *pool = NULL, *tmp; |
| 372 | |
| 373 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 374 | |
| 375 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 376 | if (tmp->md_dev == md_dev) { |
| 377 | pool = tmp; |
| 378 | break; |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | return pool; |
| 383 | } |
| 384 | |
| 385 | /*----------------------------------------------------------------*/ |
| 386 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 387 | struct dm_thin_endio_hook { |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 388 | struct thin_c *tc; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 389 | struct dm_deferred_entry *shared_read_entry; |
| 390 | struct dm_deferred_entry *all_io_entry; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 391 | struct dm_thin_new_mapping *overwrite_mapping; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 392 | struct rb_node rb_node; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 393 | }; |
| 394 | |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 395 | 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] | 396 | { |
| 397 | struct bio *bio; |
| 398 | struct bio_list bios; |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 399 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 400 | |
| 401 | bio_list_init(&bios); |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 402 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 403 | spin_lock_irqsave(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 404 | bio_list_merge(&bios, master); |
| 405 | bio_list_init(master); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 406 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 407 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 408 | while ((bio = bio_list_pop(&bios))) |
| 409 | bio_endio(bio, DM_ENDIO_REQUEUE); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 410 | } |
| 411 | |
| 412 | static void requeue_io(struct thin_c *tc) |
| 413 | { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 414 | requeue_bio_list(tc, &tc->deferred_bio_list); |
| 415 | requeue_bio_list(tc, &tc->retry_on_resume_list); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 416 | } |
| 417 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 418 | static void error_thin_retry_list(struct thin_c *tc) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 419 | { |
| 420 | struct bio *bio; |
| 421 | unsigned long flags; |
| 422 | struct bio_list bios; |
| 423 | |
| 424 | bio_list_init(&bios); |
| 425 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 426 | spin_lock_irqsave(&tc->lock, flags); |
| 427 | bio_list_merge(&bios, &tc->retry_on_resume_list); |
| 428 | bio_list_init(&tc->retry_on_resume_list); |
| 429 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 430 | |
| 431 | while ((bio = bio_list_pop(&bios))) |
| 432 | bio_io_error(bio); |
| 433 | } |
| 434 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 435 | static void error_retry_list(struct pool *pool) |
| 436 | { |
| 437 | struct thin_c *tc; |
| 438 | |
| 439 | rcu_read_lock(); |
| 440 | list_for_each_entry_rcu(tc, &pool->active_thins, list) |
| 441 | error_thin_retry_list(tc); |
| 442 | rcu_read_unlock(); |
| 443 | } |
| 444 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 445 | /* |
| 446 | * This section of code contains the logic for processing a thin device's IO. |
| 447 | * Much of the code depends on pool object resources (lists, workqueues, etc) |
| 448 | * but most is exclusively called from the thin target rather than the thin-pool |
| 449 | * target. |
| 450 | */ |
| 451 | |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 452 | static bool block_size_is_power_of_two(struct pool *pool) |
| 453 | { |
| 454 | return pool->sectors_per_block_shift >= 0; |
| 455 | } |
| 456 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 457 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
| 458 | { |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 459 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 460 | sector_t block_nr = bio->bi_iter.bi_sector; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 461 | |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 462 | if (block_size_is_power_of_two(pool)) |
| 463 | block_nr >>= pool->sectors_per_block_shift; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 464 | else |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 465 | (void) sector_div(block_nr, pool->sectors_per_block); |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 466 | |
| 467 | return block_nr; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 468 | } |
| 469 | |
| 470 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
| 471 | { |
| 472 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 473 | sector_t bi_sector = bio->bi_iter.bi_sector; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 474 | |
| 475 | bio->bi_bdev = tc->pool_dev->bdev; |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 476 | if (block_size_is_power_of_two(pool)) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 477 | bio->bi_iter.bi_sector = |
| 478 | (block << pool->sectors_per_block_shift) | |
| 479 | (bi_sector & (pool->sectors_per_block - 1)); |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 480 | else |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 481 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 482 | sector_div(bi_sector, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 483 | } |
| 484 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 485 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
| 486 | { |
| 487 | bio->bi_bdev = tc->origin_dev->bdev; |
| 488 | } |
| 489 | |
Joe Thornber | 4afdd68 | 2012-07-27 15:08:14 +0100 | [diff] [blame] | 490 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
| 491 | { |
| 492 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && |
| 493 | dm_thin_changed_this_transaction(tc->td); |
| 494 | } |
| 495 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 496 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
| 497 | { |
| 498 | struct dm_thin_endio_hook *h; |
| 499 | |
| 500 | if (bio->bi_rw & REQ_DISCARD) |
| 501 | return; |
| 502 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 503 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 504 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
| 505 | } |
| 506 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 507 | static void issue(struct thin_c *tc, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 508 | { |
| 509 | struct pool *pool = tc->pool; |
| 510 | unsigned long flags; |
| 511 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 512 | if (!bio_triggers_commit(tc, bio)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 513 | generic_make_request(bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 514 | return; |
| 515 | } |
| 516 | |
| 517 | /* |
| 518 | * Complete bio with an error if earlier I/O caused changes to |
| 519 | * the metadata that can't be committed e.g, due to I/O errors |
| 520 | * on the metadata device. |
| 521 | */ |
| 522 | if (dm_thin_aborted_changes(tc->td)) { |
| 523 | bio_io_error(bio); |
| 524 | return; |
| 525 | } |
| 526 | |
| 527 | /* |
| 528 | * Batch together any bios that trigger commits and then issue a |
| 529 | * single commit for them in process_deferred_bios(). |
| 530 | */ |
| 531 | spin_lock_irqsave(&pool->lock, flags); |
| 532 | bio_list_add(&pool->deferred_flush_bios, bio); |
| 533 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 534 | } |
| 535 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 536 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
| 537 | { |
| 538 | remap_to_origin(tc, bio); |
| 539 | issue(tc, bio); |
| 540 | } |
| 541 | |
| 542 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, |
| 543 | dm_block_t block) |
| 544 | { |
| 545 | remap(tc, bio, block); |
| 546 | issue(tc, bio); |
| 547 | } |
| 548 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 549 | /*----------------------------------------------------------------*/ |
| 550 | |
| 551 | /* |
| 552 | * Bio endio functions. |
| 553 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 554 | struct dm_thin_new_mapping { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 555 | struct list_head list; |
| 556 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 557 | bool quiesced:1; |
| 558 | bool prepared:1; |
| 559 | bool pass_discard:1; |
| 560 | bool definitely_not_shared:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 561 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 562 | int err; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 563 | struct thin_c *tc; |
| 564 | dm_block_t virt_block; |
| 565 | dm_block_t data_block; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 566 | struct dm_bio_prison_cell *cell, *cell2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 567 | |
| 568 | /* |
| 569 | * If the bio covers the whole area of a block then we can avoid |
| 570 | * zeroing or copying. Instead this bio is hooked. The bio will |
| 571 | * still be in the cell, so care has to be taken to avoid issuing |
| 572 | * the bio twice. |
| 573 | */ |
| 574 | struct bio *bio; |
| 575 | bio_end_io_t *saved_bi_end_io; |
| 576 | }; |
| 577 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 578 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 579 | { |
| 580 | struct pool *pool = m->tc->pool; |
| 581 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 582 | if (m->quiesced && m->prepared) { |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 583 | list_add_tail(&m->list, &pool->prepared_mappings); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 584 | wake_worker(pool); |
| 585 | } |
| 586 | } |
| 587 | |
| 588 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
| 589 | { |
| 590 | unsigned long flags; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 591 | struct dm_thin_new_mapping *m = context; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 592 | struct pool *pool = m->tc->pool; |
| 593 | |
| 594 | m->err = read_err || write_err ? -EIO : 0; |
| 595 | |
| 596 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 597 | m->prepared = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 598 | __maybe_add_mapping(m); |
| 599 | spin_unlock_irqrestore(&pool->lock, flags); |
| 600 | } |
| 601 | |
| 602 | static void overwrite_endio(struct bio *bio, int err) |
| 603 | { |
| 604 | unsigned long flags; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 605 | 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] | 606 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 607 | struct pool *pool = m->tc->pool; |
| 608 | |
| 609 | m->err = err; |
| 610 | |
| 611 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 612 | m->prepared = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 613 | __maybe_add_mapping(m); |
| 614 | spin_unlock_irqrestore(&pool->lock, flags); |
| 615 | } |
| 616 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 617 | /*----------------------------------------------------------------*/ |
| 618 | |
| 619 | /* |
| 620 | * Workqueue. |
| 621 | */ |
| 622 | |
| 623 | /* |
| 624 | * Prepared mapping jobs. |
| 625 | */ |
| 626 | |
| 627 | /* |
| 628 | * This sends the bios in the cell back to the deferred_bios list. |
| 629 | */ |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 630 | 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] | 631 | { |
| 632 | struct pool *pool = tc->pool; |
| 633 | unsigned long flags; |
| 634 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 635 | spin_lock_irqsave(&tc->lock, flags); |
| 636 | cell_release(pool, cell, &tc->deferred_bio_list); |
| 637 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 638 | |
| 639 | wake_worker(pool); |
| 640 | } |
| 641 | |
| 642 | /* |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 643 | * 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] | 644 | */ |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 645 | 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] | 646 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 647 | struct pool *pool = tc->pool; |
| 648 | unsigned long flags; |
| 649 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 650 | spin_lock_irqsave(&tc->lock, flags); |
| 651 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); |
| 652 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 653 | |
| 654 | wake_worker(pool); |
| 655 | } |
| 656 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 657 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
| 658 | { |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 659 | if (m->bio) { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 660 | m->bio->bi_end_io = m->saved_bi_end_io; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 661 | atomic_inc(&m->bio->bi_remaining); |
| 662 | } |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 663 | cell_error(m->tc->pool, m->cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 664 | list_del(&m->list); |
| 665 | mempool_free(m, m->tc->pool->mapping_pool); |
| 666 | } |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 667 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 668 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 669 | { |
| 670 | struct thin_c *tc = m->tc; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 671 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 672 | struct bio *bio; |
| 673 | int r; |
| 674 | |
| 675 | bio = m->bio; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 676 | if (bio) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 677 | bio->bi_end_io = m->saved_bi_end_io; |
Kent Overstreet | 196d38b | 2013-11-23 18:34:15 -0800 | [diff] [blame] | 678 | atomic_inc(&bio->bi_remaining); |
| 679 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 680 | |
| 681 | if (m->err) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 682 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 683 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 684 | } |
| 685 | |
| 686 | /* |
| 687 | * Commit the prepared block into the mapping btree. |
| 688 | * Any I/O for this block arriving after this point will get |
| 689 | * remapped to it directly. |
| 690 | */ |
| 691 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); |
| 692 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 693 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 694 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 695 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 696 | } |
| 697 | |
| 698 | /* |
| 699 | * Release any bios held while the block was being provisioned. |
| 700 | * If we are processing a write bio that completely covers the block, |
| 701 | * we already processed it so can ignore it now when processing |
| 702 | * the bios in the cell. |
| 703 | */ |
| 704 | if (bio) { |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 705 | cell_defer_no_holder(tc, m->cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 706 | bio_endio(bio, 0); |
| 707 | } else |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 708 | cell_defer(tc, m->cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 709 | |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 710 | out: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 711 | list_del(&m->list); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 712 | mempool_free(m, pool->mapping_pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 713 | } |
| 714 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 715 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 716 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 717 | struct thin_c *tc = m->tc; |
| 718 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 719 | bio_io_error(m->bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 720 | cell_defer_no_holder(tc, m->cell); |
| 721 | cell_defer_no_holder(tc, m->cell2); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 722 | mempool_free(m, tc->pool->mapping_pool); |
| 723 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 724 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 725 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) |
| 726 | { |
| 727 | struct thin_c *tc = m->tc; |
| 728 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 729 | inc_all_io_entry(tc->pool, m->bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 730 | cell_defer_no_holder(tc, m->cell); |
| 731 | cell_defer_no_holder(tc, m->cell2); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 732 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 733 | if (m->pass_discard) |
Joe Thornber | 19fa1a6 | 2013-12-17 12:09:40 -0500 | [diff] [blame] | 734 | if (m->definitely_not_shared) |
| 735 | remap_and_issue(tc, m->bio, m->data_block); |
| 736 | else { |
| 737 | bool used = false; |
| 738 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) |
| 739 | bio_endio(m->bio, 0); |
| 740 | else |
| 741 | remap_and_issue(tc, m->bio, m->data_block); |
| 742 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 743 | else |
| 744 | bio_endio(m->bio, 0); |
| 745 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 746 | mempool_free(m, tc->pool->mapping_pool); |
| 747 | } |
| 748 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 749 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
| 750 | { |
| 751 | int r; |
| 752 | struct thin_c *tc = m->tc; |
| 753 | |
| 754 | r = dm_thin_remove_block(tc->td, m->virt_block); |
| 755 | if (r) |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 756 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 757 | |
| 758 | process_prepared_discard_passdown(m); |
| 759 | } |
| 760 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 761 | static void process_prepared(struct pool *pool, struct list_head *head, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 762 | process_mapping_fn *fn) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 763 | { |
| 764 | unsigned long flags; |
| 765 | struct list_head maps; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 766 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 767 | |
| 768 | INIT_LIST_HEAD(&maps); |
| 769 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 770 | list_splice_init(head, &maps); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 771 | spin_unlock_irqrestore(&pool->lock, flags); |
| 772 | |
| 773 | list_for_each_entry_safe(m, tmp, &maps, list) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 774 | (*fn)(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 775 | } |
| 776 | |
| 777 | /* |
| 778 | * Deferred bio jobs. |
| 779 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 780 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
| 781 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 782 | return bio->bi_iter.bi_size == |
| 783 | (pool->sectors_per_block << SECTOR_SHIFT); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 784 | } |
| 785 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 786 | static int io_overwrites_block(struct pool *pool, struct bio *bio) |
| 787 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 788 | return (bio_data_dir(bio) == WRITE) && |
| 789 | io_overlaps_block(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 790 | } |
| 791 | |
| 792 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, |
| 793 | bio_end_io_t *fn) |
| 794 | { |
| 795 | *save = bio->bi_end_io; |
| 796 | bio->bi_end_io = fn; |
| 797 | } |
| 798 | |
| 799 | static int ensure_next_mapping(struct pool *pool) |
| 800 | { |
| 801 | if (pool->next_mapping) |
| 802 | return 0; |
| 803 | |
| 804 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); |
| 805 | |
| 806 | return pool->next_mapping ? 0 : -ENOMEM; |
| 807 | } |
| 808 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 809 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 810 | { |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 811 | struct dm_thin_new_mapping *m = pool->next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 812 | |
| 813 | BUG_ON(!pool->next_mapping); |
| 814 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 815 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
| 816 | INIT_LIST_HEAD(&m->list); |
| 817 | m->bio = NULL; |
| 818 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 819 | pool->next_mapping = NULL; |
| 820 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 821 | return m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 822 | } |
| 823 | |
| 824 | 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] | 825 | struct dm_dev *origin, dm_block_t data_origin, |
| 826 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 827 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 828 | { |
| 829 | int r; |
| 830 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 831 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 832 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 833 | m->tc = tc; |
| 834 | m->virt_block = virt_block; |
| 835 | m->data_block = data_dest; |
| 836 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 837 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 838 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 839 | m->quiesced = true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 840 | |
| 841 | /* |
| 842 | * IO to pool_dev remaps to the pool target's data_dev. |
| 843 | * |
| 844 | * If the whole block of data is being overwritten, we can issue the |
| 845 | * bio immediately. Otherwise we use kcopyd to clone the data first. |
| 846 | */ |
| 847 | if (io_overwrites_block(pool, bio)) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 848 | 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] | 849 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 850 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 851 | m->bio = bio; |
| 852 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 853 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 854 | remap_and_issue(tc, bio, data_dest); |
| 855 | } else { |
| 856 | struct dm_io_region from, to; |
| 857 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 858 | from.bdev = origin->bdev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 859 | from.sector = data_origin * pool->sectors_per_block; |
| 860 | from.count = pool->sectors_per_block; |
| 861 | |
| 862 | to.bdev = tc->pool_dev->bdev; |
| 863 | to.sector = data_dest * pool->sectors_per_block; |
| 864 | to.count = pool->sectors_per_block; |
| 865 | |
| 866 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, |
| 867 | 0, copy_complete, m); |
| 868 | if (r < 0) { |
| 869 | mempool_free(m, pool->mapping_pool); |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 870 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 871 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 872 | } |
| 873 | } |
| 874 | } |
| 875 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 876 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
| 877 | dm_block_t data_origin, dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 878 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 879 | { |
| 880 | schedule_copy(tc, virt_block, tc->pool_dev, |
| 881 | data_origin, data_dest, cell, bio); |
| 882 | } |
| 883 | |
| 884 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
| 885 | dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 886 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 887 | { |
| 888 | schedule_copy(tc, virt_block, tc->origin_dev, |
| 889 | virt_block, data_dest, cell, bio); |
| 890 | } |
| 891 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 892 | 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] | 893 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 894 | struct bio *bio) |
| 895 | { |
| 896 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 897 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 898 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 899 | m->quiesced = true; |
| 900 | m->prepared = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 901 | m->tc = tc; |
| 902 | m->virt_block = virt_block; |
| 903 | m->data_block = data_block; |
| 904 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 905 | |
| 906 | /* |
| 907 | * If the whole block of data is being overwritten or we are not |
| 908 | * zeroing pre-existing data, we can issue the bio immediately. |
| 909 | * Otherwise we use kcopyd to zero the data first. |
| 910 | */ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 911 | if (!pool->pf.zero_new_blocks) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 912 | process_prepared_mapping(m); |
| 913 | |
| 914 | else if (io_overwrites_block(pool, bio)) { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 915 | 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] | 916 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 917 | h->overwrite_mapping = m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 918 | m->bio = bio; |
| 919 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 920 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 921 | remap_and_issue(tc, bio, data_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 922 | } else { |
| 923 | int r; |
| 924 | struct dm_io_region to; |
| 925 | |
| 926 | to.bdev = tc->pool_dev->bdev; |
| 927 | to.sector = data_block * pool->sectors_per_block; |
| 928 | to.count = pool->sectors_per_block; |
| 929 | |
| 930 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); |
| 931 | if (r < 0) { |
| 932 | mempool_free(m, pool->mapping_pool); |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 933 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 934 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 935 | } |
| 936 | } |
| 937 | } |
| 938 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 939 | /* |
| 940 | * A non-zero return indicates read_only or fail_io mode. |
| 941 | * Many callers don't care about the return value. |
| 942 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 943 | static int commit(struct pool *pool) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 944 | { |
| 945 | int r; |
| 946 | |
Joe Thornber | 8d07e8a | 2014-05-06 16:28:14 +0100 | [diff] [blame] | 947 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 948 | return -EINVAL; |
| 949 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 950 | r = dm_pool_commit_metadata(pool->pmd); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 951 | if (r) |
| 952 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 953 | |
| 954 | return r; |
| 955 | } |
| 956 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 957 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
| 958 | { |
| 959 | unsigned long flags; |
| 960 | |
| 961 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
| 962 | DMWARN("%s: reached low water mark for data device: sending event.", |
| 963 | dm_device_name(pool->pool_md)); |
| 964 | spin_lock_irqsave(&pool->lock, flags); |
| 965 | pool->low_water_triggered = true; |
| 966 | spin_unlock_irqrestore(&pool->lock, flags); |
| 967 | dm_table_event(pool->ti->table); |
| 968 | } |
| 969 | } |
| 970 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 971 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
| 972 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 973 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
| 974 | { |
| 975 | int r; |
| 976 | dm_block_t free_blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 977 | struct pool *pool = tc->pool; |
| 978 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 979 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
Joe Thornber | 8d30abf | 2013-12-04 19:16:11 -0500 | [diff] [blame] | 980 | return -EINVAL; |
| 981 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 982 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 983 | if (r) { |
| 984 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 985 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 986 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 987 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 988 | check_low_water_mark(pool, free_blocks); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 989 | |
| 990 | if (!free_blocks) { |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 991 | /* |
| 992 | * Try to commit to see if that will free up some |
| 993 | * more space. |
| 994 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 995 | r = commit(pool); |
| 996 | if (r) |
| 997 | return r; |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 998 | |
| 999 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1000 | if (r) { |
| 1001 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1002 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1003 | } |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1004 | |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1005 | if (!free_blocks) { |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1006 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1007 | return -ENOSPC; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1008 | } |
| 1009 | } |
| 1010 | |
| 1011 | r = dm_pool_alloc_data_block(pool->pmd, result); |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 1012 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1013 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1014 | return r; |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 1015 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1016 | |
| 1017 | return 0; |
| 1018 | } |
| 1019 | |
| 1020 | /* |
| 1021 | * If we have run out of space, queue bios until the device is |
| 1022 | * resumed, presumably after having been reloaded with more space. |
| 1023 | */ |
| 1024 | static void retry_on_resume(struct bio *bio) |
| 1025 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1026 | 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] | 1027 | struct thin_c *tc = h->tc; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1028 | unsigned long flags; |
| 1029 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1030 | spin_lock_irqsave(&tc->lock, flags); |
| 1031 | bio_list_add(&tc->retry_on_resume_list, bio); |
| 1032 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1033 | } |
| 1034 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1035 | static int should_error_unserviceable_bio(struct pool *pool) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1036 | { |
| 1037 | enum pool_mode m = get_pool_mode(pool); |
| 1038 | |
| 1039 | switch (m) { |
| 1040 | case PM_WRITE: |
| 1041 | /* Shouldn't get here */ |
| 1042 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1043 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1044 | |
| 1045 | case PM_OUT_OF_DATA_SPACE: |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1046 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1047 | |
| 1048 | case PM_READ_ONLY: |
| 1049 | case PM_FAIL: |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1050 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1051 | default: |
| 1052 | /* Shouldn't get here */ |
| 1053 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1054 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1055 | } |
| 1056 | } |
| 1057 | |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1058 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
| 1059 | { |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1060 | int error = should_error_unserviceable_bio(pool); |
| 1061 | |
| 1062 | if (error) |
| 1063 | bio_endio(bio, error); |
Mike Snitzer | 6d16202 | 2013-12-20 18:09:02 -0500 | [diff] [blame] | 1064 | else |
| 1065 | retry_on_resume(bio); |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1066 | } |
| 1067 | |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1068 | 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] | 1069 | { |
| 1070 | struct bio *bio; |
| 1071 | struct bio_list bios; |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1072 | int error; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1073 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1074 | error = should_error_unserviceable_bio(pool); |
| 1075 | if (error) { |
| 1076 | cell_error_with_code(pool, cell, error); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1077 | return; |
| 1078 | } |
| 1079 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1080 | bio_list_init(&bios); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1081 | cell_release(pool, cell, &bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1082 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1083 | error = should_error_unserviceable_bio(pool); |
| 1084 | if (error) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1085 | while ((bio = bio_list_pop(&bios))) |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1086 | bio_endio(bio, error); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1087 | else |
| 1088 | while ((bio = bio_list_pop(&bios))) |
| 1089 | retry_on_resume(bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1090 | } |
| 1091 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1092 | static void process_discard(struct thin_c *tc, struct bio *bio) |
| 1093 | { |
| 1094 | int r; |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1095 | unsigned long flags; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1096 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1097 | struct dm_bio_prison_cell *cell, *cell2; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1098 | struct dm_cell_key key, key2; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1099 | dm_block_t block = get_bio_block(tc, bio); |
| 1100 | struct dm_thin_lookup_result lookup_result; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1101 | struct dm_thin_new_mapping *m; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1102 | |
| 1103 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1104 | if (bio_detain(tc->pool, &key, bio, &cell)) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1105 | return; |
| 1106 | |
| 1107 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1108 | switch (r) { |
| 1109 | case 0: |
| 1110 | /* |
| 1111 | * Check nobody is fiddling with this pool block. This can |
| 1112 | * happen if someone's in the process of breaking sharing |
| 1113 | * on this block. |
| 1114 | */ |
| 1115 | build_data_key(tc->td, lookup_result.block, &key2); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1116 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1117 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1118 | break; |
| 1119 | } |
| 1120 | |
| 1121 | if (io_overlaps_block(pool, bio)) { |
| 1122 | /* |
| 1123 | * IO may still be going to the destination block. We must |
| 1124 | * quiesce before we can do the removal. |
| 1125 | */ |
| 1126 | m = get_next_mapping(pool); |
| 1127 | m->tc = tc; |
Joe Thornber | 19fa1a6 | 2013-12-17 12:09:40 -0500 | [diff] [blame] | 1128 | m->pass_discard = pool->pf.discard_passdown; |
| 1129 | m->definitely_not_shared = !lookup_result.shared; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1130 | m->virt_block = block; |
| 1131 | m->data_block = lookup_result.block; |
| 1132 | m->cell = cell; |
| 1133 | m->cell2 = cell2; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1134 | m->bio = bio; |
| 1135 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1136 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1137 | spin_lock_irqsave(&pool->lock, flags); |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 1138 | list_add_tail(&m->list, &pool->prepared_discards); |
Mike Snitzer | c3a0ce2 | 2012-05-12 01:43:16 +0100 | [diff] [blame] | 1139 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1140 | wake_worker(pool); |
| 1141 | } |
| 1142 | } else { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1143 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1144 | cell_defer_no_holder(tc, cell); |
| 1145 | cell_defer_no_holder(tc, cell2); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1146 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1147 | /* |
Mikulas Patocka | 4929630 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1148 | * The DM core makes sure that the discard doesn't span |
| 1149 | * a block boundary. So we submit the discard of a |
| 1150 | * partial block appropriately. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1151 | */ |
Mikulas Patocka | 650d2a0 | 2012-07-20 14:25:05 +0100 | [diff] [blame] | 1152 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
| 1153 | remap_and_issue(tc, bio, lookup_result.block); |
| 1154 | else |
| 1155 | bio_endio(bio, 0); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1156 | } |
| 1157 | break; |
| 1158 | |
| 1159 | case -ENODATA: |
| 1160 | /* |
| 1161 | * It isn't provisioned, just forget it. |
| 1162 | */ |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1163 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1164 | bio_endio(bio, 0); |
| 1165 | break; |
| 1166 | |
| 1167 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1168 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1169 | __func__, r); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1170 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1171 | bio_io_error(bio); |
| 1172 | break; |
| 1173 | } |
| 1174 | } |
| 1175 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1176 | 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] | 1177 | struct dm_cell_key *key, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1178 | struct dm_thin_lookup_result *lookup_result, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1179 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1180 | { |
| 1181 | int r; |
| 1182 | dm_block_t data_block; |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1183 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1184 | |
| 1185 | r = alloc_data_block(tc, &data_block); |
| 1186 | switch (r) { |
| 1187 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1188 | schedule_internal_copy(tc, block, lookup_result->block, |
| 1189 | data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1190 | break; |
| 1191 | |
| 1192 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1193 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1194 | break; |
| 1195 | |
| 1196 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1197 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1198 | __func__, r); |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1199 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1200 | break; |
| 1201 | } |
| 1202 | } |
| 1203 | |
| 1204 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
| 1205 | dm_block_t block, |
| 1206 | struct dm_thin_lookup_result *lookup_result) |
| 1207 | { |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1208 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1209 | struct pool *pool = tc->pool; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1210 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1211 | |
| 1212 | /* |
| 1213 | * If cell is already occupied, then sharing is already in the process |
| 1214 | * of being broken so we have nothing further to do here. |
| 1215 | */ |
| 1216 | build_data_key(tc->td, lookup_result->block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1217 | if (bio_detain(pool, &key, bio, &cell)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1218 | return; |
| 1219 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1220 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1221 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
| 1222 | else { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1223 | 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] | 1224 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1225 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1226 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1227 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1228 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1229 | remap_and_issue(tc, bio, lookup_result->block); |
| 1230 | } |
| 1231 | } |
| 1232 | |
| 1233 | 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] | 1234 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1235 | { |
| 1236 | int r; |
| 1237 | dm_block_t data_block; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1238 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1239 | |
| 1240 | /* |
| 1241 | * Remap empty bios (flushes) immediately, without provisioning. |
| 1242 | */ |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1243 | if (!bio->bi_iter.bi_size) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1244 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1245 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1246 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1247 | remap_and_issue(tc, bio, 0); |
| 1248 | return; |
| 1249 | } |
| 1250 | |
| 1251 | /* |
| 1252 | * Fill read bios with zeroes and complete them immediately. |
| 1253 | */ |
| 1254 | if (bio_data_dir(bio) == READ) { |
| 1255 | zero_fill_bio(bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1256 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1257 | bio_endio(bio, 0); |
| 1258 | return; |
| 1259 | } |
| 1260 | |
| 1261 | r = alloc_data_block(tc, &data_block); |
| 1262 | switch (r) { |
| 1263 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1264 | if (tc->origin_dev) |
| 1265 | schedule_external_copy(tc, block, data_block, cell, bio); |
| 1266 | else |
| 1267 | schedule_zero(tc, block, data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1268 | break; |
| 1269 | |
| 1270 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1271 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1272 | break; |
| 1273 | |
| 1274 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1275 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1276 | __func__, r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1277 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1278 | break; |
| 1279 | } |
| 1280 | } |
| 1281 | |
| 1282 | static void process_bio(struct thin_c *tc, struct bio *bio) |
| 1283 | { |
| 1284 | int r; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1285 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1286 | dm_block_t block = get_bio_block(tc, bio); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1287 | struct dm_bio_prison_cell *cell; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1288 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1289 | struct dm_thin_lookup_result lookup_result; |
| 1290 | |
| 1291 | /* |
| 1292 | * If cell is already occupied, then the block is already |
| 1293 | * being provisioned so we have nothing further to do here. |
| 1294 | */ |
| 1295 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1296 | if (bio_detain(pool, &key, bio, &cell)) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1297 | return; |
| 1298 | |
| 1299 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1300 | switch (r) { |
| 1301 | case 0: |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1302 | if (lookup_result.shared) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1303 | process_shared_bio(tc, bio, block, &lookup_result); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1304 | 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] | 1305 | } else { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1306 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1307 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1308 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1309 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1310 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1311 | break; |
| 1312 | |
| 1313 | case -ENODATA: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1314 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1315 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1316 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1317 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1318 | remap_to_origin_and_issue(tc, bio); |
| 1319 | } else |
| 1320 | provision_block(tc, bio, block, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1321 | break; |
| 1322 | |
| 1323 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1324 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1325 | __func__, r); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1326 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1327 | bio_io_error(bio); |
| 1328 | break; |
| 1329 | } |
| 1330 | } |
| 1331 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1332 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
| 1333 | { |
| 1334 | int r; |
| 1335 | int rw = bio_data_dir(bio); |
| 1336 | dm_block_t block = get_bio_block(tc, bio); |
| 1337 | struct dm_thin_lookup_result lookup_result; |
| 1338 | |
| 1339 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1340 | switch (r) { |
| 1341 | case 0: |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1342 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1343 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1344 | else { |
| 1345 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1346 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1347 | } |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1348 | break; |
| 1349 | |
| 1350 | case -ENODATA: |
| 1351 | if (rw != READ) { |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1352 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1353 | break; |
| 1354 | } |
| 1355 | |
| 1356 | if (tc->origin_dev) { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1357 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1358 | remap_to_origin_and_issue(tc, bio); |
| 1359 | break; |
| 1360 | } |
| 1361 | |
| 1362 | zero_fill_bio(bio); |
| 1363 | bio_endio(bio, 0); |
| 1364 | break; |
| 1365 | |
| 1366 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1367 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1368 | __func__, r); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1369 | bio_io_error(bio); |
| 1370 | break; |
| 1371 | } |
| 1372 | } |
| 1373 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1374 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
| 1375 | { |
| 1376 | bio_endio(bio, 0); |
| 1377 | } |
| 1378 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1379 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
| 1380 | { |
| 1381 | bio_io_error(bio); |
| 1382 | } |
| 1383 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 1384 | /* |
| 1385 | * FIXME: should we also commit due to size of transaction, measured in |
| 1386 | * metadata blocks? |
| 1387 | */ |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1388 | static int need_commit_due_to_time(struct pool *pool) |
| 1389 | { |
| 1390 | return jiffies < pool->last_commit_jiffies || |
| 1391 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; |
| 1392 | } |
| 1393 | |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1394 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
| 1395 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) |
| 1396 | |
| 1397 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) |
| 1398 | { |
| 1399 | struct rb_node **rbp, *parent; |
| 1400 | struct dm_thin_endio_hook *pbd; |
| 1401 | sector_t bi_sector = bio->bi_iter.bi_sector; |
| 1402 | |
| 1403 | rbp = &tc->sort_bio_list.rb_node; |
| 1404 | parent = NULL; |
| 1405 | while (*rbp) { |
| 1406 | parent = *rbp; |
| 1407 | pbd = thin_pbd(parent); |
| 1408 | |
| 1409 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) |
| 1410 | rbp = &(*rbp)->rb_left; |
| 1411 | else |
| 1412 | rbp = &(*rbp)->rb_right; |
| 1413 | } |
| 1414 | |
| 1415 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
| 1416 | rb_link_node(&pbd->rb_node, parent, rbp); |
| 1417 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); |
| 1418 | } |
| 1419 | |
| 1420 | static void __extract_sorted_bios(struct thin_c *tc) |
| 1421 | { |
| 1422 | struct rb_node *node; |
| 1423 | struct dm_thin_endio_hook *pbd; |
| 1424 | struct bio *bio; |
| 1425 | |
| 1426 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { |
| 1427 | pbd = thin_pbd(node); |
| 1428 | bio = thin_bio(pbd); |
| 1429 | |
| 1430 | bio_list_add(&tc->deferred_bio_list, bio); |
| 1431 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); |
| 1432 | } |
| 1433 | |
| 1434 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); |
| 1435 | } |
| 1436 | |
| 1437 | static void __sort_thin_deferred_bios(struct thin_c *tc) |
| 1438 | { |
| 1439 | struct bio *bio; |
| 1440 | struct bio_list bios; |
| 1441 | |
| 1442 | bio_list_init(&bios); |
| 1443 | bio_list_merge(&bios, &tc->deferred_bio_list); |
| 1444 | bio_list_init(&tc->deferred_bio_list); |
| 1445 | |
| 1446 | /* Sort deferred_bio_list using rb-tree */ |
| 1447 | while ((bio = bio_list_pop(&bios))) |
| 1448 | __thin_bio_rb_add(tc, bio); |
| 1449 | |
| 1450 | /* |
| 1451 | * Transfer the sorted bios in sort_bio_list back to |
| 1452 | * deferred_bio_list to allow lockless submission of |
| 1453 | * all bios. |
| 1454 | */ |
| 1455 | __extract_sorted_bios(tc); |
| 1456 | } |
| 1457 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1458 | static void process_thin_deferred_bios(struct thin_c *tc) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1459 | { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1460 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1461 | unsigned long flags; |
| 1462 | struct bio *bio; |
| 1463 | struct bio_list bios; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1464 | struct blk_plug plug; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1465 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1466 | if (tc->requeue_mode) { |
| 1467 | requeue_bio_list(tc, &tc->deferred_bio_list); |
| 1468 | return; |
| 1469 | } |
| 1470 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1471 | bio_list_init(&bios); |
| 1472 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1473 | spin_lock_irqsave(&tc->lock, flags); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1474 | |
| 1475 | if (bio_list_empty(&tc->deferred_bio_list)) { |
| 1476 | spin_unlock_irqrestore(&tc->lock, flags); |
| 1477 | return; |
| 1478 | } |
| 1479 | |
| 1480 | __sort_thin_deferred_bios(tc); |
| 1481 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1482 | bio_list_merge(&bios, &tc->deferred_bio_list); |
| 1483 | bio_list_init(&tc->deferred_bio_list); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1484 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1485 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1486 | |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1487 | blk_start_plug(&plug); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1488 | while ((bio = bio_list_pop(&bios))) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1489 | /* |
| 1490 | * If we've got no free new_mapping structs, and processing |
| 1491 | * this bio might require one, we pause until there are some |
| 1492 | * prepared mappings to process. |
| 1493 | */ |
| 1494 | if (ensure_next_mapping(pool)) { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1495 | spin_lock_irqsave(&tc->lock, flags); |
| 1496 | bio_list_add(&tc->deferred_bio_list, bio); |
| 1497 | bio_list_merge(&tc->deferred_bio_list, &bios); |
| 1498 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1499 | break; |
| 1500 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1501 | |
| 1502 | if (bio->bi_rw & REQ_DISCARD) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1503 | pool->process_discard(tc, bio); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1504 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1505 | pool->process_bio(tc, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1506 | } |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1507 | blk_finish_plug(&plug); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1508 | } |
| 1509 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 1510 | static void thin_get(struct thin_c *tc); |
| 1511 | static void thin_put(struct thin_c *tc); |
| 1512 | |
| 1513 | /* |
| 1514 | * We can't hold rcu_read_lock() around code that can block. So we |
| 1515 | * find a thin with the rcu lock held; bump a refcount; then drop |
| 1516 | * the lock. |
| 1517 | */ |
| 1518 | static struct thin_c *get_first_thin(struct pool *pool) |
| 1519 | { |
| 1520 | struct thin_c *tc = NULL; |
| 1521 | |
| 1522 | rcu_read_lock(); |
| 1523 | if (!list_empty(&pool->active_thins)) { |
| 1524 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); |
| 1525 | thin_get(tc); |
| 1526 | } |
| 1527 | rcu_read_unlock(); |
| 1528 | |
| 1529 | return tc; |
| 1530 | } |
| 1531 | |
| 1532 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) |
| 1533 | { |
| 1534 | struct thin_c *old_tc = tc; |
| 1535 | |
| 1536 | rcu_read_lock(); |
| 1537 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { |
| 1538 | thin_get(tc); |
| 1539 | thin_put(old_tc); |
| 1540 | rcu_read_unlock(); |
| 1541 | return tc; |
| 1542 | } |
| 1543 | thin_put(old_tc); |
| 1544 | rcu_read_unlock(); |
| 1545 | |
| 1546 | return NULL; |
| 1547 | } |
| 1548 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1549 | static void process_deferred_bios(struct pool *pool) |
| 1550 | { |
| 1551 | unsigned long flags; |
| 1552 | struct bio *bio; |
| 1553 | struct bio_list bios; |
| 1554 | struct thin_c *tc; |
| 1555 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 1556 | tc = get_first_thin(pool); |
| 1557 | while (tc) { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1558 | process_thin_deferred_bios(tc); |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 1559 | tc = get_next_thin(pool, tc); |
| 1560 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1561 | |
| 1562 | /* |
| 1563 | * If there are any deferred flush bios, we must commit |
| 1564 | * the metadata before issuing them. |
| 1565 | */ |
| 1566 | bio_list_init(&bios); |
| 1567 | spin_lock_irqsave(&pool->lock, flags); |
| 1568 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
| 1569 | bio_list_init(&pool->deferred_flush_bios); |
| 1570 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1571 | |
Mike Snitzer | 4d1662a | 2014-02-06 06:08:56 -0500 | [diff] [blame] | 1572 | if (bio_list_empty(&bios) && |
| 1573 | !(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] | 1574 | return; |
| 1575 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 1576 | if (commit(pool)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1577 | while ((bio = bio_list_pop(&bios))) |
| 1578 | bio_io_error(bio); |
| 1579 | return; |
| 1580 | } |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1581 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1582 | |
| 1583 | while ((bio = bio_list_pop(&bios))) |
| 1584 | generic_make_request(bio); |
| 1585 | } |
| 1586 | |
| 1587 | static void do_worker(struct work_struct *ws) |
| 1588 | { |
| 1589 | struct pool *pool = container_of(ws, struct pool, worker); |
| 1590 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1591 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
| 1592 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1593 | process_deferred_bios(pool); |
| 1594 | } |
| 1595 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1596 | /* |
| 1597 | * We want to commit periodically so that not too much |
| 1598 | * unwritten data builds up. |
| 1599 | */ |
| 1600 | static void do_waker(struct work_struct *ws) |
| 1601 | { |
| 1602 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); |
| 1603 | wake_worker(pool); |
| 1604 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); |
| 1605 | } |
| 1606 | |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 1607 | /* |
| 1608 | * We're holding onto IO to allow userland time to react. After the |
| 1609 | * timeout either the pool will have been resized (and thus back in |
| 1610 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. |
| 1611 | */ |
| 1612 | static void do_no_space_timeout(struct work_struct *ws) |
| 1613 | { |
| 1614 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, |
| 1615 | no_space_timeout); |
| 1616 | |
| 1617 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) |
| 1618 | set_pool_mode(pool, PM_READ_ONLY); |
| 1619 | } |
| 1620 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1621 | /*----------------------------------------------------------------*/ |
| 1622 | |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1623 | struct pool_work { |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1624 | struct work_struct worker; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1625 | struct completion complete; |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1626 | }; |
| 1627 | |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1628 | static struct pool_work *to_pool_work(struct work_struct *ws) |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1629 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1630 | return container_of(ws, struct pool_work, worker); |
| 1631 | } |
| 1632 | |
| 1633 | static void pool_work_complete(struct pool_work *pw) |
| 1634 | { |
| 1635 | complete(&pw->complete); |
| 1636 | } |
| 1637 | |
| 1638 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
| 1639 | void (*fn)(struct work_struct *)) |
| 1640 | { |
| 1641 | INIT_WORK_ONSTACK(&pw->worker, fn); |
| 1642 | init_completion(&pw->complete); |
| 1643 | queue_work(pool->wq, &pw->worker); |
| 1644 | wait_for_completion(&pw->complete); |
| 1645 | } |
| 1646 | |
| 1647 | /*----------------------------------------------------------------*/ |
| 1648 | |
| 1649 | struct noflush_work { |
| 1650 | struct pool_work pw; |
| 1651 | struct thin_c *tc; |
| 1652 | }; |
| 1653 | |
| 1654 | static struct noflush_work *to_noflush(struct work_struct *ws) |
| 1655 | { |
| 1656 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1657 | } |
| 1658 | |
| 1659 | static void do_noflush_start(struct work_struct *ws) |
| 1660 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1661 | struct noflush_work *w = to_noflush(ws); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1662 | w->tc->requeue_mode = true; |
| 1663 | requeue_io(w->tc); |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1664 | pool_work_complete(&w->pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1665 | } |
| 1666 | |
| 1667 | static void do_noflush_stop(struct work_struct *ws) |
| 1668 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1669 | struct noflush_work *w = to_noflush(ws); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1670 | w->tc->requeue_mode = false; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1671 | pool_work_complete(&w->pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1672 | } |
| 1673 | |
| 1674 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) |
| 1675 | { |
| 1676 | struct noflush_work w; |
| 1677 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1678 | w.tc = tc; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 1679 | pool_work_wait(&w.pw, tc->pool, fn); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1680 | } |
| 1681 | |
| 1682 | /*----------------------------------------------------------------*/ |
| 1683 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1684 | static enum pool_mode get_pool_mode(struct pool *pool) |
| 1685 | { |
| 1686 | return pool->pf.mode; |
| 1687 | } |
| 1688 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1689 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
| 1690 | { |
| 1691 | dm_table_event(pool->ti->table); |
| 1692 | DMINFO("%s: switching pool to %s mode", |
| 1693 | dm_device_name(pool->pool_md), new_mode); |
| 1694 | } |
| 1695 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1696 | 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] | 1697 | { |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 1698 | struct pool_c *pt = pool->ti->private; |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1699 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
| 1700 | enum pool_mode old_mode = get_pool_mode(pool); |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 1701 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1702 | |
| 1703 | /* |
| 1704 | * Never allow the pool to transition to PM_WRITE mode if user |
| 1705 | * intervention is required to verify metadata and data consistency. |
| 1706 | */ |
| 1707 | if (new_mode == PM_WRITE && needs_check) { |
| 1708 | DMERR("%s: unable to switch pool to write mode until repaired.", |
| 1709 | dm_device_name(pool->pool_md)); |
| 1710 | if (old_mode != new_mode) |
| 1711 | new_mode = old_mode; |
| 1712 | else |
| 1713 | new_mode = PM_READ_ONLY; |
| 1714 | } |
| 1715 | /* |
| 1716 | * If we were in PM_FAIL mode, rollback of metadata failed. We're |
| 1717 | * not going to recover without a thin_repair. So we never let the |
| 1718 | * pool move out of the old mode. |
| 1719 | */ |
| 1720 | if (old_mode == PM_FAIL) |
| 1721 | new_mode = old_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1722 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1723 | switch (new_mode) { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1724 | case PM_FAIL: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1725 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1726 | notify_of_pool_mode_change(pool, "failure"); |
Joe Thornber | 5383ef3 | 2013-12-04 16:30:01 -0500 | [diff] [blame] | 1727 | dm_pool_metadata_read_only(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1728 | pool->process_bio = process_bio_fail; |
| 1729 | pool->process_discard = process_bio_fail; |
| 1730 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
| 1731 | pool->process_prepared_discard = process_prepared_discard_fail; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1732 | |
| 1733 | error_retry_list(pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1734 | break; |
| 1735 | |
| 1736 | case PM_READ_ONLY: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1737 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1738 | notify_of_pool_mode_change(pool, "read-only"); |
| 1739 | dm_pool_metadata_read_only(pool->pmd); |
| 1740 | pool->process_bio = process_bio_read_only; |
| 1741 | pool->process_discard = process_bio_success; |
| 1742 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
| 1743 | pool->process_prepared_discard = process_prepared_discard_passdown; |
| 1744 | |
| 1745 | error_retry_list(pool); |
| 1746 | break; |
| 1747 | |
| 1748 | case PM_OUT_OF_DATA_SPACE: |
| 1749 | /* |
| 1750 | * Ideally we'd never hit this state; the low water mark |
| 1751 | * would trigger userland to extend the pool before we |
| 1752 | * completely run out of data space. However, many small |
| 1753 | * IOs to unprovisioned space can consume data space at an |
| 1754 | * alarming rate. Adjust your low water mark if you're |
| 1755 | * frequently seeing this mode. |
| 1756 | */ |
| 1757 | if (old_mode != new_mode) |
| 1758 | notify_of_pool_mode_change(pool, "out-of-data-space"); |
| 1759 | pool->process_bio = process_bio_read_only; |
| 1760 | pool->process_discard = process_discard; |
| 1761 | pool->process_prepared_mapping = process_prepared_mapping; |
| 1762 | pool->process_prepared_discard = process_prepared_discard_passdown; |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 1763 | |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 1764 | if (!pool->pf.error_if_no_space && no_space_timeout) |
| 1765 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1766 | break; |
| 1767 | |
| 1768 | case PM_WRITE: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1769 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1770 | notify_of_pool_mode_change(pool, "write"); |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 1771 | dm_pool_metadata_read_write(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1772 | pool->process_bio = process_bio; |
| 1773 | pool->process_discard = process_discard; |
| 1774 | pool->process_prepared_mapping = process_prepared_mapping; |
| 1775 | pool->process_prepared_discard = process_prepared_discard; |
| 1776 | break; |
| 1777 | } |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 1778 | |
| 1779 | pool->pf.mode = new_mode; |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 1780 | /* |
| 1781 | * The pool mode may have changed, sync it so bind_control_target() |
| 1782 | * doesn't cause an unexpected mode transition on resume. |
| 1783 | */ |
| 1784 | pt->adjusted_pf.mode = new_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1785 | } |
| 1786 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1787 | static void abort_transaction(struct pool *pool) |
| 1788 | { |
| 1789 | const char *dev_name = dm_device_name(pool->pool_md); |
| 1790 | |
| 1791 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
| 1792 | if (dm_pool_abort_metadata(pool->pmd)) { |
| 1793 | DMERR("%s: failed to abort metadata transaction", dev_name); |
| 1794 | set_pool_mode(pool, PM_FAIL); |
| 1795 | } |
| 1796 | |
| 1797 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { |
| 1798 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
| 1799 | set_pool_mode(pool, PM_FAIL); |
| 1800 | } |
| 1801 | } |
| 1802 | |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1803 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
| 1804 | { |
| 1805 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
| 1806 | dm_device_name(pool->pool_md), op, r); |
| 1807 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 1808 | abort_transaction(pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1809 | set_pool_mode(pool, PM_READ_ONLY); |
| 1810 | } |
| 1811 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1812 | /*----------------------------------------------------------------*/ |
| 1813 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1814 | /* |
| 1815 | * Mapping functions. |
| 1816 | */ |
| 1817 | |
| 1818 | /* |
| 1819 | * Called only while mapping a thin bio to hand it over to the workqueue. |
| 1820 | */ |
| 1821 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) |
| 1822 | { |
| 1823 | unsigned long flags; |
| 1824 | struct pool *pool = tc->pool; |
| 1825 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1826 | spin_lock_irqsave(&tc->lock, flags); |
| 1827 | bio_list_add(&tc->deferred_bio_list, bio); |
| 1828 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1829 | |
| 1830 | wake_worker(pool); |
| 1831 | } |
| 1832 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1833 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1834 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1835 | 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] | 1836 | |
| 1837 | h->tc = tc; |
| 1838 | h->shared_read_entry = NULL; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1839 | h->all_io_entry = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1840 | h->overwrite_mapping = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1841 | } |
| 1842 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1843 | /* |
| 1844 | * Non-blocking function called from the thin target's map function. |
| 1845 | */ |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 1846 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1847 | { |
| 1848 | int r; |
| 1849 | struct thin_c *tc = ti->private; |
| 1850 | dm_block_t block = get_bio_block(tc, bio); |
| 1851 | struct dm_thin_device *td = tc->td; |
| 1852 | struct dm_thin_lookup_result result; |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1853 | struct dm_bio_prison_cell cell1, cell2; |
| 1854 | struct dm_bio_prison_cell *cell_result; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1855 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1856 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1857 | thin_hook_bio(tc, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1858 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 1859 | if (tc->requeue_mode) { |
| 1860 | bio_endio(bio, DM_ENDIO_REQUEUE); |
| 1861 | return DM_MAPIO_SUBMITTED; |
| 1862 | } |
| 1863 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1864 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 1865 | bio_io_error(bio); |
| 1866 | return DM_MAPIO_SUBMITTED; |
| 1867 | } |
| 1868 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1869 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1870 | thin_defer_bio(tc, bio); |
| 1871 | return DM_MAPIO_SUBMITTED; |
| 1872 | } |
| 1873 | |
| 1874 | r = dm_thin_find_block(td, block, 0, &result); |
| 1875 | |
| 1876 | /* |
| 1877 | * Note that we defer readahead too. |
| 1878 | */ |
| 1879 | switch (r) { |
| 1880 | case 0: |
| 1881 | if (unlikely(result.shared)) { |
| 1882 | /* |
| 1883 | * We have a race condition here between the |
| 1884 | * result.shared value returned by the lookup and |
| 1885 | * snapshot creation, which may cause new |
| 1886 | * sharing. |
| 1887 | * |
| 1888 | * To avoid this always quiesce the origin before |
| 1889 | * taking the snap. You want to do this anyway to |
| 1890 | * ensure a consistent application view |
| 1891 | * (i.e. lockfs). |
| 1892 | * |
| 1893 | * More distant ancestors are irrelevant. The |
| 1894 | * shared flag will be set in their case. |
| 1895 | */ |
| 1896 | thin_defer_bio(tc, bio); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1897 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1898 | } |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1899 | |
| 1900 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1901 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1902 | return DM_MAPIO_SUBMITTED; |
| 1903 | |
| 1904 | build_data_key(tc->td, result.block, &key); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1905 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
| 1906 | cell_defer_no_holder_no_free(tc, &cell1); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1907 | return DM_MAPIO_SUBMITTED; |
| 1908 | } |
| 1909 | |
| 1910 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1911 | cell_defer_no_holder_no_free(tc, &cell2); |
| 1912 | cell_defer_no_holder_no_free(tc, &cell1); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1913 | |
| 1914 | remap(tc, bio, result.block); |
| 1915 | return DM_MAPIO_REMAPPED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1916 | |
| 1917 | case -ENODATA: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1918 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
| 1919 | /* |
| 1920 | * This block isn't provisioned, and we have no way |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1921 | * of doing so. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1922 | */ |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1923 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1924 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1925 | } |
| 1926 | /* fall through */ |
| 1927 | |
| 1928 | case -EWOULDBLOCK: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1929 | /* |
| 1930 | * In future, the failed dm_thin_find_block above could |
| 1931 | * provide the hint to load the metadata into cache. |
| 1932 | */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1933 | thin_defer_bio(tc, bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1934 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1935 | |
| 1936 | default: |
| 1937 | /* |
| 1938 | * Must always call bio_io_error on failure. |
| 1939 | * dm_thin_find_block can fail with -EINVAL if the |
| 1940 | * pool is switched to fail-io mode. |
| 1941 | */ |
| 1942 | bio_io_error(bio); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1943 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1944 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1945 | } |
| 1946 | |
| 1947 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) |
| 1948 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1949 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 1950 | struct request_queue *q; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1951 | |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 1952 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
| 1953 | return 1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1954 | |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 1955 | q = bdev_get_queue(pt->data_dev->bdev); |
| 1956 | return bdi_congested(&q->backing_dev_info, bdi_bits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1957 | } |
| 1958 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1959 | static void requeue_bios(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1960 | { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1961 | unsigned long flags; |
| 1962 | struct thin_c *tc; |
| 1963 | |
| 1964 | rcu_read_lock(); |
| 1965 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { |
| 1966 | spin_lock_irqsave(&tc->lock, flags); |
| 1967 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); |
| 1968 | bio_list_init(&tc->retry_on_resume_list); |
| 1969 | spin_unlock_irqrestore(&tc->lock, flags); |
| 1970 | } |
| 1971 | rcu_read_unlock(); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1972 | } |
| 1973 | |
| 1974 | /*---------------------------------------------------------------- |
| 1975 | * Binding of control targets to a pool object |
| 1976 | *--------------------------------------------------------------*/ |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1977 | static bool data_dev_supports_discard(struct pool_c *pt) |
| 1978 | { |
| 1979 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 1980 | |
| 1981 | return q && blk_queue_discard(q); |
| 1982 | } |
| 1983 | |
Joe Thornber | 58051b9 | 2013-03-20 17:21:25 +0000 | [diff] [blame] | 1984 | static bool is_factor(sector_t block_size, uint32_t n) |
| 1985 | { |
| 1986 | return !sector_div(block_size, n); |
| 1987 | } |
| 1988 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1989 | /* |
| 1990 | * If discard_passdown was enabled verify that the data device |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1991 | * supports discards. Disable discard_passdown if not. |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1992 | */ |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1993 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 1994 | { |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 1995 | struct pool *pool = pt->pool; |
| 1996 | struct block_device *data_bdev = pt->data_dev->bdev; |
| 1997 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; |
| 1998 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; |
| 1999 | const char *reason = NULL; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2000 | char buf[BDEVNAME_SIZE]; |
| 2001 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2002 | if (!pt->adjusted_pf.discard_passdown) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2003 | return; |
| 2004 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2005 | if (!data_dev_supports_discard(pt)) |
| 2006 | reason = "discard unsupported"; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2007 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2008 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) |
| 2009 | reason = "max discard sectors smaller than a block"; |
| 2010 | |
| 2011 | else if (data_limits->discard_granularity > block_size) |
| 2012 | reason = "discard granularity larger than a block"; |
| 2013 | |
Joe Thornber | 58051b9 | 2013-03-20 17:21:25 +0000 | [diff] [blame] | 2014 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2015 | reason = "discard granularity not a factor of block size"; |
| 2016 | |
| 2017 | if (reason) { |
| 2018 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); |
| 2019 | pt->adjusted_pf.discard_passdown = false; |
| 2020 | } |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2021 | } |
| 2022 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2023 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
| 2024 | { |
| 2025 | struct pool_c *pt = ti->private; |
| 2026 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2027 | /* |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 2028 | * 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] | 2029 | */ |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2030 | enum pool_mode old_mode = get_pool_mode(pool); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2031 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2032 | |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 2033 | /* |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2034 | * Don't change the pool's mode until set_pool_mode() below. |
| 2035 | * Otherwise the pool's process_* function pointers may |
| 2036 | * not match the desired pool mode. |
| 2037 | */ |
| 2038 | pt->adjusted_pf.mode = old_mode; |
| 2039 | |
| 2040 | pool->ti = ti; |
| 2041 | pool->pf = pt->adjusted_pf; |
| 2042 | pool->low_water_blocks = pt->low_water_blocks; |
| 2043 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2044 | set_pool_mode(pool, new_mode); |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 2045 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2046 | return 0; |
| 2047 | } |
| 2048 | |
| 2049 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) |
| 2050 | { |
| 2051 | if (pool->ti == ti) |
| 2052 | pool->ti = NULL; |
| 2053 | } |
| 2054 | |
| 2055 | /*---------------------------------------------------------------- |
| 2056 | * Pool creation |
| 2057 | *--------------------------------------------------------------*/ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2058 | /* Initialize pool features. */ |
| 2059 | static void pool_features_init(struct pool_features *pf) |
| 2060 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2061 | pf->mode = PM_WRITE; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2062 | pf->zero_new_blocks = true; |
| 2063 | pf->discard_enabled = true; |
| 2064 | pf->discard_passdown = true; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2065 | pf->error_if_no_space = false; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2066 | } |
| 2067 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2068 | static void __pool_destroy(struct pool *pool) |
| 2069 | { |
| 2070 | __pool_table_remove(pool); |
| 2071 | |
| 2072 | if (dm_pool_metadata_close(pool->pmd) < 0) |
| 2073 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 2074 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2075 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2076 | dm_kcopyd_client_destroy(pool->copier); |
| 2077 | |
| 2078 | if (pool->wq) |
| 2079 | destroy_workqueue(pool->wq); |
| 2080 | |
| 2081 | if (pool->next_mapping) |
| 2082 | mempool_free(pool->next_mapping, pool->mapping_pool); |
| 2083 | mempool_destroy(pool->mapping_pool); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2084 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 2085 | dm_deferred_set_destroy(pool->all_io_ds); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2086 | kfree(pool); |
| 2087 | } |
| 2088 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2089 | static struct kmem_cache *_new_mapping_cache; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2090 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2091 | static struct pool *pool_create(struct mapped_device *pool_md, |
| 2092 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2093 | unsigned long block_size, |
| 2094 | int read_only, char **error) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2095 | { |
| 2096 | int r; |
| 2097 | void *err_p; |
| 2098 | struct pool *pool; |
| 2099 | struct dm_pool_metadata *pmd; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2100 | bool format_device = read_only ? false : true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2101 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2102 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2103 | if (IS_ERR(pmd)) { |
| 2104 | *error = "Error creating metadata object"; |
| 2105 | return (struct pool *)pmd; |
| 2106 | } |
| 2107 | |
| 2108 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); |
| 2109 | if (!pool) { |
| 2110 | *error = "Error allocating memory for pool"; |
| 2111 | err_p = ERR_PTR(-ENOMEM); |
| 2112 | goto bad_pool; |
| 2113 | } |
| 2114 | |
| 2115 | pool->pmd = pmd; |
| 2116 | pool->sectors_per_block = block_size; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 2117 | if (block_size & (block_size - 1)) |
| 2118 | pool->sectors_per_block_shift = -1; |
| 2119 | else |
| 2120 | pool->sectors_per_block_shift = __ffs(block_size); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2121 | pool->low_water_blocks = 0; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2122 | pool_features_init(&pool->pf); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2123 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2124 | if (!pool->prison) { |
| 2125 | *error = "Error creating pool's bio prison"; |
| 2126 | err_p = ERR_PTR(-ENOMEM); |
| 2127 | goto bad_prison; |
| 2128 | } |
| 2129 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 2130 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2131 | if (IS_ERR(pool->copier)) { |
| 2132 | r = PTR_ERR(pool->copier); |
| 2133 | *error = "Error creating pool's kcopyd client"; |
| 2134 | err_p = ERR_PTR(r); |
| 2135 | goto bad_kcopyd_client; |
| 2136 | } |
| 2137 | |
| 2138 | /* |
| 2139 | * Create singlethreaded workqueue that will service all devices |
| 2140 | * that use this metadata. |
| 2141 | */ |
| 2142 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); |
| 2143 | if (!pool->wq) { |
| 2144 | *error = "Error creating pool's workqueue"; |
| 2145 | err_p = ERR_PTR(-ENOMEM); |
| 2146 | goto bad_wq; |
| 2147 | } |
| 2148 | |
| 2149 | INIT_WORK(&pool->worker, do_worker); |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2150 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2151 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2152 | spin_lock_init(&pool->lock); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2153 | bio_list_init(&pool->deferred_flush_bios); |
| 2154 | INIT_LIST_HEAD(&pool->prepared_mappings); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2155 | INIT_LIST_HEAD(&pool->prepared_discards); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2156 | INIT_LIST_HEAD(&pool->active_thins); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 2157 | pool->low_water_triggered = false; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2158 | |
| 2159 | pool->shared_read_ds = dm_deferred_set_create(); |
| 2160 | if (!pool->shared_read_ds) { |
| 2161 | *error = "Error creating pool's shared read deferred set"; |
| 2162 | err_p = ERR_PTR(-ENOMEM); |
| 2163 | goto bad_shared_read_ds; |
| 2164 | } |
| 2165 | |
| 2166 | pool->all_io_ds = dm_deferred_set_create(); |
| 2167 | if (!pool->all_io_ds) { |
| 2168 | *error = "Error creating pool's all io deferred set"; |
| 2169 | err_p = ERR_PTR(-ENOMEM); |
| 2170 | goto bad_all_io_ds; |
| 2171 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2172 | |
| 2173 | pool->next_mapping = NULL; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2174 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
| 2175 | _new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2176 | if (!pool->mapping_pool) { |
| 2177 | *error = "Error creating pool's mapping mempool"; |
| 2178 | err_p = ERR_PTR(-ENOMEM); |
| 2179 | goto bad_mapping_pool; |
| 2180 | } |
| 2181 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2182 | pool->ref_count = 1; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2183 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2184 | pool->pool_md = pool_md; |
| 2185 | pool->md_dev = metadata_dev; |
| 2186 | __pool_table_insert(pool); |
| 2187 | |
| 2188 | return pool; |
| 2189 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2190 | bad_mapping_pool: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2191 | dm_deferred_set_destroy(pool->all_io_ds); |
| 2192 | bad_all_io_ds: |
| 2193 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 2194 | bad_shared_read_ds: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2195 | destroy_workqueue(pool->wq); |
| 2196 | bad_wq: |
| 2197 | dm_kcopyd_client_destroy(pool->copier); |
| 2198 | bad_kcopyd_client: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2199 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2200 | bad_prison: |
| 2201 | kfree(pool); |
| 2202 | bad_pool: |
| 2203 | if (dm_pool_metadata_close(pmd)) |
| 2204 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 2205 | |
| 2206 | return err_p; |
| 2207 | } |
| 2208 | |
| 2209 | static void __pool_inc(struct pool *pool) |
| 2210 | { |
| 2211 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2212 | pool->ref_count++; |
| 2213 | } |
| 2214 | |
| 2215 | static void __pool_dec(struct pool *pool) |
| 2216 | { |
| 2217 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2218 | BUG_ON(!pool->ref_count); |
| 2219 | if (!--pool->ref_count) |
| 2220 | __pool_destroy(pool); |
| 2221 | } |
| 2222 | |
| 2223 | static struct pool *__pool_find(struct mapped_device *pool_md, |
| 2224 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2225 | unsigned long block_size, int read_only, |
| 2226 | char **error, int *created) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2227 | { |
| 2228 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); |
| 2229 | |
| 2230 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2231 | if (pool->pool_md != pool_md) { |
| 2232 | *error = "metadata device already in use by a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2233 | return ERR_PTR(-EBUSY); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2234 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2235 | __pool_inc(pool); |
| 2236 | |
| 2237 | } else { |
| 2238 | pool = __pool_table_lookup(pool_md); |
| 2239 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2240 | if (pool->md_dev != metadata_dev) { |
| 2241 | *error = "different pool cannot replace a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2242 | return ERR_PTR(-EINVAL); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2243 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2244 | __pool_inc(pool); |
| 2245 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2246 | } else { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2247 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2248 | *created = 1; |
| 2249 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2250 | } |
| 2251 | |
| 2252 | return pool; |
| 2253 | } |
| 2254 | |
| 2255 | /*---------------------------------------------------------------- |
| 2256 | * Pool target methods |
| 2257 | *--------------------------------------------------------------*/ |
| 2258 | static void pool_dtr(struct dm_target *ti) |
| 2259 | { |
| 2260 | struct pool_c *pt = ti->private; |
| 2261 | |
| 2262 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2263 | |
| 2264 | unbind_control_target(pt->pool, ti); |
| 2265 | __pool_dec(pt->pool); |
| 2266 | dm_put_device(ti, pt->metadata_dev); |
| 2267 | dm_put_device(ti, pt->data_dev); |
| 2268 | kfree(pt); |
| 2269 | |
| 2270 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2271 | } |
| 2272 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2273 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
| 2274 | struct dm_target *ti) |
| 2275 | { |
| 2276 | int r; |
| 2277 | unsigned argc; |
| 2278 | const char *arg_name; |
| 2279 | |
| 2280 | static struct dm_arg _args[] = { |
Mike Snitzer | 74aa45c | 2014-01-15 19:07:58 -0500 | [diff] [blame] | 2281 | {0, 4, "Invalid number of pool feature arguments"}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2282 | }; |
| 2283 | |
| 2284 | /* |
| 2285 | * No feature arguments supplied. |
| 2286 | */ |
| 2287 | if (!as->argc) |
| 2288 | return 0; |
| 2289 | |
| 2290 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
| 2291 | if (r) |
| 2292 | return -EINVAL; |
| 2293 | |
| 2294 | while (argc && !r) { |
| 2295 | arg_name = dm_shift_arg(as); |
| 2296 | argc--; |
| 2297 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2298 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2299 | pf->zero_new_blocks = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2300 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2301 | else if (!strcasecmp(arg_name, "ignore_discard")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2302 | pf->discard_enabled = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2303 | |
| 2304 | else if (!strcasecmp(arg_name, "no_discard_passdown")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2305 | pf->discard_passdown = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2306 | |
| 2307 | else if (!strcasecmp(arg_name, "read_only")) |
| 2308 | pf->mode = PM_READ_ONLY; |
| 2309 | |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2310 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
| 2311 | pf->error_if_no_space = true; |
| 2312 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2313 | else { |
| 2314 | ti->error = "Unrecognised pool feature requested"; |
| 2315 | r = -EINVAL; |
| 2316 | break; |
| 2317 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2318 | } |
| 2319 | |
| 2320 | return r; |
| 2321 | } |
| 2322 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2323 | static void metadata_low_callback(void *context) |
| 2324 | { |
| 2325 | struct pool *pool = context; |
| 2326 | |
| 2327 | DMWARN("%s: reached low water mark for metadata device: sending event.", |
| 2328 | dm_device_name(pool->pool_md)); |
| 2329 | |
| 2330 | dm_table_event(pool->ti->table); |
| 2331 | } |
| 2332 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2333 | static sector_t get_dev_size(struct block_device *bdev) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2334 | { |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2335 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; |
| 2336 | } |
| 2337 | |
| 2338 | static void warn_if_metadata_device_too_big(struct block_device *bdev) |
| 2339 | { |
| 2340 | sector_t metadata_dev_size = get_dev_size(bdev); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2341 | char buffer[BDEVNAME_SIZE]; |
| 2342 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2343 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2344 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
| 2345 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2346 | } |
| 2347 | |
| 2348 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
| 2349 | { |
| 2350 | sector_t metadata_dev_size = get_dev_size(bdev); |
| 2351 | |
| 2352 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) |
| 2353 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2354 | |
| 2355 | return metadata_dev_size; |
| 2356 | } |
| 2357 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2358 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
| 2359 | { |
| 2360 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); |
| 2361 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2362 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2363 | |
| 2364 | return metadata_dev_size; |
| 2365 | } |
| 2366 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2367 | /* |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2368 | * When a metadata threshold is crossed a dm event is triggered, and |
| 2369 | * userland should respond by growing the metadata device. We could let |
| 2370 | * userland set the threshold, like we do with the data threshold, but I'm |
| 2371 | * not sure they know enough to do this well. |
| 2372 | */ |
| 2373 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) |
| 2374 | { |
| 2375 | /* |
| 2376 | * 4M is ample for all ops with the possible exception of thin |
| 2377 | * device deletion which is harmless if it fails (just retry the |
| 2378 | * delete after you've grown the device). |
| 2379 | */ |
| 2380 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; |
| 2381 | return min((dm_block_t)1024ULL /* 4M */, quarter); |
| 2382 | } |
| 2383 | |
| 2384 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2385 | * thin-pool <metadata dev> <data dev> |
| 2386 | * <data block size (sectors)> |
| 2387 | * <low water mark (blocks)> |
| 2388 | * [<#feature args> [<arg>]*] |
| 2389 | * |
| 2390 | * Optional feature arguments are: |
| 2391 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2392 | * ignore_discard: disable discard |
| 2393 | * no_discard_passdown: don't pass discards down to the data device |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2394 | * read_only: Don't allow any changes to be made to the pool metadata. |
| 2395 | * error_if_no_space: error IOs, instead of queueing, if no space. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2396 | */ |
| 2397 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 2398 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2399 | int r, pool_created = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2400 | struct pool_c *pt; |
| 2401 | struct pool *pool; |
| 2402 | struct pool_features pf; |
| 2403 | struct dm_arg_set as; |
| 2404 | struct dm_dev *data_dev; |
| 2405 | unsigned long block_size; |
| 2406 | dm_block_t low_water_blocks; |
| 2407 | struct dm_dev *metadata_dev; |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2408 | fmode_t metadata_mode; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2409 | |
| 2410 | /* |
| 2411 | * FIXME Remove validation from scope of lock. |
| 2412 | */ |
| 2413 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2414 | |
| 2415 | if (argc < 4) { |
| 2416 | ti->error = "Invalid argument count"; |
| 2417 | r = -EINVAL; |
| 2418 | goto out_unlock; |
| 2419 | } |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2420 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2421 | as.argc = argc; |
| 2422 | as.argv = argv; |
| 2423 | |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2424 | /* |
| 2425 | * Set default pool features. |
| 2426 | */ |
| 2427 | pool_features_init(&pf); |
| 2428 | |
| 2429 | dm_consume_args(&as, 4); |
| 2430 | r = parse_pool_features(&as, &pf, ti); |
| 2431 | if (r) |
| 2432 | goto out_unlock; |
| 2433 | |
| 2434 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); |
| 2435 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2436 | if (r) { |
| 2437 | ti->error = "Error opening metadata block device"; |
| 2438 | goto out_unlock; |
| 2439 | } |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2440 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2441 | |
| 2442 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); |
| 2443 | if (r) { |
| 2444 | ti->error = "Error getting data device"; |
| 2445 | goto out_metadata; |
| 2446 | } |
| 2447 | |
| 2448 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || |
| 2449 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || |
| 2450 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 2451 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2452 | ti->error = "Invalid block size"; |
| 2453 | r = -EINVAL; |
| 2454 | goto out; |
| 2455 | } |
| 2456 | |
| 2457 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { |
| 2458 | ti->error = "Invalid low water mark"; |
| 2459 | r = -EINVAL; |
| 2460 | goto out; |
| 2461 | } |
| 2462 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2463 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
| 2464 | if (!pt) { |
| 2465 | r = -ENOMEM; |
| 2466 | goto out; |
| 2467 | } |
| 2468 | |
| 2469 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2470 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2471 | if (IS_ERR(pool)) { |
| 2472 | r = PTR_ERR(pool); |
| 2473 | goto out_free_pt; |
| 2474 | } |
| 2475 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2476 | /* |
| 2477 | * 'pool_created' reflects whether this is the first table load. |
| 2478 | * Top level discard support is not allowed to be changed after |
| 2479 | * initial load. This would require a pool reload to trigger thin |
| 2480 | * device changes. |
| 2481 | */ |
| 2482 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { |
| 2483 | ti->error = "Discard support cannot be disabled once enabled"; |
| 2484 | r = -EINVAL; |
| 2485 | goto out_flags_changed; |
| 2486 | } |
| 2487 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2488 | pt->pool = pool; |
| 2489 | pt->ti = ti; |
| 2490 | pt->metadata_dev = metadata_dev; |
| 2491 | pt->data_dev = data_dev; |
| 2492 | pt->low_water_blocks = low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2493 | pt->adjusted_pf = pt->requested_pf = pf; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 2494 | ti->num_flush_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2495 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2496 | /* |
| 2497 | * Only need to enable discards if the pool should pass |
| 2498 | * them down to the data device. The thin device's discard |
| 2499 | * processing will cause mappings to be removed from the btree. |
| 2500 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 2501 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2502 | if (pf.discard_enabled && pf.discard_passdown) { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 2503 | ti->num_discard_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2504 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2505 | /* |
| 2506 | * Setting 'discards_supported' circumvents the normal |
| 2507 | * stacking of discard limits (this keeps the pool and |
| 2508 | * thin devices' discard limits consistent). |
| 2509 | */ |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 2510 | ti->discards_supported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2511 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2512 | ti->private = pt; |
| 2513 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 2514 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
| 2515 | calc_metadata_threshold(pt), |
| 2516 | metadata_low_callback, |
| 2517 | pool); |
| 2518 | if (r) |
| 2519 | goto out_free_pt; |
| 2520 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2521 | pt->callbacks.congested_fn = pool_is_congested; |
| 2522 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); |
| 2523 | |
| 2524 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2525 | |
| 2526 | return 0; |
| 2527 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2528 | out_flags_changed: |
| 2529 | __pool_dec(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2530 | out_free_pt: |
| 2531 | kfree(pt); |
| 2532 | out: |
| 2533 | dm_put_device(ti, data_dev); |
| 2534 | out_metadata: |
| 2535 | dm_put_device(ti, metadata_dev); |
| 2536 | out_unlock: |
| 2537 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2538 | |
| 2539 | return r; |
| 2540 | } |
| 2541 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 2542 | static int pool_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2543 | { |
| 2544 | int r; |
| 2545 | struct pool_c *pt = ti->private; |
| 2546 | struct pool *pool = pt->pool; |
| 2547 | unsigned long flags; |
| 2548 | |
| 2549 | /* |
| 2550 | * As this is a singleton target, ti->begin is always zero. |
| 2551 | */ |
| 2552 | spin_lock_irqsave(&pool->lock, flags); |
| 2553 | bio->bi_bdev = pt->data_dev->bdev; |
| 2554 | r = DM_MAPIO_REMAPPED; |
| 2555 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2556 | |
| 2557 | return r; |
| 2558 | } |
| 2559 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2560 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
| 2561 | { |
| 2562 | int r; |
| 2563 | struct pool_c *pt = ti->private; |
| 2564 | struct pool *pool = pt->pool; |
| 2565 | sector_t data_size = ti->len; |
| 2566 | dm_block_t sb_data_size; |
| 2567 | |
| 2568 | *need_commit = false; |
| 2569 | |
| 2570 | (void) sector_div(data_size, pool->sectors_per_block); |
| 2571 | |
| 2572 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
| 2573 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2574 | DMERR("%s: failed to retrieve data device size", |
| 2575 | dm_device_name(pool->pool_md)); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2576 | return r; |
| 2577 | } |
| 2578 | |
| 2579 | if (data_size < sb_data_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2580 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
| 2581 | dm_device_name(pool->pool_md), |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2582 | (unsigned long long)data_size, sb_data_size); |
| 2583 | return -EINVAL; |
| 2584 | |
| 2585 | } else if (data_size > sb_data_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2586 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 2587 | DMERR("%s: unable to grow the data device until repaired.", |
| 2588 | dm_device_name(pool->pool_md)); |
| 2589 | return 0; |
| 2590 | } |
| 2591 | |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 2592 | if (sb_data_size) |
| 2593 | DMINFO("%s: growing the data device from %llu to %llu blocks", |
| 2594 | dm_device_name(pool->pool_md), |
| 2595 | sb_data_size, (unsigned long long)data_size); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2596 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
| 2597 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2598 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2599 | return r; |
| 2600 | } |
| 2601 | |
| 2602 | *need_commit = true; |
| 2603 | } |
| 2604 | |
| 2605 | return 0; |
| 2606 | } |
| 2607 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2608 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
| 2609 | { |
| 2610 | int r; |
| 2611 | struct pool_c *pt = ti->private; |
| 2612 | struct pool *pool = pt->pool; |
| 2613 | dm_block_t metadata_dev_size, sb_metadata_dev_size; |
| 2614 | |
| 2615 | *need_commit = false; |
| 2616 | |
Alasdair G Kergon | 610bba8 | 2013-05-19 18:57:50 +0100 | [diff] [blame] | 2617 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2618 | |
| 2619 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); |
| 2620 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2621 | DMERR("%s: failed to retrieve metadata device size", |
| 2622 | dm_device_name(pool->pool_md)); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2623 | return r; |
| 2624 | } |
| 2625 | |
| 2626 | if (metadata_dev_size < sb_metadata_dev_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2627 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
| 2628 | dm_device_name(pool->pool_md), |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2629 | metadata_dev_size, sb_metadata_dev_size); |
| 2630 | return -EINVAL; |
| 2631 | |
| 2632 | } else if (metadata_dev_size > sb_metadata_dev_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2633 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 2634 | DMERR("%s: unable to grow the metadata device until repaired.", |
| 2635 | dm_device_name(pool->pool_md)); |
| 2636 | return 0; |
| 2637 | } |
| 2638 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 2639 | warn_if_metadata_device_too_big(pool->md_dev); |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 2640 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
| 2641 | dm_device_name(pool->pool_md), |
| 2642 | sb_metadata_dev_size, metadata_dev_size); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2643 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
| 2644 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2645 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2646 | return r; |
| 2647 | } |
| 2648 | |
| 2649 | *need_commit = true; |
| 2650 | } |
| 2651 | |
| 2652 | return 0; |
| 2653 | } |
| 2654 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2655 | /* |
| 2656 | * Retrieves the number of blocks of the data device from |
| 2657 | * the superblock and compares it to the actual device size, |
| 2658 | * thus resizing the data device in case it has grown. |
| 2659 | * |
| 2660 | * This both copes with opening preallocated data devices in the ctr |
| 2661 | * being followed by a resume |
| 2662 | * -and- |
| 2663 | * calling the resume method individually after userspace has |
| 2664 | * grown the data device in reaction to a table event. |
| 2665 | */ |
| 2666 | static int pool_preresume(struct dm_target *ti) |
| 2667 | { |
| 2668 | int r; |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2669 | bool need_commit1, need_commit2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2670 | struct pool_c *pt = ti->private; |
| 2671 | struct pool *pool = pt->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2672 | |
| 2673 | /* |
| 2674 | * Take control of the pool object. |
| 2675 | */ |
| 2676 | r = bind_control_target(pool, ti); |
| 2677 | if (r) |
| 2678 | return r; |
| 2679 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 2680 | r = maybe_resize_data_dev(ti, &need_commit1); |
| 2681 | if (r) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2682 | return r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2683 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 2684 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
| 2685 | if (r) |
| 2686 | return r; |
| 2687 | |
| 2688 | if (need_commit1 || need_commit2) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2689 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2690 | |
| 2691 | return 0; |
| 2692 | } |
| 2693 | |
| 2694 | static void pool_resume(struct dm_target *ti) |
| 2695 | { |
| 2696 | struct pool_c *pt = ti->private; |
| 2697 | struct pool *pool = pt->pool; |
| 2698 | unsigned long flags; |
| 2699 | |
| 2700 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 2701 | pool->low_water_triggered = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2702 | spin_unlock_irqrestore(&pool->lock, flags); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2703 | requeue_bios(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2704 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2705 | do_waker(&pool->waker.work); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2706 | } |
| 2707 | |
| 2708 | static void pool_postsuspend(struct dm_target *ti) |
| 2709 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2710 | struct pool_c *pt = ti->private; |
| 2711 | struct pool *pool = pt->pool; |
| 2712 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2713 | cancel_delayed_work(&pool->waker); |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2714 | cancel_delayed_work(&pool->no_space_timeout); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2715 | flush_workqueue(pool->wq); |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2716 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2717 | } |
| 2718 | |
| 2719 | static int check_arg_count(unsigned argc, unsigned args_required) |
| 2720 | { |
| 2721 | if (argc != args_required) { |
| 2722 | DMWARN("Message received with %u arguments instead of %u.", |
| 2723 | argc, args_required); |
| 2724 | return -EINVAL; |
| 2725 | } |
| 2726 | |
| 2727 | return 0; |
| 2728 | } |
| 2729 | |
| 2730 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) |
| 2731 | { |
| 2732 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && |
| 2733 | *dev_id <= MAX_DEV_ID) |
| 2734 | return 0; |
| 2735 | |
| 2736 | if (warning) |
| 2737 | DMWARN("Message received with invalid device id: %s", arg); |
| 2738 | |
| 2739 | return -EINVAL; |
| 2740 | } |
| 2741 | |
| 2742 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2743 | { |
| 2744 | dm_thin_id dev_id; |
| 2745 | int r; |
| 2746 | |
| 2747 | r = check_arg_count(argc, 2); |
| 2748 | if (r) |
| 2749 | return r; |
| 2750 | |
| 2751 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2752 | if (r) |
| 2753 | return r; |
| 2754 | |
| 2755 | r = dm_pool_create_thin(pool->pmd, dev_id); |
| 2756 | if (r) { |
| 2757 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", |
| 2758 | argv[1]); |
| 2759 | return r; |
| 2760 | } |
| 2761 | |
| 2762 | return 0; |
| 2763 | } |
| 2764 | |
| 2765 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2766 | { |
| 2767 | dm_thin_id dev_id; |
| 2768 | dm_thin_id origin_dev_id; |
| 2769 | int r; |
| 2770 | |
| 2771 | r = check_arg_count(argc, 3); |
| 2772 | if (r) |
| 2773 | return r; |
| 2774 | |
| 2775 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2776 | if (r) |
| 2777 | return r; |
| 2778 | |
| 2779 | r = read_dev_id(argv[2], &origin_dev_id, 1); |
| 2780 | if (r) |
| 2781 | return r; |
| 2782 | |
| 2783 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); |
| 2784 | if (r) { |
| 2785 | DMWARN("Creation of new snapshot %s of device %s failed.", |
| 2786 | argv[1], argv[2]); |
| 2787 | return r; |
| 2788 | } |
| 2789 | |
| 2790 | return 0; |
| 2791 | } |
| 2792 | |
| 2793 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2794 | { |
| 2795 | dm_thin_id dev_id; |
| 2796 | int r; |
| 2797 | |
| 2798 | r = check_arg_count(argc, 2); |
| 2799 | if (r) |
| 2800 | return r; |
| 2801 | |
| 2802 | r = read_dev_id(argv[1], &dev_id, 1); |
| 2803 | if (r) |
| 2804 | return r; |
| 2805 | |
| 2806 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); |
| 2807 | if (r) |
| 2808 | DMWARN("Deletion of thin device %s failed.", argv[1]); |
| 2809 | |
| 2810 | return r; |
| 2811 | } |
| 2812 | |
| 2813 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2814 | { |
| 2815 | dm_thin_id old_id, new_id; |
| 2816 | int r; |
| 2817 | |
| 2818 | r = check_arg_count(argc, 3); |
| 2819 | if (r) |
| 2820 | return r; |
| 2821 | |
| 2822 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { |
| 2823 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); |
| 2824 | return -EINVAL; |
| 2825 | } |
| 2826 | |
| 2827 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { |
| 2828 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); |
| 2829 | return -EINVAL; |
| 2830 | } |
| 2831 | |
| 2832 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); |
| 2833 | if (r) { |
| 2834 | DMWARN("Failed to change transaction id from %s to %s.", |
| 2835 | argv[1], argv[2]); |
| 2836 | return r; |
| 2837 | } |
| 2838 | |
| 2839 | return 0; |
| 2840 | } |
| 2841 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2842 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2843 | { |
| 2844 | int r; |
| 2845 | |
| 2846 | r = check_arg_count(argc, 1); |
| 2847 | if (r) |
| 2848 | return r; |
| 2849 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2850 | (void) commit(pool); |
Joe Thornber | 0d200ae | 2012-07-03 12:55:31 +0100 | [diff] [blame] | 2851 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2852 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
| 2853 | if (r) |
| 2854 | DMWARN("reserve_metadata_snap message failed."); |
| 2855 | |
| 2856 | return r; |
| 2857 | } |
| 2858 | |
| 2859 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 2860 | { |
| 2861 | int r; |
| 2862 | |
| 2863 | r = check_arg_count(argc, 1); |
| 2864 | if (r) |
| 2865 | return r; |
| 2866 | |
| 2867 | r = dm_pool_release_metadata_snap(pool->pmd); |
| 2868 | if (r) |
| 2869 | DMWARN("release_metadata_snap message failed."); |
| 2870 | |
| 2871 | return r; |
| 2872 | } |
| 2873 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2874 | /* |
| 2875 | * Messages supported: |
| 2876 | * create_thin <dev_id> |
| 2877 | * create_snap <dev_id> <origin_id> |
| 2878 | * delete <dev_id> |
| 2879 | * trim <dev_id> <new_size_in_sectors> |
| 2880 | * set_transaction_id <current_trans_id> <new_trans_id> |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2881 | * reserve_metadata_snap |
| 2882 | * release_metadata_snap |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2883 | */ |
| 2884 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) |
| 2885 | { |
| 2886 | int r = -EINVAL; |
| 2887 | struct pool_c *pt = ti->private; |
| 2888 | struct pool *pool = pt->pool; |
| 2889 | |
| 2890 | if (!strcasecmp(argv[0], "create_thin")) |
| 2891 | r = process_create_thin_mesg(argc, argv, pool); |
| 2892 | |
| 2893 | else if (!strcasecmp(argv[0], "create_snap")) |
| 2894 | r = process_create_snap_mesg(argc, argv, pool); |
| 2895 | |
| 2896 | else if (!strcasecmp(argv[0], "delete")) |
| 2897 | r = process_delete_mesg(argc, argv, pool); |
| 2898 | |
| 2899 | else if (!strcasecmp(argv[0], "set_transaction_id")) |
| 2900 | r = process_set_transaction_id_mesg(argc, argv, pool); |
| 2901 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 2902 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
| 2903 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); |
| 2904 | |
| 2905 | else if (!strcasecmp(argv[0], "release_metadata_snap")) |
| 2906 | r = process_release_metadata_snap_mesg(argc, argv, pool); |
| 2907 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2908 | else |
| 2909 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); |
| 2910 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2911 | if (!r) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2912 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2913 | |
| 2914 | return r; |
| 2915 | } |
| 2916 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2917 | static void emit_flags(struct pool_features *pf, char *result, |
| 2918 | unsigned sz, unsigned maxlen) |
| 2919 | { |
| 2920 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2921 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
| 2922 | pf->error_if_no_space; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2923 | DMEMIT("%u ", count); |
| 2924 | |
| 2925 | if (!pf->zero_new_blocks) |
| 2926 | DMEMIT("skip_block_zeroing "); |
| 2927 | |
| 2928 | if (!pf->discard_enabled) |
| 2929 | DMEMIT("ignore_discard "); |
| 2930 | |
| 2931 | if (!pf->discard_passdown) |
| 2932 | DMEMIT("no_discard_passdown "); |
| 2933 | |
| 2934 | if (pf->mode == PM_READ_ONLY) |
| 2935 | DMEMIT("read_only "); |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2936 | |
| 2937 | if (pf->error_if_no_space) |
| 2938 | DMEMIT("error_if_no_space "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2939 | } |
| 2940 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2941 | /* |
| 2942 | * Status line is: |
| 2943 | * <transaction id> <used metadata sectors>/<total metadata sectors> |
| 2944 | * <used data sectors>/<total data sectors> <held metadata root> |
| 2945 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2946 | static void pool_status(struct dm_target *ti, status_type_t type, |
| 2947 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2948 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2949 | int r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2950 | unsigned sz = 0; |
| 2951 | uint64_t transaction_id; |
| 2952 | dm_block_t nr_free_blocks_data; |
| 2953 | dm_block_t nr_free_blocks_metadata; |
| 2954 | dm_block_t nr_blocks_data; |
| 2955 | dm_block_t nr_blocks_metadata; |
| 2956 | dm_block_t held_root; |
| 2957 | char buf[BDEVNAME_SIZE]; |
| 2958 | char buf2[BDEVNAME_SIZE]; |
| 2959 | struct pool_c *pt = ti->private; |
| 2960 | struct pool *pool = pt->pool; |
| 2961 | |
| 2962 | switch (type) { |
| 2963 | case STATUSTYPE_INFO: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2964 | if (get_pool_mode(pool) == PM_FAIL) { |
| 2965 | DMEMIT("Fail"); |
| 2966 | break; |
| 2967 | } |
| 2968 | |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2969 | /* Commit to ensure statistics aren't out-of-date */ |
| 2970 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2971 | (void) commit(pool); |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2972 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2973 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
| 2974 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2975 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
| 2976 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2977 | goto err; |
| 2978 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2979 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2980 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
| 2981 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2982 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
| 2983 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2984 | goto err; |
| 2985 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2986 | |
| 2987 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2988 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2989 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
| 2990 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2991 | goto err; |
| 2992 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2993 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2994 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
| 2995 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 2996 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
| 2997 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2998 | goto err; |
| 2999 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3000 | |
| 3001 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3002 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3003 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
| 3004 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3005 | goto err; |
| 3006 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3007 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3008 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3009 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3010 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
| 3011 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3012 | goto err; |
| 3013 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3014 | |
| 3015 | DMEMIT("%llu %llu/%llu %llu/%llu ", |
| 3016 | (unsigned long long)transaction_id, |
| 3017 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
| 3018 | (unsigned long long)nr_blocks_metadata, |
| 3019 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), |
| 3020 | (unsigned long long)nr_blocks_data); |
| 3021 | |
| 3022 | if (held_root) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3023 | DMEMIT("%llu ", held_root); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3024 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3025 | DMEMIT("- "); |
| 3026 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 3027 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
| 3028 | DMEMIT("out_of_data_space "); |
| 3029 | else if (pool->pf.mode == PM_READ_ONLY) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3030 | DMEMIT("ro "); |
| 3031 | else |
| 3032 | DMEMIT("rw "); |
| 3033 | |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 3034 | if (!pool->pf.discard_enabled) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3035 | DMEMIT("ignore_discard "); |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 3036 | else if (pool->pf.discard_passdown) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3037 | DMEMIT("discard_passdown "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3038 | else |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3039 | DMEMIT("no_discard_passdown "); |
| 3040 | |
| 3041 | if (pool->pf.error_if_no_space) |
| 3042 | DMEMIT("error_if_no_space "); |
| 3043 | else |
| 3044 | DMEMIT("queue_if_no_space "); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3045 | |
| 3046 | break; |
| 3047 | |
| 3048 | case STATUSTYPE_TABLE: |
| 3049 | DMEMIT("%s %s %lu %llu ", |
| 3050 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), |
| 3051 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), |
| 3052 | (unsigned long)pool->sectors_per_block, |
| 3053 | (unsigned long long)pt->low_water_blocks); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3054 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3055 | break; |
| 3056 | } |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3057 | return; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3058 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3059 | err: |
| 3060 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3061 | } |
| 3062 | |
| 3063 | static int pool_iterate_devices(struct dm_target *ti, |
| 3064 | iterate_devices_callout_fn fn, void *data) |
| 3065 | { |
| 3066 | struct pool_c *pt = ti->private; |
| 3067 | |
| 3068 | return fn(ti, pt->data_dev, 0, ti->len, data); |
| 3069 | } |
| 3070 | |
| 3071 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
| 3072 | struct bio_vec *biovec, int max_size) |
| 3073 | { |
| 3074 | struct pool_c *pt = ti->private; |
| 3075 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 3076 | |
| 3077 | if (!q->merge_bvec_fn) |
| 3078 | return max_size; |
| 3079 | |
| 3080 | bvm->bi_bdev = pt->data_dev->bdev; |
| 3081 | |
| 3082 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); |
| 3083 | } |
| 3084 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3085 | 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] | 3086 | { |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3087 | struct pool *pool = pt->pool; |
| 3088 | struct queue_limits *data_limits; |
| 3089 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3090 | limits->max_discard_sectors = pool->sectors_per_block; |
| 3091 | |
| 3092 | /* |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3093 | * discard_granularity is just a hint, and not enforced. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3094 | */ |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3095 | if (pt->adjusted_pf.discard_passdown) { |
| 3096 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; |
Lukas Czerner | 09869de | 2014-06-11 12:28:43 -0400 | [diff] [blame] | 3097 | limits->discard_granularity = max(data_limits->discard_granularity, |
| 3098 | pool->sectors_per_block << SECTOR_SHIFT); |
Mike Snitzer | f13945d | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3099 | } else |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3100 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3101 | } |
| 3102 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3103 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 3104 | { |
| 3105 | struct pool_c *pt = ti->private; |
| 3106 | struct pool *pool = pt->pool; |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 3107 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3108 | |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 3109 | /* |
| 3110 | * If the system-determined stacked limits are compatible with the |
| 3111 | * pool's blocksize (io_opt is a factor) do not override them. |
| 3112 | */ |
| 3113 | if (io_opt_sectors < pool->sectors_per_block || |
| 3114 | do_div(io_opt_sectors, pool->sectors_per_block)) { |
| 3115 | blk_limits_io_min(limits, 0); |
| 3116 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
| 3117 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3118 | |
| 3119 | /* |
| 3120 | * pt->adjusted_pf is a staging area for the actual features to use. |
| 3121 | * They get transferred to the live pool in bind_control_target() |
| 3122 | * called from pool_preresume(). |
| 3123 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3124 | if (!pt->adjusted_pf.discard_enabled) { |
| 3125 | /* |
| 3126 | * Must explicitly disallow stacking discard limits otherwise the |
| 3127 | * block layer will stack them if pool's data device has support. |
| 3128 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the |
| 3129 | * user to see that, so make sure to set all discard limits to 0. |
| 3130 | */ |
| 3131 | limits->discard_granularity = 0; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3132 | return; |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3133 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3134 | |
| 3135 | disable_passdown_if_not_supported(pt); |
| 3136 | |
| 3137 | set_discard_limits(pt, limits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3138 | } |
| 3139 | |
| 3140 | static struct target_type pool_target = { |
| 3141 | .name = "thin-pool", |
| 3142 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | |
| 3143 | DM_TARGET_IMMUTABLE, |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 3144 | .version = {1, 12, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3145 | .module = THIS_MODULE, |
| 3146 | .ctr = pool_ctr, |
| 3147 | .dtr = pool_dtr, |
| 3148 | .map = pool_map, |
| 3149 | .postsuspend = pool_postsuspend, |
| 3150 | .preresume = pool_preresume, |
| 3151 | .resume = pool_resume, |
| 3152 | .message = pool_message, |
| 3153 | .status = pool_status, |
| 3154 | .merge = pool_merge, |
| 3155 | .iterate_devices = pool_iterate_devices, |
| 3156 | .io_hints = pool_io_hints, |
| 3157 | }; |
| 3158 | |
| 3159 | /*---------------------------------------------------------------- |
| 3160 | * Thin target methods |
| 3161 | *--------------------------------------------------------------*/ |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 3162 | static void thin_get(struct thin_c *tc) |
| 3163 | { |
| 3164 | atomic_inc(&tc->refcount); |
| 3165 | } |
| 3166 | |
| 3167 | static void thin_put(struct thin_c *tc) |
| 3168 | { |
| 3169 | if (atomic_dec_and_test(&tc->refcount)) |
| 3170 | complete(&tc->can_destroy); |
| 3171 | } |
| 3172 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3173 | static void thin_dtr(struct dm_target *ti) |
| 3174 | { |
| 3175 | struct thin_c *tc = ti->private; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3176 | unsigned long flags; |
| 3177 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 3178 | thin_put(tc); |
| 3179 | wait_for_completion(&tc->can_destroy); |
| 3180 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3181 | spin_lock_irqsave(&tc->pool->lock, flags); |
| 3182 | list_del_rcu(&tc->list); |
| 3183 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
| 3184 | synchronize_rcu(); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3185 | |
| 3186 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3187 | |
| 3188 | __pool_dec(tc->pool); |
| 3189 | dm_pool_close_thin_device(tc->td); |
| 3190 | dm_put_device(ti, tc->pool_dev); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3191 | if (tc->origin_dev) |
| 3192 | dm_put_device(ti, tc->origin_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3193 | kfree(tc); |
| 3194 | |
| 3195 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3196 | } |
| 3197 | |
| 3198 | /* |
| 3199 | * Thin target parameters: |
| 3200 | * |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3201 | * <pool_dev> <dev_id> [origin_dev] |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3202 | * |
| 3203 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) |
| 3204 | * dev_id: the internal device identifier |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3205 | * 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] | 3206 | * |
| 3207 | * If the pool device has discards disabled, they get disabled for the thin |
| 3208 | * device as well. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3209 | */ |
| 3210 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 3211 | { |
| 3212 | int r; |
| 3213 | struct thin_c *tc; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3214 | struct dm_dev *pool_dev, *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3215 | struct mapped_device *pool_md; |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 3216 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3217 | |
| 3218 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3219 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3220 | if (argc != 2 && argc != 3) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3221 | ti->error = "Invalid argument count"; |
| 3222 | r = -EINVAL; |
| 3223 | goto out_unlock; |
| 3224 | } |
| 3225 | |
| 3226 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); |
| 3227 | if (!tc) { |
| 3228 | ti->error = "Out of memory"; |
| 3229 | r = -ENOMEM; |
| 3230 | goto out_unlock; |
| 3231 | } |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3232 | spin_lock_init(&tc->lock); |
| 3233 | bio_list_init(&tc->deferred_bio_list); |
| 3234 | bio_list_init(&tc->retry_on_resume_list); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 3235 | tc->sort_bio_list = RB_ROOT; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3236 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3237 | if (argc == 3) { |
| 3238 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
| 3239 | if (r) { |
| 3240 | ti->error = "Error opening origin device"; |
| 3241 | goto bad_origin_dev; |
| 3242 | } |
| 3243 | tc->origin_dev = origin_dev; |
| 3244 | } |
| 3245 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3246 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
| 3247 | if (r) { |
| 3248 | ti->error = "Error opening pool device"; |
| 3249 | goto bad_pool_dev; |
| 3250 | } |
| 3251 | tc->pool_dev = pool_dev; |
| 3252 | |
| 3253 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { |
| 3254 | ti->error = "Invalid device id"; |
| 3255 | r = -EINVAL; |
| 3256 | goto bad_common; |
| 3257 | } |
| 3258 | |
| 3259 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); |
| 3260 | if (!pool_md) { |
| 3261 | ti->error = "Couldn't get pool mapped device"; |
| 3262 | r = -EINVAL; |
| 3263 | goto bad_common; |
| 3264 | } |
| 3265 | |
| 3266 | tc->pool = __pool_table_lookup(pool_md); |
| 3267 | if (!tc->pool) { |
| 3268 | ti->error = "Couldn't find pool object"; |
| 3269 | r = -EINVAL; |
| 3270 | goto bad_pool_lookup; |
| 3271 | } |
| 3272 | __pool_inc(tc->pool); |
| 3273 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3274 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 3275 | ti->error = "Couldn't open thin device, Pool is in fail mode"; |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3276 | r = -EINVAL; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3277 | goto bad_thin_open; |
| 3278 | } |
| 3279 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3280 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
| 3281 | if (r) { |
| 3282 | ti->error = "Couldn't open thin internal device"; |
| 3283 | goto bad_thin_open; |
| 3284 | } |
| 3285 | |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 3286 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
| 3287 | if (r) |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3288 | goto bad_target_max_io_len; |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 3289 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3290 | ti->num_flush_bios = 1; |
Joe Thornber | 16ad3d1 | 2012-07-27 15:08:07 +0100 | [diff] [blame] | 3291 | ti->flush_supported = true; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 3292 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3293 | |
| 3294 | /* In case the pool supports discards, pass them on. */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3295 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3296 | if (tc->pool->pf.discard_enabled) { |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 3297 | ti->discards_supported = true; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3298 | ti->num_discard_bios = 1; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3299 | /* Discard bios must be split on a block boundary */ |
| 3300 | ti->split_discard_bios = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3301 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3302 | |
| 3303 | dm_put(pool_md); |
| 3304 | |
| 3305 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3306 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 3307 | atomic_set(&tc->refcount, 1); |
| 3308 | init_completion(&tc->can_destroy); |
| 3309 | |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 3310 | spin_lock_irqsave(&tc->pool->lock, flags); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3311 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 3312 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3313 | /* |
| 3314 | * This synchronize_rcu() call is needed here otherwise we risk a |
| 3315 | * wake_worker() call finding no bios to process (because the newly |
| 3316 | * added tc isn't yet visible). So this reduces latency since we |
| 3317 | * aren't then dependent on the periodic commit to wake_worker(). |
| 3318 | */ |
| 3319 | synchronize_rcu(); |
| 3320 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3321 | return 0; |
| 3322 | |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 3323 | bad_target_max_io_len: |
| 3324 | dm_pool_close_thin_device(tc->td); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3325 | bad_thin_open: |
| 3326 | __pool_dec(tc->pool); |
| 3327 | bad_pool_lookup: |
| 3328 | dm_put(pool_md); |
| 3329 | bad_common: |
| 3330 | dm_put_device(ti, tc->pool_dev); |
| 3331 | bad_pool_dev: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3332 | if (tc->origin_dev) |
| 3333 | dm_put_device(ti, tc->origin_dev); |
| 3334 | bad_origin_dev: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3335 | kfree(tc); |
| 3336 | out_unlock: |
| 3337 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3338 | |
| 3339 | return r; |
| 3340 | } |
| 3341 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3342 | static int thin_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3343 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 3344 | 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] | 3345 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3346 | return thin_bio_map(ti, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3347 | } |
| 3348 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3349 | 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] | 3350 | { |
| 3351 | unsigned long flags; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 3352 | 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] | 3353 | struct list_head work; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3354 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3355 | struct pool *pool = h->tc->pool; |
| 3356 | |
| 3357 | if (h->shared_read_entry) { |
| 3358 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 3359 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3360 | |
| 3361 | spin_lock_irqsave(&pool->lock, flags); |
| 3362 | list_for_each_entry_safe(m, tmp, &work, list) { |
| 3363 | list_del(&m->list); |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 3364 | m->quiesced = true; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3365 | __maybe_add_mapping(m); |
| 3366 | } |
| 3367 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3368 | } |
| 3369 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3370 | if (h->all_io_entry) { |
| 3371 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 3372 | dm_deferred_entry_dec(h->all_io_entry, &work); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 3373 | if (!list_empty(&work)) { |
| 3374 | spin_lock_irqsave(&pool->lock, flags); |
| 3375 | list_for_each_entry_safe(m, tmp, &work, list) |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 3376 | list_add_tail(&m->list, &pool->prepared_discards); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 3377 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3378 | wake_worker(pool); |
| 3379 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3380 | } |
| 3381 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3382 | return 0; |
| 3383 | } |
| 3384 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3385 | static void thin_presuspend(struct dm_target *ti) |
| 3386 | { |
| 3387 | struct thin_c *tc = ti->private; |
| 3388 | |
| 3389 | if (dm_noflush_suspending(ti)) |
| 3390 | noflush_work(tc, do_noflush_start); |
| 3391 | } |
| 3392 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3393 | static void thin_postsuspend(struct dm_target *ti) |
| 3394 | { |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3395 | struct thin_c *tc = ti->private; |
| 3396 | |
| 3397 | /* |
| 3398 | * The dm_noflush_suspending flag has been cleared by now, so |
| 3399 | * unfortunately we must always run this. |
| 3400 | */ |
| 3401 | noflush_work(tc, do_noflush_stop); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3402 | } |
| 3403 | |
| 3404 | /* |
| 3405 | * <nr mapped sectors> <highest mapped sector> |
| 3406 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3407 | static void thin_status(struct dm_target *ti, status_type_t type, |
| 3408 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3409 | { |
| 3410 | int r; |
| 3411 | ssize_t sz = 0; |
| 3412 | dm_block_t mapped, highest; |
| 3413 | char buf[BDEVNAME_SIZE]; |
| 3414 | struct thin_c *tc = ti->private; |
| 3415 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3416 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 3417 | DMEMIT("Fail"); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3418 | return; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3419 | } |
| 3420 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3421 | if (!tc->td) |
| 3422 | DMEMIT("-"); |
| 3423 | else { |
| 3424 | switch (type) { |
| 3425 | case STATUSTYPE_INFO: |
| 3426 | r = dm_thin_get_mapped_count(tc->td, &mapped); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3427 | if (r) { |
| 3428 | DMERR("dm_thin_get_mapped_count returned %d", r); |
| 3429 | goto err; |
| 3430 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3431 | |
| 3432 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3433 | if (r < 0) { |
| 3434 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); |
| 3435 | goto err; |
| 3436 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3437 | |
| 3438 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); |
| 3439 | if (r) |
| 3440 | DMEMIT("%llu", ((highest + 1) * |
| 3441 | tc->pool->sectors_per_block) - 1); |
| 3442 | else |
| 3443 | DMEMIT("-"); |
| 3444 | break; |
| 3445 | |
| 3446 | case STATUSTYPE_TABLE: |
| 3447 | DMEMIT("%s %lu", |
| 3448 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), |
| 3449 | (unsigned long) tc->dev_id); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3450 | if (tc->origin_dev) |
| 3451 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3452 | break; |
| 3453 | } |
| 3454 | } |
| 3455 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3456 | return; |
| 3457 | |
| 3458 | err: |
| 3459 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3460 | } |
| 3461 | |
| 3462 | static int thin_iterate_devices(struct dm_target *ti, |
| 3463 | iterate_devices_callout_fn fn, void *data) |
| 3464 | { |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3465 | sector_t blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3466 | struct thin_c *tc = ti->private; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3467 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3468 | |
| 3469 | /* |
| 3470 | * We can't call dm_pool_get_data_dev_size() since that blocks. So |
| 3471 | * we follow a more convoluted path through to the pool's target. |
| 3472 | */ |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3473 | if (!pool->ti) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3474 | return 0; /* nothing is bound */ |
| 3475 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3476 | blocks = pool->ti->len; |
| 3477 | (void) sector_div(blocks, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3478 | if (blocks) |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3479 | 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] | 3480 | |
| 3481 | return 0; |
| 3482 | } |
| 3483 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3484 | static struct target_type thin_target = { |
| 3485 | .name = "thin", |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 3486 | .version = {1, 12, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3487 | .module = THIS_MODULE, |
| 3488 | .ctr = thin_ctr, |
| 3489 | .dtr = thin_dtr, |
| 3490 | .map = thin_map, |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3491 | .end_io = thin_endio, |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 3492 | .presuspend = thin_presuspend, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3493 | .postsuspend = thin_postsuspend, |
| 3494 | .status = thin_status, |
| 3495 | .iterate_devices = thin_iterate_devices, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3496 | }; |
| 3497 | |
| 3498 | /*----------------------------------------------------------------*/ |
| 3499 | |
| 3500 | static int __init dm_thin_init(void) |
| 3501 | { |
| 3502 | int r; |
| 3503 | |
| 3504 | pool_table_init(); |
| 3505 | |
| 3506 | r = dm_register_target(&thin_target); |
| 3507 | if (r) |
| 3508 | return r; |
| 3509 | |
| 3510 | r = dm_register_target(&pool_target); |
| 3511 | if (r) |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3512 | goto bad_pool_target; |
| 3513 | |
| 3514 | r = -ENOMEM; |
| 3515 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3516 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
| 3517 | if (!_new_mapping_cache) |
| 3518 | goto bad_new_mapping_cache; |
| 3519 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3520 | return 0; |
| 3521 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3522 | bad_new_mapping_cache: |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3523 | dm_unregister_target(&pool_target); |
| 3524 | bad_pool_target: |
| 3525 | dm_unregister_target(&thin_target); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3526 | |
| 3527 | return r; |
| 3528 | } |
| 3529 | |
| 3530 | static void dm_thin_exit(void) |
| 3531 | { |
| 3532 | dm_unregister_target(&thin_target); |
| 3533 | dm_unregister_target(&pool_target); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3534 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 3535 | kmem_cache_destroy(_new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3536 | } |
| 3537 | |
| 3538 | module_init(dm_thin_init); |
| 3539 | module_exit(dm_thin_exit); |
| 3540 | |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 3541 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
| 3542 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); |
| 3543 | |
Alasdair G Kergon | 7cab8bf | 2012-05-12 01:43:19 +0100 | [diff] [blame] | 3544 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3545 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 3546 | MODULE_LICENSE("GPL"); |