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> |
Manuel Schölling | 0f30af9 | 2014-05-22 22:42:37 +0200 | [diff] [blame] | 14 | #include <linux/jiffies.h> |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 15 | #include <linux/log2.h> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 16 | #include <linux/list.h> |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 17 | #include <linux/rculist.h> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 18 | #include <linux/init.h> |
| 19 | #include <linux/module.h> |
| 20 | #include <linux/slab.h> |
Joe Thornber | a822c83 | 2015-07-03 10:22:42 +0100 | [diff] [blame] | 21 | #include <linux/vmalloc.h> |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 22 | #include <linux/sort.h> |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 23 | #include <linux/rbtree.h> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 24 | |
| 25 | #define DM_MSG_PREFIX "thin" |
| 26 | |
| 27 | /* |
| 28 | * Tunable constants |
| 29 | */ |
Alasdair G Kergon | 7768ed3 | 2012-07-27 15:07:57 +0100 | [diff] [blame] | 30 | #define ENDIO_HOOK_POOL_SIZE 1024 |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 31 | #define MAPPING_POOL_SIZE 1024 |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 32 | #define COMMIT_PERIOD HZ |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 33 | #define NO_SPACE_TIMEOUT_SECS 60 |
| 34 | |
| 35 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 36 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 37 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
| 38 | "A percentage of time allocated for copy on write"); |
| 39 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 40 | /* |
| 41 | * The block size of the device holding pool data must be |
| 42 | * between 64KB and 1GB. |
| 43 | */ |
| 44 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) |
| 45 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) |
| 46 | |
| 47 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 48 | * Device id is restricted to 24 bits. |
| 49 | */ |
| 50 | #define MAX_DEV_ID ((1 << 24) - 1) |
| 51 | |
| 52 | /* |
| 53 | * How do we handle breaking sharing of data blocks? |
| 54 | * ================================================= |
| 55 | * |
| 56 | * We use a standard copy-on-write btree to store the mappings for the |
| 57 | * devices (note I'm talking about copy-on-write of the metadata here, not |
| 58 | * the data). When you take an internal snapshot you clone the root node |
| 59 | * of the origin btree. After this there is no concept of an origin or a |
| 60 | * snapshot. They are just two device trees that happen to point to the |
| 61 | * same data blocks. |
| 62 | * |
| 63 | * When we get a write in we decide if it's to a shared data block using |
| 64 | * some timestamp magic. If it is, we have to break sharing. |
| 65 | * |
| 66 | * Let's say we write to a shared block in what was the origin. The |
| 67 | * steps are: |
| 68 | * |
| 69 | * i) plug io further to this physical block. (see bio_prison code). |
| 70 | * |
| 71 | * ii) quiesce any read io to that shared data block. Obviously |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 72 | * including all devices that share this block. (see dm_deferred_set code) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 73 | * |
| 74 | * iii) copy the data block to a newly allocate block. This step can be |
| 75 | * missed out if the io covers the block. (schedule_copy). |
| 76 | * |
| 77 | * iv) insert the new mapping into the origin's btree |
Joe Thornber | fe878f3 | 2012-03-28 18:41:24 +0100 | [diff] [blame] | 78 | * (process_prepared_mapping). This act of inserting breaks some |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 79 | * sharing of btree nodes between the two devices. Breaking sharing only |
| 80 | * effects the btree of that specific device. Btrees for the other |
| 81 | * devices that share the block never change. The btree for the origin |
| 82 | * device as it was after the last commit is untouched, ie. we're using |
| 83 | * persistent data structures in the functional programming sense. |
| 84 | * |
| 85 | * v) unplug io to this physical block, including the io that triggered |
| 86 | * the breaking of sharing. |
| 87 | * |
| 88 | * Steps (ii) and (iii) occur in parallel. |
| 89 | * |
| 90 | * The metadata _doesn't_ need to be committed before the io continues. We |
| 91 | * get away with this because the io is always written to a _new_ block. |
| 92 | * If there's a crash, then: |
| 93 | * |
| 94 | * - The origin mapping will point to the old origin block (the shared |
| 95 | * one). This will contain the data as it was before the io that triggered |
| 96 | * the breaking of sharing came in. |
| 97 | * |
| 98 | * - The snap mapping still points to the old block. As it would after |
| 99 | * the commit. |
| 100 | * |
| 101 | * The downside of this scheme is the timestamp magic isn't perfect, and |
| 102 | * will continue to think that data block in the snapshot device is shared |
| 103 | * even after the write to the origin has broken sharing. I suspect data |
| 104 | * blocks will typically be shared by many different devices, so we're |
| 105 | * breaking sharing n + 1 times, rather than n, where n is the number of |
| 106 | * devices that reference this data block. At the moment I think the |
| 107 | * benefits far, far outweigh the disadvantages. |
| 108 | */ |
| 109 | |
| 110 | /*----------------------------------------------------------------*/ |
| 111 | |
| 112 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 113 | * Key building. |
| 114 | */ |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 115 | enum lock_space { |
| 116 | VIRTUAL, |
| 117 | PHYSICAL |
| 118 | }; |
| 119 | |
| 120 | static void build_key(struct dm_thin_device *td, enum lock_space ls, |
| 121 | dm_block_t b, dm_block_t e, struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 122 | { |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 123 | key->virtual = (ls == VIRTUAL); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 124 | key->dev = dm_thin_dev_id(td); |
Joe Thornber | 5f274d8 | 2014-09-17 10:17:39 +0100 | [diff] [blame] | 125 | key->block_begin = b; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 126 | key->block_end = e; |
| 127 | } |
| 128 | |
| 129 | static void build_data_key(struct dm_thin_device *td, dm_block_t b, |
| 130 | struct dm_cell_key *key) |
| 131 | { |
| 132 | build_key(td, PHYSICAL, b, b + 1llu, key); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 133 | } |
| 134 | |
| 135 | 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] | 136 | struct dm_cell_key *key) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 137 | { |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 138 | build_key(td, VIRTUAL, b, b + 1llu, key); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 139 | } |
| 140 | |
| 141 | /*----------------------------------------------------------------*/ |
| 142 | |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 143 | #define THROTTLE_THRESHOLD (1 * HZ) |
| 144 | |
| 145 | struct throttle { |
| 146 | struct rw_semaphore lock; |
| 147 | unsigned long threshold; |
| 148 | bool throttle_applied; |
| 149 | }; |
| 150 | |
| 151 | static void throttle_init(struct throttle *t) |
| 152 | { |
| 153 | init_rwsem(&t->lock); |
| 154 | t->throttle_applied = false; |
| 155 | } |
| 156 | |
| 157 | static void throttle_work_start(struct throttle *t) |
| 158 | { |
| 159 | t->threshold = jiffies + THROTTLE_THRESHOLD; |
| 160 | } |
| 161 | |
| 162 | static void throttle_work_update(struct throttle *t) |
| 163 | { |
| 164 | if (!t->throttle_applied && jiffies > t->threshold) { |
| 165 | down_write(&t->lock); |
| 166 | t->throttle_applied = true; |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | static void throttle_work_complete(struct throttle *t) |
| 171 | { |
| 172 | if (t->throttle_applied) { |
| 173 | t->throttle_applied = false; |
| 174 | up_write(&t->lock); |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | static void throttle_lock(struct throttle *t) |
| 179 | { |
| 180 | down_read(&t->lock); |
| 181 | } |
| 182 | |
| 183 | static void throttle_unlock(struct throttle *t) |
| 184 | { |
| 185 | up_read(&t->lock); |
| 186 | } |
| 187 | |
| 188 | /*----------------------------------------------------------------*/ |
| 189 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 190 | /* |
| 191 | * A pool device ties together a metadata device and a data device. It |
| 192 | * also provides the interface for creating and destroying internal |
| 193 | * devices. |
| 194 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 195 | struct dm_thin_new_mapping; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 196 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 197 | /* |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 198 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 199 | */ |
| 200 | enum pool_mode { |
| 201 | PM_WRITE, /* metadata may be changed */ |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 202 | 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] | 203 | PM_READ_ONLY, /* metadata may not be changed */ |
| 204 | PM_FAIL, /* all I/O fails */ |
| 205 | }; |
| 206 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 207 | struct pool_features { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 208 | enum pool_mode mode; |
| 209 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 210 | bool zero_new_blocks:1; |
| 211 | bool discard_enabled:1; |
| 212 | bool discard_passdown:1; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 213 | bool error_if_no_space:1; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 214 | }; |
| 215 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 216 | struct thin_c; |
| 217 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 218 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 219 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
| 220 | |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 221 | #define CELL_SORT_ARRAY_SIZE 8192 |
| 222 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 223 | struct pool { |
| 224 | struct list_head list; |
| 225 | struct dm_target *ti; /* Only set if a pool target is bound */ |
| 226 | |
| 227 | struct mapped_device *pool_md; |
| 228 | struct block_device *md_dev; |
| 229 | struct dm_pool_metadata *pmd; |
| 230 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 231 | dm_block_t low_water_blocks; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 232 | uint32_t sectors_per_block; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 233 | int sectors_per_block_shift; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 234 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 235 | struct pool_features pf; |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 236 | bool low_water_triggered:1; /* A dm event has been sent */ |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 237 | bool suspended:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 238 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 239 | struct dm_bio_prison *prison; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 240 | struct dm_kcopyd_client *copier; |
| 241 | |
| 242 | struct workqueue_struct *wq; |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 243 | struct throttle throttle; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 244 | struct work_struct worker; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 245 | struct delayed_work waker; |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 246 | struct delayed_work no_space_timeout; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 247 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 248 | unsigned long last_commit_jiffies; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 249 | unsigned ref_count; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 250 | |
| 251 | spinlock_t lock; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 252 | struct bio_list deferred_flush_bios; |
| 253 | struct list_head prepared_mappings; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 254 | struct list_head prepared_discards; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 255 | struct list_head active_thins; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 256 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 257 | struct dm_deferred_set *shared_read_ds; |
| 258 | struct dm_deferred_set *all_io_ds; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 259 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 260 | struct dm_thin_new_mapping *next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 261 | mempool_t *mapping_pool; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 262 | |
| 263 | process_bio_fn process_bio; |
| 264 | process_bio_fn process_discard; |
| 265 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 266 | process_cell_fn process_cell; |
| 267 | process_cell_fn process_discard_cell; |
| 268 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 269 | process_mapping_fn process_prepared_mapping; |
| 270 | process_mapping_fn process_prepared_discard; |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 271 | |
Joe Thornber | a822c83 | 2015-07-03 10:22:42 +0100 | [diff] [blame] | 272 | struct dm_bio_prison_cell **cell_sort_array; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 273 | }; |
| 274 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 275 | static enum pool_mode get_pool_mode(struct pool *pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 276 | 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] | 277 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 278 | /* |
| 279 | * Target context for a pool. |
| 280 | */ |
| 281 | struct pool_c { |
| 282 | struct dm_target *ti; |
| 283 | struct pool *pool; |
| 284 | struct dm_dev *data_dev; |
| 285 | struct dm_dev *metadata_dev; |
| 286 | struct dm_target_callbacks callbacks; |
| 287 | |
| 288 | dm_block_t low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 289 | struct pool_features requested_pf; /* Features requested during table load */ |
| 290 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 291 | }; |
| 292 | |
| 293 | /* |
| 294 | * Target context for a thin. |
| 295 | */ |
| 296 | struct thin_c { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 297 | struct list_head list; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 298 | struct dm_dev *pool_dev; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 299 | struct dm_dev *origin_dev; |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 300 | sector_t origin_size; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 301 | dm_thin_id dev_id; |
| 302 | |
| 303 | struct pool *pool; |
| 304 | struct dm_thin_device *td; |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 305 | struct mapped_device *thin_md; |
| 306 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 307 | bool requeue_mode:1; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 308 | spinlock_t lock; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 309 | struct list_head deferred_cells; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 310 | struct bio_list deferred_bio_list; |
| 311 | struct bio_list retry_on_resume_list; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 312 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 313 | |
| 314 | /* |
| 315 | * Ensures the thin is not destroyed until the worker has finished |
| 316 | * iterating the active_thins list. |
| 317 | */ |
| 318 | atomic_t refcount; |
| 319 | struct completion can_destroy; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 320 | }; |
| 321 | |
| 322 | /*----------------------------------------------------------------*/ |
| 323 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 324 | /** |
| 325 | * __blkdev_issue_discard_async - queue a discard with async completion |
| 326 | * @bdev: blockdev to issue discard for |
| 327 | * @sector: start sector |
| 328 | * @nr_sects: number of sectors to discard |
| 329 | * @gfp_mask: memory allocation flags (for bio_alloc) |
| 330 | * @flags: BLKDEV_IFL_* flags to control behaviour |
| 331 | * @parent_bio: parent discard bio that all sub discards get chained to |
| 332 | * |
| 333 | * Description: |
| 334 | * Asynchronously issue a discard request for the sectors in question. |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 335 | */ |
| 336 | static int __blkdev_issue_discard_async(struct block_device *bdev, sector_t sector, |
| 337 | sector_t nr_sects, gfp_t gfp_mask, unsigned long flags, |
| 338 | struct bio *parent_bio) |
| 339 | { |
| 340 | struct request_queue *q = bdev_get_queue(bdev); |
| 341 | int type = REQ_WRITE | REQ_DISCARD; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 342 | struct bio *bio; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 343 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 344 | if (!q || !nr_sects) |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 345 | return -ENXIO; |
| 346 | |
| 347 | if (!blk_queue_discard(q)) |
| 348 | return -EOPNOTSUPP; |
| 349 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 350 | if (flags & BLKDEV_DISCARD_SECURE) { |
| 351 | if (!blk_queue_secdiscard(q)) |
| 352 | return -EOPNOTSUPP; |
| 353 | type |= REQ_SECURE; |
| 354 | } |
| 355 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 356 | /* |
| 357 | * Required bio_put occurs in bio_endio thanks to bio_chain below |
| 358 | */ |
| 359 | bio = bio_alloc(gfp_mask, 1); |
| 360 | if (!bio) |
| 361 | return -ENOMEM; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 362 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 363 | bio_chain(bio, parent_bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 364 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 365 | bio->bi_iter.bi_sector = sector; |
| 366 | bio->bi_bdev = bdev; |
| 367 | bio->bi_iter.bi_size = nr_sects << 9; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 368 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 369 | submit_bio(type, bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 370 | |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 371 | return 0; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 372 | } |
| 373 | |
| 374 | static bool block_size_is_power_of_two(struct pool *pool) |
| 375 | { |
| 376 | return pool->sectors_per_block_shift >= 0; |
| 377 | } |
| 378 | |
| 379 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) |
| 380 | { |
| 381 | return block_size_is_power_of_two(pool) ? |
| 382 | (b << pool->sectors_per_block_shift) : |
| 383 | (b * pool->sectors_per_block); |
| 384 | } |
| 385 | |
| 386 | static int issue_discard(struct thin_c *tc, dm_block_t data_b, dm_block_t data_e, |
| 387 | struct bio *parent_bio) |
| 388 | { |
| 389 | sector_t s = block_to_sectors(tc->pool, data_b); |
| 390 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); |
| 391 | |
| 392 | return __blkdev_issue_discard_async(tc->pool_dev->bdev, s, len, |
| 393 | GFP_NOWAIT, 0, parent_bio); |
| 394 | } |
| 395 | |
| 396 | /*----------------------------------------------------------------*/ |
| 397 | |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 398 | /* |
| 399 | * wake_worker() is used when new work is queued and when pool_resume is |
| 400 | * ready to continue deferred IO processing. |
| 401 | */ |
| 402 | static void wake_worker(struct pool *pool) |
| 403 | { |
| 404 | queue_work(pool->wq, &pool->worker); |
| 405 | } |
| 406 | |
| 407 | /*----------------------------------------------------------------*/ |
| 408 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 409 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
| 410 | struct dm_bio_prison_cell **cell_result) |
| 411 | { |
| 412 | int r; |
| 413 | struct dm_bio_prison_cell *cell_prealloc; |
| 414 | |
| 415 | /* |
| 416 | * Allocate a cell from the prison's mempool. |
| 417 | * This might block but it can't fail. |
| 418 | */ |
| 419 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); |
| 420 | |
| 421 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); |
| 422 | if (r) |
| 423 | /* |
| 424 | * We reused an old cell; we can get rid of |
| 425 | * the new one. |
| 426 | */ |
| 427 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); |
| 428 | |
| 429 | return r; |
| 430 | } |
| 431 | |
| 432 | static void cell_release(struct pool *pool, |
| 433 | struct dm_bio_prison_cell *cell, |
| 434 | struct bio_list *bios) |
| 435 | { |
| 436 | dm_cell_release(pool->prison, cell, bios); |
| 437 | dm_bio_prison_free_cell(pool->prison, cell); |
| 438 | } |
| 439 | |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 440 | static void cell_visit_release(struct pool *pool, |
| 441 | void (*fn)(void *, struct dm_bio_prison_cell *), |
| 442 | void *context, |
| 443 | struct dm_bio_prison_cell *cell) |
| 444 | { |
| 445 | dm_cell_visit_release(pool->prison, fn, context, cell); |
| 446 | dm_bio_prison_free_cell(pool->prison, cell); |
| 447 | } |
| 448 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 449 | static void cell_release_no_holder(struct pool *pool, |
| 450 | struct dm_bio_prison_cell *cell, |
| 451 | struct bio_list *bios) |
| 452 | { |
| 453 | dm_cell_release_no_holder(pool->prison, cell, bios); |
| 454 | dm_bio_prison_free_cell(pool->prison, cell); |
| 455 | } |
| 456 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 457 | static void cell_error_with_code(struct pool *pool, |
| 458 | struct dm_bio_prison_cell *cell, int error_code) |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 459 | { |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 460 | dm_cell_error(pool->prison, cell, error_code); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 461 | dm_bio_prison_free_cell(pool->prison, cell); |
| 462 | } |
| 463 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 464 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
| 465 | { |
| 466 | cell_error_with_code(pool, cell, -EIO); |
| 467 | } |
| 468 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 469 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
| 470 | { |
| 471 | cell_error_with_code(pool, cell, 0); |
| 472 | } |
| 473 | |
| 474 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) |
| 475 | { |
| 476 | cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE); |
| 477 | } |
| 478 | |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 479 | /*----------------------------------------------------------------*/ |
| 480 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 481 | /* |
| 482 | * A global list of pools that uses a struct mapped_device as a key. |
| 483 | */ |
| 484 | static struct dm_thin_pool_table { |
| 485 | struct mutex mutex; |
| 486 | struct list_head pools; |
| 487 | } dm_thin_pool_table; |
| 488 | |
| 489 | static void pool_table_init(void) |
| 490 | { |
| 491 | mutex_init(&dm_thin_pool_table.mutex); |
| 492 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); |
| 493 | } |
| 494 | |
| 495 | static void __pool_table_insert(struct pool *pool) |
| 496 | { |
| 497 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 498 | list_add(&pool->list, &dm_thin_pool_table.pools); |
| 499 | } |
| 500 | |
| 501 | static void __pool_table_remove(struct pool *pool) |
| 502 | { |
| 503 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 504 | list_del(&pool->list); |
| 505 | } |
| 506 | |
| 507 | static struct pool *__pool_table_lookup(struct mapped_device *md) |
| 508 | { |
| 509 | struct pool *pool = NULL, *tmp; |
| 510 | |
| 511 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 512 | |
| 513 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 514 | if (tmp->pool_md == md) { |
| 515 | pool = tmp; |
| 516 | break; |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | return pool; |
| 521 | } |
| 522 | |
| 523 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) |
| 524 | { |
| 525 | struct pool *pool = NULL, *tmp; |
| 526 | |
| 527 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 528 | |
| 529 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { |
| 530 | if (tmp->md_dev == md_dev) { |
| 531 | pool = tmp; |
| 532 | break; |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | return pool; |
| 537 | } |
| 538 | |
| 539 | /*----------------------------------------------------------------*/ |
| 540 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 541 | struct dm_thin_endio_hook { |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 542 | struct thin_c *tc; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 543 | struct dm_deferred_entry *shared_read_entry; |
| 544 | struct dm_deferred_entry *all_io_entry; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 545 | struct dm_thin_new_mapping *overwrite_mapping; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 546 | struct rb_node rb_node; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 547 | struct dm_bio_prison_cell *cell; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 548 | }; |
| 549 | |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 550 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
| 551 | { |
| 552 | bio_list_merge(bios, master); |
| 553 | bio_list_init(master); |
| 554 | } |
| 555 | |
| 556 | static void error_bio_list(struct bio_list *bios, int error) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 557 | { |
| 558 | struct bio *bio; |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 559 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 560 | while ((bio = bio_list_pop(bios))) { |
| 561 | bio->bi_error = error; |
| 562 | bio_endio(bio); |
| 563 | } |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 564 | } |
| 565 | |
| 566 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error) |
| 567 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 568 | struct bio_list bios; |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 569 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 570 | |
| 571 | bio_list_init(&bios); |
Joe Thornber | 18adc57 | 2014-03-03 15:46:42 +0000 | [diff] [blame] | 572 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 573 | spin_lock_irqsave(&tc->lock, flags); |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 574 | __merge_bio_list(&bios, master); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 575 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 576 | |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 577 | error_bio_list(&bios, error); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 578 | } |
| 579 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 580 | static void requeue_deferred_cells(struct thin_c *tc) |
| 581 | { |
| 582 | struct pool *pool = tc->pool; |
| 583 | unsigned long flags; |
| 584 | struct list_head cells; |
| 585 | struct dm_bio_prison_cell *cell, *tmp; |
| 586 | |
| 587 | INIT_LIST_HEAD(&cells); |
| 588 | |
| 589 | spin_lock_irqsave(&tc->lock, flags); |
| 590 | list_splice_init(&tc->deferred_cells, &cells); |
| 591 | spin_unlock_irqrestore(&tc->lock, flags); |
| 592 | |
| 593 | list_for_each_entry_safe(cell, tmp, &cells, user_list) |
| 594 | cell_requeue(pool, cell); |
| 595 | } |
| 596 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 597 | static void requeue_io(struct thin_c *tc) |
| 598 | { |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 599 | struct bio_list bios; |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 600 | unsigned long flags; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 601 | |
| 602 | bio_list_init(&bios); |
| 603 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 604 | spin_lock_irqsave(&tc->lock, flags); |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 605 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
| 606 | __merge_bio_list(&bios, &tc->retry_on_resume_list); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 607 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 608 | |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 609 | error_bio_list(&bios, DM_ENDIO_REQUEUE); |
| 610 | requeue_deferred_cells(tc); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 611 | } |
| 612 | |
Mike Snitzer | 0a927c2 | 2015-07-21 13:20:46 -0400 | [diff] [blame] | 613 | static void error_retry_list_with_code(struct pool *pool, int error) |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 614 | { |
| 615 | struct thin_c *tc; |
| 616 | |
| 617 | rcu_read_lock(); |
| 618 | list_for_each_entry_rcu(tc, &pool->active_thins, list) |
Mike Snitzer | 0a927c2 | 2015-07-21 13:20:46 -0400 | [diff] [blame] | 619 | error_thin_bio_list(tc, &tc->retry_on_resume_list, error); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 620 | rcu_read_unlock(); |
| 621 | } |
| 622 | |
Mike Snitzer | 0a927c2 | 2015-07-21 13:20:46 -0400 | [diff] [blame] | 623 | static void error_retry_list(struct pool *pool) |
| 624 | { |
| 625 | return error_retry_list_with_code(pool, -EIO); |
| 626 | } |
| 627 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 628 | /* |
| 629 | * This section of code contains the logic for processing a thin device's IO. |
| 630 | * Much of the code depends on pool object resources (lists, workqueues, etc) |
| 631 | * but most is exclusively called from the thin target rather than the thin-pool |
| 632 | * target. |
| 633 | */ |
| 634 | |
| 635 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
| 636 | { |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 637 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 638 | sector_t block_nr = bio->bi_iter.bi_sector; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 639 | |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 640 | if (block_size_is_power_of_two(pool)) |
| 641 | block_nr >>= pool->sectors_per_block_shift; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 642 | else |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 643 | (void) sector_div(block_nr, pool->sectors_per_block); |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 644 | |
| 645 | return block_nr; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 646 | } |
| 647 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 648 | /* |
| 649 | * Returns the _complete_ blocks that this bio covers. |
| 650 | */ |
| 651 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, |
| 652 | dm_block_t *begin, dm_block_t *end) |
| 653 | { |
| 654 | struct pool *pool = tc->pool; |
| 655 | sector_t b = bio->bi_iter.bi_sector; |
| 656 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); |
| 657 | |
| 658 | b += pool->sectors_per_block - 1ull; /* so we round up */ |
| 659 | |
| 660 | if (block_size_is_power_of_two(pool)) { |
| 661 | b >>= pool->sectors_per_block_shift; |
| 662 | e >>= pool->sectors_per_block_shift; |
| 663 | } else { |
| 664 | (void) sector_div(b, pool->sectors_per_block); |
| 665 | (void) sector_div(e, pool->sectors_per_block); |
| 666 | } |
| 667 | |
| 668 | if (e < b) |
| 669 | /* Can happen if the bio is within a single block. */ |
| 670 | e = b; |
| 671 | |
| 672 | *begin = b; |
| 673 | *end = e; |
| 674 | } |
| 675 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 676 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
| 677 | { |
| 678 | struct pool *pool = tc->pool; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 679 | sector_t bi_sector = bio->bi_iter.bi_sector; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 680 | |
| 681 | bio->bi_bdev = tc->pool_dev->bdev; |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 682 | if (block_size_is_power_of_two(pool)) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 683 | bio->bi_iter.bi_sector = |
| 684 | (block << pool->sectors_per_block_shift) | |
| 685 | (bi_sector & (pool->sectors_per_block - 1)); |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 686 | else |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 687 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
Mike Snitzer | 58f77a2 | 2013-03-01 22:45:45 +0000 | [diff] [blame] | 688 | sector_div(bi_sector, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 689 | } |
| 690 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 691 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
| 692 | { |
| 693 | bio->bi_bdev = tc->origin_dev->bdev; |
| 694 | } |
| 695 | |
Joe Thornber | 4afdd68 | 2012-07-27 15:08:14 +0100 | [diff] [blame] | 696 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
| 697 | { |
| 698 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && |
| 699 | dm_thin_changed_this_transaction(tc->td); |
| 700 | } |
| 701 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 702 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
| 703 | { |
| 704 | struct dm_thin_endio_hook *h; |
| 705 | |
| 706 | if (bio->bi_rw & REQ_DISCARD) |
| 707 | return; |
| 708 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 709 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 710 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
| 711 | } |
| 712 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 713 | static void issue(struct thin_c *tc, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 714 | { |
| 715 | struct pool *pool = tc->pool; |
| 716 | unsigned long flags; |
| 717 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 718 | if (!bio_triggers_commit(tc, bio)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 719 | generic_make_request(bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 720 | return; |
| 721 | } |
| 722 | |
| 723 | /* |
| 724 | * Complete bio with an error if earlier I/O caused changes to |
| 725 | * the metadata that can't be committed e.g, due to I/O errors |
| 726 | * on the metadata device. |
| 727 | */ |
| 728 | if (dm_thin_aborted_changes(tc->td)) { |
| 729 | bio_io_error(bio); |
| 730 | return; |
| 731 | } |
| 732 | |
| 733 | /* |
| 734 | * Batch together any bios that trigger commits and then issue a |
| 735 | * single commit for them in process_deferred_bios(). |
| 736 | */ |
| 737 | spin_lock_irqsave(&pool->lock, flags); |
| 738 | bio_list_add(&pool->deferred_flush_bios, bio); |
| 739 | spin_unlock_irqrestore(&pool->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 740 | } |
| 741 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 742 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
| 743 | { |
| 744 | remap_to_origin(tc, bio); |
| 745 | issue(tc, bio); |
| 746 | } |
| 747 | |
| 748 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, |
| 749 | dm_block_t block) |
| 750 | { |
| 751 | remap(tc, bio, block); |
| 752 | issue(tc, bio); |
| 753 | } |
| 754 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 755 | /*----------------------------------------------------------------*/ |
| 756 | |
| 757 | /* |
| 758 | * Bio endio functions. |
| 759 | */ |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 760 | struct dm_thin_new_mapping { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 761 | struct list_head list; |
| 762 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 763 | bool pass_discard:1; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 764 | bool maybe_shared:1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 765 | |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 766 | /* |
| 767 | * Track quiescing, copying and zeroing preparation actions. When this |
| 768 | * counter hits zero the block is prepared and can be inserted into the |
| 769 | * btree. |
| 770 | */ |
| 771 | atomic_t prepare_actions; |
| 772 | |
Mike Snitzer | 7f21466 | 2013-12-17 13:43:31 -0500 | [diff] [blame] | 773 | int err; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 774 | struct thin_c *tc; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 775 | dm_block_t virt_begin, virt_end; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 776 | dm_block_t data_block; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 777 | struct dm_bio_prison_cell *cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 778 | |
| 779 | /* |
| 780 | * If the bio covers the whole area of a block then we can avoid |
| 781 | * zeroing or copying. Instead this bio is hooked. The bio will |
| 782 | * still be in the cell, so care has to be taken to avoid issuing |
| 783 | * the bio twice. |
| 784 | */ |
| 785 | struct bio *bio; |
| 786 | bio_end_io_t *saved_bi_end_io; |
| 787 | }; |
| 788 | |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 789 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 790 | { |
| 791 | struct pool *pool = m->tc->pool; |
| 792 | |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 793 | if (atomic_dec_and_test(&m->prepare_actions)) { |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 794 | list_add_tail(&m->list, &pool->prepared_mappings); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 795 | wake_worker(pool); |
| 796 | } |
| 797 | } |
| 798 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 799 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 800 | { |
| 801 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 802 | struct pool *pool = m->tc->pool; |
| 803 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 804 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 805 | __complete_mapping_preparation(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 806 | spin_unlock_irqrestore(&pool->lock, flags); |
| 807 | } |
| 808 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 809 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
| 810 | { |
| 811 | struct dm_thin_new_mapping *m = context; |
| 812 | |
| 813 | m->err = read_err || write_err ? -EIO : 0; |
| 814 | complete_mapping_preparation(m); |
| 815 | } |
| 816 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 817 | static void overwrite_endio(struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 818 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 819 | 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] | 820 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 821 | |
Mike Snitzer | 8b908f8 | 2015-05-13 17:53:13 -0400 | [diff] [blame] | 822 | bio->bi_end_io = m->saved_bi_end_io; |
| 823 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 824 | m->err = bio->bi_error; |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 825 | complete_mapping_preparation(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 826 | } |
| 827 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 828 | /*----------------------------------------------------------------*/ |
| 829 | |
| 830 | /* |
| 831 | * Workqueue. |
| 832 | */ |
| 833 | |
| 834 | /* |
| 835 | * Prepared mapping jobs. |
| 836 | */ |
| 837 | |
| 838 | /* |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 839 | * This sends the bios in the cell, except the original holder, back |
| 840 | * to the deferred_bios list. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 841 | */ |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 842 | 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] | 843 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 844 | struct pool *pool = tc->pool; |
| 845 | unsigned long flags; |
| 846 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 847 | spin_lock_irqsave(&tc->lock, flags); |
| 848 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); |
| 849 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 850 | |
| 851 | wake_worker(pool); |
| 852 | } |
| 853 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 854 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
| 855 | |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 856 | struct remap_info { |
| 857 | struct thin_c *tc; |
| 858 | struct bio_list defer_bios; |
| 859 | struct bio_list issue_bios; |
| 860 | }; |
| 861 | |
| 862 | static void __inc_remap_and_issue_cell(void *context, |
| 863 | struct dm_bio_prison_cell *cell) |
| 864 | { |
| 865 | struct remap_info *info = context; |
| 866 | struct bio *bio; |
| 867 | |
| 868 | while ((bio = bio_list_pop(&cell->bios))) { |
| 869 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) |
| 870 | bio_list_add(&info->defer_bios, bio); |
| 871 | else { |
| 872 | inc_all_io_entry(info->tc->pool, bio); |
| 873 | |
| 874 | /* |
| 875 | * We can't issue the bios with the bio prison lock |
| 876 | * held, so we add them to a list to issue on |
| 877 | * return from this function. |
| 878 | */ |
| 879 | bio_list_add(&info->issue_bios, bio); |
| 880 | } |
| 881 | } |
| 882 | } |
| 883 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 884 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
| 885 | struct dm_bio_prison_cell *cell, |
| 886 | dm_block_t block) |
| 887 | { |
| 888 | struct bio *bio; |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 889 | struct remap_info info; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 890 | |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 891 | info.tc = tc; |
| 892 | bio_list_init(&info.defer_bios); |
| 893 | bio_list_init(&info.issue_bios); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 894 | |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 895 | /* |
| 896 | * We have to be careful to inc any bios we're about to issue |
| 897 | * before the cell is released, and avoid a race with new bios |
| 898 | * being added to the cell. |
| 899 | */ |
| 900 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, |
| 901 | &info, cell); |
| 902 | |
| 903 | while ((bio = bio_list_pop(&info.defer_bios))) |
| 904 | thin_defer_bio(tc, bio); |
| 905 | |
| 906 | while ((bio = bio_list_pop(&info.issue_bios))) |
| 907 | remap_and_issue(info.tc, bio, block); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 908 | } |
| 909 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 910 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
| 911 | { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 912 | cell_error(m->tc->pool, m->cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 913 | list_del(&m->list); |
| 914 | mempool_free(m, m->tc->pool->mapping_pool); |
| 915 | } |
Joe Thornber | 025b968 | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 916 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 917 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 918 | { |
| 919 | struct thin_c *tc = m->tc; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 920 | struct pool *pool = tc->pool; |
Mike Snitzer | 8b908f8 | 2015-05-13 17:53:13 -0400 | [diff] [blame] | 921 | struct bio *bio = m->bio; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 922 | int r; |
| 923 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 924 | if (m->err) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 925 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 926 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 927 | } |
| 928 | |
| 929 | /* |
| 930 | * Commit the prepared block into the mapping btree. |
| 931 | * Any I/O for this block arriving after this point will get |
| 932 | * remapped to it directly. |
| 933 | */ |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 934 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 935 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 936 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 937 | cell_error(pool, m->cell); |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 938 | goto out; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 939 | } |
| 940 | |
| 941 | /* |
| 942 | * Release any bios held while the block was being provisioned. |
| 943 | * If we are processing a write bio that completely covers the block, |
| 944 | * we already processed it so can ignore it now when processing |
| 945 | * the bios in the cell. |
| 946 | */ |
| 947 | if (bio) { |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 948 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 949 | bio_endio(bio); |
Joe Thornber | 2d759a4 | 2014-10-10 15:27:16 +0100 | [diff] [blame] | 950 | } else { |
| 951 | inc_all_io_entry(tc->pool, m->cell->holder); |
| 952 | remap_and_issue(tc, m->cell->holder, m->data_block); |
| 953 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
| 954 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 955 | |
Joe Thornber | 905386f | 2012-07-27 15:08:05 +0100 | [diff] [blame] | 956 | out: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 957 | list_del(&m->list); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 958 | mempool_free(m, pool->mapping_pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 959 | } |
| 960 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 961 | /*----------------------------------------------------------------*/ |
| 962 | |
| 963 | static void free_discard_mapping(struct dm_thin_new_mapping *m) |
| 964 | { |
| 965 | struct thin_c *tc = m->tc; |
| 966 | if (m->cell) |
| 967 | cell_defer_no_holder(tc, m->cell); |
| 968 | mempool_free(m, tc->pool->mapping_pool); |
| 969 | } |
| 970 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 971 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 972 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 973 | bio_io_error(m->bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 974 | free_discard_mapping(m); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 975 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 976 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 977 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 978 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 979 | bio_endio(m->bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 980 | free_discard_mapping(m); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 981 | } |
| 982 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 983 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 984 | { |
| 985 | int r; |
| 986 | struct thin_c *tc = m->tc; |
| 987 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 988 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); |
| 989 | if (r) { |
| 990 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); |
| 991 | bio_io_error(m->bio); |
| 992 | } else |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 993 | bio_endio(m->bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 994 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 995 | cell_defer_no_holder(tc, m->cell); |
| 996 | mempool_free(m, tc->pool->mapping_pool); |
| 997 | } |
| 998 | |
| 999 | static int passdown_double_checking_shared_status(struct dm_thin_new_mapping *m) |
| 1000 | { |
| 1001 | /* |
| 1002 | * We've already unmapped this range of blocks, but before we |
| 1003 | * passdown we have to check that these blocks are now unused. |
| 1004 | */ |
| 1005 | int r; |
| 1006 | bool used = true; |
| 1007 | struct thin_c *tc = m->tc; |
| 1008 | struct pool *pool = tc->pool; |
| 1009 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; |
| 1010 | |
| 1011 | while (b != end) { |
| 1012 | /* find start of unmapped run */ |
| 1013 | for (; b < end; b++) { |
| 1014 | r = dm_pool_block_is_used(pool->pmd, b, &used); |
| 1015 | if (r) |
| 1016 | return r; |
| 1017 | |
| 1018 | if (!used) |
| 1019 | break; |
| 1020 | } |
| 1021 | |
| 1022 | if (b == end) |
| 1023 | break; |
| 1024 | |
| 1025 | /* find end of run */ |
| 1026 | for (e = b + 1; e != end; e++) { |
| 1027 | r = dm_pool_block_is_used(pool->pmd, e, &used); |
| 1028 | if (r) |
| 1029 | return r; |
| 1030 | |
| 1031 | if (used) |
| 1032 | break; |
| 1033 | } |
| 1034 | |
| 1035 | r = issue_discard(tc, b, e, m->bio); |
| 1036 | if (r) |
| 1037 | return r; |
| 1038 | |
| 1039 | b = e; |
| 1040 | } |
| 1041 | |
| 1042 | return 0; |
| 1043 | } |
| 1044 | |
| 1045 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) |
| 1046 | { |
| 1047 | int r; |
| 1048 | struct thin_c *tc = m->tc; |
| 1049 | struct pool *pool = tc->pool; |
| 1050 | |
| 1051 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
| 1052 | if (r) |
| 1053 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
| 1054 | |
| 1055 | else if (m->maybe_shared) |
| 1056 | r = passdown_double_checking_shared_status(m); |
| 1057 | else |
| 1058 | r = issue_discard(tc, m->data_block, m->data_block + (m->virt_end - m->virt_begin), m->bio); |
| 1059 | |
| 1060 | /* |
| 1061 | * Even if r is set, there could be sub discards in flight that we |
| 1062 | * need to wait for. |
| 1063 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1064 | m->bio->bi_error = r; |
| 1065 | bio_endio(m->bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1066 | cell_defer_no_holder(tc, m->cell); |
| 1067 | mempool_free(m, pool->mapping_pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1068 | } |
| 1069 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1070 | static void process_prepared(struct pool *pool, struct list_head *head, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1071 | process_mapping_fn *fn) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1072 | { |
| 1073 | unsigned long flags; |
| 1074 | struct list_head maps; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1075 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1076 | |
| 1077 | INIT_LIST_HEAD(&maps); |
| 1078 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1079 | list_splice_init(head, &maps); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1080 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1081 | |
| 1082 | list_for_each_entry_safe(m, tmp, &maps, list) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1083 | (*fn)(m); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1084 | } |
| 1085 | |
| 1086 | /* |
| 1087 | * Deferred bio jobs. |
| 1088 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1089 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
| 1090 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1091 | return bio->bi_iter.bi_size == |
| 1092 | (pool->sectors_per_block << SECTOR_SHIFT); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1093 | } |
| 1094 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1095 | static int io_overwrites_block(struct pool *pool, struct bio *bio) |
| 1096 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1097 | return (bio_data_dir(bio) == WRITE) && |
| 1098 | io_overlaps_block(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1099 | } |
| 1100 | |
| 1101 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, |
| 1102 | bio_end_io_t *fn) |
| 1103 | { |
| 1104 | *save = bio->bi_end_io; |
| 1105 | bio->bi_end_io = fn; |
| 1106 | } |
| 1107 | |
| 1108 | static int ensure_next_mapping(struct pool *pool) |
| 1109 | { |
| 1110 | if (pool->next_mapping) |
| 1111 | return 0; |
| 1112 | |
| 1113 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); |
| 1114 | |
| 1115 | return pool->next_mapping ? 0 : -ENOMEM; |
| 1116 | } |
| 1117 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1118 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1119 | { |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 1120 | struct dm_thin_new_mapping *m = pool->next_mapping; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1121 | |
| 1122 | BUG_ON(!pool->next_mapping); |
| 1123 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 1124 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
| 1125 | INIT_LIST_HEAD(&m->list); |
| 1126 | m->bio = NULL; |
| 1127 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1128 | pool->next_mapping = NULL; |
| 1129 | |
Mike Snitzer | 16961b0 | 2013-12-17 13:19:11 -0500 | [diff] [blame] | 1130 | return m; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1131 | } |
| 1132 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1133 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
| 1134 | sector_t begin, sector_t end) |
| 1135 | { |
| 1136 | int r; |
| 1137 | struct dm_io_region to; |
| 1138 | |
| 1139 | to.bdev = tc->pool_dev->bdev; |
| 1140 | to.sector = begin; |
| 1141 | to.count = end - begin; |
| 1142 | |
| 1143 | r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); |
| 1144 | if (r < 0) { |
| 1145 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
| 1146 | copy_complete(1, 1, m); |
| 1147 | } |
| 1148 | } |
| 1149 | |
Mike Snitzer | 452d7a6 | 2014-10-09 19:20:21 -0400 | [diff] [blame] | 1150 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1151 | dm_block_t data_begin, |
Mike Snitzer | 452d7a6 | 2014-10-09 19:20:21 -0400 | [diff] [blame] | 1152 | struct dm_thin_new_mapping *m) |
| 1153 | { |
| 1154 | struct pool *pool = tc->pool; |
| 1155 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
| 1156 | |
| 1157 | h->overwrite_mapping = m; |
| 1158 | m->bio = bio; |
| 1159 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); |
| 1160 | inc_all_io_entry(pool, bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1161 | remap_and_issue(tc, bio, data_begin); |
Mike Snitzer | 452d7a6 | 2014-10-09 19:20:21 -0400 | [diff] [blame] | 1162 | } |
| 1163 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1164 | /* |
| 1165 | * A partial copy also needs to zero the uncopied region. |
| 1166 | */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1167 | 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] | 1168 | struct dm_dev *origin, dm_block_t data_origin, |
| 1169 | dm_block_t data_dest, |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1170 | struct dm_bio_prison_cell *cell, struct bio *bio, |
| 1171 | sector_t len) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1172 | { |
| 1173 | int r; |
| 1174 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1175 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1176 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1177 | m->tc = tc; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1178 | m->virt_begin = virt_block; |
| 1179 | m->virt_end = virt_block + 1u; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1180 | m->data_block = data_dest; |
| 1181 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1182 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1183 | /* |
| 1184 | * quiesce action + copy action + an extra reference held for the |
| 1185 | * duration of this function (we may need to inc later for a |
| 1186 | * partial zero). |
| 1187 | */ |
| 1188 | atomic_set(&m->prepare_actions, 3); |
| 1189 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1190 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1191 | complete_mapping_preparation(m); /* already quiesced */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1192 | |
| 1193 | /* |
| 1194 | * IO to pool_dev remaps to the pool target's data_dev. |
| 1195 | * |
| 1196 | * If the whole block of data is being overwritten, we can issue the |
| 1197 | * bio immediately. Otherwise we use kcopyd to clone the data first. |
| 1198 | */ |
Mike Snitzer | 452d7a6 | 2014-10-09 19:20:21 -0400 | [diff] [blame] | 1199 | if (io_overwrites_block(pool, bio)) |
| 1200 | remap_and_issue_overwrite(tc, bio, data_dest, m); |
| 1201 | else { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1202 | struct dm_io_region from, to; |
| 1203 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1204 | from.bdev = origin->bdev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1205 | from.sector = data_origin * pool->sectors_per_block; |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1206 | from.count = len; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1207 | |
| 1208 | to.bdev = tc->pool_dev->bdev; |
| 1209 | to.sector = data_dest * pool->sectors_per_block; |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1210 | to.count = len; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1211 | |
| 1212 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, |
| 1213 | 0, copy_complete, m); |
| 1214 | if (r < 0) { |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1215 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1216 | copy_complete(1, 1, m); |
| 1217 | |
| 1218 | /* |
| 1219 | * We allow the zero to be issued, to simplify the |
| 1220 | * error path. Otherwise we'd need to start |
| 1221 | * worrying about decrementing the prepare_actions |
| 1222 | * counter. |
| 1223 | */ |
| 1224 | } |
| 1225 | |
| 1226 | /* |
| 1227 | * Do we need to zero a tail region? |
| 1228 | */ |
| 1229 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { |
| 1230 | atomic_inc(&m->prepare_actions); |
| 1231 | ll_zero(tc, m, |
| 1232 | data_dest * pool->sectors_per_block + len, |
| 1233 | (data_dest + 1) * pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1234 | } |
| 1235 | } |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1236 | |
| 1237 | complete_mapping_preparation(m); /* drop our ref */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1238 | } |
| 1239 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1240 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
| 1241 | dm_block_t data_origin, dm_block_t data_dest, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1242 | struct dm_bio_prison_cell *cell, struct bio *bio) |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1243 | { |
| 1244 | schedule_copy(tc, virt_block, tc->pool_dev, |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1245 | data_origin, data_dest, cell, bio, |
| 1246 | tc->pool->sectors_per_block); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1247 | } |
| 1248 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1249 | 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] | 1250 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1251 | struct bio *bio) |
| 1252 | { |
| 1253 | struct pool *pool = tc->pool; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1254 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1255 | |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 1256 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1257 | m->tc = tc; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1258 | m->virt_begin = virt_block; |
| 1259 | m->virt_end = virt_block + 1u; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1260 | m->data_block = data_block; |
| 1261 | m->cell = cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1262 | |
| 1263 | /* |
| 1264 | * If the whole block of data is being overwritten or we are not |
| 1265 | * zeroing pre-existing data, we can issue the bio immediately. |
| 1266 | * Otherwise we use kcopyd to zero the data first. |
| 1267 | */ |
Mike Snitzer | f8ae752 | 2015-05-14 11:28:37 -0400 | [diff] [blame] | 1268 | if (pool->pf.zero_new_blocks) { |
| 1269 | if (io_overwrites_block(pool, bio)) |
| 1270 | remap_and_issue_overwrite(tc, bio, data_block, m); |
| 1271 | else |
| 1272 | ll_zero(tc, m, data_block * pool->sectors_per_block, |
| 1273 | (data_block + 1) * pool->sectors_per_block); |
| 1274 | } else |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1275 | process_prepared_mapping(m); |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1276 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1277 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1278 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
| 1279 | dm_block_t data_dest, |
| 1280 | struct dm_bio_prison_cell *cell, struct bio *bio) |
| 1281 | { |
| 1282 | struct pool *pool = tc->pool; |
| 1283 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; |
| 1284 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; |
| 1285 | |
| 1286 | if (virt_block_end <= tc->origin_size) |
| 1287 | schedule_copy(tc, virt_block, tc->origin_dev, |
| 1288 | virt_block, data_dest, cell, bio, |
| 1289 | pool->sectors_per_block); |
| 1290 | |
| 1291 | else if (virt_block_begin < tc->origin_size) |
| 1292 | schedule_copy(tc, virt_block, tc->origin_dev, |
| 1293 | virt_block, data_dest, cell, bio, |
| 1294 | tc->origin_size - virt_block_begin); |
| 1295 | |
| 1296 | else |
| 1297 | schedule_zero(tc, virt_block, data_dest, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1298 | } |
| 1299 | |
Joe Thornber | 2c43fd2 | 2014-12-11 11:12:19 +0000 | [diff] [blame] | 1300 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
| 1301 | |
| 1302 | static void check_for_space(struct pool *pool) |
| 1303 | { |
| 1304 | int r; |
| 1305 | dm_block_t nr_free; |
| 1306 | |
| 1307 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) |
| 1308 | return; |
| 1309 | |
| 1310 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); |
| 1311 | if (r) |
| 1312 | return; |
| 1313 | |
| 1314 | if (nr_free) |
| 1315 | set_pool_mode(pool, PM_WRITE); |
| 1316 | } |
| 1317 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1318 | /* |
| 1319 | * A non-zero return indicates read_only or fail_io mode. |
| 1320 | * Many callers don't care about the return value. |
| 1321 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 1322 | static int commit(struct pool *pool) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1323 | { |
| 1324 | int r; |
| 1325 | |
Joe Thornber | 8d07e8a | 2014-05-06 16:28:14 +0100 | [diff] [blame] | 1326 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1327 | return -EINVAL; |
| 1328 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 1329 | r = dm_pool_commit_metadata(pool->pmd); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1330 | if (r) |
| 1331 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); |
Joe Thornber | 2c43fd2 | 2014-12-11 11:12:19 +0000 | [diff] [blame] | 1332 | else |
| 1333 | check_for_space(pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1334 | |
| 1335 | return r; |
| 1336 | } |
| 1337 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 1338 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
| 1339 | { |
| 1340 | unsigned long flags; |
| 1341 | |
| 1342 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
| 1343 | DMWARN("%s: reached low water mark for data device: sending event.", |
| 1344 | dm_device_name(pool->pool_md)); |
| 1345 | spin_lock_irqsave(&pool->lock, flags); |
| 1346 | pool->low_water_triggered = true; |
| 1347 | spin_unlock_irqrestore(&pool->lock, flags); |
| 1348 | dm_table_event(pool->ti->table); |
| 1349 | } |
| 1350 | } |
| 1351 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1352 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
| 1353 | { |
| 1354 | int r; |
| 1355 | dm_block_t free_blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1356 | struct pool *pool = tc->pool; |
| 1357 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1358 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
Joe Thornber | 8d30abf | 2013-12-04 19:16:11 -0500 | [diff] [blame] | 1359 | return -EINVAL; |
| 1360 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1361 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1362 | if (r) { |
| 1363 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1364 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1365 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1366 | |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 1367 | check_low_water_mark(pool, free_blocks); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1368 | |
| 1369 | if (!free_blocks) { |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1370 | /* |
| 1371 | * Try to commit to see if that will free up some |
| 1372 | * more space. |
| 1373 | */ |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 1374 | r = commit(pool); |
| 1375 | if (r) |
| 1376 | return r; |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1377 | |
| 1378 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1379 | if (r) { |
| 1380 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1381 | return r; |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1382 | } |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1383 | |
Mike Snitzer | 94563ba | 2013-08-22 09:56:18 -0400 | [diff] [blame] | 1384 | if (!free_blocks) { |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1385 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1386 | return -ENOSPC; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1387 | } |
| 1388 | } |
| 1389 | |
| 1390 | r = dm_pool_alloc_data_block(pool->pmd, result); |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 1391 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 1392 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1393 | return r; |
Mike Snitzer | 4a02b34 | 2013-12-03 12:20:57 -0500 | [diff] [blame] | 1394 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1395 | |
| 1396 | return 0; |
| 1397 | } |
| 1398 | |
| 1399 | /* |
| 1400 | * If we have run out of space, queue bios until the device is |
| 1401 | * resumed, presumably after having been reloaded with more space. |
| 1402 | */ |
| 1403 | static void retry_on_resume(struct bio *bio) |
| 1404 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1405 | 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] | 1406 | struct thin_c *tc = h->tc; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1407 | unsigned long flags; |
| 1408 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1409 | spin_lock_irqsave(&tc->lock, flags); |
| 1410 | bio_list_add(&tc->retry_on_resume_list, bio); |
| 1411 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1412 | } |
| 1413 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1414 | static int should_error_unserviceable_bio(struct pool *pool) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1415 | { |
| 1416 | enum pool_mode m = get_pool_mode(pool); |
| 1417 | |
| 1418 | switch (m) { |
| 1419 | case PM_WRITE: |
| 1420 | /* Shouldn't get here */ |
| 1421 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1422 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1423 | |
| 1424 | case PM_OUT_OF_DATA_SPACE: |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1425 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1426 | |
| 1427 | case PM_READ_ONLY: |
| 1428 | case PM_FAIL: |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1429 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1430 | default: |
| 1431 | /* Shouldn't get here */ |
| 1432 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1433 | return -EIO; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1434 | } |
| 1435 | } |
| 1436 | |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1437 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
| 1438 | { |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1439 | int error = should_error_unserviceable_bio(pool); |
| 1440 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1441 | if (error) { |
| 1442 | bio->bi_error = error; |
| 1443 | bio_endio(bio); |
| 1444 | } else |
Mike Snitzer | 6d16202 | 2013-12-20 18:09:02 -0500 | [diff] [blame] | 1445 | retry_on_resume(bio); |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1446 | } |
| 1447 | |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1448 | 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] | 1449 | { |
| 1450 | struct bio *bio; |
| 1451 | struct bio_list bios; |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1452 | int error; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1453 | |
Mike Snitzer | af91805 | 2014-05-22 14:32:51 -0400 | [diff] [blame] | 1454 | error = should_error_unserviceable_bio(pool); |
| 1455 | if (error) { |
| 1456 | cell_error_with_code(pool, cell, error); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1457 | return; |
| 1458 | } |
| 1459 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1460 | bio_list_init(&bios); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1461 | cell_release(pool, cell, &bios); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1462 | |
Mike Snitzer | 9d094ee | 2014-10-19 08:23:09 -0400 | [diff] [blame] | 1463 | while ((bio = bio_list_pop(&bios))) |
| 1464 | retry_on_resume(bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1465 | } |
| 1466 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1467 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
| 1468 | struct dm_bio_prison_cell *virt_cell) |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1469 | { |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1470 | struct pool *pool = tc->pool; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1471 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
| 1472 | |
| 1473 | /* |
| 1474 | * We don't need to lock the data blocks, since there's no |
| 1475 | * passdown. We only lock data blocks for allocation and breaking sharing. |
| 1476 | */ |
| 1477 | m->tc = tc; |
| 1478 | m->virt_begin = virt_cell->key.block_begin; |
| 1479 | m->virt_end = virt_cell->key.block_end; |
| 1480 | m->cell = virt_cell; |
| 1481 | m->bio = virt_cell->holder; |
| 1482 | |
| 1483 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
| 1484 | pool->process_prepared_discard(m); |
| 1485 | } |
| 1486 | |
| 1487 | /* |
Mike Snitzer | 84f8bd8 | 2015-08-18 10:31:09 -0400 | [diff] [blame] | 1488 | * __bio_inc_remaining() is used to defer parent bios's end_io until |
| 1489 | * we _know_ all chained sub range discard bios have completed. |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1490 | */ |
| 1491 | static inline void __bio_inc_remaining(struct bio *bio) |
| 1492 | { |
| 1493 | bio->bi_flags |= (1 << BIO_CHAIN); |
| 1494 | smp_mb__before_atomic(); |
| 1495 | atomic_inc(&bio->__bi_remaining); |
| 1496 | } |
| 1497 | |
| 1498 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
| 1499 | struct bio *bio) |
| 1500 | { |
| 1501 | struct pool *pool = tc->pool; |
| 1502 | |
| 1503 | int r; |
| 1504 | bool maybe_shared; |
| 1505 | struct dm_cell_key data_key; |
| 1506 | struct dm_bio_prison_cell *data_cell; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1507 | struct dm_thin_new_mapping *m; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1508 | dm_block_t virt_begin, virt_end, data_begin; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1509 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1510 | while (begin != end) { |
| 1511 | r = ensure_next_mapping(pool); |
| 1512 | if (r) |
| 1513 | /* we did our best */ |
| 1514 | return; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1515 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1516 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
| 1517 | &data_begin, &maybe_shared); |
| 1518 | if (r) |
| 1519 | /* |
| 1520 | * Silently fail, letting any mappings we've |
| 1521 | * created complete. |
| 1522 | */ |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1523 | break; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1524 | |
| 1525 | build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); |
| 1526 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { |
| 1527 | /* contention, we'll give up with this range */ |
| 1528 | begin = virt_end; |
| 1529 | continue; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1530 | } |
| 1531 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1532 | /* |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1533 | * IO may still be going to the destination block. We must |
| 1534 | * quiesce before we can do the removal. |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1535 | */ |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1536 | m = get_next_mapping(pool); |
| 1537 | m->tc = tc; |
| 1538 | m->maybe_shared = maybe_shared; |
| 1539 | m->virt_begin = virt_begin; |
| 1540 | m->virt_end = virt_end; |
| 1541 | m->data_block = data_begin; |
| 1542 | m->cell = data_cell; |
| 1543 | m->bio = bio; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1544 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1545 | /* |
| 1546 | * The parent bio must not complete before sub discard bios are |
| 1547 | * chained to it (see __blkdev_issue_discard_async's bio_chain)! |
| 1548 | * |
| 1549 | * This per-mapping bi_remaining increment is paired with |
| 1550 | * the implicit decrement that occurs via bio_endio() in |
| 1551 | * process_prepared_discard_{passdown,no_passdown}. |
| 1552 | */ |
| 1553 | __bio_inc_remaining(bio); |
| 1554 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
| 1555 | pool->process_prepared_discard(m); |
| 1556 | |
| 1557 | begin = virt_end; |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1558 | } |
| 1559 | } |
| 1560 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1561 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
| 1562 | { |
| 1563 | struct bio *bio = virt_cell->holder; |
| 1564 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
| 1565 | |
| 1566 | /* |
| 1567 | * The virt_cell will only get freed once the origin bio completes. |
| 1568 | * This means it will remain locked while all the individual |
| 1569 | * passdown bios are in flight. |
| 1570 | */ |
| 1571 | h->cell = virt_cell; |
| 1572 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); |
| 1573 | |
| 1574 | /* |
| 1575 | * We complete the bio now, knowing that the bi_remaining field |
| 1576 | * will prevent completion until the sub range discards have |
| 1577 | * completed. |
| 1578 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1579 | bio_endio(bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1580 | } |
| 1581 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1582 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
| 1583 | { |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1584 | dm_block_t begin, end; |
| 1585 | struct dm_cell_key virt_key; |
| 1586 | struct dm_bio_prison_cell *virt_cell; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1587 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1588 | get_bio_block_range(tc, bio, &begin, &end); |
| 1589 | if (begin == end) { |
| 1590 | /* |
| 1591 | * The discard covers less than a block. |
| 1592 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1593 | bio_endio(bio); |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1594 | return; |
| 1595 | } |
| 1596 | |
| 1597 | build_key(tc->td, VIRTUAL, begin, end, &virt_key); |
| 1598 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) |
| 1599 | /* |
| 1600 | * Potential starvation issue: We're relying on the |
| 1601 | * fs/application being well behaved, and not trying to |
| 1602 | * send IO to a region at the same time as discarding it. |
| 1603 | * If they do this persistently then it's possible this |
| 1604 | * cell will never be granted. |
| 1605 | */ |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1606 | return; |
| 1607 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 1608 | tc->pool->process_discard_cell(tc, virt_cell); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1609 | } |
| 1610 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1611 | 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] | 1612 | struct dm_cell_key *key, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1613 | struct dm_thin_lookup_result *lookup_result, |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 1614 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1615 | { |
| 1616 | int r; |
| 1617 | dm_block_t data_block; |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1618 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1619 | |
| 1620 | r = alloc_data_block(tc, &data_block); |
| 1621 | switch (r) { |
| 1622 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1623 | schedule_internal_copy(tc, block, lookup_result->block, |
| 1624 | data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1625 | break; |
| 1626 | |
| 1627 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1628 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1629 | break; |
| 1630 | |
| 1631 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1632 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1633 | __func__, r); |
Mike Snitzer | d6fc204 | 2013-08-21 17:40:11 -0400 | [diff] [blame] | 1634 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1635 | break; |
| 1636 | } |
| 1637 | } |
| 1638 | |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1639 | static void __remap_and_issue_shared_cell(void *context, |
| 1640 | struct dm_bio_prison_cell *cell) |
| 1641 | { |
| 1642 | struct remap_info *info = context; |
| 1643 | struct bio *bio; |
| 1644 | |
| 1645 | while ((bio = bio_list_pop(&cell->bios))) { |
| 1646 | if ((bio_data_dir(bio) == WRITE) || |
| 1647 | (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))) |
| 1648 | bio_list_add(&info->defer_bios, bio); |
| 1649 | else { |
| 1650 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));; |
| 1651 | |
| 1652 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); |
| 1653 | inc_all_io_entry(info->tc->pool, bio); |
| 1654 | bio_list_add(&info->issue_bios, bio); |
| 1655 | } |
| 1656 | } |
| 1657 | } |
| 1658 | |
| 1659 | static void remap_and_issue_shared_cell(struct thin_c *tc, |
| 1660 | struct dm_bio_prison_cell *cell, |
| 1661 | dm_block_t block) |
| 1662 | { |
| 1663 | struct bio *bio; |
| 1664 | struct remap_info info; |
| 1665 | |
| 1666 | info.tc = tc; |
| 1667 | bio_list_init(&info.defer_bios); |
| 1668 | bio_list_init(&info.issue_bios); |
| 1669 | |
| 1670 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, |
| 1671 | &info, cell); |
| 1672 | |
| 1673 | while ((bio = bio_list_pop(&info.defer_bios))) |
| 1674 | thin_defer_bio(tc, bio); |
| 1675 | |
| 1676 | while ((bio = bio_list_pop(&info.issue_bios))) |
| 1677 | remap_and_issue(tc, bio, block); |
| 1678 | } |
| 1679 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1680 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
| 1681 | dm_block_t block, |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1682 | struct dm_thin_lookup_result *lookup_result, |
| 1683 | struct dm_bio_prison_cell *virt_cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1684 | { |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1685 | struct dm_bio_prison_cell *data_cell; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1686 | struct pool *pool = tc->pool; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1687 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1688 | |
| 1689 | /* |
| 1690 | * If cell is already occupied, then sharing is already in the process |
| 1691 | * of being broken so we have nothing further to do here. |
| 1692 | */ |
| 1693 | build_data_key(tc->td, lookup_result->block, &key); |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1694 | if (bio_detain(pool, &key, bio, &data_cell)) { |
| 1695 | cell_defer_no_holder(tc, virt_cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1696 | return; |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1697 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1698 | |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1699 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
| 1700 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); |
| 1701 | cell_defer_no_holder(tc, virt_cell); |
| 1702 | } else { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 1703 | 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] | 1704 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 1705 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1706 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1707 | remap_and_issue(tc, bio, lookup_result->block); |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1708 | |
| 1709 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); |
| 1710 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1711 | } |
| 1712 | } |
| 1713 | |
| 1714 | 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] | 1715 | struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1716 | { |
| 1717 | int r; |
| 1718 | dm_block_t data_block; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1719 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1720 | |
| 1721 | /* |
| 1722 | * Remap empty bios (flushes) immediately, without provisioning. |
| 1723 | */ |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1724 | if (!bio->bi_iter.bi_size) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1725 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1726 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1727 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1728 | remap_and_issue(tc, bio, 0); |
| 1729 | return; |
| 1730 | } |
| 1731 | |
| 1732 | /* |
| 1733 | * Fill read bios with zeroes and complete them immediately. |
| 1734 | */ |
| 1735 | if (bio_data_dir(bio) == READ) { |
| 1736 | zero_fill_bio(bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1737 | cell_defer_no_holder(tc, cell); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1738 | bio_endio(bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1739 | return; |
| 1740 | } |
| 1741 | |
| 1742 | r = alloc_data_block(tc, &data_block); |
| 1743 | switch (r) { |
| 1744 | case 0: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1745 | if (tc->origin_dev) |
| 1746 | schedule_external_copy(tc, block, data_block, cell, bio); |
| 1747 | else |
| 1748 | schedule_zero(tc, block, data_block, cell, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1749 | break; |
| 1750 | |
| 1751 | case -ENOSPC: |
Mike Snitzer | 399cadd | 2013-12-05 16:03:33 -0500 | [diff] [blame] | 1752 | retry_bios_on_resume(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1753 | break; |
| 1754 | |
| 1755 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1756 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
| 1757 | __func__, r); |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1758 | cell_error(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1759 | break; |
| 1760 | } |
| 1761 | } |
| 1762 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1763 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1764 | { |
| 1765 | int r; |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1766 | struct pool *pool = tc->pool; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1767 | struct bio *bio = cell->holder; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1768 | dm_block_t block = get_bio_block(tc, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1769 | struct dm_thin_lookup_result lookup_result; |
| 1770 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1771 | if (tc->requeue_mode) { |
| 1772 | cell_requeue(pool, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1773 | return; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1774 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1775 | |
| 1776 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1777 | switch (r) { |
| 1778 | case 0: |
Joe Thornber | 23ca2bb | 2014-10-15 14:46:58 +0100 | [diff] [blame] | 1779 | if (lookup_result.shared) |
| 1780 | process_shared_bio(tc, bio, block, &lookup_result, cell); |
| 1781 | else { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1782 | inc_all_io_entry(pool, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1783 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1784 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1785 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1786 | break; |
| 1787 | |
| 1788 | case -ENODATA: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1789 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
Joe Thornber | 6beca5e | 2013-03-01 22:45:50 +0000 | [diff] [blame] | 1790 | inc_all_io_entry(pool, bio); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1791 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1792 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1793 | if (bio_end_sector(bio) <= tc->origin_size) |
| 1794 | remap_to_origin_and_issue(tc, bio); |
| 1795 | |
| 1796 | else if (bio->bi_iter.bi_sector < tc->origin_size) { |
| 1797 | zero_fill_bio(bio); |
| 1798 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; |
| 1799 | remap_to_origin_and_issue(tc, bio); |
| 1800 | |
| 1801 | } else { |
| 1802 | zero_fill_bio(bio); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1803 | bio_endio(bio); |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 1804 | } |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 1805 | } else |
| 1806 | provision_block(tc, bio, block, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1807 | break; |
| 1808 | |
| 1809 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1810 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1811 | __func__, r); |
Joe Thornber | f286ba0 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 1812 | cell_defer_no_holder(tc, cell); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1813 | bio_io_error(bio); |
| 1814 | break; |
| 1815 | } |
| 1816 | } |
| 1817 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1818 | static void process_bio(struct thin_c *tc, struct bio *bio) |
| 1819 | { |
| 1820 | struct pool *pool = tc->pool; |
| 1821 | dm_block_t block = get_bio_block(tc, bio); |
| 1822 | struct dm_bio_prison_cell *cell; |
| 1823 | struct dm_cell_key key; |
| 1824 | |
| 1825 | /* |
| 1826 | * If cell is already occupied, then the block is already |
| 1827 | * being provisioned so we have nothing further to do here. |
| 1828 | */ |
| 1829 | build_virtual_key(tc->td, block, &key); |
| 1830 | if (bio_detain(pool, &key, bio, &cell)) |
| 1831 | return; |
| 1832 | |
| 1833 | process_cell(tc, cell); |
| 1834 | } |
| 1835 | |
| 1836 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, |
| 1837 | struct dm_bio_prison_cell *cell) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1838 | { |
| 1839 | int r; |
| 1840 | int rw = bio_data_dir(bio); |
| 1841 | dm_block_t block = get_bio_block(tc, bio); |
| 1842 | struct dm_thin_lookup_result lookup_result; |
| 1843 | |
| 1844 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); |
| 1845 | switch (r) { |
| 1846 | case 0: |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1847 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1848 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1849 | if (cell) |
| 1850 | cell_defer_no_holder(tc, cell); |
| 1851 | } else { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1852 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1853 | remap_and_issue(tc, bio, lookup_result.block); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1854 | if (cell) |
| 1855 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1856 | } |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1857 | break; |
| 1858 | |
| 1859 | case -ENODATA: |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1860 | if (cell) |
| 1861 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1862 | if (rw != READ) { |
Mike Snitzer | 8c0f0e8 | 2013-12-05 15:47:24 -0500 | [diff] [blame] | 1863 | handle_unserviceable_bio(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1864 | break; |
| 1865 | } |
| 1866 | |
| 1867 | if (tc->origin_dev) { |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 1868 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1869 | remap_to_origin_and_issue(tc, bio); |
| 1870 | break; |
| 1871 | } |
| 1872 | |
| 1873 | zero_fill_bio(bio); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1874 | bio_endio(bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1875 | break; |
| 1876 | |
| 1877 | default: |
Mike Snitzer | c397741 | 2012-12-21 20:23:34 +0000 | [diff] [blame] | 1878 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
| 1879 | __func__, r); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1880 | if (cell) |
| 1881 | cell_defer_no_holder(tc, cell); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1882 | bio_io_error(bio); |
| 1883 | break; |
| 1884 | } |
| 1885 | } |
| 1886 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1887 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
| 1888 | { |
| 1889 | __process_bio_read_only(tc, bio, NULL); |
| 1890 | } |
| 1891 | |
| 1892 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
| 1893 | { |
| 1894 | __process_bio_read_only(tc, cell->holder, cell); |
| 1895 | } |
| 1896 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1897 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
| 1898 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1899 | bio_endio(bio); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 1900 | } |
| 1901 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 1902 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
| 1903 | { |
| 1904 | bio_io_error(bio); |
| 1905 | } |
| 1906 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 1907 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
| 1908 | { |
| 1909 | cell_success(tc->pool, cell); |
| 1910 | } |
| 1911 | |
| 1912 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
| 1913 | { |
| 1914 | cell_error(tc->pool, cell); |
| 1915 | } |
| 1916 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 1917 | /* |
| 1918 | * FIXME: should we also commit due to size of transaction, measured in |
| 1919 | * metadata blocks? |
| 1920 | */ |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1921 | static int need_commit_due_to_time(struct pool *pool) |
| 1922 | { |
Manuel Schölling | 0f30af9 | 2014-05-22 22:42:37 +0200 | [diff] [blame] | 1923 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
| 1924 | pool->last_commit_jiffies + COMMIT_PERIOD); |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 1925 | } |
| 1926 | |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1927 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
| 1928 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) |
| 1929 | |
| 1930 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) |
| 1931 | { |
| 1932 | struct rb_node **rbp, *parent; |
| 1933 | struct dm_thin_endio_hook *pbd; |
| 1934 | sector_t bi_sector = bio->bi_iter.bi_sector; |
| 1935 | |
| 1936 | rbp = &tc->sort_bio_list.rb_node; |
| 1937 | parent = NULL; |
| 1938 | while (*rbp) { |
| 1939 | parent = *rbp; |
| 1940 | pbd = thin_pbd(parent); |
| 1941 | |
| 1942 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) |
| 1943 | rbp = &(*rbp)->rb_left; |
| 1944 | else |
| 1945 | rbp = &(*rbp)->rb_right; |
| 1946 | } |
| 1947 | |
| 1948 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
| 1949 | rb_link_node(&pbd->rb_node, parent, rbp); |
| 1950 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); |
| 1951 | } |
| 1952 | |
| 1953 | static void __extract_sorted_bios(struct thin_c *tc) |
| 1954 | { |
| 1955 | struct rb_node *node; |
| 1956 | struct dm_thin_endio_hook *pbd; |
| 1957 | struct bio *bio; |
| 1958 | |
| 1959 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { |
| 1960 | pbd = thin_pbd(node); |
| 1961 | bio = thin_bio(pbd); |
| 1962 | |
| 1963 | bio_list_add(&tc->deferred_bio_list, bio); |
| 1964 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); |
| 1965 | } |
| 1966 | |
| 1967 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); |
| 1968 | } |
| 1969 | |
| 1970 | static void __sort_thin_deferred_bios(struct thin_c *tc) |
| 1971 | { |
| 1972 | struct bio *bio; |
| 1973 | struct bio_list bios; |
| 1974 | |
| 1975 | bio_list_init(&bios); |
| 1976 | bio_list_merge(&bios, &tc->deferred_bio_list); |
| 1977 | bio_list_init(&tc->deferred_bio_list); |
| 1978 | |
| 1979 | /* Sort deferred_bio_list using rb-tree */ |
| 1980 | while ((bio = bio_list_pop(&bios))) |
| 1981 | __thin_bio_rb_add(tc, bio); |
| 1982 | |
| 1983 | /* |
| 1984 | * Transfer the sorted bios in sort_bio_list back to |
| 1985 | * deferred_bio_list to allow lockless submission of |
| 1986 | * all bios. |
| 1987 | */ |
| 1988 | __extract_sorted_bios(tc); |
| 1989 | } |
| 1990 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1991 | static void process_thin_deferred_bios(struct thin_c *tc) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1992 | { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 1993 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1994 | unsigned long flags; |
| 1995 | struct bio *bio; |
| 1996 | struct bio_list bios; |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 1997 | struct blk_plug plug; |
Joe Thornber | 8a01a6a | 2014-10-06 15:28:30 +0100 | [diff] [blame] | 1998 | unsigned count = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 1999 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2000 | if (tc->requeue_mode) { |
Mike Snitzer | 42d6a8c | 2014-10-19 07:52:44 -0400 | [diff] [blame] | 2001 | error_thin_bio_list(tc, &tc->deferred_bio_list, DM_ENDIO_REQUEUE); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2002 | return; |
| 2003 | } |
| 2004 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2005 | bio_list_init(&bios); |
| 2006 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2007 | spin_lock_irqsave(&tc->lock, flags); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 2008 | |
| 2009 | if (bio_list_empty(&tc->deferred_bio_list)) { |
| 2010 | spin_unlock_irqrestore(&tc->lock, flags); |
| 2011 | return; |
| 2012 | } |
| 2013 | |
| 2014 | __sort_thin_deferred_bios(tc); |
| 2015 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2016 | bio_list_merge(&bios, &tc->deferred_bio_list); |
| 2017 | bio_list_init(&tc->deferred_bio_list); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 2018 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2019 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2020 | |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 2021 | blk_start_plug(&plug); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2022 | while ((bio = bio_list_pop(&bios))) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2023 | /* |
| 2024 | * If we've got no free new_mapping structs, and processing |
| 2025 | * this bio might require one, we pause until there are some |
| 2026 | * prepared mappings to process. |
| 2027 | */ |
| 2028 | if (ensure_next_mapping(pool)) { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2029 | spin_lock_irqsave(&tc->lock, flags); |
| 2030 | bio_list_add(&tc->deferred_bio_list, bio); |
| 2031 | bio_list_merge(&tc->deferred_bio_list, &bios); |
| 2032 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2033 | break; |
| 2034 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2035 | |
| 2036 | if (bio->bi_rw & REQ_DISCARD) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2037 | pool->process_discard(tc, bio); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2038 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2039 | pool->process_bio(tc, bio); |
Joe Thornber | 8a01a6a | 2014-10-06 15:28:30 +0100 | [diff] [blame] | 2040 | |
| 2041 | if ((count++ & 127) == 0) { |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2042 | throttle_work_update(&pool->throttle); |
Joe Thornber | 8a01a6a | 2014-10-06 15:28:30 +0100 | [diff] [blame] | 2043 | dm_pool_issue_prefetches(pool->pmd); |
| 2044 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2045 | } |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 2046 | blk_finish_plug(&plug); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2047 | } |
| 2048 | |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 2049 | static int cmp_cells(const void *lhs, const void *rhs) |
| 2050 | { |
| 2051 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); |
| 2052 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); |
| 2053 | |
| 2054 | BUG_ON(!lhs_cell->holder); |
| 2055 | BUG_ON(!rhs_cell->holder); |
| 2056 | |
| 2057 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) |
| 2058 | return -1; |
| 2059 | |
| 2060 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) |
| 2061 | return 1; |
| 2062 | |
| 2063 | return 0; |
| 2064 | } |
| 2065 | |
| 2066 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) |
| 2067 | { |
| 2068 | unsigned count = 0; |
| 2069 | struct dm_bio_prison_cell *cell, *tmp; |
| 2070 | |
| 2071 | list_for_each_entry_safe(cell, tmp, cells, user_list) { |
| 2072 | if (count >= CELL_SORT_ARRAY_SIZE) |
| 2073 | break; |
| 2074 | |
| 2075 | pool->cell_sort_array[count++] = cell; |
| 2076 | list_del(&cell->user_list); |
| 2077 | } |
| 2078 | |
| 2079 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); |
| 2080 | |
| 2081 | return count; |
| 2082 | } |
| 2083 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2084 | static void process_thin_deferred_cells(struct thin_c *tc) |
| 2085 | { |
| 2086 | struct pool *pool = tc->pool; |
| 2087 | unsigned long flags; |
| 2088 | struct list_head cells; |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 2089 | struct dm_bio_prison_cell *cell; |
| 2090 | unsigned i, j, count; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2091 | |
| 2092 | INIT_LIST_HEAD(&cells); |
| 2093 | |
| 2094 | spin_lock_irqsave(&tc->lock, flags); |
| 2095 | list_splice_init(&tc->deferred_cells, &cells); |
| 2096 | spin_unlock_irqrestore(&tc->lock, flags); |
| 2097 | |
| 2098 | if (list_empty(&cells)) |
| 2099 | return; |
| 2100 | |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 2101 | do { |
| 2102 | count = sort_cells(tc->pool, &cells); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2103 | |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 2104 | for (i = 0; i < count; i++) { |
| 2105 | cell = pool->cell_sort_array[i]; |
| 2106 | BUG_ON(!cell->holder); |
| 2107 | |
| 2108 | /* |
| 2109 | * If we've got no free new_mapping structs, and processing |
| 2110 | * this bio might require one, we pause until there are some |
| 2111 | * prepared mappings to process. |
| 2112 | */ |
| 2113 | if (ensure_next_mapping(pool)) { |
| 2114 | for (j = i; j < count; j++) |
| 2115 | list_add(&pool->cell_sort_array[j]->user_list, &cells); |
| 2116 | |
| 2117 | spin_lock_irqsave(&tc->lock, flags); |
| 2118 | list_splice(&cells, &tc->deferred_cells); |
| 2119 | spin_unlock_irqrestore(&tc->lock, flags); |
| 2120 | return; |
| 2121 | } |
| 2122 | |
| 2123 | if (cell->holder->bi_rw & REQ_DISCARD) |
| 2124 | pool->process_discard_cell(tc, cell); |
| 2125 | else |
| 2126 | pool->process_cell(tc, cell); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2127 | } |
Joe Thornber | ac4c3f3 | 2014-10-10 16:42:10 +0100 | [diff] [blame] | 2128 | } while (!list_empty(&cells)); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2129 | } |
| 2130 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 2131 | static void thin_get(struct thin_c *tc); |
| 2132 | static void thin_put(struct thin_c *tc); |
| 2133 | |
| 2134 | /* |
| 2135 | * We can't hold rcu_read_lock() around code that can block. So we |
| 2136 | * find a thin with the rcu lock held; bump a refcount; then drop |
| 2137 | * the lock. |
| 2138 | */ |
| 2139 | static struct thin_c *get_first_thin(struct pool *pool) |
| 2140 | { |
| 2141 | struct thin_c *tc = NULL; |
| 2142 | |
| 2143 | rcu_read_lock(); |
| 2144 | if (!list_empty(&pool->active_thins)) { |
| 2145 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); |
| 2146 | thin_get(tc); |
| 2147 | } |
| 2148 | rcu_read_unlock(); |
| 2149 | |
| 2150 | return tc; |
| 2151 | } |
| 2152 | |
| 2153 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) |
| 2154 | { |
| 2155 | struct thin_c *old_tc = tc; |
| 2156 | |
| 2157 | rcu_read_lock(); |
| 2158 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { |
| 2159 | thin_get(tc); |
| 2160 | thin_put(old_tc); |
| 2161 | rcu_read_unlock(); |
| 2162 | return tc; |
| 2163 | } |
| 2164 | thin_put(old_tc); |
| 2165 | rcu_read_unlock(); |
| 2166 | |
| 2167 | return NULL; |
| 2168 | } |
| 2169 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2170 | static void process_deferred_bios(struct pool *pool) |
| 2171 | { |
| 2172 | unsigned long flags; |
| 2173 | struct bio *bio; |
| 2174 | struct bio_list bios; |
| 2175 | struct thin_c *tc; |
| 2176 | |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 2177 | tc = get_first_thin(pool); |
| 2178 | while (tc) { |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2179 | process_thin_deferred_cells(tc); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2180 | process_thin_deferred_bios(tc); |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 2181 | tc = get_next_thin(pool, tc); |
| 2182 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2183 | |
| 2184 | /* |
| 2185 | * If there are any deferred flush bios, we must commit |
| 2186 | * the metadata before issuing them. |
| 2187 | */ |
| 2188 | bio_list_init(&bios); |
| 2189 | spin_lock_irqsave(&pool->lock, flags); |
| 2190 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
| 2191 | bio_list_init(&pool->deferred_flush_bios); |
| 2192 | spin_unlock_irqrestore(&pool->lock, flags); |
| 2193 | |
Mike Snitzer | 4d1662a | 2014-02-06 06:08:56 -0500 | [diff] [blame] | 2194 | if (bio_list_empty(&bios) && |
| 2195 | !(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] | 2196 | return; |
| 2197 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 2198 | if (commit(pool)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2199 | while ((bio = bio_list_pop(&bios))) |
| 2200 | bio_io_error(bio); |
| 2201 | return; |
| 2202 | } |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2203 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2204 | |
| 2205 | while ((bio = bio_list_pop(&bios))) |
| 2206 | generic_make_request(bio); |
| 2207 | } |
| 2208 | |
| 2209 | static void do_worker(struct work_struct *ws) |
| 2210 | { |
| 2211 | struct pool *pool = container_of(ws, struct pool, worker); |
| 2212 | |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2213 | throttle_work_start(&pool->throttle); |
Joe Thornber | 8a01a6a | 2014-10-06 15:28:30 +0100 | [diff] [blame] | 2214 | dm_pool_issue_prefetches(pool->pmd); |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2215 | throttle_work_update(&pool->throttle); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2216 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2217 | throttle_work_update(&pool->throttle); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2218 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2219 | throttle_work_update(&pool->throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2220 | process_deferred_bios(pool); |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2221 | throttle_work_complete(&pool->throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2222 | } |
| 2223 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2224 | /* |
| 2225 | * We want to commit periodically so that not too much |
| 2226 | * unwritten data builds up. |
| 2227 | */ |
| 2228 | static void do_waker(struct work_struct *ws) |
| 2229 | { |
| 2230 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); |
| 2231 | wake_worker(pool); |
| 2232 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); |
| 2233 | } |
| 2234 | |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2235 | static void notify_of_pool_mode_change_to_oods(struct pool *pool); |
| 2236 | |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2237 | /* |
| 2238 | * We're holding onto IO to allow userland time to react. After the |
| 2239 | * timeout either the pool will have been resized (and thus back in |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2240 | * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space. |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2241 | */ |
| 2242 | static void do_no_space_timeout(struct work_struct *ws) |
| 2243 | { |
| 2244 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, |
| 2245 | no_space_timeout); |
| 2246 | |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2247 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) { |
| 2248 | pool->pf.error_if_no_space = true; |
| 2249 | notify_of_pool_mode_change_to_oods(pool); |
Mike Snitzer | 0a927c2 | 2015-07-21 13:20:46 -0400 | [diff] [blame] | 2250 | error_retry_list_with_code(pool, -ENOSPC); |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2251 | } |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2252 | } |
| 2253 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2254 | /*----------------------------------------------------------------*/ |
| 2255 | |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2256 | struct pool_work { |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2257 | struct work_struct worker; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2258 | struct completion complete; |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2259 | }; |
| 2260 | |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2261 | static struct pool_work *to_pool_work(struct work_struct *ws) |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2262 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2263 | return container_of(ws, struct pool_work, worker); |
| 2264 | } |
| 2265 | |
| 2266 | static void pool_work_complete(struct pool_work *pw) |
| 2267 | { |
| 2268 | complete(&pw->complete); |
| 2269 | } |
| 2270 | |
| 2271 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
| 2272 | void (*fn)(struct work_struct *)) |
| 2273 | { |
| 2274 | INIT_WORK_ONSTACK(&pw->worker, fn); |
| 2275 | init_completion(&pw->complete); |
| 2276 | queue_work(pool->wq, &pw->worker); |
| 2277 | wait_for_completion(&pw->complete); |
| 2278 | } |
| 2279 | |
| 2280 | /*----------------------------------------------------------------*/ |
| 2281 | |
| 2282 | struct noflush_work { |
| 2283 | struct pool_work pw; |
| 2284 | struct thin_c *tc; |
| 2285 | }; |
| 2286 | |
| 2287 | static struct noflush_work *to_noflush(struct work_struct *ws) |
| 2288 | { |
| 2289 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2290 | } |
| 2291 | |
| 2292 | static void do_noflush_start(struct work_struct *ws) |
| 2293 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2294 | struct noflush_work *w = to_noflush(ws); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2295 | w->tc->requeue_mode = true; |
| 2296 | requeue_io(w->tc); |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2297 | pool_work_complete(&w->pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2298 | } |
| 2299 | |
| 2300 | static void do_noflush_stop(struct work_struct *ws) |
| 2301 | { |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2302 | struct noflush_work *w = to_noflush(ws); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2303 | w->tc->requeue_mode = false; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2304 | pool_work_complete(&w->pw); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2305 | } |
| 2306 | |
| 2307 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) |
| 2308 | { |
| 2309 | struct noflush_work w; |
| 2310 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2311 | w.tc = tc; |
Joe Thornber | e7a3e87 | 2014-05-13 16:14:14 -0400 | [diff] [blame] | 2312 | pool_work_wait(&w.pw, tc->pool, fn); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2313 | } |
| 2314 | |
| 2315 | /*----------------------------------------------------------------*/ |
| 2316 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2317 | static enum pool_mode get_pool_mode(struct pool *pool) |
| 2318 | { |
| 2319 | return pool->pf.mode; |
| 2320 | } |
| 2321 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2322 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
| 2323 | { |
| 2324 | dm_table_event(pool->ti->table); |
| 2325 | DMINFO("%s: switching pool to %s mode", |
| 2326 | dm_device_name(pool->pool_md), new_mode); |
| 2327 | } |
| 2328 | |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2329 | static void notify_of_pool_mode_change_to_oods(struct pool *pool) |
| 2330 | { |
| 2331 | if (!pool->pf.error_if_no_space) |
| 2332 | notify_of_pool_mode_change(pool, "out-of-data-space (queue IO)"); |
| 2333 | else |
| 2334 | notify_of_pool_mode_change(pool, "out-of-data-space (error IO)"); |
| 2335 | } |
| 2336 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 2337 | static bool passdown_enabled(struct pool_c *pt) |
| 2338 | { |
| 2339 | return pt->adjusted_pf.discard_passdown; |
| 2340 | } |
| 2341 | |
| 2342 | static void set_discard_callbacks(struct pool *pool) |
| 2343 | { |
| 2344 | struct pool_c *pt = pool->ti->private; |
| 2345 | |
| 2346 | if (passdown_enabled(pt)) { |
| 2347 | pool->process_discard_cell = process_discard_cell_passdown; |
| 2348 | pool->process_prepared_discard = process_prepared_discard_passdown; |
| 2349 | } else { |
| 2350 | pool->process_discard_cell = process_discard_cell_no_passdown; |
| 2351 | pool->process_prepared_discard = process_prepared_discard_no_passdown; |
| 2352 | } |
| 2353 | } |
| 2354 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2355 | 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] | 2356 | { |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 2357 | struct pool_c *pt = pool->ti->private; |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2358 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
| 2359 | enum pool_mode old_mode = get_pool_mode(pool); |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 2360 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2361 | |
| 2362 | /* |
| 2363 | * Never allow the pool to transition to PM_WRITE mode if user |
| 2364 | * intervention is required to verify metadata and data consistency. |
| 2365 | */ |
| 2366 | if (new_mode == PM_WRITE && needs_check) { |
| 2367 | DMERR("%s: unable to switch pool to write mode until repaired.", |
| 2368 | dm_device_name(pool->pool_md)); |
| 2369 | if (old_mode != new_mode) |
| 2370 | new_mode = old_mode; |
| 2371 | else |
| 2372 | new_mode = PM_READ_ONLY; |
| 2373 | } |
| 2374 | /* |
| 2375 | * If we were in PM_FAIL mode, rollback of metadata failed. We're |
| 2376 | * not going to recover without a thin_repair. So we never let the |
| 2377 | * pool move out of the old mode. |
| 2378 | */ |
| 2379 | if (old_mode == PM_FAIL) |
| 2380 | new_mode = old_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2381 | |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2382 | switch (new_mode) { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2383 | case PM_FAIL: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2384 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2385 | notify_of_pool_mode_change(pool, "failure"); |
Joe Thornber | 5383ef3 | 2013-12-04 16:30:01 -0500 | [diff] [blame] | 2386 | dm_pool_metadata_read_only(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2387 | pool->process_bio = process_bio_fail; |
| 2388 | pool->process_discard = process_bio_fail; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2389 | pool->process_cell = process_cell_fail; |
| 2390 | pool->process_discard_cell = process_cell_fail; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2391 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
| 2392 | pool->process_prepared_discard = process_prepared_discard_fail; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2393 | |
| 2394 | error_retry_list(pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2395 | break; |
| 2396 | |
| 2397 | case PM_READ_ONLY: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2398 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2399 | notify_of_pool_mode_change(pool, "read-only"); |
| 2400 | dm_pool_metadata_read_only(pool->pmd); |
| 2401 | pool->process_bio = process_bio_read_only; |
| 2402 | pool->process_discard = process_bio_success; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2403 | pool->process_cell = process_cell_read_only; |
| 2404 | pool->process_discard_cell = process_cell_success; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2405 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 2406 | pool->process_prepared_discard = process_prepared_discard_success; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2407 | |
| 2408 | error_retry_list(pool); |
| 2409 | break; |
| 2410 | |
| 2411 | case PM_OUT_OF_DATA_SPACE: |
| 2412 | /* |
| 2413 | * Ideally we'd never hit this state; the low water mark |
| 2414 | * would trigger userland to extend the pool before we |
| 2415 | * completely run out of data space. However, many small |
| 2416 | * IOs to unprovisioned space can consume data space at an |
| 2417 | * alarming rate. Adjust your low water mark if you're |
| 2418 | * frequently seeing this mode. |
| 2419 | */ |
| 2420 | if (old_mode != new_mode) |
Mike Snitzer | bcc696f | 2015-07-15 16:52:04 -0400 | [diff] [blame] | 2421 | notify_of_pool_mode_change_to_oods(pool); |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2422 | pool->process_bio = process_bio_read_only; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2423 | pool->process_discard = process_discard_bio; |
| 2424 | pool->process_cell = process_cell_read_only; |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2425 | pool->process_prepared_mapping = process_prepared_mapping; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 2426 | set_discard_callbacks(pool); |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2427 | |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 2428 | if (!pool->pf.error_if_no_space && no_space_timeout) |
| 2429 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2430 | break; |
| 2431 | |
| 2432 | case PM_WRITE: |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2433 | if (old_mode != new_mode) |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 2434 | notify_of_pool_mode_change(pool, "write"); |
Mike Snitzer | 172c238 | 2015-11-06 10:53:01 -0500 | [diff] [blame] | 2435 | pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space; |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 2436 | dm_pool_metadata_read_write(pool->pmd); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2437 | pool->process_bio = process_bio; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2438 | pool->process_discard = process_discard_bio; |
| 2439 | pool->process_cell = process_cell; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2440 | pool->process_prepared_mapping = process_prepared_mapping; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 2441 | set_discard_callbacks(pool); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2442 | break; |
| 2443 | } |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2444 | |
| 2445 | pool->pf.mode = new_mode; |
Mike Snitzer | cdc2b41 | 2014-02-14 18:10:55 -0500 | [diff] [blame] | 2446 | /* |
| 2447 | * The pool mode may have changed, sync it so bind_control_target() |
| 2448 | * doesn't cause an unexpected mode transition on resume. |
| 2449 | */ |
| 2450 | pt->adjusted_pf.mode = new_mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2451 | } |
| 2452 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2453 | static void abort_transaction(struct pool *pool) |
| 2454 | { |
| 2455 | const char *dev_name = dm_device_name(pool->pool_md); |
| 2456 | |
| 2457 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); |
| 2458 | if (dm_pool_abort_metadata(pool->pmd)) { |
| 2459 | DMERR("%s: failed to abort metadata transaction", dev_name); |
| 2460 | set_pool_mode(pool, PM_FAIL); |
| 2461 | } |
| 2462 | |
| 2463 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { |
| 2464 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); |
| 2465 | set_pool_mode(pool, PM_FAIL); |
| 2466 | } |
| 2467 | } |
| 2468 | |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2469 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
| 2470 | { |
| 2471 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
| 2472 | dm_device_name(pool->pool_md), op, r); |
| 2473 | |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2474 | abort_transaction(pool); |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 2475 | set_pool_mode(pool, PM_READ_ONLY); |
| 2476 | } |
| 2477 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2478 | /*----------------------------------------------------------------*/ |
| 2479 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2480 | /* |
| 2481 | * Mapping functions. |
| 2482 | */ |
| 2483 | |
| 2484 | /* |
| 2485 | * Called only while mapping a thin bio to hand it over to the workqueue. |
| 2486 | */ |
| 2487 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) |
| 2488 | { |
| 2489 | unsigned long flags; |
| 2490 | struct pool *pool = tc->pool; |
| 2491 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2492 | spin_lock_irqsave(&tc->lock, flags); |
| 2493 | bio_list_add(&tc->deferred_bio_list, bio); |
| 2494 | spin_unlock_irqrestore(&tc->lock, flags); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2495 | |
| 2496 | wake_worker(pool); |
| 2497 | } |
| 2498 | |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2499 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
| 2500 | { |
| 2501 | struct pool *pool = tc->pool; |
| 2502 | |
| 2503 | throttle_lock(&pool->throttle); |
| 2504 | thin_defer_bio(tc, bio); |
| 2505 | throttle_unlock(&pool->throttle); |
| 2506 | } |
| 2507 | |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2508 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
| 2509 | { |
| 2510 | unsigned long flags; |
| 2511 | struct pool *pool = tc->pool; |
| 2512 | |
| 2513 | throttle_lock(&pool->throttle); |
| 2514 | spin_lock_irqsave(&tc->lock, flags); |
| 2515 | list_add_tail(&cell->user_list, &tc->deferred_cells); |
| 2516 | spin_unlock_irqrestore(&tc->lock, flags); |
| 2517 | throttle_unlock(&pool->throttle); |
| 2518 | |
| 2519 | wake_worker(pool); |
| 2520 | } |
| 2521 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 2522 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2523 | { |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 2524 | 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] | 2525 | |
| 2526 | h->tc = tc; |
| 2527 | h->shared_read_entry = NULL; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2528 | h->all_io_entry = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2529 | h->overwrite_mapping = NULL; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 2530 | h->cell = NULL; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2531 | } |
| 2532 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2533 | /* |
| 2534 | * Non-blocking function called from the thin target's map function. |
| 2535 | */ |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 2536 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2537 | { |
| 2538 | int r; |
| 2539 | struct thin_c *tc = ti->private; |
| 2540 | dm_block_t block = get_bio_block(tc, bio); |
| 2541 | struct dm_thin_device *td = tc->td; |
| 2542 | struct dm_thin_lookup_result result; |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2543 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2544 | struct dm_cell_key key; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2545 | |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 2546 | thin_hook_bio(tc, bio); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2547 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2548 | if (tc->requeue_mode) { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2549 | bio->bi_error = DM_ENDIO_REQUEUE; |
| 2550 | bio_endio(bio); |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 2551 | return DM_MAPIO_SUBMITTED; |
| 2552 | } |
| 2553 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2554 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 2555 | bio_io_error(bio); |
| 2556 | return DM_MAPIO_SUBMITTED; |
| 2557 | } |
| 2558 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2559 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2560 | thin_defer_bio_with_throttle(tc, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2561 | return DM_MAPIO_SUBMITTED; |
| 2562 | } |
| 2563 | |
Joe Thornber | c822ed9 | 2014-10-10 09:41:09 +0100 | [diff] [blame] | 2564 | /* |
| 2565 | * We must hold the virtual cell before doing the lookup, otherwise |
| 2566 | * there's a race with discard. |
| 2567 | */ |
| 2568 | build_virtual_key(tc->td, block, &key); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2569 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
Joe Thornber | c822ed9 | 2014-10-10 09:41:09 +0100 | [diff] [blame] | 2570 | return DM_MAPIO_SUBMITTED; |
| 2571 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2572 | r = dm_thin_find_block(td, block, 0, &result); |
| 2573 | |
| 2574 | /* |
| 2575 | * Note that we defer readahead too. |
| 2576 | */ |
| 2577 | switch (r) { |
| 2578 | case 0: |
| 2579 | if (unlikely(result.shared)) { |
| 2580 | /* |
| 2581 | * We have a race condition here between the |
| 2582 | * result.shared value returned by the lookup and |
| 2583 | * snapshot creation, which may cause new |
| 2584 | * sharing. |
| 2585 | * |
| 2586 | * To avoid this always quiesce the origin before |
| 2587 | * taking the snap. You want to do this anyway to |
| 2588 | * ensure a consistent application view |
| 2589 | * (i.e. lockfs). |
| 2590 | * |
| 2591 | * More distant ancestors are irrelevant. The |
| 2592 | * shared flag will be set in their case. |
| 2593 | */ |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2594 | thin_defer_cell(tc, virt_cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2595 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2596 | } |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2597 | |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2598 | build_data_key(tc->td, result.block, &key); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2599 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
| 2600 | cell_defer_no_holder(tc, virt_cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2601 | return DM_MAPIO_SUBMITTED; |
| 2602 | } |
| 2603 | |
| 2604 | inc_all_io_entry(tc->pool, bio); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2605 | cell_defer_no_holder(tc, data_cell); |
| 2606 | cell_defer_no_holder(tc, virt_cell); |
Joe Thornber | e808807 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 2607 | |
| 2608 | remap(tc, bio, result.block); |
| 2609 | return DM_MAPIO_REMAPPED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2610 | |
| 2611 | case -ENODATA: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2612 | case -EWOULDBLOCK: |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2613 | thin_defer_cell(tc, virt_cell); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 2614 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2615 | |
| 2616 | default: |
| 2617 | /* |
| 2618 | * Must always call bio_io_error on failure. |
| 2619 | * dm_thin_find_block can fail with -EINVAL if the |
| 2620 | * pool is switched to fail-io mode. |
| 2621 | */ |
| 2622 | bio_io_error(bio); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 2623 | cell_defer_no_holder(tc, virt_cell); |
Joe Thornber | 2aab385 | 2012-12-21 20:23:33 +0000 | [diff] [blame] | 2624 | return DM_MAPIO_SUBMITTED; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2625 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2626 | } |
| 2627 | |
| 2628 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) |
| 2629 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2630 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 2631 | struct request_queue *q; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2632 | |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 2633 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
| 2634 | return 1; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2635 | |
Mike Snitzer | 760fe67 | 2014-03-20 08:36:47 -0400 | [diff] [blame] | 2636 | q = bdev_get_queue(pt->data_dev->bdev); |
| 2637 | return bdi_congested(&q->backing_dev_info, bdi_bits); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2638 | } |
| 2639 | |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2640 | static void requeue_bios(struct pool *pool) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2641 | { |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2642 | unsigned long flags; |
| 2643 | struct thin_c *tc; |
| 2644 | |
| 2645 | rcu_read_lock(); |
| 2646 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { |
| 2647 | spin_lock_irqsave(&tc->lock, flags); |
| 2648 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); |
| 2649 | bio_list_init(&tc->retry_on_resume_list); |
| 2650 | spin_unlock_irqrestore(&tc->lock, flags); |
| 2651 | } |
| 2652 | rcu_read_unlock(); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2653 | } |
| 2654 | |
| 2655 | /*---------------------------------------------------------------- |
| 2656 | * Binding of control targets to a pool object |
| 2657 | *--------------------------------------------------------------*/ |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2658 | static bool data_dev_supports_discard(struct pool_c *pt) |
| 2659 | { |
| 2660 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); |
| 2661 | |
| 2662 | return q && blk_queue_discard(q); |
| 2663 | } |
| 2664 | |
Joe Thornber | 58051b9 | 2013-03-20 17:21:25 +0000 | [diff] [blame] | 2665 | static bool is_factor(sector_t block_size, uint32_t n) |
| 2666 | { |
| 2667 | return !sector_div(block_size, n); |
| 2668 | } |
| 2669 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2670 | /* |
| 2671 | * If discard_passdown was enabled verify that the data device |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2672 | * supports discards. Disable discard_passdown if not. |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2673 | */ |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2674 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2675 | { |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2676 | struct pool *pool = pt->pool; |
| 2677 | struct block_device *data_bdev = pt->data_dev->bdev; |
| 2678 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2679 | const char *reason = NULL; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2680 | char buf[BDEVNAME_SIZE]; |
| 2681 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2682 | if (!pt->adjusted_pf.discard_passdown) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2683 | return; |
| 2684 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2685 | if (!data_dev_supports_discard(pt)) |
| 2686 | reason = "discard unsupported"; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2687 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2688 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) |
| 2689 | reason = "max discard sectors smaller than a block"; |
| 2690 | |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2691 | if (reason) { |
| 2692 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); |
| 2693 | pt->adjusted_pf.discard_passdown = false; |
| 2694 | } |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2695 | } |
| 2696 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2697 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
| 2698 | { |
| 2699 | struct pool_c *pt = ti->private; |
| 2700 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2701 | /* |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 2702 | * 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] | 2703 | */ |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 2704 | enum pool_mode old_mode = get_pool_mode(pool); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 2705 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2706 | |
Joe Thornber | 9b7aaa6 | 2013-12-04 16:58:19 -0500 | [diff] [blame] | 2707 | /* |
Mike Snitzer | 8b64e88 | 2013-12-20 14:27:28 -0500 | [diff] [blame] | 2708 | * Don't change the pool's mode until set_pool_mode() below. |
| 2709 | * Otherwise the pool's process_* function pointers may |
| 2710 | * not match the desired pool mode. |
| 2711 | */ |
| 2712 | pt->adjusted_pf.mode = old_mode; |
| 2713 | |
| 2714 | pool->ti = ti; |
| 2715 | pool->pf = pt->adjusted_pf; |
| 2716 | pool->low_water_blocks = pt->low_water_blocks; |
| 2717 | |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2718 | set_pool_mode(pool, new_mode); |
Mike Snitzer | f402693 | 2012-05-19 01:01:01 +0100 | [diff] [blame] | 2719 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2720 | return 0; |
| 2721 | } |
| 2722 | |
| 2723 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) |
| 2724 | { |
| 2725 | if (pool->ti == ti) |
| 2726 | pool->ti = NULL; |
| 2727 | } |
| 2728 | |
| 2729 | /*---------------------------------------------------------------- |
| 2730 | * Pool creation |
| 2731 | *--------------------------------------------------------------*/ |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2732 | /* Initialize pool features. */ |
| 2733 | static void pool_features_init(struct pool_features *pf) |
| 2734 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2735 | pf->mode = PM_WRITE; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2736 | pf->zero_new_blocks = true; |
| 2737 | pf->discard_enabled = true; |
| 2738 | pf->discard_passdown = true; |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2739 | pf->error_if_no_space = false; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2740 | } |
| 2741 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2742 | static void __pool_destroy(struct pool *pool) |
| 2743 | { |
| 2744 | __pool_table_remove(pool); |
| 2745 | |
Joe Thornber | a822c83 | 2015-07-03 10:22:42 +0100 | [diff] [blame] | 2746 | vfree(pool->cell_sort_array); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2747 | if (dm_pool_metadata_close(pool->pmd) < 0) |
| 2748 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 2749 | |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2750 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2751 | dm_kcopyd_client_destroy(pool->copier); |
| 2752 | |
| 2753 | if (pool->wq) |
| 2754 | destroy_workqueue(pool->wq); |
| 2755 | |
| 2756 | if (pool->next_mapping) |
| 2757 | mempool_free(pool->next_mapping, pool->mapping_pool); |
| 2758 | mempool_destroy(pool->mapping_pool); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2759 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 2760 | dm_deferred_set_destroy(pool->all_io_ds); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2761 | kfree(pool); |
| 2762 | } |
| 2763 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2764 | static struct kmem_cache *_new_mapping_cache; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2765 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2766 | static struct pool *pool_create(struct mapped_device *pool_md, |
| 2767 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2768 | unsigned long block_size, |
| 2769 | int read_only, char **error) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2770 | { |
| 2771 | int r; |
| 2772 | void *err_p; |
| 2773 | struct pool *pool; |
| 2774 | struct dm_pool_metadata *pmd; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2775 | bool format_device = read_only ? false : true; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2776 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2777 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2778 | if (IS_ERR(pmd)) { |
| 2779 | *error = "Error creating metadata object"; |
| 2780 | return (struct pool *)pmd; |
| 2781 | } |
| 2782 | |
| 2783 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); |
| 2784 | if (!pool) { |
| 2785 | *error = "Error allocating memory for pool"; |
| 2786 | err_p = ERR_PTR(-ENOMEM); |
| 2787 | goto bad_pool; |
| 2788 | } |
| 2789 | |
| 2790 | pool->pmd = pmd; |
| 2791 | pool->sectors_per_block = block_size; |
Mikulas Patocka | f9a8e0c | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 2792 | if (block_size & (block_size - 1)) |
| 2793 | pool->sectors_per_block_shift = -1; |
| 2794 | else |
| 2795 | pool->sectors_per_block_shift = __ffs(block_size); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2796 | pool->low_water_blocks = 0; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2797 | pool_features_init(&pool->pf); |
Joe Thornber | a195db2 | 2014-10-06 16:30:06 -0400 | [diff] [blame] | 2798 | pool->prison = dm_bio_prison_create(); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2799 | if (!pool->prison) { |
| 2800 | *error = "Error creating pool's bio prison"; |
| 2801 | err_p = ERR_PTR(-ENOMEM); |
| 2802 | goto bad_prison; |
| 2803 | } |
| 2804 | |
Mikulas Patocka | df5d2e9 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 2805 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2806 | if (IS_ERR(pool->copier)) { |
| 2807 | r = PTR_ERR(pool->copier); |
| 2808 | *error = "Error creating pool's kcopyd client"; |
| 2809 | err_p = ERR_PTR(r); |
| 2810 | goto bad_kcopyd_client; |
| 2811 | } |
| 2812 | |
| 2813 | /* |
| 2814 | * Create singlethreaded workqueue that will service all devices |
| 2815 | * that use this metadata. |
| 2816 | */ |
| 2817 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); |
| 2818 | if (!pool->wq) { |
| 2819 | *error = "Error creating pool's workqueue"; |
| 2820 | err_p = ERR_PTR(-ENOMEM); |
| 2821 | goto bad_wq; |
| 2822 | } |
| 2823 | |
Joe Thornber | 7d327fe | 2014-10-06 15:45:59 +0100 | [diff] [blame] | 2824 | throttle_init(&pool->throttle); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2825 | INIT_WORK(&pool->worker, do_worker); |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2826 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
Joe Thornber | 85ad643b | 2014-05-09 15:59:38 +0100 | [diff] [blame] | 2827 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2828 | spin_lock_init(&pool->lock); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2829 | bio_list_init(&pool->deferred_flush_bios); |
| 2830 | INIT_LIST_HEAD(&pool->prepared_mappings); |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 2831 | INIT_LIST_HEAD(&pool->prepared_discards); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 2832 | INIT_LIST_HEAD(&pool->active_thins); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 2833 | pool->low_water_triggered = false; |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 2834 | pool->suspended = true; |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2835 | |
| 2836 | pool->shared_read_ds = dm_deferred_set_create(); |
| 2837 | if (!pool->shared_read_ds) { |
| 2838 | *error = "Error creating pool's shared read deferred set"; |
| 2839 | err_p = ERR_PTR(-ENOMEM); |
| 2840 | goto bad_shared_read_ds; |
| 2841 | } |
| 2842 | |
| 2843 | pool->all_io_ds = dm_deferred_set_create(); |
| 2844 | if (!pool->all_io_ds) { |
| 2845 | *error = "Error creating pool's all io deferred set"; |
| 2846 | err_p = ERR_PTR(-ENOMEM); |
| 2847 | goto bad_all_io_ds; |
| 2848 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2849 | |
| 2850 | pool->next_mapping = NULL; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 2851 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
| 2852 | _new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2853 | if (!pool->mapping_pool) { |
| 2854 | *error = "Error creating pool's mapping mempool"; |
| 2855 | err_p = ERR_PTR(-ENOMEM); |
| 2856 | goto bad_mapping_pool; |
| 2857 | } |
| 2858 | |
Joe Thornber | a822c83 | 2015-07-03 10:22:42 +0100 | [diff] [blame] | 2859 | pool->cell_sort_array = vmalloc(sizeof(*pool->cell_sort_array) * CELL_SORT_ARRAY_SIZE); |
| 2860 | if (!pool->cell_sort_array) { |
| 2861 | *error = "Error allocating cell sort array"; |
| 2862 | err_p = ERR_PTR(-ENOMEM); |
| 2863 | goto bad_sort_array; |
| 2864 | } |
| 2865 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2866 | pool->ref_count = 1; |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 2867 | pool->last_commit_jiffies = jiffies; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2868 | pool->pool_md = pool_md; |
| 2869 | pool->md_dev = metadata_dev; |
| 2870 | __pool_table_insert(pool); |
| 2871 | |
| 2872 | return pool; |
| 2873 | |
Joe Thornber | a822c83 | 2015-07-03 10:22:42 +0100 | [diff] [blame] | 2874 | bad_sort_array: |
| 2875 | mempool_destroy(pool->mapping_pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2876 | bad_mapping_pool: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2877 | dm_deferred_set_destroy(pool->all_io_ds); |
| 2878 | bad_all_io_ds: |
| 2879 | dm_deferred_set_destroy(pool->shared_read_ds); |
| 2880 | bad_shared_read_ds: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2881 | destroy_workqueue(pool->wq); |
| 2882 | bad_wq: |
| 2883 | dm_kcopyd_client_destroy(pool->copier); |
| 2884 | bad_kcopyd_client: |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 2885 | dm_bio_prison_destroy(pool->prison); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2886 | bad_prison: |
| 2887 | kfree(pool); |
| 2888 | bad_pool: |
| 2889 | if (dm_pool_metadata_close(pmd)) |
| 2890 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); |
| 2891 | |
| 2892 | return err_p; |
| 2893 | } |
| 2894 | |
| 2895 | static void __pool_inc(struct pool *pool) |
| 2896 | { |
| 2897 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2898 | pool->ref_count++; |
| 2899 | } |
| 2900 | |
| 2901 | static void __pool_dec(struct pool *pool) |
| 2902 | { |
| 2903 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); |
| 2904 | BUG_ON(!pool->ref_count); |
| 2905 | if (!--pool->ref_count) |
| 2906 | __pool_destroy(pool); |
| 2907 | } |
| 2908 | |
| 2909 | static struct pool *__pool_find(struct mapped_device *pool_md, |
| 2910 | struct block_device *metadata_dev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2911 | unsigned long block_size, int read_only, |
| 2912 | char **error, int *created) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2913 | { |
| 2914 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); |
| 2915 | |
| 2916 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2917 | if (pool->pool_md != pool_md) { |
| 2918 | *error = "metadata device already in use by a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2919 | return ERR_PTR(-EBUSY); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2920 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2921 | __pool_inc(pool); |
| 2922 | |
| 2923 | } else { |
| 2924 | pool = __pool_table_lookup(pool_md); |
| 2925 | if (pool) { |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2926 | if (pool->md_dev != metadata_dev) { |
| 2927 | *error = "different pool cannot replace a pool"; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2928 | return ERR_PTR(-EINVAL); |
Mike Snitzer | f09996c | 2012-07-27 15:07:59 +0100 | [diff] [blame] | 2929 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2930 | __pool_inc(pool); |
| 2931 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2932 | } else { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2933 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 2934 | *created = 1; |
| 2935 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2936 | } |
| 2937 | |
| 2938 | return pool; |
| 2939 | } |
| 2940 | |
| 2941 | /*---------------------------------------------------------------- |
| 2942 | * Pool target methods |
| 2943 | *--------------------------------------------------------------*/ |
| 2944 | static void pool_dtr(struct dm_target *ti) |
| 2945 | { |
| 2946 | struct pool_c *pt = ti->private; |
| 2947 | |
| 2948 | mutex_lock(&dm_thin_pool_table.mutex); |
| 2949 | |
| 2950 | unbind_control_target(pt->pool, ti); |
| 2951 | __pool_dec(pt->pool); |
| 2952 | dm_put_device(ti, pt->metadata_dev); |
| 2953 | dm_put_device(ti, pt->data_dev); |
| 2954 | kfree(pt); |
| 2955 | |
| 2956 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 2957 | } |
| 2958 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2959 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
| 2960 | struct dm_target *ti) |
| 2961 | { |
| 2962 | int r; |
| 2963 | unsigned argc; |
| 2964 | const char *arg_name; |
| 2965 | |
| 2966 | static struct dm_arg _args[] = { |
Mike Snitzer | 74aa45c | 2014-01-15 19:07:58 -0500 | [diff] [blame] | 2967 | {0, 4, "Invalid number of pool feature arguments"}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2968 | }; |
| 2969 | |
| 2970 | /* |
| 2971 | * No feature arguments supplied. |
| 2972 | */ |
| 2973 | if (!as->argc) |
| 2974 | return 0; |
| 2975 | |
| 2976 | r = dm_read_arg_group(_args, as, &argc, &ti->error); |
| 2977 | if (r) |
| 2978 | return -EINVAL; |
| 2979 | |
| 2980 | while (argc && !r) { |
| 2981 | arg_name = dm_shift_arg(as); |
| 2982 | argc--; |
| 2983 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2984 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2985 | pf->zero_new_blocks = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 2986 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2987 | else if (!strcasecmp(arg_name, "ignore_discard")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2988 | pf->discard_enabled = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2989 | |
| 2990 | else if (!strcasecmp(arg_name, "no_discard_passdown")) |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 2991 | pf->discard_passdown = false; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2992 | |
| 2993 | else if (!strcasecmp(arg_name, "read_only")) |
| 2994 | pf->mode = PM_READ_ONLY; |
| 2995 | |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 2996 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
| 2997 | pf->error_if_no_space = true; |
| 2998 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 2999 | else { |
| 3000 | ti->error = "Unrecognised pool feature requested"; |
| 3001 | r = -EINVAL; |
| 3002 | break; |
| 3003 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3004 | } |
| 3005 | |
| 3006 | return r; |
| 3007 | } |
| 3008 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 3009 | static void metadata_low_callback(void *context) |
| 3010 | { |
| 3011 | struct pool *pool = context; |
| 3012 | |
| 3013 | DMWARN("%s: reached low water mark for metadata device: sending event.", |
| 3014 | dm_device_name(pool->pool_md)); |
| 3015 | |
| 3016 | dm_table_event(pool->ti->table); |
| 3017 | } |
| 3018 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3019 | static sector_t get_dev_size(struct block_device *bdev) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3020 | { |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3021 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; |
| 3022 | } |
| 3023 | |
| 3024 | static void warn_if_metadata_device_too_big(struct block_device *bdev) |
| 3025 | { |
| 3026 | sector_t metadata_dev_size = get_dev_size(bdev); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3027 | char buffer[BDEVNAME_SIZE]; |
| 3028 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3029 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3030 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
| 3031 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3032 | } |
| 3033 | |
| 3034 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
| 3035 | { |
| 3036 | sector_t metadata_dev_size = get_dev_size(bdev); |
| 3037 | |
| 3038 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) |
| 3039 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3040 | |
| 3041 | return metadata_dev_size; |
| 3042 | } |
| 3043 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3044 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
| 3045 | { |
| 3046 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); |
| 3047 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3048 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3049 | |
| 3050 | return metadata_dev_size; |
| 3051 | } |
| 3052 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3053 | /* |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 3054 | * When a metadata threshold is crossed a dm event is triggered, and |
| 3055 | * userland should respond by growing the metadata device. We could let |
| 3056 | * userland set the threshold, like we do with the data threshold, but I'm |
| 3057 | * not sure they know enough to do this well. |
| 3058 | */ |
| 3059 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) |
| 3060 | { |
| 3061 | /* |
| 3062 | * 4M is ample for all ops with the possible exception of thin |
| 3063 | * device deletion which is harmless if it fails (just retry the |
| 3064 | * delete after you've grown the device). |
| 3065 | */ |
| 3066 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; |
| 3067 | return min((dm_block_t)1024ULL /* 4M */, quarter); |
| 3068 | } |
| 3069 | |
| 3070 | /* |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3071 | * thin-pool <metadata dev> <data dev> |
| 3072 | * <data block size (sectors)> |
| 3073 | * <low water mark (blocks)> |
| 3074 | * [<#feature args> [<arg>]*] |
| 3075 | * |
| 3076 | * Optional feature arguments are: |
| 3077 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3078 | * ignore_discard: disable discard |
| 3079 | * no_discard_passdown: don't pass discards down to the data device |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3080 | * read_only: Don't allow any changes to be made to the pool metadata. |
| 3081 | * error_if_no_space: error IOs, instead of queueing, if no space. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3082 | */ |
| 3083 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 3084 | { |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3085 | int r, pool_created = 0; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3086 | struct pool_c *pt; |
| 3087 | struct pool *pool; |
| 3088 | struct pool_features pf; |
| 3089 | struct dm_arg_set as; |
| 3090 | struct dm_dev *data_dev; |
| 3091 | unsigned long block_size; |
| 3092 | dm_block_t low_water_blocks; |
| 3093 | struct dm_dev *metadata_dev; |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3094 | fmode_t metadata_mode; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3095 | |
| 3096 | /* |
| 3097 | * FIXME Remove validation from scope of lock. |
| 3098 | */ |
| 3099 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3100 | |
| 3101 | if (argc < 4) { |
| 3102 | ti->error = "Invalid argument count"; |
| 3103 | r = -EINVAL; |
| 3104 | goto out_unlock; |
| 3105 | } |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3106 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3107 | as.argc = argc; |
| 3108 | as.argv = argv; |
| 3109 | |
Joe Thornber | 5d0db96 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3110 | /* |
| 3111 | * Set default pool features. |
| 3112 | */ |
| 3113 | pool_features_init(&pf); |
| 3114 | |
| 3115 | dm_consume_args(&as, 4); |
| 3116 | r = parse_pool_features(&as, &pf, ti); |
| 3117 | if (r) |
| 3118 | goto out_unlock; |
| 3119 | |
| 3120 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); |
| 3121 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3122 | if (r) { |
| 3123 | ti->error = "Error opening metadata block device"; |
| 3124 | goto out_unlock; |
| 3125 | } |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3126 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3127 | |
| 3128 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); |
| 3129 | if (r) { |
| 3130 | ti->error = "Error getting data device"; |
| 3131 | goto out_metadata; |
| 3132 | } |
| 3133 | |
| 3134 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || |
| 3135 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || |
| 3136 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 3137 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3138 | ti->error = "Invalid block size"; |
| 3139 | r = -EINVAL; |
| 3140 | goto out; |
| 3141 | } |
| 3142 | |
| 3143 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { |
| 3144 | ti->error = "Invalid low water mark"; |
| 3145 | r = -EINVAL; |
| 3146 | goto out; |
| 3147 | } |
| 3148 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3149 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
| 3150 | if (!pt) { |
| 3151 | r = -ENOMEM; |
| 3152 | goto out; |
| 3153 | } |
| 3154 | |
| 3155 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3156 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3157 | if (IS_ERR(pool)) { |
| 3158 | r = PTR_ERR(pool); |
| 3159 | goto out_free_pt; |
| 3160 | } |
| 3161 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3162 | /* |
| 3163 | * 'pool_created' reflects whether this is the first table load. |
| 3164 | * Top level discard support is not allowed to be changed after |
| 3165 | * initial load. This would require a pool reload to trigger thin |
| 3166 | * device changes. |
| 3167 | */ |
| 3168 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { |
| 3169 | ti->error = "Discard support cannot be disabled once enabled"; |
| 3170 | r = -EINVAL; |
| 3171 | goto out_flags_changed; |
| 3172 | } |
| 3173 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3174 | pt->pool = pool; |
| 3175 | pt->ti = ti; |
| 3176 | pt->metadata_dev = metadata_dev; |
| 3177 | pt->data_dev = data_dev; |
| 3178 | pt->low_water_blocks = low_water_blocks; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3179 | pt->adjusted_pf = pt->requested_pf = pf; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3180 | ti->num_flush_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 3181 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3182 | /* |
| 3183 | * Only need to enable discards if the pool should pass |
| 3184 | * them down to the data device. The thin device's discard |
| 3185 | * processing will cause mappings to be removed from the btree. |
| 3186 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3187 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3188 | if (pf.discard_enabled && pf.discard_passdown) { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 3189 | ti->num_discard_bios = 1; |
Mike Snitzer | 9bc142d | 2012-09-26 23:45:46 +0100 | [diff] [blame] | 3190 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3191 | /* |
| 3192 | * Setting 'discards_supported' circumvents the normal |
| 3193 | * stacking of discard limits (this keeps the pool and |
| 3194 | * thin devices' discard limits consistent). |
| 3195 | */ |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 3196 | ti->discards_supported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3197 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3198 | ti->private = pt; |
| 3199 | |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 3200 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
| 3201 | calc_metadata_threshold(pt), |
| 3202 | metadata_low_callback, |
| 3203 | pool); |
| 3204 | if (r) |
Mike Snitzer | ba30670 | 2015-10-13 12:04:28 -0400 | [diff] [blame] | 3205 | goto out_flags_changed; |
Joe Thornber | ac8c3f3 | 2013-05-10 14:37:21 +0100 | [diff] [blame] | 3206 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3207 | pt->callbacks.congested_fn = pool_is_congested; |
| 3208 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); |
| 3209 | |
| 3210 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3211 | |
| 3212 | return 0; |
| 3213 | |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 3214 | out_flags_changed: |
| 3215 | __pool_dec(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3216 | out_free_pt: |
| 3217 | kfree(pt); |
| 3218 | out: |
| 3219 | dm_put_device(ti, data_dev); |
| 3220 | out_metadata: |
| 3221 | dm_put_device(ti, metadata_dev); |
| 3222 | out_unlock: |
| 3223 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3224 | |
| 3225 | return r; |
| 3226 | } |
| 3227 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 3228 | static int pool_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3229 | { |
| 3230 | int r; |
| 3231 | struct pool_c *pt = ti->private; |
| 3232 | struct pool *pool = pt->pool; |
| 3233 | unsigned long flags; |
| 3234 | |
| 3235 | /* |
| 3236 | * As this is a singleton target, ti->begin is always zero. |
| 3237 | */ |
| 3238 | spin_lock_irqsave(&pool->lock, flags); |
| 3239 | bio->bi_bdev = pt->data_dev->bdev; |
| 3240 | r = DM_MAPIO_REMAPPED; |
| 3241 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3242 | |
| 3243 | return r; |
| 3244 | } |
| 3245 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3246 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
| 3247 | { |
| 3248 | int r; |
| 3249 | struct pool_c *pt = ti->private; |
| 3250 | struct pool *pool = pt->pool; |
| 3251 | sector_t data_size = ti->len; |
| 3252 | dm_block_t sb_data_size; |
| 3253 | |
| 3254 | *need_commit = false; |
| 3255 | |
| 3256 | (void) sector_div(data_size, pool->sectors_per_block); |
| 3257 | |
| 3258 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
| 3259 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3260 | DMERR("%s: failed to retrieve data device size", |
| 3261 | dm_device_name(pool->pool_md)); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3262 | return r; |
| 3263 | } |
| 3264 | |
| 3265 | if (data_size < sb_data_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3266 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
| 3267 | dm_device_name(pool->pool_md), |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3268 | (unsigned long long)data_size, sb_data_size); |
| 3269 | return -EINVAL; |
| 3270 | |
| 3271 | } else if (data_size > sb_data_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 3272 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 3273 | DMERR("%s: unable to grow the data device until repaired.", |
| 3274 | dm_device_name(pool->pool_md)); |
| 3275 | return 0; |
| 3276 | } |
| 3277 | |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 3278 | if (sb_data_size) |
| 3279 | DMINFO("%s: growing the data device from %llu to %llu blocks", |
| 3280 | dm_device_name(pool->pool_md), |
| 3281 | sb_data_size, (unsigned long long)data_size); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3282 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
| 3283 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 3284 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3285 | return r; |
| 3286 | } |
| 3287 | |
| 3288 | *need_commit = true; |
| 3289 | } |
| 3290 | |
| 3291 | return 0; |
| 3292 | } |
| 3293 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3294 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
| 3295 | { |
| 3296 | int r; |
| 3297 | struct pool_c *pt = ti->private; |
| 3298 | struct pool *pool = pt->pool; |
| 3299 | dm_block_t metadata_dev_size, sb_metadata_dev_size; |
| 3300 | |
| 3301 | *need_commit = false; |
| 3302 | |
Alasdair G Kergon | 610bba8 | 2013-05-19 18:57:50 +0100 | [diff] [blame] | 3303 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3304 | |
| 3305 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); |
| 3306 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3307 | DMERR("%s: failed to retrieve metadata device size", |
| 3308 | dm_device_name(pool->pool_md)); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3309 | return r; |
| 3310 | } |
| 3311 | |
| 3312 | if (metadata_dev_size < sb_metadata_dev_size) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3313 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
| 3314 | dm_device_name(pool->pool_md), |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3315 | metadata_dev_size, sb_metadata_dev_size); |
| 3316 | return -EINVAL; |
| 3317 | |
| 3318 | } else if (metadata_dev_size > sb_metadata_dev_size) { |
Mike Snitzer | 07f2b6e | 2014-02-14 11:58:41 -0500 | [diff] [blame] | 3319 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
| 3320 | DMERR("%s: unable to grow the metadata device until repaired.", |
| 3321 | dm_device_name(pool->pool_md)); |
| 3322 | return 0; |
| 3323 | } |
| 3324 | |
Mike Snitzer | 7d48935 | 2014-02-12 23:58:15 -0500 | [diff] [blame] | 3325 | warn_if_metadata_device_too_big(pool->md_dev); |
Mike Snitzer | 6f7f51d | 2013-12-04 10:25:53 -0500 | [diff] [blame] | 3326 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
| 3327 | dm_device_name(pool->pool_md), |
| 3328 | sb_metadata_dev_size, metadata_dev_size); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3329 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
| 3330 | if (r) { |
Joe Thornber | b533065 | 2013-12-04 19:51:33 -0500 | [diff] [blame] | 3331 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3332 | return r; |
| 3333 | } |
| 3334 | |
| 3335 | *need_commit = true; |
| 3336 | } |
| 3337 | |
| 3338 | return 0; |
| 3339 | } |
| 3340 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3341 | /* |
| 3342 | * Retrieves the number of blocks of the data device from |
| 3343 | * the superblock and compares it to the actual device size, |
| 3344 | * thus resizing the data device in case it has grown. |
| 3345 | * |
| 3346 | * This both copes with opening preallocated data devices in the ctr |
| 3347 | * being followed by a resume |
| 3348 | * -and- |
| 3349 | * calling the resume method individually after userspace has |
| 3350 | * grown the data device in reaction to a table event. |
| 3351 | */ |
| 3352 | static int pool_preresume(struct dm_target *ti) |
| 3353 | { |
| 3354 | int r; |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3355 | bool need_commit1, need_commit2; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3356 | struct pool_c *pt = ti->private; |
| 3357 | struct pool *pool = pt->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3358 | |
| 3359 | /* |
| 3360 | * Take control of the pool object. |
| 3361 | */ |
| 3362 | r = bind_control_target(pool, ti); |
| 3363 | if (r) |
| 3364 | return r; |
| 3365 | |
Joe Thornber | b17446d | 2013-05-10 14:37:18 +0100 | [diff] [blame] | 3366 | r = maybe_resize_data_dev(ti, &need_commit1); |
| 3367 | if (r) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3368 | return r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3369 | |
Joe Thornber | 24347e9 | 2013-05-10 14:37:19 +0100 | [diff] [blame] | 3370 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
| 3371 | if (r) |
| 3372 | return r; |
| 3373 | |
| 3374 | if (need_commit1 || need_commit2) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 3375 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3376 | |
| 3377 | return 0; |
| 3378 | } |
| 3379 | |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 3380 | static void pool_suspend_active_thins(struct pool *pool) |
| 3381 | { |
| 3382 | struct thin_c *tc; |
| 3383 | |
| 3384 | /* Suspend all active thin devices */ |
| 3385 | tc = get_first_thin(pool); |
| 3386 | while (tc) { |
| 3387 | dm_internal_suspend_noflush(tc->thin_md); |
| 3388 | tc = get_next_thin(pool, tc); |
| 3389 | } |
| 3390 | } |
| 3391 | |
| 3392 | static void pool_resume_active_thins(struct pool *pool) |
| 3393 | { |
| 3394 | struct thin_c *tc; |
| 3395 | |
| 3396 | /* Resume all active thin devices */ |
| 3397 | tc = get_first_thin(pool); |
| 3398 | while (tc) { |
| 3399 | dm_internal_resume(tc->thin_md); |
| 3400 | tc = get_next_thin(pool, tc); |
| 3401 | } |
| 3402 | } |
| 3403 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3404 | static void pool_resume(struct dm_target *ti) |
| 3405 | { |
| 3406 | struct pool_c *pt = ti->private; |
| 3407 | struct pool *pool = pt->pool; |
| 3408 | unsigned long flags; |
| 3409 | |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 3410 | /* |
| 3411 | * Must requeue active_thins' bios and then resume |
| 3412 | * active_thins _before_ clearing 'suspend' flag. |
| 3413 | */ |
| 3414 | requeue_bios(pool); |
| 3415 | pool_resume_active_thins(pool); |
| 3416 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3417 | spin_lock_irqsave(&pool->lock, flags); |
Joe Thornber | 88a6621 | 2013-12-04 20:16:12 -0500 | [diff] [blame] | 3418 | pool->low_water_triggered = false; |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3419 | pool->suspended = false; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3420 | spin_unlock_irqrestore(&pool->lock, flags); |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3421 | |
Joe Thornber | 905e51b | 2012-03-28 18:41:27 +0100 | [diff] [blame] | 3422 | do_waker(&pool->waker.work); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3423 | } |
| 3424 | |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3425 | static void pool_presuspend(struct dm_target *ti) |
| 3426 | { |
| 3427 | struct pool_c *pt = ti->private; |
| 3428 | struct pool *pool = pt->pool; |
| 3429 | unsigned long flags; |
| 3430 | |
| 3431 | spin_lock_irqsave(&pool->lock, flags); |
| 3432 | pool->suspended = true; |
| 3433 | spin_unlock_irqrestore(&pool->lock, flags); |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 3434 | |
| 3435 | pool_suspend_active_thins(pool); |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3436 | } |
| 3437 | |
| 3438 | static void pool_presuspend_undo(struct dm_target *ti) |
| 3439 | { |
| 3440 | struct pool_c *pt = ti->private; |
| 3441 | struct pool *pool = pt->pool; |
| 3442 | unsigned long flags; |
| 3443 | |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 3444 | pool_resume_active_thins(pool); |
| 3445 | |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3446 | spin_lock_irqsave(&pool->lock, flags); |
| 3447 | pool->suspended = false; |
| 3448 | spin_unlock_irqrestore(&pool->lock, flags); |
| 3449 | } |
| 3450 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3451 | static void pool_postsuspend(struct dm_target *ti) |
| 3452 | { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3453 | struct pool_c *pt = ti->private; |
| 3454 | struct pool *pool = pt->pool; |
| 3455 | |
Nikolay Borisov | 18d03e8 | 2015-12-17 18:03:35 +0200 | [diff] [blame] | 3456 | cancel_delayed_work_sync(&pool->waker); |
| 3457 | cancel_delayed_work_sync(&pool->no_space_timeout); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3458 | flush_workqueue(pool->wq); |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 3459 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3460 | } |
| 3461 | |
| 3462 | static int check_arg_count(unsigned argc, unsigned args_required) |
| 3463 | { |
| 3464 | if (argc != args_required) { |
| 3465 | DMWARN("Message received with %u arguments instead of %u.", |
| 3466 | argc, args_required); |
| 3467 | return -EINVAL; |
| 3468 | } |
| 3469 | |
| 3470 | return 0; |
| 3471 | } |
| 3472 | |
| 3473 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) |
| 3474 | { |
| 3475 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && |
| 3476 | *dev_id <= MAX_DEV_ID) |
| 3477 | return 0; |
| 3478 | |
| 3479 | if (warning) |
| 3480 | DMWARN("Message received with invalid device id: %s", arg); |
| 3481 | |
| 3482 | return -EINVAL; |
| 3483 | } |
| 3484 | |
| 3485 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3486 | { |
| 3487 | dm_thin_id dev_id; |
| 3488 | int r; |
| 3489 | |
| 3490 | r = check_arg_count(argc, 2); |
| 3491 | if (r) |
| 3492 | return r; |
| 3493 | |
| 3494 | r = read_dev_id(argv[1], &dev_id, 1); |
| 3495 | if (r) |
| 3496 | return r; |
| 3497 | |
| 3498 | r = dm_pool_create_thin(pool->pmd, dev_id); |
| 3499 | if (r) { |
| 3500 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", |
| 3501 | argv[1]); |
| 3502 | return r; |
| 3503 | } |
| 3504 | |
| 3505 | return 0; |
| 3506 | } |
| 3507 | |
| 3508 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3509 | { |
| 3510 | dm_thin_id dev_id; |
| 3511 | dm_thin_id origin_dev_id; |
| 3512 | int r; |
| 3513 | |
| 3514 | r = check_arg_count(argc, 3); |
| 3515 | if (r) |
| 3516 | return r; |
| 3517 | |
| 3518 | r = read_dev_id(argv[1], &dev_id, 1); |
| 3519 | if (r) |
| 3520 | return r; |
| 3521 | |
| 3522 | r = read_dev_id(argv[2], &origin_dev_id, 1); |
| 3523 | if (r) |
| 3524 | return r; |
| 3525 | |
| 3526 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); |
| 3527 | if (r) { |
| 3528 | DMWARN("Creation of new snapshot %s of device %s failed.", |
| 3529 | argv[1], argv[2]); |
| 3530 | return r; |
| 3531 | } |
| 3532 | |
| 3533 | return 0; |
| 3534 | } |
| 3535 | |
| 3536 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3537 | { |
| 3538 | dm_thin_id dev_id; |
| 3539 | int r; |
| 3540 | |
| 3541 | r = check_arg_count(argc, 2); |
| 3542 | if (r) |
| 3543 | return r; |
| 3544 | |
| 3545 | r = read_dev_id(argv[1], &dev_id, 1); |
| 3546 | if (r) |
| 3547 | return r; |
| 3548 | |
| 3549 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); |
| 3550 | if (r) |
| 3551 | DMWARN("Deletion of thin device %s failed.", argv[1]); |
| 3552 | |
| 3553 | return r; |
| 3554 | } |
| 3555 | |
| 3556 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3557 | { |
| 3558 | dm_thin_id old_id, new_id; |
| 3559 | int r; |
| 3560 | |
| 3561 | r = check_arg_count(argc, 3); |
| 3562 | if (r) |
| 3563 | return r; |
| 3564 | |
| 3565 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { |
| 3566 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); |
| 3567 | return -EINVAL; |
| 3568 | } |
| 3569 | |
| 3570 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { |
| 3571 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); |
| 3572 | return -EINVAL; |
| 3573 | } |
| 3574 | |
| 3575 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); |
| 3576 | if (r) { |
| 3577 | DMWARN("Failed to change transaction id from %s to %s.", |
| 3578 | argv[1], argv[2]); |
| 3579 | return r; |
| 3580 | } |
| 3581 | |
| 3582 | return 0; |
| 3583 | } |
| 3584 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3585 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3586 | { |
| 3587 | int r; |
| 3588 | |
| 3589 | r = check_arg_count(argc, 1); |
| 3590 | if (r) |
| 3591 | return r; |
| 3592 | |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 3593 | (void) commit(pool); |
Joe Thornber | 0d200ae | 2012-07-03 12:55:31 +0100 | [diff] [blame] | 3594 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3595 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
| 3596 | if (r) |
| 3597 | DMWARN("reserve_metadata_snap message failed."); |
| 3598 | |
| 3599 | return r; |
| 3600 | } |
| 3601 | |
| 3602 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
| 3603 | { |
| 3604 | int r; |
| 3605 | |
| 3606 | r = check_arg_count(argc, 1); |
| 3607 | if (r) |
| 3608 | return r; |
| 3609 | |
| 3610 | r = dm_pool_release_metadata_snap(pool->pmd); |
| 3611 | if (r) |
| 3612 | DMWARN("release_metadata_snap message failed."); |
| 3613 | |
| 3614 | return r; |
| 3615 | } |
| 3616 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3617 | /* |
| 3618 | * Messages supported: |
| 3619 | * create_thin <dev_id> |
| 3620 | * create_snap <dev_id> <origin_id> |
| 3621 | * delete <dev_id> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3622 | * set_transaction_id <current_trans_id> <new_trans_id> |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3623 | * reserve_metadata_snap |
| 3624 | * release_metadata_snap |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3625 | */ |
| 3626 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) |
| 3627 | { |
| 3628 | int r = -EINVAL; |
| 3629 | struct pool_c *pt = ti->private; |
| 3630 | struct pool *pool = pt->pool; |
| 3631 | |
Joe Thornber | 2a7eaea | 2015-01-26 11:38:21 +0000 | [diff] [blame] | 3632 | if (get_pool_mode(pool) >= PM_READ_ONLY) { |
| 3633 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", |
| 3634 | dm_device_name(pool->pool_md)); |
Mike Snitzer | fd46769 | 2015-06-09 12:31:26 -0400 | [diff] [blame] | 3635 | return -EOPNOTSUPP; |
Joe Thornber | 2a7eaea | 2015-01-26 11:38:21 +0000 | [diff] [blame] | 3636 | } |
| 3637 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3638 | if (!strcasecmp(argv[0], "create_thin")) |
| 3639 | r = process_create_thin_mesg(argc, argv, pool); |
| 3640 | |
| 3641 | else if (!strcasecmp(argv[0], "create_snap")) |
| 3642 | r = process_create_snap_mesg(argc, argv, pool); |
| 3643 | |
| 3644 | else if (!strcasecmp(argv[0], "delete")) |
| 3645 | r = process_delete_mesg(argc, argv, pool); |
| 3646 | |
| 3647 | else if (!strcasecmp(argv[0], "set_transaction_id")) |
| 3648 | r = process_set_transaction_id_mesg(argc, argv, pool); |
| 3649 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3650 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
| 3651 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); |
| 3652 | |
| 3653 | else if (!strcasecmp(argv[0], "release_metadata_snap")) |
| 3654 | r = process_release_metadata_snap_mesg(argc, argv, pool); |
| 3655 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3656 | else |
| 3657 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); |
| 3658 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3659 | if (!r) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 3660 | (void) commit(pool); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3661 | |
| 3662 | return r; |
| 3663 | } |
| 3664 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3665 | static void emit_flags(struct pool_features *pf, char *result, |
| 3666 | unsigned sz, unsigned maxlen) |
| 3667 | { |
| 3668 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3669 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
| 3670 | pf->error_if_no_space; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3671 | DMEMIT("%u ", count); |
| 3672 | |
| 3673 | if (!pf->zero_new_blocks) |
| 3674 | DMEMIT("skip_block_zeroing "); |
| 3675 | |
| 3676 | if (!pf->discard_enabled) |
| 3677 | DMEMIT("ignore_discard "); |
| 3678 | |
| 3679 | if (!pf->discard_passdown) |
| 3680 | DMEMIT("no_discard_passdown "); |
| 3681 | |
| 3682 | if (pf->mode == PM_READ_ONLY) |
| 3683 | DMEMIT("read_only "); |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3684 | |
| 3685 | if (pf->error_if_no_space) |
| 3686 | DMEMIT("error_if_no_space "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3687 | } |
| 3688 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3689 | /* |
| 3690 | * Status line is: |
| 3691 | * <transaction id> <used metadata sectors>/<total metadata sectors> |
| 3692 | * <used data sectors>/<total data sectors> <held metadata root> |
Mike Snitzer | e4c78e2 | 2015-07-15 11:40:24 -0400 | [diff] [blame] | 3693 | * <pool mode> <discard config> <no space config> <needs_check> |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3694 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3695 | static void pool_status(struct dm_target *ti, status_type_t type, |
| 3696 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3697 | { |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3698 | int r; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3699 | unsigned sz = 0; |
| 3700 | uint64_t transaction_id; |
| 3701 | dm_block_t nr_free_blocks_data; |
| 3702 | dm_block_t nr_free_blocks_metadata; |
| 3703 | dm_block_t nr_blocks_data; |
| 3704 | dm_block_t nr_blocks_metadata; |
| 3705 | dm_block_t held_root; |
| 3706 | char buf[BDEVNAME_SIZE]; |
| 3707 | char buf2[BDEVNAME_SIZE]; |
| 3708 | struct pool_c *pt = ti->private; |
| 3709 | struct pool *pool = pt->pool; |
| 3710 | |
| 3711 | switch (type) { |
| 3712 | case STATUSTYPE_INFO: |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3713 | if (get_pool_mode(pool) == PM_FAIL) { |
| 3714 | DMEMIT("Fail"); |
| 3715 | break; |
| 3716 | } |
| 3717 | |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3718 | /* Commit to ensure statistics aren't out-of-date */ |
| 3719 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) |
Joe Thornber | 020cc3b | 2013-12-04 15:05:36 -0500 | [diff] [blame] | 3720 | (void) commit(pool); |
Alasdair G Kergon | 1f4e0ff | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3721 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3722 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
| 3723 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3724 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
| 3725 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3726 | goto err; |
| 3727 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3728 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3729 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
| 3730 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3731 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
| 3732 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3733 | goto err; |
| 3734 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3735 | |
| 3736 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3737 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3738 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
| 3739 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3740 | goto err; |
| 3741 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3742 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3743 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
| 3744 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3745 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
| 3746 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3747 | goto err; |
| 3748 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3749 | |
| 3750 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3751 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3752 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
| 3753 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3754 | goto err; |
| 3755 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3756 | |
Joe Thornber | cc8394d | 2012-06-03 00:30:01 +0100 | [diff] [blame] | 3757 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3758 | if (r) { |
Mike Snitzer | 4fa5971 | 2013-08-21 17:30:40 -0400 | [diff] [blame] | 3759 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
| 3760 | dm_device_name(pool->pool_md), r); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3761 | goto err; |
| 3762 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3763 | |
| 3764 | DMEMIT("%llu %llu/%llu %llu/%llu ", |
| 3765 | (unsigned long long)transaction_id, |
| 3766 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), |
| 3767 | (unsigned long long)nr_blocks_metadata, |
| 3768 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), |
| 3769 | (unsigned long long)nr_blocks_data); |
| 3770 | |
| 3771 | if (held_root) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3772 | DMEMIT("%llu ", held_root); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3773 | else |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3774 | DMEMIT("- "); |
| 3775 | |
Joe Thornber | 3e1a069 | 2014-03-03 16:03:26 +0000 | [diff] [blame] | 3776 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
| 3777 | DMEMIT("out_of_data_space "); |
| 3778 | else if (pool->pf.mode == PM_READ_ONLY) |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3779 | DMEMIT("ro "); |
| 3780 | else |
| 3781 | DMEMIT("rw "); |
| 3782 | |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 3783 | if (!pool->pf.discard_enabled) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3784 | DMEMIT("ignore_discard "); |
Mike Snitzer | 018debe | 2012-12-21 20:23:32 +0000 | [diff] [blame] | 3785 | else if (pool->pf.discard_passdown) |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3786 | DMEMIT("discard_passdown "); |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 3787 | else |
Mike Snitzer | 787a996c | 2013-12-06 16:21:43 -0500 | [diff] [blame] | 3788 | DMEMIT("no_discard_passdown "); |
| 3789 | |
| 3790 | if (pool->pf.error_if_no_space) |
| 3791 | DMEMIT("error_if_no_space "); |
| 3792 | else |
| 3793 | DMEMIT("queue_if_no_space "); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3794 | |
Mike Snitzer | e4c78e2 | 2015-07-15 11:40:24 -0400 | [diff] [blame] | 3795 | if (dm_pool_metadata_needs_check(pool->pmd)) |
| 3796 | DMEMIT("needs_check "); |
| 3797 | else |
| 3798 | DMEMIT("- "); |
| 3799 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3800 | break; |
| 3801 | |
| 3802 | case STATUSTYPE_TABLE: |
| 3803 | DMEMIT("%s %s %lu %llu ", |
| 3804 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), |
| 3805 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), |
| 3806 | (unsigned long)pool->sectors_per_block, |
| 3807 | (unsigned long long)pt->low_water_blocks); |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3808 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3809 | break; |
| 3810 | } |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3811 | return; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3812 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 3813 | err: |
| 3814 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3815 | } |
| 3816 | |
| 3817 | static int pool_iterate_devices(struct dm_target *ti, |
| 3818 | iterate_devices_callout_fn fn, void *data) |
| 3819 | { |
| 3820 | struct pool_c *pt = ti->private; |
| 3821 | |
| 3822 | return fn(ti, pt->data_dev, 0, ti->len, data); |
| 3823 | } |
| 3824 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3825 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 3826 | { |
| 3827 | struct pool_c *pt = ti->private; |
| 3828 | struct pool *pool = pt->pool; |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 3829 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
| 3830 | |
| 3831 | /* |
Mike Snitzer | d200c30 | 2014-11-20 18:07:43 -0500 | [diff] [blame] | 3832 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
| 3833 | * to the highest possible power-of-2 factor of pool->sectors_per_block. |
| 3834 | * This is especially beneficial when the pool's data device is a RAID |
| 3835 | * device that has a full stripe width that matches pool->sectors_per_block |
| 3836 | * -- because even though partial RAID stripe-sized IOs will be issued to a |
| 3837 | * single RAID stripe; when aggregated they will end on a full RAID stripe |
| 3838 | * boundary.. which avoids additional partial RAID stripe writes cascading |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 3839 | */ |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 3840 | if (limits->max_sectors < pool->sectors_per_block) { |
| 3841 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { |
| 3842 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) |
| 3843 | limits->max_sectors--; |
| 3844 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); |
| 3845 | } |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 3846 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3847 | |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 3848 | /* |
| 3849 | * If the system-determined stacked limits are compatible with the |
| 3850 | * pool's blocksize (io_opt is a factor) do not override them. |
| 3851 | */ |
| 3852 | if (io_opt_sectors < pool->sectors_per_block || |
Mike Snitzer | 604ea90 | 2014-10-09 18:43:25 -0400 | [diff] [blame] | 3853 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
| 3854 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) |
| 3855 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); |
| 3856 | else |
| 3857 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); |
Mike Snitzer | 0cc67cd | 2013-08-20 15:02:41 -0400 | [diff] [blame] | 3858 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
| 3859 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3860 | |
| 3861 | /* |
| 3862 | * pt->adjusted_pf is a staging area for the actual features to use. |
| 3863 | * They get transferred to the live pool in bind_control_target() |
| 3864 | * called from pool_preresume(). |
| 3865 | */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3866 | if (!pt->adjusted_pf.discard_enabled) { |
| 3867 | /* |
| 3868 | * Must explicitly disallow stacking discard limits otherwise the |
| 3869 | * block layer will stack them if pool's data device has support. |
| 3870 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the |
| 3871 | * user to see that, so make sure to set all discard limits to 0. |
| 3872 | */ |
| 3873 | limits->discard_granularity = 0; |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3874 | return; |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 3875 | } |
Mike Snitzer | 0424caa | 2012-09-26 23:45:47 +0100 | [diff] [blame] | 3876 | |
| 3877 | disable_passdown_if_not_supported(pt); |
| 3878 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 3879 | /* |
| 3880 | * The pool uses the same discard limits as the underlying data |
| 3881 | * device. DM core has already set this up. |
| 3882 | */ |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3883 | } |
| 3884 | |
| 3885 | static struct target_type pool_target = { |
| 3886 | .name = "thin-pool", |
| 3887 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | |
| 3888 | DM_TARGET_IMMUTABLE, |
Mike Snitzer | 1c2e54e | 2016-01-06 14:13:40 -0500 | [diff] [blame] | 3889 | .version = {1, 17, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3890 | .module = THIS_MODULE, |
| 3891 | .ctr = pool_ctr, |
| 3892 | .dtr = pool_dtr, |
| 3893 | .map = pool_map, |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 3894 | .presuspend = pool_presuspend, |
| 3895 | .presuspend_undo = pool_presuspend_undo, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3896 | .postsuspend = pool_postsuspend, |
| 3897 | .preresume = pool_preresume, |
| 3898 | .resume = pool_resume, |
| 3899 | .message = pool_message, |
| 3900 | .status = pool_status, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3901 | .iterate_devices = pool_iterate_devices, |
| 3902 | .io_hints = pool_io_hints, |
| 3903 | }; |
| 3904 | |
| 3905 | /*---------------------------------------------------------------- |
| 3906 | * Thin target methods |
| 3907 | *--------------------------------------------------------------*/ |
Joe Thornber | b10ebd3 | 2014-04-08 11:29:01 +0100 | [diff] [blame] | 3908 | static void thin_get(struct thin_c *tc) |
| 3909 | { |
| 3910 | atomic_inc(&tc->refcount); |
| 3911 | } |
| 3912 | |
| 3913 | static void thin_put(struct thin_c *tc) |
| 3914 | { |
| 3915 | if (atomic_dec_and_test(&tc->refcount)) |
| 3916 | complete(&tc->can_destroy); |
| 3917 | } |
| 3918 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3919 | static void thin_dtr(struct dm_target *ti) |
| 3920 | { |
| 3921 | struct thin_c *tc = ti->private; |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3922 | unsigned long flags; |
| 3923 | |
| 3924 | spin_lock_irqsave(&tc->pool->lock, flags); |
| 3925 | list_del_rcu(&tc->list); |
| 3926 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
| 3927 | synchronize_rcu(); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3928 | |
Mikulas Patocka | 17181fb | 2014-11-05 17:00:13 -0500 | [diff] [blame] | 3929 | thin_put(tc); |
| 3930 | wait_for_completion(&tc->can_destroy); |
| 3931 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3932 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3933 | |
| 3934 | __pool_dec(tc->pool); |
| 3935 | dm_pool_close_thin_device(tc->td); |
| 3936 | dm_put_device(ti, tc->pool_dev); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3937 | if (tc->origin_dev) |
| 3938 | dm_put_device(ti, tc->origin_dev); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3939 | kfree(tc); |
| 3940 | |
| 3941 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 3942 | } |
| 3943 | |
| 3944 | /* |
| 3945 | * Thin target parameters: |
| 3946 | * |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3947 | * <pool_dev> <dev_id> [origin_dev] |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3948 | * |
| 3949 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) |
| 3950 | * dev_id: the internal device identifier |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3951 | * 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] | 3952 | * |
| 3953 | * If the pool device has discards disabled, they get disabled for the thin |
| 3954 | * device as well. |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3955 | */ |
| 3956 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) |
| 3957 | { |
| 3958 | int r; |
| 3959 | struct thin_c *tc; |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3960 | struct dm_dev *pool_dev, *origin_dev; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3961 | struct mapped_device *pool_md; |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 3962 | unsigned long flags; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3963 | |
| 3964 | mutex_lock(&dm_thin_pool_table.mutex); |
| 3965 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3966 | if (argc != 2 && argc != 3) { |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3967 | ti->error = "Invalid argument count"; |
| 3968 | r = -EINVAL; |
| 3969 | goto out_unlock; |
| 3970 | } |
| 3971 | |
| 3972 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); |
| 3973 | if (!tc) { |
| 3974 | ti->error = "Out of memory"; |
| 3975 | r = -ENOMEM; |
| 3976 | goto out_unlock; |
| 3977 | } |
Mike Snitzer | 583024d | 2014-10-28 20:58:45 -0400 | [diff] [blame] | 3978 | tc->thin_md = dm_table_get_md(ti->table); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3979 | spin_lock_init(&tc->lock); |
Joe Thornber | a374bb2 | 2014-10-10 13:43:14 +0100 | [diff] [blame] | 3980 | INIT_LIST_HEAD(&tc->deferred_cells); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 3981 | bio_list_init(&tc->deferred_bio_list); |
| 3982 | bio_list_init(&tc->retry_on_resume_list); |
Mike Snitzer | 67324ea | 2014-03-21 18:33:41 -0400 | [diff] [blame] | 3983 | tc->sort_bio_list = RB_ROOT; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3984 | |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 3985 | if (argc == 3) { |
| 3986 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
| 3987 | if (r) { |
| 3988 | ti->error = "Error opening origin device"; |
| 3989 | goto bad_origin_dev; |
| 3990 | } |
| 3991 | tc->origin_dev = origin_dev; |
| 3992 | } |
| 3993 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 3994 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
| 3995 | if (r) { |
| 3996 | ti->error = "Error opening pool device"; |
| 3997 | goto bad_pool_dev; |
| 3998 | } |
| 3999 | tc->pool_dev = pool_dev; |
| 4000 | |
| 4001 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { |
| 4002 | ti->error = "Invalid device id"; |
| 4003 | r = -EINVAL; |
| 4004 | goto bad_common; |
| 4005 | } |
| 4006 | |
| 4007 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); |
| 4008 | if (!pool_md) { |
| 4009 | ti->error = "Couldn't get pool mapped device"; |
| 4010 | r = -EINVAL; |
| 4011 | goto bad_common; |
| 4012 | } |
| 4013 | |
| 4014 | tc->pool = __pool_table_lookup(pool_md); |
| 4015 | if (!tc->pool) { |
| 4016 | ti->error = "Couldn't find pool object"; |
| 4017 | r = -EINVAL; |
| 4018 | goto bad_pool_lookup; |
| 4019 | } |
| 4020 | __pool_inc(tc->pool); |
| 4021 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 4022 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 4023 | ti->error = "Couldn't open thin device, Pool is in fail mode"; |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 4024 | r = -EINVAL; |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4025 | goto bad_pool; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 4026 | } |
| 4027 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4028 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
| 4029 | if (r) { |
| 4030 | ti->error = "Couldn't open thin internal device"; |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4031 | goto bad_pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4032 | } |
| 4033 | |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 4034 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
| 4035 | if (r) |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4036 | goto bad; |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 4037 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 4038 | ti->num_flush_bios = 1; |
Joe Thornber | 16ad3d1 | 2012-07-27 15:08:07 +0100 | [diff] [blame] | 4039 | ti->flush_supported = true; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 4040 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 4041 | |
| 4042 | /* In case the pool supports discards, pass them on. */ |
Mike Snitzer | b60ab99 | 2013-09-19 18:49:11 -0400 | [diff] [blame] | 4043 | ti->discard_zeroes_data_unsupported = true; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 4044 | if (tc->pool->pf.discard_enabled) { |
Alasdair G Kergon | 0ac5548 | 2012-07-27 15:08:08 +0100 | [diff] [blame] | 4045 | ti->discards_supported = true; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 4046 | ti->num_discard_bios = 1; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 4047 | ti->split_discard_bios = false; |
Joe Thornber | 67e2e2b | 2012-03-28 18:41:29 +0100 | [diff] [blame] | 4048 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4049 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4050 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 4051 | |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 4052 | spin_lock_irqsave(&tc->pool->lock, flags); |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4053 | if (tc->pool->suspended) { |
| 4054 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
| 4055 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ |
| 4056 | ti->error = "Unable to activate thin device while pool is suspended"; |
| 4057 | r = -EINVAL; |
| 4058 | goto bad; |
| 4059 | } |
Marc Dionne | 2b94e89 | 2014-12-17 07:59:59 -0500 | [diff] [blame] | 4060 | atomic_set(&tc->refcount, 1); |
| 4061 | init_completion(&tc->can_destroy); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 4062 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
Joe Thornber | 5e3283e | 2014-04-08 11:08:41 +0100 | [diff] [blame] | 4063 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
Mike Snitzer | c140e1c | 2014-03-20 21:17:14 -0400 | [diff] [blame] | 4064 | /* |
| 4065 | * This synchronize_rcu() call is needed here otherwise we risk a |
| 4066 | * wake_worker() call finding no bios to process (because the newly |
| 4067 | * added tc isn't yet visible). So this reduces latency since we |
| 4068 | * aren't then dependent on the periodic commit to wake_worker(). |
| 4069 | */ |
| 4070 | synchronize_rcu(); |
| 4071 | |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4072 | dm_put(pool_md); |
| 4073 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4074 | return 0; |
| 4075 | |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4076 | bad: |
Mike Snitzer | 1acacc0 | 2014-02-19 20:32:33 -0500 | [diff] [blame] | 4077 | dm_pool_close_thin_device(tc->td); |
Mike Snitzer | 80e96c5 | 2014-11-07 15:09:46 -0500 | [diff] [blame] | 4078 | bad_pool: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4079 | __pool_dec(tc->pool); |
| 4080 | bad_pool_lookup: |
| 4081 | dm_put(pool_md); |
| 4082 | bad_common: |
| 4083 | dm_put_device(ti, tc->pool_dev); |
| 4084 | bad_pool_dev: |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4085 | if (tc->origin_dev) |
| 4086 | dm_put_device(ti, tc->origin_dev); |
| 4087 | bad_origin_dev: |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4088 | kfree(tc); |
| 4089 | out_unlock: |
| 4090 | mutex_unlock(&dm_thin_pool_table.mutex); |
| 4091 | |
| 4092 | return r; |
| 4093 | } |
| 4094 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 4095 | static int thin_map(struct dm_target *ti, struct bio *bio) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4096 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 4097 | 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] | 4098 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 4099 | return thin_bio_map(ti, bio); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4100 | } |
| 4101 | |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 4102 | 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] | 4103 | { |
| 4104 | unsigned long flags; |
Mikulas Patocka | 59c3d2c | 2012-12-21 20:23:40 +0000 | [diff] [blame] | 4105 | 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] | 4106 | struct list_head work; |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4107 | struct dm_thin_new_mapping *m, *tmp; |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4108 | struct pool *pool = h->tc->pool; |
| 4109 | |
| 4110 | if (h->shared_read_entry) { |
| 4111 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 4112 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4113 | |
| 4114 | spin_lock_irqsave(&pool->lock, flags); |
| 4115 | list_for_each_entry_safe(m, tmp, &work, list) { |
| 4116 | list_del(&m->list); |
Joe Thornber | 50f3c3e | 2014-06-13 13:57:09 +0100 | [diff] [blame] | 4117 | __complete_mapping_preparation(m); |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4118 | } |
| 4119 | spin_unlock_irqrestore(&pool->lock, flags); |
| 4120 | } |
| 4121 | |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4122 | if (h->all_io_entry) { |
| 4123 | INIT_LIST_HEAD(&work); |
Mike Snitzer | 44feb38 | 2012-10-12 21:02:10 +0100 | [diff] [blame] | 4124 | dm_deferred_entry_dec(h->all_io_entry, &work); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 4125 | if (!list_empty(&work)) { |
| 4126 | spin_lock_irqsave(&pool->lock, flags); |
| 4127 | list_for_each_entry_safe(m, tmp, &work, list) |
Mike Snitzer | daec338 | 2013-12-11 14:01:20 -0500 | [diff] [blame] | 4128 | list_add_tail(&m->list, &pool->prepared_discards); |
Joe Thornber | 563af18 | 2012-12-21 20:23:31 +0000 | [diff] [blame] | 4129 | spin_unlock_irqrestore(&pool->lock, flags); |
| 4130 | wake_worker(pool); |
| 4131 | } |
Joe Thornber | 104655f | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4132 | } |
| 4133 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 4134 | if (h->cell) |
| 4135 | cell_defer_no_holder(h->tc, h->cell); |
| 4136 | |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4137 | return 0; |
| 4138 | } |
| 4139 | |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 4140 | static void thin_presuspend(struct dm_target *ti) |
| 4141 | { |
| 4142 | struct thin_c *tc = ti->private; |
| 4143 | |
| 4144 | if (dm_noflush_suspending(ti)) |
| 4145 | noflush_work(tc, do_noflush_start); |
| 4146 | } |
| 4147 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4148 | static void thin_postsuspend(struct dm_target *ti) |
| 4149 | { |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 4150 | struct thin_c *tc = ti->private; |
| 4151 | |
| 4152 | /* |
| 4153 | * The dm_noflush_suspending flag has been cleared by now, so |
| 4154 | * unfortunately we must always run this. |
| 4155 | */ |
| 4156 | noflush_work(tc, do_noflush_stop); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4157 | } |
| 4158 | |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 4159 | static int thin_preresume(struct dm_target *ti) |
| 4160 | { |
| 4161 | struct thin_c *tc = ti->private; |
| 4162 | |
| 4163 | if (tc->origin_dev) |
| 4164 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); |
| 4165 | |
| 4166 | return 0; |
| 4167 | } |
| 4168 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4169 | /* |
| 4170 | * <nr mapped sectors> <highest mapped sector> |
| 4171 | */ |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 4172 | static void thin_status(struct dm_target *ti, status_type_t type, |
| 4173 | unsigned status_flags, char *result, unsigned maxlen) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4174 | { |
| 4175 | int r; |
| 4176 | ssize_t sz = 0; |
| 4177 | dm_block_t mapped, highest; |
| 4178 | char buf[BDEVNAME_SIZE]; |
| 4179 | struct thin_c *tc = ti->private; |
| 4180 | |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 4181 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
| 4182 | DMEMIT("Fail"); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 4183 | return; |
Joe Thornber | e49e582 | 2012-07-27 15:08:16 +0100 | [diff] [blame] | 4184 | } |
| 4185 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4186 | if (!tc->td) |
| 4187 | DMEMIT("-"); |
| 4188 | else { |
| 4189 | switch (type) { |
| 4190 | case STATUSTYPE_INFO: |
| 4191 | r = dm_thin_get_mapped_count(tc->td, &mapped); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 4192 | if (r) { |
| 4193 | DMERR("dm_thin_get_mapped_count returned %d", r); |
| 4194 | goto err; |
| 4195 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4196 | |
| 4197 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 4198 | if (r < 0) { |
| 4199 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); |
| 4200 | goto err; |
| 4201 | } |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4202 | |
| 4203 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); |
| 4204 | if (r) |
| 4205 | DMEMIT("%llu", ((highest + 1) * |
| 4206 | tc->pool->sectors_per_block) - 1); |
| 4207 | else |
| 4208 | DMEMIT("-"); |
| 4209 | break; |
| 4210 | |
| 4211 | case STATUSTYPE_TABLE: |
| 4212 | DMEMIT("%s %lu", |
| 4213 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), |
| 4214 | (unsigned long) tc->dev_id); |
Joe Thornber | 2dd9c25 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4215 | if (tc->origin_dev) |
| 4216 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4217 | break; |
| 4218 | } |
| 4219 | } |
| 4220 | |
Mikulas Patocka | fd7c092 | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 4221 | return; |
| 4222 | |
| 4223 | err: |
| 4224 | DMEMIT("Error"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4225 | } |
| 4226 | |
| 4227 | static int thin_iterate_devices(struct dm_target *ti, |
| 4228 | iterate_devices_callout_fn fn, void *data) |
| 4229 | { |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 4230 | sector_t blocks; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4231 | struct thin_c *tc = ti->private; |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 4232 | struct pool *pool = tc->pool; |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4233 | |
| 4234 | /* |
| 4235 | * We can't call dm_pool_get_data_dev_size() since that blocks. So |
| 4236 | * we follow a more convoluted path through to the pool's target. |
| 4237 | */ |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 4238 | if (!pool->ti) |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4239 | return 0; /* nothing is bound */ |
| 4240 | |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 4241 | blocks = pool->ti->len; |
| 4242 | (void) sector_div(blocks, pool->sectors_per_block); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4243 | if (blocks) |
Mike Snitzer | 55f2b8b | 2012-07-27 15:08:02 +0100 | [diff] [blame] | 4244 | 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] | 4245 | |
| 4246 | return 0; |
| 4247 | } |
| 4248 | |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 4249 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 4250 | { |
| 4251 | struct thin_c *tc = ti->private; |
| 4252 | struct pool *pool = tc->pool; |
Mike Snitzer | 2160767 | 2015-09-08 08:56:13 -0400 | [diff] [blame] | 4253 | |
Mike Snitzer | 0fcb04d | 2015-11-23 13:44:38 -0500 | [diff] [blame] | 4254 | if (!pool->pf.discard_enabled) |
| 4255 | return; |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 4256 | |
| 4257 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
| 4258 | limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ |
| 4259 | } |
| 4260 | |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4261 | static struct target_type thin_target = { |
| 4262 | .name = "thin", |
Mike Snitzer | 1c2e54e | 2016-01-06 14:13:40 -0500 | [diff] [blame] | 4263 | .version = {1, 17, 0}, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4264 | .module = THIS_MODULE, |
| 4265 | .ctr = thin_ctr, |
| 4266 | .dtr = thin_dtr, |
| 4267 | .map = thin_map, |
Joe Thornber | eb2aa48 | 2012-03-28 18:41:28 +0100 | [diff] [blame] | 4268 | .end_io = thin_endio, |
Joe Thornber | e5aea7b | 2014-06-13 14:47:24 +0100 | [diff] [blame] | 4269 | .preresume = thin_preresume, |
Joe Thornber | 738211f | 2014-03-03 15:52:28 +0000 | [diff] [blame] | 4270 | .presuspend = thin_presuspend, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4271 | .postsuspend = thin_postsuspend, |
| 4272 | .status = thin_status, |
| 4273 | .iterate_devices = thin_iterate_devices, |
Joe Thornber | 34fbcf6 | 2015-04-16 12:58:35 +0100 | [diff] [blame] | 4274 | .io_hints = thin_io_hints, |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4275 | }; |
| 4276 | |
| 4277 | /*----------------------------------------------------------------*/ |
| 4278 | |
| 4279 | static int __init dm_thin_init(void) |
| 4280 | { |
| 4281 | int r; |
| 4282 | |
| 4283 | pool_table_init(); |
| 4284 | |
| 4285 | r = dm_register_target(&thin_target); |
| 4286 | if (r) |
| 4287 | return r; |
| 4288 | |
| 4289 | r = dm_register_target(&pool_target); |
| 4290 | if (r) |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4291 | goto bad_pool_target; |
| 4292 | |
| 4293 | r = -ENOMEM; |
| 4294 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4295 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
| 4296 | if (!_new_mapping_cache) |
| 4297 | goto bad_new_mapping_cache; |
| 4298 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4299 | return 0; |
| 4300 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4301 | bad_new_mapping_cache: |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4302 | dm_unregister_target(&pool_target); |
| 4303 | bad_pool_target: |
| 4304 | dm_unregister_target(&thin_target); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4305 | |
| 4306 | return r; |
| 4307 | } |
| 4308 | |
| 4309 | static void dm_thin_exit(void) |
| 4310 | { |
| 4311 | dm_unregister_target(&thin_target); |
| 4312 | dm_unregister_target(&pool_target); |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4313 | |
Mike Snitzer | a24c256 | 2012-06-03 00:30:00 +0100 | [diff] [blame] | 4314 | kmem_cache_destroy(_new_mapping_cache); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4315 | } |
| 4316 | |
| 4317 | module_init(dm_thin_init); |
| 4318 | module_exit(dm_thin_exit); |
| 4319 | |
Mike Snitzer | 80c5789 | 2014-05-20 13:38:33 -0400 | [diff] [blame] | 4320 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
| 4321 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); |
| 4322 | |
Alasdair G Kergon | 7cab8bf | 2012-05-12 01:43:19 +0100 | [diff] [blame] | 4323 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
Joe Thornber | 991d9fa | 2011-10-31 20:21:18 +0000 | [diff] [blame] | 4324 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 4325 | MODULE_LICENSE("GPL"); |