Commit | Line | Data |
---|---|---|
991d9fa0 | 1 | /* |
e49e5829 | 2 | * Copyright (C) 2011-2012 Red Hat UK. |
991d9fa0 JT |
3 | * |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-thin-metadata.h" | |
742c8fdc | 8 | #include "dm-bio-prison-v1.h" |
1f4e0ff0 | 9 | #include "dm.h" |
991d9fa0 JT |
10 | |
11 | #include <linux/device-mapper.h> | |
12 | #include <linux/dm-io.h> | |
13 | #include <linux/dm-kcopyd.h> | |
0f30af98 | 14 | #include <linux/jiffies.h> |
604ea906 | 15 | #include <linux/log2.h> |
991d9fa0 | 16 | #include <linux/list.h> |
c140e1c4 | 17 | #include <linux/rculist.h> |
991d9fa0 JT |
18 | #include <linux/init.h> |
19 | #include <linux/module.h> | |
20 | #include <linux/slab.h> | |
a822c83e | 21 | #include <linux/vmalloc.h> |
ac4c3f34 | 22 | #include <linux/sort.h> |
67324ea1 | 23 | #include <linux/rbtree.h> |
991d9fa0 JT |
24 | |
25 | #define DM_MSG_PREFIX "thin" | |
26 | ||
27 | /* | |
28 | * Tunable constants | |
29 | */ | |
7768ed33 | 30 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 31 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 32 | #define COMMIT_PERIOD HZ |
80c57893 MS |
33 | #define NO_SPACE_TIMEOUT_SECS 60 |
34 | ||
35 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 36 | |
df5d2e90 MP |
37 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
38 | "A percentage of time allocated for copy on write"); | |
39 | ||
991d9fa0 JT |
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 | ||
991d9fa0 JT |
47 | /* |
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 | |
44feb387 | 72 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
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 | |
fe878f34 | 78 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
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 | ||
991d9fa0 JT |
112 | /* |
113 | * Key building. | |
114 | */ | |
34fbcf62 JT |
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) | |
991d9fa0 | 122 | { |
34fbcf62 | 123 | key->virtual = (ls == VIRTUAL); |
991d9fa0 | 124 | key->dev = dm_thin_dev_id(td); |
5f274d88 | 125 | key->block_begin = b; |
34fbcf62 JT |
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); | |
991d9fa0 JT |
133 | } |
134 | ||
135 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 136 | struct dm_cell_key *key) |
991d9fa0 | 137 | { |
34fbcf62 | 138 | build_key(td, VIRTUAL, b, b + 1llu, key); |
991d9fa0 JT |
139 | } |
140 | ||
141 | /*----------------------------------------------------------------*/ | |
142 | ||
7d327fe0 JT |
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 | { | |
8ca8b1e1 | 164 | if (!t->throttle_applied && time_is_before_jiffies(t->threshold)) { |
7d327fe0 JT |
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 | ||
991d9fa0 JT |
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 | */ | |
a24c2569 | 195 | struct dm_thin_new_mapping; |
67e2e2b2 | 196 | |
e49e5829 | 197 | /* |
f6c36758 | 198 | * The pool runs in various modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
199 | */ |
200 | enum pool_mode { | |
201 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 202 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
3ab91828 JT |
203 | |
204 | /* | |
205 | * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY. | |
206 | */ | |
207 | PM_OUT_OF_METADATA_SPACE, | |
e49e5829 | 208 | PM_READ_ONLY, /* metadata may not be changed */ |
3ab91828 | 209 | |
e49e5829 JT |
210 | PM_FAIL, /* all I/O fails */ |
211 | }; | |
212 | ||
67e2e2b2 | 213 | struct pool_features { |
e49e5829 JT |
214 | enum pool_mode mode; |
215 | ||
9bc142dd MS |
216 | bool zero_new_blocks:1; |
217 | bool discard_enabled:1; | |
218 | bool discard_passdown:1; | |
787a996c | 219 | bool error_if_no_space:1; |
67e2e2b2 JT |
220 | }; |
221 | ||
e49e5829 JT |
222 | struct thin_c; |
223 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 224 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
225 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
226 | ||
ac4c3f34 JT |
227 | #define CELL_SORT_ARRAY_SIZE 8192 |
228 | ||
991d9fa0 JT |
229 | struct pool { |
230 | struct list_head list; | |
231 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
232 | ||
233 | struct mapped_device *pool_md; | |
873937e7 | 234 | struct block_device *data_dev; |
991d9fa0 JT |
235 | struct block_device *md_dev; |
236 | struct dm_pool_metadata *pmd; | |
237 | ||
991d9fa0 | 238 | dm_block_t low_water_blocks; |
55f2b8bd | 239 | uint32_t sectors_per_block; |
f9a8e0cd | 240 | int sectors_per_block_shift; |
991d9fa0 | 241 | |
67e2e2b2 | 242 | struct pool_features pf; |
88a6621b | 243 | bool low_water_triggered:1; /* A dm event has been sent */ |
80e96c54 | 244 | bool suspended:1; |
c3667cc6 | 245 | bool out_of_data_space:1; |
991d9fa0 | 246 | |
44feb387 | 247 | struct dm_bio_prison *prison; |
991d9fa0 JT |
248 | struct dm_kcopyd_client *copier; |
249 | ||
72d711c8 | 250 | struct work_struct worker; |
991d9fa0 | 251 | struct workqueue_struct *wq; |
7d327fe0 | 252 | struct throttle throttle; |
905e51b3 | 253 | struct delayed_work waker; |
85ad643b | 254 | struct delayed_work no_space_timeout; |
991d9fa0 | 255 | |
905e51b3 | 256 | unsigned long last_commit_jiffies; |
55f2b8bd | 257 | unsigned ref_count; |
991d9fa0 JT |
258 | |
259 | spinlock_t lock; | |
991d9fa0 | 260 | struct bio_list deferred_flush_bios; |
4ae280b4 | 261 | struct bio_list deferred_flush_completions; |
991d9fa0 | 262 | struct list_head prepared_mappings; |
104655fd | 263 | struct list_head prepared_discards; |
2a0fbffb | 264 | struct list_head prepared_discards_pt2; |
c140e1c4 | 265 | struct list_head active_thins; |
991d9fa0 | 266 | |
44feb387 MS |
267 | struct dm_deferred_set *shared_read_ds; |
268 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 269 | |
a24c2569 | 270 | struct dm_thin_new_mapping *next_mapping; |
e49e5829 JT |
271 | |
272 | process_bio_fn process_bio; | |
273 | process_bio_fn process_discard; | |
274 | ||
a374bb21 JT |
275 | process_cell_fn process_cell; |
276 | process_cell_fn process_discard_cell; | |
277 | ||
e49e5829 JT |
278 | process_mapping_fn process_prepared_mapping; |
279 | process_mapping_fn process_prepared_discard; | |
2a0fbffb | 280 | process_mapping_fn process_prepared_discard_pt2; |
ac4c3f34 | 281 | |
a822c83e | 282 | struct dm_bio_prison_cell **cell_sort_array; |
72d711c8 MS |
283 | |
284 | mempool_t mapping_pool; | |
991d9fa0 JT |
285 | }; |
286 | ||
b5330655 | 287 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 288 | |
f6c36758 MS |
289 | static enum pool_mode get_pool_mode(struct pool *pool) |
290 | { | |
291 | return pool->pf.mode; | |
292 | } | |
293 | ||
294 | static void notify_of_pool_mode_change(struct pool *pool) | |
295 | { | |
296 | const char *descs[] = { | |
297 | "write", | |
298 | "out-of-data-space", | |
299 | "read-only", | |
300 | "read-only", | |
301 | "fail" | |
302 | }; | |
303 | const char *extra_desc = NULL; | |
304 | enum pool_mode mode = get_pool_mode(pool); | |
305 | ||
306 | if (mode == PM_OUT_OF_DATA_SPACE) { | |
307 | if (!pool->pf.error_if_no_space) | |
308 | extra_desc = " (queue IO)"; | |
309 | else | |
310 | extra_desc = " (error IO)"; | |
311 | } | |
312 | ||
313 | dm_table_event(pool->ti->table); | |
314 | DMINFO("%s: switching pool to %s%s mode", | |
315 | dm_device_name(pool->pool_md), | |
316 | descs[(int)mode], extra_desc ? : ""); | |
317 | } | |
318 | ||
991d9fa0 JT |
319 | /* |
320 | * Target context for a pool. | |
321 | */ | |
322 | struct pool_c { | |
323 | struct dm_target *ti; | |
324 | struct pool *pool; | |
325 | struct dm_dev *data_dev; | |
326 | struct dm_dev *metadata_dev; | |
991d9fa0 JT |
327 | |
328 | dm_block_t low_water_blocks; | |
0424caa1 MS |
329 | struct pool_features requested_pf; /* Features requested during table load */ |
330 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
331 | }; |
332 | ||
333 | /* | |
334 | * Target context for a thin. | |
335 | */ | |
336 | struct thin_c { | |
c140e1c4 | 337 | struct list_head list; |
991d9fa0 | 338 | struct dm_dev *pool_dev; |
2dd9c257 | 339 | struct dm_dev *origin_dev; |
e5aea7b4 | 340 | sector_t origin_size; |
991d9fa0 JT |
341 | dm_thin_id dev_id; |
342 | ||
343 | struct pool *pool; | |
344 | struct dm_thin_device *td; | |
583024d2 MS |
345 | struct mapped_device *thin_md; |
346 | ||
738211f7 | 347 | bool requeue_mode:1; |
c140e1c4 | 348 | spinlock_t lock; |
a374bb21 | 349 | struct list_head deferred_cells; |
c140e1c4 MS |
350 | struct bio_list deferred_bio_list; |
351 | struct bio_list retry_on_resume_list; | |
67324ea1 | 352 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
353 | |
354 | /* | |
355 | * Ensures the thin is not destroyed until the worker has finished | |
356 | * iterating the active_thins list. | |
357 | */ | |
22d4c291 | 358 | refcount_t refcount; |
b10ebd34 | 359 | struct completion can_destroy; |
991d9fa0 JT |
360 | }; |
361 | ||
362 | /*----------------------------------------------------------------*/ | |
363 | ||
34fbcf62 JT |
364 | static bool block_size_is_power_of_two(struct pool *pool) |
365 | { | |
366 | return pool->sectors_per_block_shift >= 0; | |
367 | } | |
368 | ||
369 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) | |
370 | { | |
371 | return block_size_is_power_of_two(pool) ? | |
372 | (b << pool->sectors_per_block_shift) : | |
373 | (b * pool->sectors_per_block); | |
374 | } | |
375 | ||
202bae52 JT |
376 | /*----------------------------------------------------------------*/ |
377 | ||
378 | struct discard_op { | |
379 | struct thin_c *tc; | |
380 | struct blk_plug plug; | |
381 | struct bio *parent_bio; | |
382 | struct bio *bio; | |
383 | }; | |
384 | ||
385 | static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent) | |
386 | { | |
387 | BUG_ON(!parent); | |
388 | ||
389 | op->tc = tc; | |
390 | blk_start_plug(&op->plug); | |
391 | op->parent_bio = parent; | |
392 | op->bio = NULL; | |
393 | } | |
394 | ||
395 | static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e) | |
34fbcf62 | 396 | { |
202bae52 | 397 | struct thin_c *tc = op->tc; |
34fbcf62 JT |
398 | sector_t s = block_to_sectors(tc->pool, data_b); |
399 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); | |
3dba53a9 | 400 | |
202bae52 | 401 | return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, |
469e3216 | 402 | GFP_NOWAIT, 0, &op->bio); |
202bae52 JT |
403 | } |
404 | ||
405 | static void end_discard(struct discard_op *op, int r) | |
406 | { | |
407 | if (op->bio) { | |
408 | /* | |
409 | * Even if one of the calls to issue_discard failed, we | |
410 | * need to wait for the chain to complete. | |
411 | */ | |
412 | bio_chain(op->bio, op->parent_bio); | |
e6047149 | 413 | bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0); |
4e49ea4a | 414 | submit_bio(op->bio); |
3dba53a9 | 415 | } |
34fbcf62 | 416 | |
202bae52 JT |
417 | blk_finish_plug(&op->plug); |
418 | ||
419 | /* | |
420 | * Even if r is set, there could be sub discards in flight that we | |
421 | * need to wait for. | |
422 | */ | |
4e4cbee9 CH |
423 | if (r && !op->parent_bio->bi_status) |
424 | op->parent_bio->bi_status = errno_to_blk_status(r); | |
202bae52 | 425 | bio_endio(op->parent_bio); |
34fbcf62 JT |
426 | } |
427 | ||
428 | /*----------------------------------------------------------------*/ | |
429 | ||
025b9685 JT |
430 | /* |
431 | * wake_worker() is used when new work is queued and when pool_resume is | |
432 | * ready to continue deferred IO processing. | |
433 | */ | |
434 | static void wake_worker(struct pool *pool) | |
435 | { | |
436 | queue_work(pool->wq, &pool->worker); | |
437 | } | |
438 | ||
439 | /*----------------------------------------------------------------*/ | |
440 | ||
6beca5eb JT |
441 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
442 | struct dm_bio_prison_cell **cell_result) | |
443 | { | |
444 | int r; | |
445 | struct dm_bio_prison_cell *cell_prealloc; | |
446 | ||
447 | /* | |
448 | * Allocate a cell from the prison's mempool. | |
449 | * This might block but it can't fail. | |
450 | */ | |
451 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
452 | ||
453 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
454 | if (r) | |
455 | /* | |
456 | * We reused an old cell; we can get rid of | |
457 | * the new one. | |
458 | */ | |
459 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
460 | ||
461 | return r; | |
462 | } | |
463 | ||
464 | static void cell_release(struct pool *pool, | |
465 | struct dm_bio_prison_cell *cell, | |
466 | struct bio_list *bios) | |
467 | { | |
468 | dm_cell_release(pool->prison, cell, bios); | |
469 | dm_bio_prison_free_cell(pool->prison, cell); | |
470 | } | |
471 | ||
2d759a46 JT |
472 | static void cell_visit_release(struct pool *pool, |
473 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
474 | void *context, | |
475 | struct dm_bio_prison_cell *cell) | |
476 | { | |
477 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
478 | dm_bio_prison_free_cell(pool->prison, cell); | |
479 | } | |
480 | ||
6beca5eb JT |
481 | static void cell_release_no_holder(struct pool *pool, |
482 | struct dm_bio_prison_cell *cell, | |
483 | struct bio_list *bios) | |
484 | { | |
485 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
486 | dm_bio_prison_free_cell(pool->prison, cell); | |
487 | } | |
488 | ||
af91805a | 489 | static void cell_error_with_code(struct pool *pool, |
4e4cbee9 | 490 | struct dm_bio_prison_cell *cell, blk_status_t error_code) |
6beca5eb | 491 | { |
af91805a | 492 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
493 | dm_bio_prison_free_cell(pool->prison, cell); |
494 | } | |
495 | ||
4e4cbee9 | 496 | static blk_status_t get_pool_io_error_code(struct pool *pool) |
c3667cc6 | 497 | { |
4e4cbee9 | 498 | return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR; |
c3667cc6 MS |
499 | } |
500 | ||
af91805a MS |
501 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
502 | { | |
4e4cbee9 | 503 | cell_error_with_code(pool, cell, get_pool_io_error_code(pool)); |
af91805a MS |
504 | } |
505 | ||
a374bb21 JT |
506 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
507 | { | |
508 | cell_error_with_code(pool, cell, 0); | |
509 | } | |
510 | ||
511 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
512 | { | |
4e4cbee9 | 513 | cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE); |
a374bb21 JT |
514 | } |
515 | ||
6beca5eb JT |
516 | /*----------------------------------------------------------------*/ |
517 | ||
991d9fa0 JT |
518 | /* |
519 | * A global list of pools that uses a struct mapped_device as a key. | |
520 | */ | |
521 | static struct dm_thin_pool_table { | |
522 | struct mutex mutex; | |
523 | struct list_head pools; | |
524 | } dm_thin_pool_table; | |
525 | ||
526 | static void pool_table_init(void) | |
527 | { | |
528 | mutex_init(&dm_thin_pool_table.mutex); | |
529 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
530 | } | |
531 | ||
d5ffebdd MS |
532 | static void pool_table_exit(void) |
533 | { | |
534 | mutex_destroy(&dm_thin_pool_table.mutex); | |
535 | } | |
536 | ||
991d9fa0 JT |
537 | static void __pool_table_insert(struct pool *pool) |
538 | { | |
539 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
540 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
541 | } | |
542 | ||
543 | static void __pool_table_remove(struct pool *pool) | |
544 | { | |
545 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
546 | list_del(&pool->list); | |
547 | } | |
548 | ||
549 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
550 | { | |
551 | struct pool *pool = NULL, *tmp; | |
552 | ||
553 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
554 | ||
555 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
556 | if (tmp->pool_md == md) { | |
557 | pool = tmp; | |
558 | break; | |
559 | } | |
560 | } | |
561 | ||
562 | return pool; | |
563 | } | |
564 | ||
565 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
566 | { | |
567 | struct pool *pool = NULL, *tmp; | |
568 | ||
569 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
570 | ||
571 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
572 | if (tmp->md_dev == md_dev) { | |
573 | pool = tmp; | |
574 | break; | |
575 | } | |
576 | } | |
577 | ||
578 | return pool; | |
579 | } | |
580 | ||
581 | /*----------------------------------------------------------------*/ | |
582 | ||
a24c2569 | 583 | struct dm_thin_endio_hook { |
eb2aa48d | 584 | struct thin_c *tc; |
44feb387 MS |
585 | struct dm_deferred_entry *shared_read_entry; |
586 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 587 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 588 | struct rb_node rb_node; |
34fbcf62 | 589 | struct dm_bio_prison_cell *cell; |
eb2aa48d JT |
590 | }; |
591 | ||
42d6a8ce MS |
592 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
593 | { | |
594 | bio_list_merge(bios, master); | |
595 | bio_list_init(master); | |
596 | } | |
597 | ||
4e4cbee9 | 598 | static void error_bio_list(struct bio_list *bios, blk_status_t error) |
991d9fa0 JT |
599 | { |
600 | struct bio *bio; | |
42d6a8ce | 601 | |
4246a0b6 | 602 | while ((bio = bio_list_pop(bios))) { |
4e4cbee9 | 603 | bio->bi_status = error; |
4246a0b6 CH |
604 | bio_endio(bio); |
605 | } | |
42d6a8ce MS |
606 | } |
607 | ||
4e4cbee9 CH |
608 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, |
609 | blk_status_t error) | |
42d6a8ce | 610 | { |
991d9fa0 JT |
611 | struct bio_list bios; |
612 | ||
613 | bio_list_init(&bios); | |
18adc577 | 614 | |
8e0c9dac | 615 | spin_lock_irq(&tc->lock); |
42d6a8ce | 616 | __merge_bio_list(&bios, master); |
8e0c9dac | 617 | spin_unlock_irq(&tc->lock); |
991d9fa0 | 618 | |
42d6a8ce | 619 | error_bio_list(&bios, error); |
991d9fa0 JT |
620 | } |
621 | ||
a374bb21 JT |
622 | static void requeue_deferred_cells(struct thin_c *tc) |
623 | { | |
624 | struct pool *pool = tc->pool; | |
a374bb21 JT |
625 | struct list_head cells; |
626 | struct dm_bio_prison_cell *cell, *tmp; | |
627 | ||
628 | INIT_LIST_HEAD(&cells); | |
629 | ||
8e0c9dac | 630 | spin_lock_irq(&tc->lock); |
a374bb21 | 631 | list_splice_init(&tc->deferred_cells, &cells); |
8e0c9dac | 632 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
633 | |
634 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
635 | cell_requeue(pool, cell); | |
636 | } | |
637 | ||
991d9fa0 JT |
638 | static void requeue_io(struct thin_c *tc) |
639 | { | |
3e1a0699 JT |
640 | struct bio_list bios; |
641 | ||
642 | bio_list_init(&bios); | |
643 | ||
8e0c9dac | 644 | spin_lock_irq(&tc->lock); |
42d6a8ce MS |
645 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
646 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
8e0c9dac | 647 | spin_unlock_irq(&tc->lock); |
3e1a0699 | 648 | |
4e4cbee9 | 649 | error_bio_list(&bios, BLK_STS_DM_REQUEUE); |
42d6a8ce | 650 | requeue_deferred_cells(tc); |
3e1a0699 JT |
651 | } |
652 | ||
4e4cbee9 | 653 | static void error_retry_list_with_code(struct pool *pool, blk_status_t error) |
c140e1c4 MS |
654 | { |
655 | struct thin_c *tc; | |
656 | ||
657 | rcu_read_lock(); | |
658 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
0a927c2f | 659 | error_thin_bio_list(tc, &tc->retry_on_resume_list, error); |
c140e1c4 MS |
660 | rcu_read_unlock(); |
661 | } | |
662 | ||
0a927c2f MS |
663 | static void error_retry_list(struct pool *pool) |
664 | { | |
4e4cbee9 | 665 | error_retry_list_with_code(pool, get_pool_io_error_code(pool)); |
0a927c2f MS |
666 | } |
667 | ||
991d9fa0 JT |
668 | /* |
669 | * This section of code contains the logic for processing a thin device's IO. | |
670 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
671 | * but most is exclusively called from the thin target rather than the thin-pool | |
672 | * target. | |
673 | */ | |
674 | ||
675 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
676 | { | |
58f77a21 | 677 | struct pool *pool = tc->pool; |
4f024f37 | 678 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 679 | |
58f77a21 MS |
680 | if (block_size_is_power_of_two(pool)) |
681 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 682 | else |
58f77a21 | 683 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
684 | |
685 | return block_nr; | |
991d9fa0 JT |
686 | } |
687 | ||
34fbcf62 JT |
688 | /* |
689 | * Returns the _complete_ blocks that this bio covers. | |
690 | */ | |
691 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, | |
692 | dm_block_t *begin, dm_block_t *end) | |
693 | { | |
694 | struct pool *pool = tc->pool; | |
695 | sector_t b = bio->bi_iter.bi_sector; | |
696 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
697 | ||
698 | b += pool->sectors_per_block - 1ull; /* so we round up */ | |
699 | ||
700 | if (block_size_is_power_of_two(pool)) { | |
701 | b >>= pool->sectors_per_block_shift; | |
702 | e >>= pool->sectors_per_block_shift; | |
703 | } else { | |
704 | (void) sector_div(b, pool->sectors_per_block); | |
705 | (void) sector_div(e, pool->sectors_per_block); | |
706 | } | |
707 | ||
708 | if (e < b) | |
709 | /* Can happen if the bio is within a single block. */ | |
710 | e = b; | |
711 | ||
712 | *begin = b; | |
713 | *end = e; | |
714 | } | |
715 | ||
991d9fa0 JT |
716 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
717 | { | |
718 | struct pool *pool = tc->pool; | |
4f024f37 | 719 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 | 720 | |
74d46992 | 721 | bio_set_dev(bio, tc->pool_dev->bdev); |
58f77a21 | 722 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
723 | bio->bi_iter.bi_sector = |
724 | (block << pool->sectors_per_block_shift) | | |
725 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 726 | else |
4f024f37 | 727 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 728 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
729 | } |
730 | ||
2dd9c257 JT |
731 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
732 | { | |
74d46992 | 733 | bio_set_dev(bio, tc->origin_dev->bdev); |
2dd9c257 JT |
734 | } |
735 | ||
4afdd680 JT |
736 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
737 | { | |
f73f44eb | 738 | return op_is_flush(bio->bi_opf) && |
4afdd680 JT |
739 | dm_thin_changed_this_transaction(tc->td); |
740 | } | |
741 | ||
e8088073 JT |
742 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
743 | { | |
744 | struct dm_thin_endio_hook *h; | |
745 | ||
e6047149 | 746 | if (bio_op(bio) == REQ_OP_DISCARD) |
e8088073 JT |
747 | return; |
748 | ||
59c3d2c6 | 749 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
750 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
751 | } | |
752 | ||
2dd9c257 | 753 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
754 | { |
755 | struct pool *pool = tc->pool; | |
991d9fa0 | 756 | |
e49e5829 | 757 | if (!bio_triggers_commit(tc, bio)) { |
ed00aabd | 758 | submit_bio_noacct(bio); |
e49e5829 JT |
759 | return; |
760 | } | |
761 | ||
991d9fa0 | 762 | /* |
e49e5829 JT |
763 | * Complete bio with an error if earlier I/O caused changes to |
764 | * the metadata that can't be committed e.g, due to I/O errors | |
765 | * on the metadata device. | |
991d9fa0 | 766 | */ |
e49e5829 JT |
767 | if (dm_thin_aborted_changes(tc->td)) { |
768 | bio_io_error(bio); | |
769 | return; | |
770 | } | |
771 | ||
772 | /* | |
773 | * Batch together any bios that trigger commits and then issue a | |
774 | * single commit for them in process_deferred_bios(). | |
775 | */ | |
8e0c9dac | 776 | spin_lock_irq(&pool->lock); |
e49e5829 | 777 | bio_list_add(&pool->deferred_flush_bios, bio); |
8e0c9dac | 778 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
779 | } |
780 | ||
2dd9c257 JT |
781 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
782 | { | |
783 | remap_to_origin(tc, bio); | |
784 | issue(tc, bio); | |
785 | } | |
786 | ||
787 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
788 | dm_block_t block) | |
789 | { | |
790 | remap(tc, bio, block); | |
791 | issue(tc, bio); | |
792 | } | |
793 | ||
991d9fa0 JT |
794 | /*----------------------------------------------------------------*/ |
795 | ||
796 | /* | |
797 | * Bio endio functions. | |
798 | */ | |
a24c2569 | 799 | struct dm_thin_new_mapping { |
991d9fa0 JT |
800 | struct list_head list; |
801 | ||
7f214665 | 802 | bool pass_discard:1; |
34fbcf62 | 803 | bool maybe_shared:1; |
991d9fa0 | 804 | |
50f3c3ef JT |
805 | /* |
806 | * Track quiescing, copying and zeroing preparation actions. When this | |
807 | * counter hits zero the block is prepared and can be inserted into the | |
808 | * btree. | |
809 | */ | |
810 | atomic_t prepare_actions; | |
811 | ||
4e4cbee9 | 812 | blk_status_t status; |
991d9fa0 | 813 | struct thin_c *tc; |
34fbcf62 | 814 | dm_block_t virt_begin, virt_end; |
991d9fa0 | 815 | dm_block_t data_block; |
34fbcf62 | 816 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
817 | |
818 | /* | |
819 | * If the bio covers the whole area of a block then we can avoid | |
820 | * zeroing or copying. Instead this bio is hooked. The bio will | |
821 | * still be in the cell, so care has to be taken to avoid issuing | |
822 | * the bio twice. | |
823 | */ | |
824 | struct bio *bio; | |
825 | bio_end_io_t *saved_bi_end_io; | |
826 | }; | |
827 | ||
50f3c3ef | 828 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
829 | { |
830 | struct pool *pool = m->tc->pool; | |
831 | ||
50f3c3ef | 832 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 833 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
834 | wake_worker(pool); |
835 | } | |
836 | } | |
837 | ||
e5aea7b4 | 838 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
839 | { |
840 | unsigned long flags; | |
991d9fa0 JT |
841 | struct pool *pool = m->tc->pool; |
842 | ||
991d9fa0 | 843 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 844 | __complete_mapping_preparation(m); |
991d9fa0 JT |
845 | spin_unlock_irqrestore(&pool->lock, flags); |
846 | } | |
847 | ||
e5aea7b4 JT |
848 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
849 | { | |
850 | struct dm_thin_new_mapping *m = context; | |
851 | ||
4e4cbee9 | 852 | m->status = read_err || write_err ? BLK_STS_IOERR : 0; |
e5aea7b4 JT |
853 | complete_mapping_preparation(m); |
854 | } | |
855 | ||
4246a0b6 | 856 | static void overwrite_endio(struct bio *bio) |
991d9fa0 | 857 | { |
59c3d2c6 | 858 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 859 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 | 860 | |
8b908f8e MS |
861 | bio->bi_end_io = m->saved_bi_end_io; |
862 | ||
4e4cbee9 | 863 | m->status = bio->bi_status; |
e5aea7b4 | 864 | complete_mapping_preparation(m); |
991d9fa0 JT |
865 | } |
866 | ||
991d9fa0 JT |
867 | /*----------------------------------------------------------------*/ |
868 | ||
869 | /* | |
870 | * Workqueue. | |
871 | */ | |
872 | ||
873 | /* | |
874 | * Prepared mapping jobs. | |
875 | */ | |
876 | ||
877 | /* | |
2d759a46 JT |
878 | * This sends the bios in the cell, except the original holder, back |
879 | * to the deferred_bios list. | |
991d9fa0 | 880 | */ |
f286ba0e | 881 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 882 | { |
991d9fa0 JT |
883 | struct pool *pool = tc->pool; |
884 | unsigned long flags; | |
d256d796 | 885 | int has_work; |
991d9fa0 | 886 | |
c140e1c4 MS |
887 | spin_lock_irqsave(&tc->lock, flags); |
888 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
d256d796 | 889 | has_work = !bio_list_empty(&tc->deferred_bio_list); |
c140e1c4 | 890 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 891 | |
d256d796 JX |
892 | if (has_work) |
893 | wake_worker(pool); | |
991d9fa0 JT |
894 | } |
895 | ||
a374bb21 JT |
896 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
897 | ||
2d759a46 JT |
898 | struct remap_info { |
899 | struct thin_c *tc; | |
900 | struct bio_list defer_bios; | |
901 | struct bio_list issue_bios; | |
902 | }; | |
903 | ||
904 | static void __inc_remap_and_issue_cell(void *context, | |
905 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 906 | { |
2d759a46 | 907 | struct remap_info *info = context; |
a374bb21 | 908 | struct bio *bio; |
a374bb21 | 909 | |
2d759a46 | 910 | while ((bio = bio_list_pop(&cell->bios))) { |
f73f44eb | 911 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) |
2d759a46 | 912 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 913 | else { |
2d759a46 JT |
914 | inc_all_io_entry(info->tc->pool, bio); |
915 | ||
916 | /* | |
917 | * We can't issue the bios with the bio prison lock | |
918 | * held, so we add them to a list to issue on | |
919 | * return from this function. | |
920 | */ | |
921 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
922 | } |
923 | } | |
924 | } | |
925 | ||
2d759a46 JT |
926 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
927 | struct dm_bio_prison_cell *cell, | |
928 | dm_block_t block) | |
929 | { | |
930 | struct bio *bio; | |
931 | struct remap_info info; | |
932 | ||
933 | info.tc = tc; | |
934 | bio_list_init(&info.defer_bios); | |
935 | bio_list_init(&info.issue_bios); | |
936 | ||
937 | /* | |
938 | * We have to be careful to inc any bios we're about to issue | |
939 | * before the cell is released, and avoid a race with new bios | |
940 | * being added to the cell. | |
941 | */ | |
942 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
943 | &info, cell); | |
944 | ||
945 | while ((bio = bio_list_pop(&info.defer_bios))) | |
946 | thin_defer_bio(tc, bio); | |
947 | ||
948 | while ((bio = bio_list_pop(&info.issue_bios))) | |
949 | remap_and_issue(info.tc, bio, block); | |
950 | } | |
951 | ||
e49e5829 JT |
952 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
953 | { | |
6beca5eb | 954 | cell_error(m->tc->pool, m->cell); |
e49e5829 | 955 | list_del(&m->list); |
6f1c819c | 956 | mempool_free(m, &m->tc->pool->mapping_pool); |
e49e5829 | 957 | } |
025b9685 | 958 | |
4ae280b4 NT |
959 | static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio) |
960 | { | |
961 | struct pool *pool = tc->pool; | |
4ae280b4 NT |
962 | |
963 | /* | |
964 | * If the bio has the REQ_FUA flag set we must commit the metadata | |
965 | * before signaling its completion. | |
966 | */ | |
967 | if (!bio_triggers_commit(tc, bio)) { | |
968 | bio_endio(bio); | |
969 | return; | |
970 | } | |
971 | ||
972 | /* | |
973 | * Complete bio with an error if earlier I/O caused changes to the | |
974 | * metadata that can't be committed, e.g, due to I/O errors on the | |
975 | * metadata device. | |
976 | */ | |
977 | if (dm_thin_aborted_changes(tc->td)) { | |
978 | bio_io_error(bio); | |
979 | return; | |
980 | } | |
981 | ||
982 | /* | |
983 | * Batch together any bios that trigger commits and then issue a | |
984 | * single commit for them in process_deferred_bios(). | |
985 | */ | |
8e0c9dac | 986 | spin_lock_irq(&pool->lock); |
4ae280b4 | 987 | bio_list_add(&pool->deferred_flush_completions, bio); |
8e0c9dac | 988 | spin_unlock_irq(&pool->lock); |
4ae280b4 NT |
989 | } |
990 | ||
a24c2569 | 991 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
992 | { |
993 | struct thin_c *tc = m->tc; | |
6beca5eb | 994 | struct pool *pool = tc->pool; |
8b908f8e | 995 | struct bio *bio = m->bio; |
991d9fa0 JT |
996 | int r; |
997 | ||
4e4cbee9 | 998 | if (m->status) { |
6beca5eb | 999 | cell_error(pool, m->cell); |
905386f8 | 1000 | goto out; |
991d9fa0 JT |
1001 | } |
1002 | ||
1003 | /* | |
1004 | * Commit the prepared block into the mapping btree. | |
1005 | * Any I/O for this block arriving after this point will get | |
1006 | * remapped to it directly. | |
1007 | */ | |
34fbcf62 | 1008 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
991d9fa0 | 1009 | if (r) { |
b5330655 | 1010 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 1011 | cell_error(pool, m->cell); |
905386f8 | 1012 | goto out; |
991d9fa0 JT |
1013 | } |
1014 | ||
1015 | /* | |
1016 | * Release any bios held while the block was being provisioned. | |
1017 | * If we are processing a write bio that completely covers the block, | |
1018 | * we already processed it so can ignore it now when processing | |
1019 | * the bios in the cell. | |
1020 | */ | |
1021 | if (bio) { | |
2d759a46 | 1022 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
4ae280b4 | 1023 | complete_overwrite_bio(tc, bio); |
2d759a46 JT |
1024 | } else { |
1025 | inc_all_io_entry(tc->pool, m->cell->holder); | |
1026 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
1027 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
1028 | } | |
991d9fa0 | 1029 | |
905386f8 | 1030 | out: |
991d9fa0 | 1031 | list_del(&m->list); |
6f1c819c | 1032 | mempool_free(m, &pool->mapping_pool); |
991d9fa0 JT |
1033 | } |
1034 | ||
34fbcf62 JT |
1035 | /*----------------------------------------------------------------*/ |
1036 | ||
1037 | static void free_discard_mapping(struct dm_thin_new_mapping *m) | |
104655fd | 1038 | { |
104655fd | 1039 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1040 | if (m->cell) |
1041 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1042 | mempool_free(m, &tc->pool->mapping_pool); |
34fbcf62 | 1043 | } |
104655fd | 1044 | |
34fbcf62 JT |
1045 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
1046 | { | |
e49e5829 | 1047 | bio_io_error(m->bio); |
34fbcf62 JT |
1048 | free_discard_mapping(m); |
1049 | } | |
1050 | ||
1051 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) | |
1052 | { | |
4246a0b6 | 1053 | bio_endio(m->bio); |
34fbcf62 JT |
1054 | free_discard_mapping(m); |
1055 | } | |
1056 | ||
1057 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) | |
1058 | { | |
1059 | int r; | |
1060 | struct thin_c *tc = m->tc; | |
1061 | ||
1062 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); | |
1063 | if (r) { | |
1064 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); | |
1065 | bio_io_error(m->bio); | |
1066 | } else | |
4246a0b6 | 1067 | bio_endio(m->bio); |
34fbcf62 | 1068 | |
f286ba0e | 1069 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1070 | mempool_free(m, &tc->pool->mapping_pool); |
e49e5829 JT |
1071 | } |
1072 | ||
202bae52 JT |
1073 | /*----------------------------------------------------------------*/ |
1074 | ||
2a0fbffb JT |
1075 | static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m, |
1076 | struct bio *discard_parent) | |
e49e5829 | 1077 | { |
34fbcf62 JT |
1078 | /* |
1079 | * We've already unmapped this range of blocks, but before we | |
1080 | * passdown we have to check that these blocks are now unused. | |
1081 | */ | |
202bae52 | 1082 | int r = 0; |
d445bd9c | 1083 | bool shared = true; |
e49e5829 | 1084 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1085 | struct pool *pool = tc->pool; |
1086 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; | |
202bae52 | 1087 | struct discard_op op; |
104655fd | 1088 | |
2a0fbffb | 1089 | begin_discard(&op, tc, discard_parent); |
34fbcf62 JT |
1090 | while (b != end) { |
1091 | /* find start of unmapped run */ | |
1092 | for (; b < end; b++) { | |
d445bd9c | 1093 | r = dm_pool_block_is_shared(pool->pmd, b, &shared); |
34fbcf62 | 1094 | if (r) |
202bae52 | 1095 | goto out; |
e8088073 | 1096 | |
d445bd9c | 1097 | if (!shared) |
34fbcf62 | 1098 | break; |
19fa1a67 | 1099 | } |
104655fd | 1100 | |
34fbcf62 JT |
1101 | if (b == end) |
1102 | break; | |
1103 | ||
1104 | /* find end of run */ | |
1105 | for (e = b + 1; e != end; e++) { | |
d445bd9c | 1106 | r = dm_pool_block_is_shared(pool->pmd, e, &shared); |
34fbcf62 | 1107 | if (r) |
202bae52 | 1108 | goto out; |
34fbcf62 | 1109 | |
d445bd9c | 1110 | if (shared) |
34fbcf62 JT |
1111 | break; |
1112 | } | |
1113 | ||
202bae52 | 1114 | r = issue_discard(&op, b, e); |
34fbcf62 | 1115 | if (r) |
202bae52 | 1116 | goto out; |
34fbcf62 JT |
1117 | |
1118 | b = e; | |
1119 | } | |
202bae52 JT |
1120 | out: |
1121 | end_discard(&op, r); | |
104655fd JT |
1122 | } |
1123 | ||
2a0fbffb JT |
1124 | static void queue_passdown_pt2(struct dm_thin_new_mapping *m) |
1125 | { | |
1126 | unsigned long flags; | |
1127 | struct pool *pool = m->tc->pool; | |
1128 | ||
1129 | spin_lock_irqsave(&pool->lock, flags); | |
1130 | list_add_tail(&m->list, &pool->prepared_discards_pt2); | |
1131 | spin_unlock_irqrestore(&pool->lock, flags); | |
1132 | wake_worker(pool); | |
1133 | } | |
1134 | ||
1135 | static void passdown_endio(struct bio *bio) | |
1136 | { | |
1137 | /* | |
1138 | * It doesn't matter if the passdown discard failed, we still want | |
1139 | * to unmap (we ignore err). | |
1140 | */ | |
1141 | queue_passdown_pt2(bio->bi_private); | |
948f581a | 1142 | bio_put(bio); |
2a0fbffb JT |
1143 | } |
1144 | ||
1145 | static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m) | |
e49e5829 JT |
1146 | { |
1147 | int r; | |
1148 | struct thin_c *tc = m->tc; | |
34fbcf62 | 1149 | struct pool *pool = tc->pool; |
2a0fbffb JT |
1150 | struct bio *discard_parent; |
1151 | dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin); | |
e49e5829 | 1152 | |
2a0fbffb JT |
1153 | /* |
1154 | * Only this thread allocates blocks, so we can be sure that the | |
1155 | * newly unmapped blocks will not be allocated before the end of | |
1156 | * the function. | |
1157 | */ | |
34fbcf62 | 1158 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
202bae52 | 1159 | if (r) { |
34fbcf62 | 1160 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
202bae52 | 1161 | bio_io_error(m->bio); |
2a0fbffb | 1162 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1163 | mempool_free(m, &pool->mapping_pool); |
2a0fbffb JT |
1164 | return; |
1165 | } | |
34fbcf62 | 1166 | |
00a0ea33 VV |
1167 | /* |
1168 | * Increment the unmapped blocks. This prevents a race between the | |
1169 | * passdown io and reallocation of freed blocks. | |
1170 | */ | |
1171 | r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end); | |
1172 | if (r) { | |
1173 | metadata_operation_failed(pool, "dm_pool_inc_data_range", r); | |
1174 | bio_io_error(m->bio); | |
1175 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1176 | mempool_free(m, &pool->mapping_pool); |
00a0ea33 VV |
1177 | return; |
1178 | } | |
1179 | ||
07888c66 | 1180 | discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO); |
53db984e CH |
1181 | discard_parent->bi_end_io = passdown_endio; |
1182 | discard_parent->bi_private = m; | |
07888c66 CH |
1183 | if (m->maybe_shared) |
1184 | passdown_double_checking_shared_status(m, discard_parent); | |
1185 | else { | |
53db984e | 1186 | struct discard_op op; |
2a0fbffb | 1187 | |
53db984e CH |
1188 | begin_discard(&op, tc, discard_parent); |
1189 | r = issue_discard(&op, m->data_block, data_end); | |
1190 | end_discard(&op, r); | |
202bae52 | 1191 | } |
2a0fbffb JT |
1192 | } |
1193 | ||
1194 | static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m) | |
1195 | { | |
1196 | int r; | |
1197 | struct thin_c *tc = m->tc; | |
1198 | struct pool *pool = tc->pool; | |
1199 | ||
1200 | /* | |
1201 | * The passdown has completed, so now we can decrement all those | |
1202 | * unmapped blocks. | |
1203 | */ | |
1204 | r = dm_pool_dec_data_range(pool->pmd, m->data_block, | |
1205 | m->data_block + (m->virt_end - m->virt_begin)); | |
1206 | if (r) { | |
1207 | metadata_operation_failed(pool, "dm_pool_dec_data_range", r); | |
1208 | bio_io_error(m->bio); | |
1209 | } else | |
1210 | bio_endio(m->bio); | |
1211 | ||
34fbcf62 | 1212 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1213 | mempool_free(m, &pool->mapping_pool); |
e49e5829 JT |
1214 | } |
1215 | ||
104655fd | 1216 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 1217 | process_mapping_fn *fn) |
991d9fa0 | 1218 | { |
991d9fa0 | 1219 | struct list_head maps; |
a24c2569 | 1220 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
1221 | |
1222 | INIT_LIST_HEAD(&maps); | |
8e0c9dac | 1223 | spin_lock_irq(&pool->lock); |
104655fd | 1224 | list_splice_init(head, &maps); |
8e0c9dac | 1225 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
1226 | |
1227 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1228 | (*fn)(m); |
991d9fa0 JT |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * Deferred bio jobs. | |
1233 | */ | |
104655fd | 1234 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1235 | { |
4f024f37 KO |
1236 | return bio->bi_iter.bi_size == |
1237 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
1238 | } |
1239 | ||
1240 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1241 | { | |
1242 | return (bio_data_dir(bio) == WRITE) && | |
1243 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1244 | } |
1245 | ||
1246 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1247 | bio_end_io_t *fn) | |
1248 | { | |
1249 | *save = bio->bi_end_io; | |
1250 | bio->bi_end_io = fn; | |
1251 | } | |
1252 | ||
1253 | static int ensure_next_mapping(struct pool *pool) | |
1254 | { | |
1255 | if (pool->next_mapping) | |
1256 | return 0; | |
1257 | ||
6f1c819c | 1258 | pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC); |
991d9fa0 JT |
1259 | |
1260 | return pool->next_mapping ? 0 : -ENOMEM; | |
1261 | } | |
1262 | ||
a24c2569 | 1263 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1264 | { |
16961b04 | 1265 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
1266 | |
1267 | BUG_ON(!pool->next_mapping); | |
1268 | ||
16961b04 MS |
1269 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
1270 | INIT_LIST_HEAD(&m->list); | |
1271 | m->bio = NULL; | |
1272 | ||
991d9fa0 JT |
1273 | pool->next_mapping = NULL; |
1274 | ||
16961b04 | 1275 | return m; |
991d9fa0 JT |
1276 | } |
1277 | ||
e5aea7b4 JT |
1278 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
1279 | sector_t begin, sector_t end) | |
1280 | { | |
e5aea7b4 JT |
1281 | struct dm_io_region to; |
1282 | ||
1283 | to.bdev = tc->pool_dev->bdev; | |
1284 | to.sector = begin; | |
1285 | to.count = end - begin; | |
1286 | ||
7209049d | 1287 | dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); |
e5aea7b4 JT |
1288 | } |
1289 | ||
452d7a62 | 1290 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
34fbcf62 | 1291 | dm_block_t data_begin, |
452d7a62 MS |
1292 | struct dm_thin_new_mapping *m) |
1293 | { | |
1294 | struct pool *pool = tc->pool; | |
1295 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1296 | ||
1297 | h->overwrite_mapping = m; | |
1298 | m->bio = bio; | |
1299 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
1300 | inc_all_io_entry(pool, bio); | |
34fbcf62 | 1301 | remap_and_issue(tc, bio, data_begin); |
452d7a62 MS |
1302 | } |
1303 | ||
e5aea7b4 JT |
1304 | /* |
1305 | * A partial copy also needs to zero the uncopied region. | |
1306 | */ | |
991d9fa0 | 1307 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
1308 | struct dm_dev *origin, dm_block_t data_origin, |
1309 | dm_block_t data_dest, | |
e5aea7b4 JT |
1310 | struct dm_bio_prison_cell *cell, struct bio *bio, |
1311 | sector_t len) | |
991d9fa0 | 1312 | { |
991d9fa0 | 1313 | struct pool *pool = tc->pool; |
a24c2569 | 1314 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1315 | |
991d9fa0 | 1316 | m->tc = tc; |
34fbcf62 JT |
1317 | m->virt_begin = virt_block; |
1318 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1319 | m->data_block = data_dest; |
1320 | m->cell = cell; | |
991d9fa0 | 1321 | |
e5aea7b4 JT |
1322 | /* |
1323 | * quiesce action + copy action + an extra reference held for the | |
1324 | * duration of this function (we may need to inc later for a | |
1325 | * partial zero). | |
1326 | */ | |
1327 | atomic_set(&m->prepare_actions, 3); | |
1328 | ||
44feb387 | 1329 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1330 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1331 | |
1332 | /* | |
1333 | * IO to pool_dev remaps to the pool target's data_dev. | |
1334 | * | |
1335 | * If the whole block of data is being overwritten, we can issue the | |
1336 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1337 | */ | |
452d7a62 MS |
1338 | if (io_overwrites_block(pool, bio)) |
1339 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1340 | else { | |
991d9fa0 JT |
1341 | struct dm_io_region from, to; |
1342 | ||
2dd9c257 | 1343 | from.bdev = origin->bdev; |
991d9fa0 | 1344 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1345 | from.count = len; |
991d9fa0 JT |
1346 | |
1347 | to.bdev = tc->pool_dev->bdev; | |
1348 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1349 | to.count = len; |
991d9fa0 | 1350 | |
7209049d MS |
1351 | dm_kcopyd_copy(pool->copier, &from, 1, &to, |
1352 | 0, copy_complete, m); | |
e5aea7b4 JT |
1353 | |
1354 | /* | |
1355 | * Do we need to zero a tail region? | |
1356 | */ | |
1357 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1358 | atomic_inc(&m->prepare_actions); | |
1359 | ll_zero(tc, m, | |
1360 | data_dest * pool->sectors_per_block + len, | |
1361 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1362 | } |
1363 | } | |
e5aea7b4 JT |
1364 | |
1365 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1366 | } |
1367 | ||
2dd9c257 JT |
1368 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1369 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1370 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1371 | { |
1372 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1373 | data_origin, data_dest, cell, bio, |
1374 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1375 | } |
1376 | ||
991d9fa0 | 1377 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1378 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1379 | struct bio *bio) |
1380 | { | |
1381 | struct pool *pool = tc->pool; | |
a24c2569 | 1382 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1383 | |
50f3c3ef | 1384 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 | 1385 | m->tc = tc; |
34fbcf62 JT |
1386 | m->virt_begin = virt_block; |
1387 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1388 | m->data_block = data_block; |
1389 | m->cell = cell; | |
991d9fa0 JT |
1390 | |
1391 | /* | |
1392 | * If the whole block of data is being overwritten or we are not | |
1393 | * zeroing pre-existing data, we can issue the bio immediately. | |
1394 | * Otherwise we use kcopyd to zero the data first. | |
1395 | */ | |
f8ae7525 MS |
1396 | if (pool->pf.zero_new_blocks) { |
1397 | if (io_overwrites_block(pool, bio)) | |
1398 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
1399 | else | |
1400 | ll_zero(tc, m, data_block * pool->sectors_per_block, | |
1401 | (data_block + 1) * pool->sectors_per_block); | |
1402 | } else | |
991d9fa0 | 1403 | process_prepared_mapping(m); |
e5aea7b4 | 1404 | } |
991d9fa0 | 1405 | |
e5aea7b4 JT |
1406 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1407 | dm_block_t data_dest, | |
1408 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1409 | { | |
1410 | struct pool *pool = tc->pool; | |
1411 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1412 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1413 | ||
1414 | if (virt_block_end <= tc->origin_size) | |
1415 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1416 | virt_block, data_dest, cell, bio, | |
1417 | pool->sectors_per_block); | |
1418 | ||
1419 | else if (virt_block_begin < tc->origin_size) | |
1420 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1421 | virt_block, data_dest, cell, bio, | |
1422 | tc->origin_size - virt_block_begin); | |
1423 | ||
1424 | else | |
1425 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1426 | } |
1427 | ||
2c43fd26 JT |
1428 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1429 | ||
a685557f MS |
1430 | static void requeue_bios(struct pool *pool); |
1431 | ||
3ab91828 JT |
1432 | static bool is_read_only_pool_mode(enum pool_mode mode) |
1433 | { | |
1434 | return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY); | |
1435 | } | |
1436 | ||
1437 | static bool is_read_only(struct pool *pool) | |
1438 | { | |
1439 | return is_read_only_pool_mode(get_pool_mode(pool)); | |
1440 | } | |
1441 | ||
1442 | static void check_for_metadata_space(struct pool *pool) | |
1443 | { | |
1444 | int r; | |
1445 | const char *ooms_reason = NULL; | |
1446 | dm_block_t nr_free; | |
1447 | ||
1448 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free); | |
1449 | if (r) | |
1450 | ooms_reason = "Could not get free metadata blocks"; | |
1451 | else if (!nr_free) | |
1452 | ooms_reason = "No free metadata blocks"; | |
1453 | ||
1454 | if (ooms_reason && !is_read_only(pool)) { | |
1455 | DMERR("%s", ooms_reason); | |
1456 | set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE); | |
1457 | } | |
1458 | } | |
1459 | ||
1460 | static void check_for_data_space(struct pool *pool) | |
2c43fd26 JT |
1461 | { |
1462 | int r; | |
1463 | dm_block_t nr_free; | |
1464 | ||
1465 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) | |
1466 | return; | |
1467 | ||
1468 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); | |
1469 | if (r) | |
1470 | return; | |
1471 | ||
a685557f | 1472 | if (nr_free) { |
2c43fd26 | 1473 | set_pool_mode(pool, PM_WRITE); |
a685557f MS |
1474 | requeue_bios(pool); |
1475 | } | |
2c43fd26 JT |
1476 | } |
1477 | ||
e49e5829 JT |
1478 | /* |
1479 | * A non-zero return indicates read_only or fail_io mode. | |
1480 | * Many callers don't care about the return value. | |
1481 | */ | |
020cc3b5 | 1482 | static int commit(struct pool *pool) |
e49e5829 JT |
1483 | { |
1484 | int r; | |
1485 | ||
3ab91828 | 1486 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) |
e49e5829 JT |
1487 | return -EINVAL; |
1488 | ||
020cc3b5 | 1489 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1490 | if (r) |
1491 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
3ab91828 JT |
1492 | else { |
1493 | check_for_metadata_space(pool); | |
1494 | check_for_data_space(pool); | |
1495 | } | |
e49e5829 JT |
1496 | |
1497 | return r; | |
1498 | } | |
1499 | ||
88a6621b JT |
1500 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1501 | { | |
88a6621b JT |
1502 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
1503 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1504 | dm_device_name(pool->pool_md)); | |
8e0c9dac | 1505 | spin_lock_irq(&pool->lock); |
88a6621b | 1506 | pool->low_water_triggered = true; |
8e0c9dac | 1507 | spin_unlock_irq(&pool->lock); |
88a6621b JT |
1508 | dm_table_event(pool->ti->table); |
1509 | } | |
1510 | } | |
1511 | ||
991d9fa0 JT |
1512 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1513 | { | |
1514 | int r; | |
1515 | dm_block_t free_blocks; | |
991d9fa0 JT |
1516 | struct pool *pool = tc->pool; |
1517 | ||
3e1a0699 | 1518 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1519 | return -EINVAL; |
1520 | ||
991d9fa0 | 1521 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1522 | if (r) { |
1523 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1524 | return r; |
b5330655 | 1525 | } |
991d9fa0 | 1526 | |
88a6621b | 1527 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1528 | |
1529 | if (!free_blocks) { | |
94563bad MS |
1530 | /* |
1531 | * Try to commit to see if that will free up some | |
1532 | * more space. | |
1533 | */ | |
020cc3b5 JT |
1534 | r = commit(pool); |
1535 | if (r) | |
1536 | return r; | |
991d9fa0 | 1537 | |
94563bad | 1538 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1539 | if (r) { |
1540 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1541 | return r; |
b5330655 | 1542 | } |
991d9fa0 | 1543 | |
94563bad | 1544 | if (!free_blocks) { |
3e1a0699 | 1545 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1546 | return -ENOSPC; |
991d9fa0 JT |
1547 | } |
1548 | } | |
1549 | ||
1550 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1551 | if (r) { |
a685557f MS |
1552 | if (r == -ENOSPC) |
1553 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); | |
1554 | else | |
1555 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); | |
991d9fa0 | 1556 | return r; |
4a02b34e | 1557 | } |
991d9fa0 | 1558 | |
3ab91828 JT |
1559 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks); |
1560 | if (r) { | |
1561 | metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r); | |
1562 | return r; | |
1563 | } | |
1564 | ||
1565 | if (!free_blocks) { | |
1566 | /* Let's commit before we use up the metadata reserve. */ | |
1567 | r = commit(pool); | |
1568 | if (r) | |
1569 | return r; | |
1570 | } | |
1571 | ||
991d9fa0 JT |
1572 | return 0; |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * If we have run out of space, queue bios until the device is | |
1577 | * resumed, presumably after having been reloaded with more space. | |
1578 | */ | |
1579 | static void retry_on_resume(struct bio *bio) | |
1580 | { | |
59c3d2c6 | 1581 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1582 | struct thin_c *tc = h->tc; |
991d9fa0 | 1583 | |
8e0c9dac | 1584 | spin_lock_irq(&tc->lock); |
c140e1c4 | 1585 | bio_list_add(&tc->retry_on_resume_list, bio); |
8e0c9dac | 1586 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
1587 | } |
1588 | ||
4e4cbee9 | 1589 | static blk_status_t should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1590 | { |
3e1a0699 JT |
1591 | enum pool_mode m = get_pool_mode(pool); |
1592 | ||
1593 | switch (m) { | |
1594 | case PM_WRITE: | |
1595 | /* Shouldn't get here */ | |
1596 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
4e4cbee9 | 1597 | return BLK_STS_IOERR; |
3e1a0699 JT |
1598 | |
1599 | case PM_OUT_OF_DATA_SPACE: | |
4e4cbee9 | 1600 | return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0; |
3e1a0699 | 1601 | |
3ab91828 | 1602 | case PM_OUT_OF_METADATA_SPACE: |
3e1a0699 JT |
1603 | case PM_READ_ONLY: |
1604 | case PM_FAIL: | |
4e4cbee9 | 1605 | return BLK_STS_IOERR; |
3e1a0699 JT |
1606 | default: |
1607 | /* Shouldn't get here */ | |
1608 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
4e4cbee9 | 1609 | return BLK_STS_IOERR; |
3e1a0699 JT |
1610 | } |
1611 | } | |
8c0f0e8c | 1612 | |
3e1a0699 JT |
1613 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1614 | { | |
4e4cbee9 | 1615 | blk_status_t error = should_error_unserviceable_bio(pool); |
af91805a | 1616 | |
4246a0b6 | 1617 | if (error) { |
4e4cbee9 | 1618 | bio->bi_status = error; |
4246a0b6 CH |
1619 | bio_endio(bio); |
1620 | } else | |
6d16202b | 1621 | retry_on_resume(bio); |
8c0f0e8c MS |
1622 | } |
1623 | ||
399caddf | 1624 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1625 | { |
1626 | struct bio *bio; | |
1627 | struct bio_list bios; | |
4e4cbee9 | 1628 | blk_status_t error; |
991d9fa0 | 1629 | |
af91805a MS |
1630 | error = should_error_unserviceable_bio(pool); |
1631 | if (error) { | |
1632 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1633 | return; |
1634 | } | |
1635 | ||
991d9fa0 | 1636 | bio_list_init(&bios); |
6beca5eb | 1637 | cell_release(pool, cell, &bios); |
991d9fa0 | 1638 | |
9d094eeb MS |
1639 | while ((bio = bio_list_pop(&bios))) |
1640 | retry_on_resume(bio); | |
991d9fa0 JT |
1641 | } |
1642 | ||
34fbcf62 JT |
1643 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
1644 | struct dm_bio_prison_cell *virt_cell) | |
104655fd | 1645 | { |
104655fd | 1646 | struct pool *pool = tc->pool; |
34fbcf62 | 1647 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
104655fd | 1648 | |
34fbcf62 JT |
1649 | /* |
1650 | * We don't need to lock the data blocks, since there's no | |
1651 | * passdown. We only lock data blocks for allocation and breaking sharing. | |
1652 | */ | |
1653 | m->tc = tc; | |
1654 | m->virt_begin = virt_cell->key.block_begin; | |
1655 | m->virt_end = virt_cell->key.block_end; | |
1656 | m->cell = virt_cell; | |
1657 | m->bio = virt_cell->holder; | |
104655fd | 1658 | |
34fbcf62 JT |
1659 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1660 | pool->process_prepared_discard(m); | |
1661 | } | |
104655fd | 1662 | |
34fbcf62 JT |
1663 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
1664 | struct bio *bio) | |
1665 | { | |
1666 | struct pool *pool = tc->pool; | |
1667 | ||
1668 | int r; | |
1669 | bool maybe_shared; | |
1670 | struct dm_cell_key data_key; | |
1671 | struct dm_bio_prison_cell *data_cell; | |
1672 | struct dm_thin_new_mapping *m; | |
1673 | dm_block_t virt_begin, virt_end, data_begin; | |
1674 | ||
1675 | while (begin != end) { | |
1676 | r = ensure_next_mapping(pool); | |
1677 | if (r) | |
1678 | /* we did our best */ | |
1679 | return; | |
e8088073 | 1680 | |
34fbcf62 JT |
1681 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
1682 | &data_begin, &maybe_shared); | |
1683 | if (r) | |
104655fd | 1684 | /* |
34fbcf62 JT |
1685 | * Silently fail, letting any mappings we've |
1686 | * created complete. | |
104655fd | 1687 | */ |
34fbcf62 JT |
1688 | break; |
1689 | ||
1690 | build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); | |
1691 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { | |
1692 | /* contention, we'll give up with this range */ | |
1693 | begin = virt_end; | |
1694 | continue; | |
104655fd | 1695 | } |
104655fd | 1696 | |
104655fd | 1697 | /* |
34fbcf62 JT |
1698 | * IO may still be going to the destination block. We must |
1699 | * quiesce before we can do the removal. | |
104655fd | 1700 | */ |
34fbcf62 JT |
1701 | m = get_next_mapping(pool); |
1702 | m->tc = tc; | |
1703 | m->maybe_shared = maybe_shared; | |
1704 | m->virt_begin = virt_begin; | |
1705 | m->virt_end = virt_end; | |
1706 | m->data_block = data_begin; | |
1707 | m->cell = data_cell; | |
1708 | m->bio = bio; | |
104655fd | 1709 | |
34fbcf62 JT |
1710 | /* |
1711 | * The parent bio must not complete before sub discard bios are | |
202bae52 | 1712 | * chained to it (see end_discard's bio_chain)! |
34fbcf62 JT |
1713 | * |
1714 | * This per-mapping bi_remaining increment is paired with | |
1715 | * the implicit decrement that occurs via bio_endio() in | |
202bae52 | 1716 | * end_discard(). |
34fbcf62 | 1717 | */ |
13e4f8a6 | 1718 | bio_inc_remaining(bio); |
34fbcf62 JT |
1719 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1720 | pool->process_prepared_discard(m); | |
1721 | ||
1722 | begin = virt_end; | |
104655fd JT |
1723 | } |
1724 | } | |
1725 | ||
34fbcf62 JT |
1726 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
1727 | { | |
1728 | struct bio *bio = virt_cell->holder; | |
1729 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1730 | ||
1731 | /* | |
1732 | * The virt_cell will only get freed once the origin bio completes. | |
1733 | * This means it will remain locked while all the individual | |
1734 | * passdown bios are in flight. | |
1735 | */ | |
1736 | h->cell = virt_cell; | |
1737 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); | |
1738 | ||
1739 | /* | |
1740 | * We complete the bio now, knowing that the bi_remaining field | |
1741 | * will prevent completion until the sub range discards have | |
1742 | * completed. | |
1743 | */ | |
4246a0b6 | 1744 | bio_endio(bio); |
34fbcf62 JT |
1745 | } |
1746 | ||
a374bb21 JT |
1747 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1748 | { | |
34fbcf62 JT |
1749 | dm_block_t begin, end; |
1750 | struct dm_cell_key virt_key; | |
1751 | struct dm_bio_prison_cell *virt_cell; | |
a374bb21 | 1752 | |
34fbcf62 JT |
1753 | get_bio_block_range(tc, bio, &begin, &end); |
1754 | if (begin == end) { | |
1755 | /* | |
1756 | * The discard covers less than a block. | |
1757 | */ | |
4246a0b6 | 1758 | bio_endio(bio); |
a374bb21 | 1759 | return; |
34fbcf62 | 1760 | } |
a374bb21 | 1761 | |
34fbcf62 JT |
1762 | build_key(tc->td, VIRTUAL, begin, end, &virt_key); |
1763 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) | |
1764 | /* | |
1765 | * Potential starvation issue: We're relying on the | |
1766 | * fs/application being well behaved, and not trying to | |
1767 | * send IO to a region at the same time as discarding it. | |
1768 | * If they do this persistently then it's possible this | |
1769 | * cell will never be granted. | |
1770 | */ | |
1771 | return; | |
1772 | ||
1773 | tc->pool->process_discard_cell(tc, virt_cell); | |
a374bb21 JT |
1774 | } |
1775 | ||
991d9fa0 | 1776 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1777 | struct dm_cell_key *key, |
991d9fa0 | 1778 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1779 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1780 | { |
1781 | int r; | |
1782 | dm_block_t data_block; | |
d6fc2042 | 1783 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1784 | |
1785 | r = alloc_data_block(tc, &data_block); | |
1786 | switch (r) { | |
1787 | case 0: | |
2dd9c257 JT |
1788 | schedule_internal_copy(tc, block, lookup_result->block, |
1789 | data_block, cell, bio); | |
991d9fa0 JT |
1790 | break; |
1791 | ||
1792 | case -ENOSPC: | |
399caddf | 1793 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1794 | break; |
1795 | ||
1796 | default: | |
c397741c MS |
1797 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1798 | __func__, r); | |
d6fc2042 | 1799 | cell_error(pool, cell); |
991d9fa0 JT |
1800 | break; |
1801 | } | |
1802 | } | |
1803 | ||
23ca2bb6 JT |
1804 | static void __remap_and_issue_shared_cell(void *context, |
1805 | struct dm_bio_prison_cell *cell) | |
1806 | { | |
1807 | struct remap_info *info = context; | |
1808 | struct bio *bio; | |
1809 | ||
1810 | while ((bio = bio_list_pop(&cell->bios))) { | |
f73f44eb CH |
1811 | if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) || |
1812 | bio_op(bio) == REQ_OP_DISCARD) | |
23ca2bb6 JT |
1813 | bio_list_add(&info->defer_bios, bio); |
1814 | else { | |
bd6d1e0a | 1815 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
23ca2bb6 JT |
1816 | |
1817 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1818 | inc_all_io_entry(info->tc->pool, bio); | |
1819 | bio_list_add(&info->issue_bios, bio); | |
1820 | } | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1825 | struct dm_bio_prison_cell *cell, | |
1826 | dm_block_t block) | |
1827 | { | |
1828 | struct bio *bio; | |
1829 | struct remap_info info; | |
1830 | ||
1831 | info.tc = tc; | |
1832 | bio_list_init(&info.defer_bios); | |
1833 | bio_list_init(&info.issue_bios); | |
1834 | ||
1835 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1836 | &info, cell); | |
1837 | ||
1838 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1839 | thin_defer_bio(tc, bio); | |
1840 | ||
1841 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1842 | remap_and_issue(tc, bio, block); | |
1843 | } | |
1844 | ||
991d9fa0 JT |
1845 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1846 | dm_block_t block, | |
23ca2bb6 JT |
1847 | struct dm_thin_lookup_result *lookup_result, |
1848 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1849 | { |
23ca2bb6 | 1850 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1851 | struct pool *pool = tc->pool; |
44feb387 | 1852 | struct dm_cell_key key; |
991d9fa0 JT |
1853 | |
1854 | /* | |
1855 | * If cell is already occupied, then sharing is already in the process | |
1856 | * of being broken so we have nothing further to do here. | |
1857 | */ | |
1858 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1859 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1860 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1861 | return; |
23ca2bb6 | 1862 | } |
991d9fa0 | 1863 | |
23ca2bb6 JT |
1864 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1865 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1866 | cell_defer_no_holder(tc, virt_cell); | |
1867 | } else { | |
59c3d2c6 | 1868 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1869 | |
44feb387 | 1870 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1871 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1872 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1873 | |
1874 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1875 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1876 | } |
1877 | } | |
1878 | ||
1879 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1880 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1881 | { |
1882 | int r; | |
1883 | dm_block_t data_block; | |
6beca5eb | 1884 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1885 | |
1886 | /* | |
1887 | * Remap empty bios (flushes) immediately, without provisioning. | |
1888 | */ | |
4f024f37 | 1889 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1890 | inc_all_io_entry(pool, bio); |
f286ba0e | 1891 | cell_defer_no_holder(tc, cell); |
e8088073 | 1892 | |
991d9fa0 JT |
1893 | remap_and_issue(tc, bio, 0); |
1894 | return; | |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * Fill read bios with zeroes and complete them immediately. | |
1899 | */ | |
1900 | if (bio_data_dir(bio) == READ) { | |
1901 | zero_fill_bio(bio); | |
f286ba0e | 1902 | cell_defer_no_holder(tc, cell); |
4246a0b6 | 1903 | bio_endio(bio); |
991d9fa0 JT |
1904 | return; |
1905 | } | |
1906 | ||
1907 | r = alloc_data_block(tc, &data_block); | |
1908 | switch (r) { | |
1909 | case 0: | |
2dd9c257 JT |
1910 | if (tc->origin_dev) |
1911 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1912 | else | |
1913 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1914 | break; |
1915 | ||
1916 | case -ENOSPC: | |
399caddf | 1917 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1918 | break; |
1919 | ||
1920 | default: | |
c397741c MS |
1921 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1922 | __func__, r); | |
6beca5eb | 1923 | cell_error(pool, cell); |
991d9fa0 JT |
1924 | break; |
1925 | } | |
1926 | } | |
1927 | ||
a374bb21 | 1928 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1929 | { |
1930 | int r; | |
6beca5eb | 1931 | struct pool *pool = tc->pool; |
a374bb21 | 1932 | struct bio *bio = cell->holder; |
991d9fa0 | 1933 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1934 | struct dm_thin_lookup_result lookup_result; |
1935 | ||
a374bb21 JT |
1936 | if (tc->requeue_mode) { |
1937 | cell_requeue(pool, cell); | |
991d9fa0 | 1938 | return; |
a374bb21 | 1939 | } |
991d9fa0 JT |
1940 | |
1941 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1942 | switch (r) { | |
1943 | case 0: | |
23ca2bb6 JT |
1944 | if (lookup_result.shared) |
1945 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1946 | else { | |
6beca5eb | 1947 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1948 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1949 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1950 | } |
991d9fa0 JT |
1951 | break; |
1952 | ||
1953 | case -ENODATA: | |
2dd9c257 | 1954 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1955 | inc_all_io_entry(pool, bio); |
f286ba0e | 1956 | cell_defer_no_holder(tc, cell); |
e8088073 | 1957 | |
e5aea7b4 JT |
1958 | if (bio_end_sector(bio) <= tc->origin_size) |
1959 | remap_to_origin_and_issue(tc, bio); | |
1960 | ||
1961 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1962 | zero_fill_bio(bio); | |
1963 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1964 | remap_to_origin_and_issue(tc, bio); | |
1965 | ||
1966 | } else { | |
1967 | zero_fill_bio(bio); | |
4246a0b6 | 1968 | bio_endio(bio); |
e5aea7b4 | 1969 | } |
2dd9c257 JT |
1970 | } else |
1971 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1972 | break; |
1973 | ||
1974 | default: | |
c397741c MS |
1975 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1976 | __func__, r); | |
f286ba0e | 1977 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1978 | bio_io_error(bio); |
1979 | break; | |
1980 | } | |
1981 | } | |
1982 | ||
a374bb21 JT |
1983 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1984 | { | |
1985 | struct pool *pool = tc->pool; | |
1986 | dm_block_t block = get_bio_block(tc, bio); | |
1987 | struct dm_bio_prison_cell *cell; | |
1988 | struct dm_cell_key key; | |
1989 | ||
1990 | /* | |
1991 | * If cell is already occupied, then the block is already | |
1992 | * being provisioned so we have nothing further to do here. | |
1993 | */ | |
1994 | build_virtual_key(tc->td, block, &key); | |
1995 | if (bio_detain(pool, &key, bio, &cell)) | |
1996 | return; | |
1997 | ||
1998 | process_cell(tc, cell); | |
1999 | } | |
2000 | ||
2001 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
2002 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
2003 | { |
2004 | int r; | |
2005 | int rw = bio_data_dir(bio); | |
2006 | dm_block_t block = get_bio_block(tc, bio); | |
2007 | struct dm_thin_lookup_result lookup_result; | |
2008 | ||
2009 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
2010 | switch (r) { | |
2011 | case 0: | |
a374bb21 | 2012 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 2013 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
2014 | if (cell) |
2015 | cell_defer_no_holder(tc, cell); | |
2016 | } else { | |
e8088073 | 2017 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 2018 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
2019 | if (cell) |
2020 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 2021 | } |
e49e5829 JT |
2022 | break; |
2023 | ||
2024 | case -ENODATA: | |
a374bb21 JT |
2025 | if (cell) |
2026 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 2027 | if (rw != READ) { |
8c0f0e8c | 2028 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
2029 | break; |
2030 | } | |
2031 | ||
2032 | if (tc->origin_dev) { | |
e8088073 | 2033 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
2034 | remap_to_origin_and_issue(tc, bio); |
2035 | break; | |
2036 | } | |
2037 | ||
2038 | zero_fill_bio(bio); | |
4246a0b6 | 2039 | bio_endio(bio); |
e49e5829 JT |
2040 | break; |
2041 | ||
2042 | default: | |
c397741c MS |
2043 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
2044 | __func__, r); | |
a374bb21 JT |
2045 | if (cell) |
2046 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
2047 | bio_io_error(bio); |
2048 | break; | |
2049 | } | |
2050 | } | |
2051 | ||
a374bb21 JT |
2052 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
2053 | { | |
2054 | __process_bio_read_only(tc, bio, NULL); | |
2055 | } | |
2056 | ||
2057 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2058 | { | |
2059 | __process_bio_read_only(tc, cell->holder, cell); | |
2060 | } | |
2061 | ||
3e1a0699 JT |
2062 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
2063 | { | |
4246a0b6 | 2064 | bio_endio(bio); |
3e1a0699 JT |
2065 | } |
2066 | ||
e49e5829 JT |
2067 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
2068 | { | |
2069 | bio_io_error(bio); | |
2070 | } | |
2071 | ||
a374bb21 JT |
2072 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2073 | { | |
2074 | cell_success(tc->pool, cell); | |
2075 | } | |
2076 | ||
2077 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2078 | { | |
2079 | cell_error(tc->pool, cell); | |
2080 | } | |
2081 | ||
ac8c3f3d JT |
2082 | /* |
2083 | * FIXME: should we also commit due to size of transaction, measured in | |
2084 | * metadata blocks? | |
2085 | */ | |
905e51b3 JT |
2086 | static int need_commit_due_to_time(struct pool *pool) |
2087 | { | |
0f30af98 MS |
2088 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
2089 | pool->last_commit_jiffies + COMMIT_PERIOD); | |
905e51b3 JT |
2090 | } |
2091 | ||
67324ea1 MS |
2092 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
2093 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
2094 | ||
2095 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
2096 | { | |
2097 | struct rb_node **rbp, *parent; | |
2098 | struct dm_thin_endio_hook *pbd; | |
2099 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
2100 | ||
2101 | rbp = &tc->sort_bio_list.rb_node; | |
2102 | parent = NULL; | |
2103 | while (*rbp) { | |
2104 | parent = *rbp; | |
2105 | pbd = thin_pbd(parent); | |
2106 | ||
2107 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
2108 | rbp = &(*rbp)->rb_left; | |
2109 | else | |
2110 | rbp = &(*rbp)->rb_right; | |
2111 | } | |
2112 | ||
2113 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
2114 | rb_link_node(&pbd->rb_node, parent, rbp); | |
2115 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
2116 | } | |
2117 | ||
2118 | static void __extract_sorted_bios(struct thin_c *tc) | |
2119 | { | |
2120 | struct rb_node *node; | |
2121 | struct dm_thin_endio_hook *pbd; | |
2122 | struct bio *bio; | |
2123 | ||
2124 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
2125 | pbd = thin_pbd(node); | |
2126 | bio = thin_bio(pbd); | |
2127 | ||
2128 | bio_list_add(&tc->deferred_bio_list, bio); | |
2129 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
2130 | } | |
2131 | ||
2132 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
2133 | } | |
2134 | ||
2135 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
2136 | { | |
2137 | struct bio *bio; | |
2138 | struct bio_list bios; | |
2139 | ||
2140 | bio_list_init(&bios); | |
2141 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
2142 | bio_list_init(&tc->deferred_bio_list); | |
2143 | ||
2144 | /* Sort deferred_bio_list using rb-tree */ | |
2145 | while ((bio = bio_list_pop(&bios))) | |
2146 | __thin_bio_rb_add(tc, bio); | |
2147 | ||
2148 | /* | |
2149 | * Transfer the sorted bios in sort_bio_list back to | |
2150 | * deferred_bio_list to allow lockless submission of | |
2151 | * all bios. | |
2152 | */ | |
2153 | __extract_sorted_bios(tc); | |
2154 | } | |
2155 | ||
c140e1c4 | 2156 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 2157 | { |
c140e1c4 | 2158 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2159 | struct bio *bio; |
2160 | struct bio_list bios; | |
67324ea1 | 2161 | struct blk_plug plug; |
8a01a6af | 2162 | unsigned count = 0; |
991d9fa0 | 2163 | |
c140e1c4 | 2164 | if (tc->requeue_mode) { |
4e4cbee9 CH |
2165 | error_thin_bio_list(tc, &tc->deferred_bio_list, |
2166 | BLK_STS_DM_REQUEUE); | |
c140e1c4 MS |
2167 | return; |
2168 | } | |
2169 | ||
991d9fa0 JT |
2170 | bio_list_init(&bios); |
2171 | ||
8e0c9dac | 2172 | spin_lock_irq(&tc->lock); |
67324ea1 MS |
2173 | |
2174 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
8e0c9dac | 2175 | spin_unlock_irq(&tc->lock); |
67324ea1 MS |
2176 | return; |
2177 | } | |
2178 | ||
2179 | __sort_thin_deferred_bios(tc); | |
2180 | ||
c140e1c4 MS |
2181 | bio_list_merge(&bios, &tc->deferred_bio_list); |
2182 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 2183 | |
8e0c9dac | 2184 | spin_unlock_irq(&tc->lock); |
991d9fa0 | 2185 | |
67324ea1 | 2186 | blk_start_plug(&plug); |
991d9fa0 | 2187 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
2188 | /* |
2189 | * If we've got no free new_mapping structs, and processing | |
2190 | * this bio might require one, we pause until there are some | |
2191 | * prepared mappings to process. | |
2192 | */ | |
2193 | if (ensure_next_mapping(pool)) { | |
8e0c9dac | 2194 | spin_lock_irq(&tc->lock); |
c140e1c4 MS |
2195 | bio_list_add(&tc->deferred_bio_list, bio); |
2196 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
8e0c9dac | 2197 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
2198 | break; |
2199 | } | |
104655fd | 2200 | |
e6047149 | 2201 | if (bio_op(bio) == REQ_OP_DISCARD) |
e49e5829 | 2202 | pool->process_discard(tc, bio); |
104655fd | 2203 | else |
e49e5829 | 2204 | pool->process_bio(tc, bio); |
8a01a6af JT |
2205 | |
2206 | if ((count++ & 127) == 0) { | |
7d327fe0 | 2207 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
2208 | dm_pool_issue_prefetches(pool->pmd); |
2209 | } | |
991d9fa0 | 2210 | } |
67324ea1 | 2211 | blk_finish_plug(&plug); |
c140e1c4 MS |
2212 | } |
2213 | ||
ac4c3f34 JT |
2214 | static int cmp_cells(const void *lhs, const void *rhs) |
2215 | { | |
2216 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
2217 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
2218 | ||
2219 | BUG_ON(!lhs_cell->holder); | |
2220 | BUG_ON(!rhs_cell->holder); | |
2221 | ||
2222 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
2223 | return -1; | |
2224 | ||
2225 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
2226 | return 1; | |
2227 | ||
2228 | return 0; | |
2229 | } | |
2230 | ||
2231 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) | |
2232 | { | |
2233 | unsigned count = 0; | |
2234 | struct dm_bio_prison_cell *cell, *tmp; | |
2235 | ||
2236 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
2237 | if (count >= CELL_SORT_ARRAY_SIZE) | |
2238 | break; | |
2239 | ||
2240 | pool->cell_sort_array[count++] = cell; | |
2241 | list_del(&cell->user_list); | |
2242 | } | |
2243 | ||
2244 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
2245 | ||
2246 | return count; | |
2247 | } | |
2248 | ||
a374bb21 JT |
2249 | static void process_thin_deferred_cells(struct thin_c *tc) |
2250 | { | |
2251 | struct pool *pool = tc->pool; | |
a374bb21 | 2252 | struct list_head cells; |
ac4c3f34 JT |
2253 | struct dm_bio_prison_cell *cell; |
2254 | unsigned i, j, count; | |
a374bb21 JT |
2255 | |
2256 | INIT_LIST_HEAD(&cells); | |
2257 | ||
8e0c9dac | 2258 | spin_lock_irq(&tc->lock); |
a374bb21 | 2259 | list_splice_init(&tc->deferred_cells, &cells); |
8e0c9dac | 2260 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
2261 | |
2262 | if (list_empty(&cells)) | |
2263 | return; | |
2264 | ||
ac4c3f34 JT |
2265 | do { |
2266 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 2267 | |
ac4c3f34 JT |
2268 | for (i = 0; i < count; i++) { |
2269 | cell = pool->cell_sort_array[i]; | |
2270 | BUG_ON(!cell->holder); | |
a374bb21 | 2271 | |
ac4c3f34 JT |
2272 | /* |
2273 | * If we've got no free new_mapping structs, and processing | |
2274 | * this bio might require one, we pause until there are some | |
2275 | * prepared mappings to process. | |
2276 | */ | |
2277 | if (ensure_next_mapping(pool)) { | |
2278 | for (j = i; j < count; j++) | |
2279 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
2280 | ||
8e0c9dac | 2281 | spin_lock_irq(&tc->lock); |
ac4c3f34 | 2282 | list_splice(&cells, &tc->deferred_cells); |
8e0c9dac | 2283 | spin_unlock_irq(&tc->lock); |
ac4c3f34 JT |
2284 | return; |
2285 | } | |
2286 | ||
e6047149 | 2287 | if (bio_op(cell->holder) == REQ_OP_DISCARD) |
ac4c3f34 JT |
2288 | pool->process_discard_cell(tc, cell); |
2289 | else | |
2290 | pool->process_cell(tc, cell); | |
2291 | } | |
2292 | } while (!list_empty(&cells)); | |
a374bb21 JT |
2293 | } |
2294 | ||
b10ebd34 JT |
2295 | static void thin_get(struct thin_c *tc); |
2296 | static void thin_put(struct thin_c *tc); | |
2297 | ||
2298 | /* | |
2299 | * We can't hold rcu_read_lock() around code that can block. So we | |
2300 | * find a thin with the rcu lock held; bump a refcount; then drop | |
2301 | * the lock. | |
2302 | */ | |
2303 | static struct thin_c *get_first_thin(struct pool *pool) | |
2304 | { | |
2305 | struct thin_c *tc = NULL; | |
2306 | ||
2307 | rcu_read_lock(); | |
2308 | if (!list_empty(&pool->active_thins)) { | |
2309 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
2310 | thin_get(tc); | |
2311 | } | |
2312 | rcu_read_unlock(); | |
2313 | ||
2314 | return tc; | |
2315 | } | |
2316 | ||
2317 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
2318 | { | |
2319 | struct thin_c *old_tc = tc; | |
2320 | ||
2321 | rcu_read_lock(); | |
2322 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
2323 | thin_get(tc); | |
2324 | thin_put(old_tc); | |
2325 | rcu_read_unlock(); | |
2326 | return tc; | |
2327 | } | |
2328 | thin_put(old_tc); | |
2329 | rcu_read_unlock(); | |
2330 | ||
2331 | return NULL; | |
2332 | } | |
2333 | ||
c140e1c4 MS |
2334 | static void process_deferred_bios(struct pool *pool) |
2335 | { | |
c140e1c4 | 2336 | struct bio *bio; |
4ae280b4 | 2337 | struct bio_list bios, bio_completions; |
c140e1c4 MS |
2338 | struct thin_c *tc; |
2339 | ||
b10ebd34 JT |
2340 | tc = get_first_thin(pool); |
2341 | while (tc) { | |
a374bb21 | 2342 | process_thin_deferred_cells(tc); |
c140e1c4 | 2343 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
2344 | tc = get_next_thin(pool, tc); |
2345 | } | |
991d9fa0 JT |
2346 | |
2347 | /* | |
4ae280b4 NT |
2348 | * If there are any deferred flush bios, we must commit the metadata |
2349 | * before issuing them or signaling their completion. | |
991d9fa0 JT |
2350 | */ |
2351 | bio_list_init(&bios); | |
4ae280b4 NT |
2352 | bio_list_init(&bio_completions); |
2353 | ||
8e0c9dac | 2354 | spin_lock_irq(&pool->lock); |
991d9fa0 JT |
2355 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
2356 | bio_list_init(&pool->deferred_flush_bios); | |
4ae280b4 NT |
2357 | |
2358 | bio_list_merge(&bio_completions, &pool->deferred_flush_completions); | |
2359 | bio_list_init(&pool->deferred_flush_completions); | |
8e0c9dac | 2360 | spin_unlock_irq(&pool->lock); |
991d9fa0 | 2361 | |
4ae280b4 | 2362 | if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) && |
4d1662a3 | 2363 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) |
991d9fa0 JT |
2364 | return; |
2365 | ||
020cc3b5 | 2366 | if (commit(pool)) { |
4ae280b4 NT |
2367 | bio_list_merge(&bios, &bio_completions); |
2368 | ||
991d9fa0 JT |
2369 | while ((bio = bio_list_pop(&bios))) |
2370 | bio_io_error(bio); | |
2371 | return; | |
2372 | } | |
905e51b3 | 2373 | pool->last_commit_jiffies = jiffies; |
991d9fa0 | 2374 | |
4ae280b4 NT |
2375 | while ((bio = bio_list_pop(&bio_completions))) |
2376 | bio_endio(bio); | |
2377 | ||
694cfe7f NT |
2378 | while ((bio = bio_list_pop(&bios))) { |
2379 | /* | |
2380 | * The data device was flushed as part of metadata commit, | |
2381 | * so complete redundant flushes immediately. | |
2382 | */ | |
2383 | if (bio->bi_opf & REQ_PREFLUSH) | |
2384 | bio_endio(bio); | |
2385 | else | |
ed00aabd | 2386 | submit_bio_noacct(bio); |
694cfe7f | 2387 | } |
991d9fa0 JT |
2388 | } |
2389 | ||
2390 | static void do_worker(struct work_struct *ws) | |
2391 | { | |
2392 | struct pool *pool = container_of(ws, struct pool, worker); | |
2393 | ||
7d327fe0 | 2394 | throttle_work_start(&pool->throttle); |
8a01a6af | 2395 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 2396 | throttle_work_update(&pool->throttle); |
e49e5829 | 2397 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 2398 | throttle_work_update(&pool->throttle); |
e49e5829 | 2399 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 2400 | throttle_work_update(&pool->throttle); |
2a0fbffb JT |
2401 | process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2); |
2402 | throttle_work_update(&pool->throttle); | |
991d9fa0 | 2403 | process_deferred_bios(pool); |
7d327fe0 | 2404 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
2405 | } |
2406 | ||
905e51b3 JT |
2407 | /* |
2408 | * We want to commit periodically so that not too much | |
2409 | * unwritten data builds up. | |
2410 | */ | |
2411 | static void do_waker(struct work_struct *ws) | |
2412 | { | |
2413 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
2414 | wake_worker(pool); | |
2415 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
2416 | } | |
2417 | ||
85ad643b JT |
2418 | /* |
2419 | * We're holding onto IO to allow userland time to react. After the | |
2420 | * timeout either the pool will have been resized (and thus back in | |
bcc696fa | 2421 | * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space. |
85ad643b JT |
2422 | */ |
2423 | static void do_no_space_timeout(struct work_struct *ws) | |
2424 | { | |
2425 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2426 | no_space_timeout); | |
2427 | ||
bcc696fa MS |
2428 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) { |
2429 | pool->pf.error_if_no_space = true; | |
f6c36758 | 2430 | notify_of_pool_mode_change(pool); |
4e4cbee9 | 2431 | error_retry_list_with_code(pool, BLK_STS_NOSPC); |
bcc696fa | 2432 | } |
85ad643b JT |
2433 | } |
2434 | ||
991d9fa0 JT |
2435 | /*----------------------------------------------------------------*/ |
2436 | ||
e7a3e871 | 2437 | struct pool_work { |
738211f7 | 2438 | struct work_struct worker; |
e7a3e871 JT |
2439 | struct completion complete; |
2440 | }; | |
2441 | ||
2442 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2443 | { | |
2444 | return container_of(ws, struct pool_work, worker); | |
2445 | } | |
2446 | ||
2447 | static void pool_work_complete(struct pool_work *pw) | |
2448 | { | |
2449 | complete(&pw->complete); | |
2450 | } | |
738211f7 | 2451 | |
e7a3e871 JT |
2452 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2453 | void (*fn)(struct work_struct *)) | |
2454 | { | |
2455 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2456 | init_completion(&pw->complete); | |
2457 | queue_work(pool->wq, &pw->worker); | |
2458 | wait_for_completion(&pw->complete); | |
2459 | } | |
2460 | ||
2461 | /*----------------------------------------------------------------*/ | |
2462 | ||
2463 | struct noflush_work { | |
2464 | struct pool_work pw; | |
2465 | struct thin_c *tc; | |
738211f7 JT |
2466 | }; |
2467 | ||
e7a3e871 | 2468 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2469 | { |
e7a3e871 | 2470 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2471 | } |
2472 | ||
2473 | static void do_noflush_start(struct work_struct *ws) | |
2474 | { | |
e7a3e871 | 2475 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2476 | w->tc->requeue_mode = true; |
2477 | requeue_io(w->tc); | |
e7a3e871 | 2478 | pool_work_complete(&w->pw); |
738211f7 JT |
2479 | } |
2480 | ||
2481 | static void do_noflush_stop(struct work_struct *ws) | |
2482 | { | |
e7a3e871 | 2483 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2484 | w->tc->requeue_mode = false; |
e7a3e871 | 2485 | pool_work_complete(&w->pw); |
738211f7 JT |
2486 | } |
2487 | ||
2488 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2489 | { | |
2490 | struct noflush_work w; | |
2491 | ||
738211f7 | 2492 | w.tc = tc; |
e7a3e871 | 2493 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2494 | } |
2495 | ||
2496 | /*----------------------------------------------------------------*/ | |
2497 | ||
34fbcf62 JT |
2498 | static bool passdown_enabled(struct pool_c *pt) |
2499 | { | |
2500 | return pt->adjusted_pf.discard_passdown; | |
2501 | } | |
2502 | ||
2503 | static void set_discard_callbacks(struct pool *pool) | |
2504 | { | |
2505 | struct pool_c *pt = pool->ti->private; | |
2506 | ||
2507 | if (passdown_enabled(pt)) { | |
2508 | pool->process_discard_cell = process_discard_cell_passdown; | |
2a0fbffb JT |
2509 | pool->process_prepared_discard = process_prepared_discard_passdown_pt1; |
2510 | pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2; | |
34fbcf62 JT |
2511 | } else { |
2512 | pool->process_discard_cell = process_discard_cell_no_passdown; | |
2513 | pool->process_prepared_discard = process_prepared_discard_no_passdown; | |
2514 | } | |
2515 | } | |
2516 | ||
8b64e881 | 2517 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2518 | { |
cdc2b415 | 2519 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2520 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2521 | enum pool_mode old_mode = get_pool_mode(pool); | |
6aa7de05 | 2522 | unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2523 | |
2524 | /* | |
2525 | * Never allow the pool to transition to PM_WRITE mode if user | |
2526 | * intervention is required to verify metadata and data consistency. | |
2527 | */ | |
2528 | if (new_mode == PM_WRITE && needs_check) { | |
2529 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2530 | dm_device_name(pool->pool_md)); | |
2531 | if (old_mode != new_mode) | |
2532 | new_mode = old_mode; | |
2533 | else | |
2534 | new_mode = PM_READ_ONLY; | |
2535 | } | |
2536 | /* | |
2537 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2538 | * not going to recover without a thin_repair. So we never let the | |
2539 | * pool move out of the old mode. | |
2540 | */ | |
2541 | if (old_mode == PM_FAIL) | |
2542 | new_mode = old_mode; | |
e49e5829 | 2543 | |
8b64e881 | 2544 | switch (new_mode) { |
e49e5829 | 2545 | case PM_FAIL: |
5383ef3a | 2546 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2547 | pool->process_bio = process_bio_fail; |
2548 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2549 | pool->process_cell = process_cell_fail; |
2550 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2551 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2552 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2553 | |
2554 | error_retry_list(pool); | |
e49e5829 JT |
2555 | break; |
2556 | ||
3ab91828 | 2557 | case PM_OUT_OF_METADATA_SPACE: |
e49e5829 | 2558 | case PM_READ_ONLY: |
3e1a0699 JT |
2559 | dm_pool_metadata_read_only(pool->pmd); |
2560 | pool->process_bio = process_bio_read_only; | |
2561 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2562 | pool->process_cell = process_cell_read_only; |
2563 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 | 2564 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
34fbcf62 | 2565 | pool->process_prepared_discard = process_prepared_discard_success; |
3e1a0699 JT |
2566 | |
2567 | error_retry_list(pool); | |
2568 | break; | |
2569 | ||
2570 | case PM_OUT_OF_DATA_SPACE: | |
2571 | /* | |
2572 | * Ideally we'd never hit this state; the low water mark | |
2573 | * would trigger userland to extend the pool before we | |
2574 | * completely run out of data space. However, many small | |
2575 | * IOs to unprovisioned space can consume data space at an | |
2576 | * alarming rate. Adjust your low water mark if you're | |
2577 | * frequently seeing this mode. | |
2578 | */ | |
c3667cc6 | 2579 | pool->out_of_data_space = true; |
3e1a0699 | 2580 | pool->process_bio = process_bio_read_only; |
a374bb21 JT |
2581 | pool->process_discard = process_discard_bio; |
2582 | pool->process_cell = process_cell_read_only; | |
3e1a0699 | 2583 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2584 | set_discard_callbacks(pool); |
85ad643b | 2585 | |
80c57893 MS |
2586 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2587 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2588 | break; |
2589 | ||
2590 | case PM_WRITE: | |
75294442 HT |
2591 | if (old_mode == PM_OUT_OF_DATA_SPACE) |
2592 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
c3667cc6 | 2593 | pool->out_of_data_space = false; |
172c2386 | 2594 | pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space; |
9b7aaa64 | 2595 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2596 | pool->process_bio = process_bio; |
a374bb21 JT |
2597 | pool->process_discard = process_discard_bio; |
2598 | pool->process_cell = process_cell; | |
e49e5829 | 2599 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2600 | set_discard_callbacks(pool); |
e49e5829 JT |
2601 | break; |
2602 | } | |
8b64e881 MS |
2603 | |
2604 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2605 | /* |
2606 | * The pool mode may have changed, sync it so bind_control_target() | |
2607 | * doesn't cause an unexpected mode transition on resume. | |
2608 | */ | |
2609 | pt->adjusted_pf.mode = new_mode; | |
f6c36758 MS |
2610 | |
2611 | if (old_mode != new_mode) | |
2612 | notify_of_pool_mode_change(pool); | |
e49e5829 JT |
2613 | } |
2614 | ||
07f2b6e0 | 2615 | static void abort_transaction(struct pool *pool) |
b5330655 | 2616 | { |
07f2b6e0 MS |
2617 | const char *dev_name = dm_device_name(pool->pool_md); |
2618 | ||
2619 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2620 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2621 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2622 | set_pool_mode(pool, PM_FAIL); | |
2623 | } | |
2624 | ||
2625 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2626 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2627 | set_pool_mode(pool, PM_FAIL); | |
2628 | } | |
2629 | } | |
399caddf | 2630 | |
07f2b6e0 MS |
2631 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2632 | { | |
b5330655 JT |
2633 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2634 | dm_device_name(pool->pool_md), op, r); | |
2635 | ||
07f2b6e0 | 2636 | abort_transaction(pool); |
b5330655 JT |
2637 | set_pool_mode(pool, PM_READ_ONLY); |
2638 | } | |
2639 | ||
e49e5829 JT |
2640 | /*----------------------------------------------------------------*/ |
2641 | ||
991d9fa0 JT |
2642 | /* |
2643 | * Mapping functions. | |
2644 | */ | |
2645 | ||
2646 | /* | |
2647 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2648 | */ | |
2649 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2650 | { | |
991d9fa0 JT |
2651 | struct pool *pool = tc->pool; |
2652 | ||
8e0c9dac | 2653 | spin_lock_irq(&tc->lock); |
c140e1c4 | 2654 | bio_list_add(&tc->deferred_bio_list, bio); |
8e0c9dac | 2655 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
2656 | |
2657 | wake_worker(pool); | |
2658 | } | |
2659 | ||
7d327fe0 JT |
2660 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2661 | { | |
2662 | struct pool *pool = tc->pool; | |
2663 | ||
2664 | throttle_lock(&pool->throttle); | |
2665 | thin_defer_bio(tc, bio); | |
2666 | throttle_unlock(&pool->throttle); | |
2667 | } | |
2668 | ||
a374bb21 JT |
2669 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2670 | { | |
a374bb21 JT |
2671 | struct pool *pool = tc->pool; |
2672 | ||
2673 | throttle_lock(&pool->throttle); | |
8e0c9dac | 2674 | spin_lock_irq(&tc->lock); |
a374bb21 | 2675 | list_add_tail(&cell->user_list, &tc->deferred_cells); |
8e0c9dac | 2676 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
2677 | throttle_unlock(&pool->throttle); |
2678 | ||
2679 | wake_worker(pool); | |
2680 | } | |
2681 | ||
59c3d2c6 | 2682 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2683 | { |
59c3d2c6 | 2684 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2685 | |
2686 | h->tc = tc; | |
2687 | h->shared_read_entry = NULL; | |
e8088073 | 2688 | h->all_io_entry = NULL; |
eb2aa48d | 2689 | h->overwrite_mapping = NULL; |
34fbcf62 | 2690 | h->cell = NULL; |
eb2aa48d JT |
2691 | } |
2692 | ||
991d9fa0 JT |
2693 | /* |
2694 | * Non-blocking function called from the thin target's map function. | |
2695 | */ | |
7de3ee57 | 2696 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2697 | { |
2698 | int r; | |
2699 | struct thin_c *tc = ti->private; | |
2700 | dm_block_t block = get_bio_block(tc, bio); | |
2701 | struct dm_thin_device *td = tc->td; | |
2702 | struct dm_thin_lookup_result result; | |
a374bb21 | 2703 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2704 | struct dm_cell_key key; |
991d9fa0 | 2705 | |
59c3d2c6 | 2706 | thin_hook_bio(tc, bio); |
e49e5829 | 2707 | |
738211f7 | 2708 | if (tc->requeue_mode) { |
4e4cbee9 | 2709 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 | 2710 | bio_endio(bio); |
738211f7 JT |
2711 | return DM_MAPIO_SUBMITTED; |
2712 | } | |
2713 | ||
e49e5829 JT |
2714 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2715 | bio_io_error(bio); | |
2716 | return DM_MAPIO_SUBMITTED; | |
2717 | } | |
2718 | ||
f73f44eb | 2719 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) { |
7d327fe0 | 2720 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2721 | return DM_MAPIO_SUBMITTED; |
2722 | } | |
2723 | ||
c822ed96 JT |
2724 | /* |
2725 | * We must hold the virtual cell before doing the lookup, otherwise | |
2726 | * there's a race with discard. | |
2727 | */ | |
2728 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2729 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2730 | return DM_MAPIO_SUBMITTED; |
2731 | ||
991d9fa0 JT |
2732 | r = dm_thin_find_block(td, block, 0, &result); |
2733 | ||
2734 | /* | |
2735 | * Note that we defer readahead too. | |
2736 | */ | |
2737 | switch (r) { | |
2738 | case 0: | |
2739 | if (unlikely(result.shared)) { | |
2740 | /* | |
2741 | * We have a race condition here between the | |
2742 | * result.shared value returned by the lookup and | |
2743 | * snapshot creation, which may cause new | |
2744 | * sharing. | |
2745 | * | |
2746 | * To avoid this always quiesce the origin before | |
2747 | * taking the snap. You want to do this anyway to | |
2748 | * ensure a consistent application view | |
2749 | * (i.e. lockfs). | |
2750 | * | |
2751 | * More distant ancestors are irrelevant. The | |
2752 | * shared flag will be set in their case. | |
2753 | */ | |
a374bb21 | 2754 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2755 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2756 | } |
e8088073 | 2757 | |
e8088073 | 2758 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2759 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2760 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2761 | return DM_MAPIO_SUBMITTED; |
2762 | } | |
2763 | ||
2764 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2765 | cell_defer_no_holder(tc, data_cell); |
2766 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2767 | |
2768 | remap(tc, bio, result.block); | |
2769 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2770 | |
2771 | case -ENODATA: | |
e49e5829 | 2772 | case -EWOULDBLOCK: |
a374bb21 | 2773 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2774 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2775 | |
2776 | default: | |
2777 | /* | |
2778 | * Must always call bio_io_error on failure. | |
2779 | * dm_thin_find_block can fail with -EINVAL if the | |
2780 | * pool is switched to fail-io mode. | |
2781 | */ | |
2782 | bio_io_error(bio); | |
a374bb21 | 2783 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2784 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2785 | } |
991d9fa0 JT |
2786 | } |
2787 | ||
c140e1c4 | 2788 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2789 | { |
c140e1c4 MS |
2790 | struct thin_c *tc; |
2791 | ||
2792 | rcu_read_lock(); | |
2793 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
8e0c9dac | 2794 | spin_lock_irq(&tc->lock); |
c140e1c4 MS |
2795 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); |
2796 | bio_list_init(&tc->retry_on_resume_list); | |
8e0c9dac | 2797 | spin_unlock_irq(&tc->lock); |
c140e1c4 MS |
2798 | } |
2799 | rcu_read_unlock(); | |
991d9fa0 JT |
2800 | } |
2801 | ||
2802 | /*---------------------------------------------------------------- | |
2803 | * Binding of control targets to a pool object | |
2804 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2805 | static bool data_dev_supports_discard(struct pool_c *pt) |
2806 | { | |
2807 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2808 | ||
695902bb | 2809 | return blk_queue_discard(q); |
9bc142dd MS |
2810 | } |
2811 | ||
58051b94 JT |
2812 | static bool is_factor(sector_t block_size, uint32_t n) |
2813 | { | |
2814 | return !sector_div(block_size, n); | |
2815 | } | |
2816 | ||
9bc142dd MS |
2817 | /* |
2818 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2819 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2820 | */ |
0424caa1 | 2821 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2822 | { |
0424caa1 MS |
2823 | struct pool *pool = pt->pool; |
2824 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2825 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
0424caa1 | 2826 | const char *reason = NULL; |
9bc142dd | 2827 | |
0424caa1 | 2828 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2829 | return; |
2830 | ||
0424caa1 MS |
2831 | if (!data_dev_supports_discard(pt)) |
2832 | reason = "discard unsupported"; | |
2833 | ||
2834 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2835 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2836 | |
0424caa1 | 2837 | if (reason) { |
385411ff | 2838 | DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason); |
0424caa1 MS |
2839 | pt->adjusted_pf.discard_passdown = false; |
2840 | } | |
9bc142dd MS |
2841 | } |
2842 | ||
991d9fa0 JT |
2843 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2844 | { | |
2845 | struct pool_c *pt = ti->private; | |
2846 | ||
e49e5829 | 2847 | /* |
9b7aaa64 | 2848 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2849 | */ |
07f2b6e0 | 2850 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2851 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2852 | |
8b64e881 MS |
2853 | /* |
2854 | * Don't change the pool's mode until set_pool_mode() below. | |
2855 | * Otherwise the pool's process_* function pointers may | |
2856 | * not match the desired pool mode. | |
2857 | */ | |
2858 | pt->adjusted_pf.mode = old_mode; | |
2859 | ||
2860 | pool->ti = ti; | |
2861 | pool->pf = pt->adjusted_pf; | |
2862 | pool->low_water_blocks = pt->low_water_blocks; | |
2863 | ||
9bc142dd | 2864 | set_pool_mode(pool, new_mode); |
f402693d | 2865 | |
991d9fa0 JT |
2866 | return 0; |
2867 | } | |
2868 | ||
2869 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2870 | { | |
2871 | if (pool->ti == ti) | |
2872 | pool->ti = NULL; | |
2873 | } | |
2874 | ||
2875 | /*---------------------------------------------------------------- | |
2876 | * Pool creation | |
2877 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2878 | /* Initialize pool features. */ |
2879 | static void pool_features_init(struct pool_features *pf) | |
2880 | { | |
e49e5829 | 2881 | pf->mode = PM_WRITE; |
9bc142dd MS |
2882 | pf->zero_new_blocks = true; |
2883 | pf->discard_enabled = true; | |
2884 | pf->discard_passdown = true; | |
787a996c | 2885 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2886 | } |
2887 | ||
991d9fa0 JT |
2888 | static void __pool_destroy(struct pool *pool) |
2889 | { | |
2890 | __pool_table_remove(pool); | |
2891 | ||
a822c83e | 2892 | vfree(pool->cell_sort_array); |
991d9fa0 JT |
2893 | if (dm_pool_metadata_close(pool->pmd) < 0) |
2894 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2895 | ||
44feb387 | 2896 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2897 | dm_kcopyd_client_destroy(pool->copier); |
2898 | ||
2899 | if (pool->wq) | |
2900 | destroy_workqueue(pool->wq); | |
2901 | ||
2902 | if (pool->next_mapping) | |
6f1c819c KO |
2903 | mempool_free(pool->next_mapping, &pool->mapping_pool); |
2904 | mempool_exit(&pool->mapping_pool); | |
44feb387 MS |
2905 | dm_deferred_set_destroy(pool->shared_read_ds); |
2906 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2907 | kfree(pool); |
2908 | } | |
2909 | ||
a24c2569 | 2910 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2911 | |
991d9fa0 JT |
2912 | static struct pool *pool_create(struct mapped_device *pool_md, |
2913 | struct block_device *metadata_dev, | |
873937e7 | 2914 | struct block_device *data_dev, |
e49e5829 JT |
2915 | unsigned long block_size, |
2916 | int read_only, char **error) | |
991d9fa0 JT |
2917 | { |
2918 | int r; | |
2919 | void *err_p; | |
2920 | struct pool *pool; | |
2921 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2922 | bool format_device = read_only ? false : true; |
991d9fa0 | 2923 | |
e49e5829 | 2924 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2925 | if (IS_ERR(pmd)) { |
2926 | *error = "Error creating metadata object"; | |
2927 | return (struct pool *)pmd; | |
2928 | } | |
2929 | ||
d3775354 | 2930 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
991d9fa0 JT |
2931 | if (!pool) { |
2932 | *error = "Error allocating memory for pool"; | |
2933 | err_p = ERR_PTR(-ENOMEM); | |
2934 | goto bad_pool; | |
2935 | } | |
2936 | ||
2937 | pool->pmd = pmd; | |
2938 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2939 | if (block_size & (block_size - 1)) |
2940 | pool->sectors_per_block_shift = -1; | |
2941 | else | |
2942 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2943 | pool->low_water_blocks = 0; |
67e2e2b2 | 2944 | pool_features_init(&pool->pf); |
a195db2d | 2945 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2946 | if (!pool->prison) { |
2947 | *error = "Error creating pool's bio prison"; | |
2948 | err_p = ERR_PTR(-ENOMEM); | |
2949 | goto bad_prison; | |
2950 | } | |
2951 | ||
df5d2e90 | 2952 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2953 | if (IS_ERR(pool->copier)) { |
2954 | r = PTR_ERR(pool->copier); | |
2955 | *error = "Error creating pool's kcopyd client"; | |
2956 | err_p = ERR_PTR(r); | |
2957 | goto bad_kcopyd_client; | |
2958 | } | |
2959 | ||
2960 | /* | |
2961 | * Create singlethreaded workqueue that will service all devices | |
2962 | * that use this metadata. | |
2963 | */ | |
2964 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2965 | if (!pool->wq) { | |
2966 | *error = "Error creating pool's workqueue"; | |
2967 | err_p = ERR_PTR(-ENOMEM); | |
2968 | goto bad_wq; | |
2969 | } | |
2970 | ||
7d327fe0 | 2971 | throttle_init(&pool->throttle); |
991d9fa0 | 2972 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 2973 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2974 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2975 | spin_lock_init(&pool->lock); |
991d9fa0 | 2976 | bio_list_init(&pool->deferred_flush_bios); |
4ae280b4 | 2977 | bio_list_init(&pool->deferred_flush_completions); |
991d9fa0 | 2978 | INIT_LIST_HEAD(&pool->prepared_mappings); |
104655fd | 2979 | INIT_LIST_HEAD(&pool->prepared_discards); |
2a0fbffb | 2980 | INIT_LIST_HEAD(&pool->prepared_discards_pt2); |
c140e1c4 | 2981 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2982 | pool->low_water_triggered = false; |
80e96c54 | 2983 | pool->suspended = true; |
c3667cc6 | 2984 | pool->out_of_data_space = false; |
44feb387 MS |
2985 | |
2986 | pool->shared_read_ds = dm_deferred_set_create(); | |
2987 | if (!pool->shared_read_ds) { | |
2988 | *error = "Error creating pool's shared read deferred set"; | |
2989 | err_p = ERR_PTR(-ENOMEM); | |
2990 | goto bad_shared_read_ds; | |
2991 | } | |
2992 | ||
2993 | pool->all_io_ds = dm_deferred_set_create(); | |
2994 | if (!pool->all_io_ds) { | |
2995 | *error = "Error creating pool's all io deferred set"; | |
2996 | err_p = ERR_PTR(-ENOMEM); | |
2997 | goto bad_all_io_ds; | |
2998 | } | |
991d9fa0 JT |
2999 | |
3000 | pool->next_mapping = NULL; | |
6f1c819c KO |
3001 | r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE, |
3002 | _new_mapping_cache); | |
3003 | if (r) { | |
991d9fa0 | 3004 | *error = "Error creating pool's mapping mempool"; |
6f1c819c | 3005 | err_p = ERR_PTR(r); |
991d9fa0 JT |
3006 | goto bad_mapping_pool; |
3007 | } | |
3008 | ||
42bc47b3 KC |
3009 | pool->cell_sort_array = |
3010 | vmalloc(array_size(CELL_SORT_ARRAY_SIZE, | |
3011 | sizeof(*pool->cell_sort_array))); | |
a822c83e JT |
3012 | if (!pool->cell_sort_array) { |
3013 | *error = "Error allocating cell sort array"; | |
3014 | err_p = ERR_PTR(-ENOMEM); | |
3015 | goto bad_sort_array; | |
3016 | } | |
3017 | ||
991d9fa0 | 3018 | pool->ref_count = 1; |
905e51b3 | 3019 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
3020 | pool->pool_md = pool_md; |
3021 | pool->md_dev = metadata_dev; | |
873937e7 | 3022 | pool->data_dev = data_dev; |
991d9fa0 JT |
3023 | __pool_table_insert(pool); |
3024 | ||
3025 | return pool; | |
3026 | ||
a822c83e | 3027 | bad_sort_array: |
6f1c819c | 3028 | mempool_exit(&pool->mapping_pool); |
991d9fa0 | 3029 | bad_mapping_pool: |
44feb387 MS |
3030 | dm_deferred_set_destroy(pool->all_io_ds); |
3031 | bad_all_io_ds: | |
3032 | dm_deferred_set_destroy(pool->shared_read_ds); | |
3033 | bad_shared_read_ds: | |
991d9fa0 JT |
3034 | destroy_workqueue(pool->wq); |
3035 | bad_wq: | |
3036 | dm_kcopyd_client_destroy(pool->copier); | |
3037 | bad_kcopyd_client: | |
44feb387 | 3038 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
3039 | bad_prison: |
3040 | kfree(pool); | |
3041 | bad_pool: | |
3042 | if (dm_pool_metadata_close(pmd)) | |
3043 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
3044 | ||
3045 | return err_p; | |
3046 | } | |
3047 | ||
3048 | static void __pool_inc(struct pool *pool) | |
3049 | { | |
3050 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
3051 | pool->ref_count++; | |
3052 | } | |
3053 | ||
3054 | static void __pool_dec(struct pool *pool) | |
3055 | { | |
3056 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
3057 | BUG_ON(!pool->ref_count); | |
3058 | if (!--pool->ref_count) | |
3059 | __pool_destroy(pool); | |
3060 | } | |
3061 | ||
3062 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
3063 | struct block_device *metadata_dev, | |
873937e7 | 3064 | struct block_device *data_dev, |
e49e5829 JT |
3065 | unsigned long block_size, int read_only, |
3066 | char **error, int *created) | |
991d9fa0 JT |
3067 | { |
3068 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
3069 | ||
3070 | if (pool) { | |
f09996c9 MS |
3071 | if (pool->pool_md != pool_md) { |
3072 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 3073 | return ERR_PTR(-EBUSY); |
f09996c9 | 3074 | } |
873937e7 MP |
3075 | if (pool->data_dev != data_dev) { |
3076 | *error = "data device already in use by a pool"; | |
3077 | return ERR_PTR(-EBUSY); | |
3078 | } | |
991d9fa0 JT |
3079 | __pool_inc(pool); |
3080 | ||
3081 | } else { | |
3082 | pool = __pool_table_lookup(pool_md); | |
3083 | if (pool) { | |
873937e7 | 3084 | if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) { |
f09996c9 | 3085 | *error = "different pool cannot replace a pool"; |
991d9fa0 | 3086 | return ERR_PTR(-EINVAL); |
f09996c9 | 3087 | } |
991d9fa0 JT |
3088 | __pool_inc(pool); |
3089 | ||
67e2e2b2 | 3090 | } else { |
873937e7 | 3091 | pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error); |
67e2e2b2 JT |
3092 | *created = 1; |
3093 | } | |
991d9fa0 JT |
3094 | } |
3095 | ||
3096 | return pool; | |
3097 | } | |
3098 | ||
3099 | /*---------------------------------------------------------------- | |
3100 | * Pool target methods | |
3101 | *--------------------------------------------------------------*/ | |
3102 | static void pool_dtr(struct dm_target *ti) | |
3103 | { | |
3104 | struct pool_c *pt = ti->private; | |
3105 | ||
3106 | mutex_lock(&dm_thin_pool_table.mutex); | |
3107 | ||
3108 | unbind_control_target(pt->pool, ti); | |
3109 | __pool_dec(pt->pool); | |
3110 | dm_put_device(ti, pt->metadata_dev); | |
3111 | dm_put_device(ti, pt->data_dev); | |
3112 | kfree(pt); | |
3113 | ||
3114 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3115 | } | |
3116 | ||
991d9fa0 JT |
3117 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
3118 | struct dm_target *ti) | |
3119 | { | |
3120 | int r; | |
3121 | unsigned argc; | |
3122 | const char *arg_name; | |
3123 | ||
5916a22b | 3124 | static const struct dm_arg _args[] = { |
74aa45c3 | 3125 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
3126 | }; |
3127 | ||
3128 | /* | |
3129 | * No feature arguments supplied. | |
3130 | */ | |
3131 | if (!as->argc) | |
3132 | return 0; | |
3133 | ||
3134 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
3135 | if (r) | |
3136 | return -EINVAL; | |
3137 | ||
3138 | while (argc && !r) { | |
3139 | arg_name = dm_shift_arg(as); | |
3140 | argc--; | |
3141 | ||
e49e5829 | 3142 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 3143 | pf->zero_new_blocks = false; |
e49e5829 JT |
3144 | |
3145 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 3146 | pf->discard_enabled = false; |
e49e5829 JT |
3147 | |
3148 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 3149 | pf->discard_passdown = false; |
991d9fa0 | 3150 | |
e49e5829 JT |
3151 | else if (!strcasecmp(arg_name, "read_only")) |
3152 | pf->mode = PM_READ_ONLY; | |
3153 | ||
787a996c MS |
3154 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
3155 | pf->error_if_no_space = true; | |
3156 | ||
e49e5829 JT |
3157 | else { |
3158 | ti->error = "Unrecognised pool feature requested"; | |
3159 | r = -EINVAL; | |
3160 | break; | |
3161 | } | |
991d9fa0 JT |
3162 | } |
3163 | ||
3164 | return r; | |
3165 | } | |
3166 | ||
ac8c3f3d JT |
3167 | static void metadata_low_callback(void *context) |
3168 | { | |
3169 | struct pool *pool = context; | |
3170 | ||
3171 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
3172 | dm_device_name(pool->pool_md)); | |
3173 | ||
3174 | dm_table_event(pool->ti->table); | |
3175 | } | |
3176 | ||
694cfe7f NT |
3177 | /* |
3178 | * We need to flush the data device **before** committing the metadata. | |
3179 | * | |
3180 | * This ensures that the data blocks of any newly inserted mappings are | |
3181 | * properly written to non-volatile storage and won't be lost in case of a | |
3182 | * crash. | |
3183 | * | |
3184 | * Failure to do so can result in data corruption in the case of internal or | |
3185 | * external snapshots and in the case of newly provisioned blocks, when block | |
3186 | * zeroing is enabled. | |
3187 | */ | |
3188 | static int metadata_pre_commit_callback(void *context) | |
3189 | { | |
f06c03d1 | 3190 | struct pool *pool = context; |
694cfe7f | 3191 | |
28d7d128 | 3192 | return blkdev_issue_flush(pool->data_dev); |
694cfe7f NT |
3193 | } |
3194 | ||
7d48935e MS |
3195 | static sector_t get_dev_size(struct block_device *bdev) |
3196 | { | |
6dcbb52c | 3197 | return bdev_nr_sectors(bdev); |
7d48935e MS |
3198 | } |
3199 | ||
3200 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 3201 | { |
7d48935e | 3202 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df | 3203 | |
7d48935e | 3204 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
385411ff CH |
3205 | DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.", |
3206 | bdev, THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
3207 | } |
3208 | ||
3209 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
3210 | { | |
3211 | sector_t metadata_dev_size = get_dev_size(bdev); | |
3212 | ||
3213 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
3214 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
3215 | |
3216 | return metadata_dev_size; | |
3217 | } | |
3218 | ||
24347e95 JT |
3219 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
3220 | { | |
3221 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
3222 | ||
7d48935e | 3223 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
3224 | |
3225 | return metadata_dev_size; | |
3226 | } | |
3227 | ||
ac8c3f3d JT |
3228 | /* |
3229 | * When a metadata threshold is crossed a dm event is triggered, and | |
3230 | * userland should respond by growing the metadata device. We could let | |
3231 | * userland set the threshold, like we do with the data threshold, but I'm | |
3232 | * not sure they know enough to do this well. | |
3233 | */ | |
3234 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
3235 | { | |
3236 | /* | |
3237 | * 4M is ample for all ops with the possible exception of thin | |
3238 | * device deletion which is harmless if it fails (just retry the | |
3239 | * delete after you've grown the device). | |
3240 | */ | |
3241 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
3242 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
3243 | } | |
3244 | ||
991d9fa0 JT |
3245 | /* |
3246 | * thin-pool <metadata dev> <data dev> | |
3247 | * <data block size (sectors)> | |
3248 | * <low water mark (blocks)> | |
3249 | * [<#feature args> [<arg>]*] | |
3250 | * | |
3251 | * Optional feature arguments are: | |
3252 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
3253 | * ignore_discard: disable discard |
3254 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
3255 | * read_only: Don't allow any changes to be made to the pool metadata. |
3256 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
3257 | */ |
3258 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3259 | { | |
67e2e2b2 | 3260 | int r, pool_created = 0; |
991d9fa0 JT |
3261 | struct pool_c *pt; |
3262 | struct pool *pool; | |
3263 | struct pool_features pf; | |
3264 | struct dm_arg_set as; | |
3265 | struct dm_dev *data_dev; | |
3266 | unsigned long block_size; | |
3267 | dm_block_t low_water_blocks; | |
3268 | struct dm_dev *metadata_dev; | |
5d0db96d | 3269 | fmode_t metadata_mode; |
991d9fa0 JT |
3270 | |
3271 | /* | |
3272 | * FIXME Remove validation from scope of lock. | |
3273 | */ | |
3274 | mutex_lock(&dm_thin_pool_table.mutex); | |
3275 | ||
3276 | if (argc < 4) { | |
3277 | ti->error = "Invalid argument count"; | |
3278 | r = -EINVAL; | |
3279 | goto out_unlock; | |
3280 | } | |
5d0db96d | 3281 | |
991d9fa0 JT |
3282 | as.argc = argc; |
3283 | as.argv = argv; | |
3284 | ||
70de2cbd JCXF |
3285 | /* make sure metadata and data are different devices */ |
3286 | if (!strcmp(argv[0], argv[1])) { | |
3287 | ti->error = "Error setting metadata or data device"; | |
3288 | r = -EINVAL; | |
3289 | goto out_unlock; | |
3290 | } | |
3291 | ||
5d0db96d JT |
3292 | /* |
3293 | * Set default pool features. | |
3294 | */ | |
3295 | pool_features_init(&pf); | |
3296 | ||
3297 | dm_consume_args(&as, 4); | |
3298 | r = parse_pool_features(&as, &pf, ti); | |
3299 | if (r) | |
3300 | goto out_unlock; | |
3301 | ||
3302 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
3303 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
3304 | if (r) { |
3305 | ti->error = "Error opening metadata block device"; | |
3306 | goto out_unlock; | |
3307 | } | |
7d48935e | 3308 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
3309 | |
3310 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
3311 | if (r) { | |
3312 | ti->error = "Error getting data device"; | |
3313 | goto out_metadata; | |
3314 | } | |
3315 | ||
3316 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
3317 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
3318 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 3319 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
3320 | ti->error = "Invalid block size"; |
3321 | r = -EINVAL; | |
3322 | goto out; | |
3323 | } | |
3324 | ||
3325 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
3326 | ti->error = "Invalid low water mark"; | |
3327 | r = -EINVAL; | |
3328 | goto out; | |
3329 | } | |
3330 | ||
991d9fa0 JT |
3331 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
3332 | if (!pt) { | |
3333 | r = -ENOMEM; | |
3334 | goto out; | |
3335 | } | |
3336 | ||
873937e7 | 3337 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev, |
e49e5829 | 3338 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
3339 | if (IS_ERR(pool)) { |
3340 | r = PTR_ERR(pool); | |
3341 | goto out_free_pt; | |
3342 | } | |
3343 | ||
67e2e2b2 JT |
3344 | /* |
3345 | * 'pool_created' reflects whether this is the first table load. | |
3346 | * Top level discard support is not allowed to be changed after | |
3347 | * initial load. This would require a pool reload to trigger thin | |
3348 | * device changes. | |
3349 | */ | |
3350 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
3351 | ti->error = "Discard support cannot be disabled once enabled"; | |
3352 | r = -EINVAL; | |
3353 | goto out_flags_changed; | |
3354 | } | |
3355 | ||
991d9fa0 JT |
3356 | pt->pool = pool; |
3357 | pt->ti = ti; | |
3358 | pt->metadata_dev = metadata_dev; | |
3359 | pt->data_dev = data_dev; | |
3360 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 3361 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 3362 | ti->num_flush_bios = 1; |
9bc142dd | 3363 | |
67e2e2b2 JT |
3364 | /* |
3365 | * Only need to enable discards if the pool should pass | |
3366 | * them down to the data device. The thin device's discard | |
3367 | * processing will cause mappings to be removed from the btree. | |
3368 | */ | |
3369 | if (pf.discard_enabled && pf.discard_passdown) { | |
55a62eef | 3370 | ti->num_discard_bios = 1; |
9bc142dd | 3371 | |
67e2e2b2 JT |
3372 | /* |
3373 | * Setting 'discards_supported' circumvents the normal | |
3374 | * stacking of discard limits (this keeps the pool and | |
3375 | * thin devices' discard limits consistent). | |
3376 | */ | |
0ac55489 | 3377 | ti->discards_supported = true; |
67e2e2b2 | 3378 | } |
991d9fa0 JT |
3379 | ti->private = pt; |
3380 | ||
ac8c3f3d JT |
3381 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
3382 | calc_metadata_threshold(pt), | |
3383 | metadata_low_callback, | |
3384 | pool); | |
3385 | if (r) | |
ba30670f | 3386 | goto out_flags_changed; |
ac8c3f3d | 3387 | |
f06c03d1 MP |
3388 | dm_pool_register_pre_commit_callback(pool->pmd, |
3389 | metadata_pre_commit_callback, pool); | |
3390 | ||
991d9fa0 JT |
3391 | mutex_unlock(&dm_thin_pool_table.mutex); |
3392 | ||
3393 | return 0; | |
3394 | ||
67e2e2b2 JT |
3395 | out_flags_changed: |
3396 | __pool_dec(pool); | |
991d9fa0 JT |
3397 | out_free_pt: |
3398 | kfree(pt); | |
3399 | out: | |
3400 | dm_put_device(ti, data_dev); | |
3401 | out_metadata: | |
3402 | dm_put_device(ti, metadata_dev); | |
3403 | out_unlock: | |
3404 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3405 | ||
3406 | return r; | |
3407 | } | |
3408 | ||
7de3ee57 | 3409 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
3410 | { |
3411 | int r; | |
3412 | struct pool_c *pt = ti->private; | |
3413 | struct pool *pool = pt->pool; | |
991d9fa0 JT |
3414 | |
3415 | /* | |
3416 | * As this is a singleton target, ti->begin is always zero. | |
3417 | */ | |
8e0c9dac | 3418 | spin_lock_irq(&pool->lock); |
74d46992 | 3419 | bio_set_dev(bio, pt->data_dev->bdev); |
991d9fa0 | 3420 | r = DM_MAPIO_REMAPPED; |
8e0c9dac | 3421 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
3422 | |
3423 | return r; | |
3424 | } | |
3425 | ||
b17446df | 3426 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
3427 | { |
3428 | int r; | |
3429 | struct pool_c *pt = ti->private; | |
3430 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
3431 | sector_t data_size = ti->len; |
3432 | dm_block_t sb_data_size; | |
991d9fa0 | 3433 | |
b17446df | 3434 | *need_commit = false; |
991d9fa0 | 3435 | |
55f2b8bd MS |
3436 | (void) sector_div(data_size, pool->sectors_per_block); |
3437 | ||
991d9fa0 JT |
3438 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
3439 | if (r) { | |
4fa5971a MS |
3440 | DMERR("%s: failed to retrieve data device size", |
3441 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
3442 | return r; |
3443 | } | |
3444 | ||
3445 | if (data_size < sb_data_size) { | |
4fa5971a MS |
3446 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
3447 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3448 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3449 | return -EINVAL; |
3450 | ||
3451 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3452 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3453 | DMERR("%s: unable to grow the data device until repaired.", | |
3454 | dm_device_name(pool->pool_md)); | |
3455 | return 0; | |
3456 | } | |
3457 | ||
6f7f51d4 MS |
3458 | if (sb_data_size) |
3459 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3460 | dm_device_name(pool->pool_md), | |
3461 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3462 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3463 | if (r) { | |
b5330655 | 3464 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3465 | return r; |
3466 | } | |
3467 | ||
b17446df | 3468 | *need_commit = true; |
991d9fa0 JT |
3469 | } |
3470 | ||
3471 | return 0; | |
3472 | } | |
3473 | ||
24347e95 JT |
3474 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3475 | { | |
3476 | int r; | |
3477 | struct pool_c *pt = ti->private; | |
3478 | struct pool *pool = pt->pool; | |
3479 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3480 | ||
3481 | *need_commit = false; | |
3482 | ||
610bba8b | 3483 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3484 | |
3485 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3486 | if (r) { | |
4fa5971a MS |
3487 | DMERR("%s: failed to retrieve metadata device size", |
3488 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3489 | return r; |
3490 | } | |
3491 | ||
3492 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3493 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3494 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3495 | metadata_dev_size, sb_metadata_dev_size); |
3496 | return -EINVAL; | |
3497 | ||
3498 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3499 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3500 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3501 | dm_device_name(pool->pool_md)); | |
3502 | return 0; | |
3503 | } | |
3504 | ||
7d48935e | 3505 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3506 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3507 | dm_device_name(pool->pool_md), | |
3508 | sb_metadata_dev_size, metadata_dev_size); | |
3ab91828 JT |
3509 | |
3510 | if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE) | |
3511 | set_pool_mode(pool, PM_WRITE); | |
3512 | ||
24347e95 JT |
3513 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3514 | if (r) { | |
b5330655 | 3515 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3516 | return r; |
3517 | } | |
3518 | ||
3519 | *need_commit = true; | |
3520 | } | |
3521 | ||
3522 | return 0; | |
3523 | } | |
3524 | ||
b17446df JT |
3525 | /* |
3526 | * Retrieves the number of blocks of the data device from | |
3527 | * the superblock and compares it to the actual device size, | |
3528 | * thus resizing the data device in case it has grown. | |
3529 | * | |
3530 | * This both copes with opening preallocated data devices in the ctr | |
3531 | * being followed by a resume | |
3532 | * -and- | |
3533 | * calling the resume method individually after userspace has | |
3534 | * grown the data device in reaction to a table event. | |
3535 | */ | |
3536 | static int pool_preresume(struct dm_target *ti) | |
3537 | { | |
3538 | int r; | |
24347e95 | 3539 | bool need_commit1, need_commit2; |
b17446df JT |
3540 | struct pool_c *pt = ti->private; |
3541 | struct pool *pool = pt->pool; | |
3542 | ||
3543 | /* | |
3544 | * Take control of the pool object. | |
3545 | */ | |
3546 | r = bind_control_target(pool, ti); | |
3547 | if (r) | |
3548 | return r; | |
3549 | ||
3550 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3551 | if (r) | |
3552 | return r; | |
3553 | ||
24347e95 JT |
3554 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3555 | if (r) | |
3556 | return r; | |
3557 | ||
3558 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3559 | (void) commit(pool); |
b17446df JT |
3560 | |
3561 | return 0; | |
3562 | } | |
3563 | ||
583024d2 MS |
3564 | static void pool_suspend_active_thins(struct pool *pool) |
3565 | { | |
3566 | struct thin_c *tc; | |
3567 | ||
3568 | /* Suspend all active thin devices */ | |
3569 | tc = get_first_thin(pool); | |
3570 | while (tc) { | |
3571 | dm_internal_suspend_noflush(tc->thin_md); | |
3572 | tc = get_next_thin(pool, tc); | |
3573 | } | |
3574 | } | |
3575 | ||
3576 | static void pool_resume_active_thins(struct pool *pool) | |
3577 | { | |
3578 | struct thin_c *tc; | |
3579 | ||
3580 | /* Resume all active thin devices */ | |
3581 | tc = get_first_thin(pool); | |
3582 | while (tc) { | |
3583 | dm_internal_resume(tc->thin_md); | |
3584 | tc = get_next_thin(pool, tc); | |
3585 | } | |
3586 | } | |
3587 | ||
991d9fa0 JT |
3588 | static void pool_resume(struct dm_target *ti) |
3589 | { | |
3590 | struct pool_c *pt = ti->private; | |
3591 | struct pool *pool = pt->pool; | |
991d9fa0 | 3592 | |
583024d2 MS |
3593 | /* |
3594 | * Must requeue active_thins' bios and then resume | |
3595 | * active_thins _before_ clearing 'suspend' flag. | |
3596 | */ | |
3597 | requeue_bios(pool); | |
3598 | pool_resume_active_thins(pool); | |
3599 | ||
8e0c9dac | 3600 | spin_lock_irq(&pool->lock); |
88a6621b | 3601 | pool->low_water_triggered = false; |
80e96c54 | 3602 | pool->suspended = false; |
8e0c9dac | 3603 | spin_unlock_irq(&pool->lock); |
80e96c54 | 3604 | |
905e51b3 | 3605 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3606 | } |
3607 | ||
80e96c54 MS |
3608 | static void pool_presuspend(struct dm_target *ti) |
3609 | { | |
3610 | struct pool_c *pt = ti->private; | |
3611 | struct pool *pool = pt->pool; | |
80e96c54 | 3612 | |
8e0c9dac | 3613 | spin_lock_irq(&pool->lock); |
80e96c54 | 3614 | pool->suspended = true; |
8e0c9dac | 3615 | spin_unlock_irq(&pool->lock); |
583024d2 MS |
3616 | |
3617 | pool_suspend_active_thins(pool); | |
80e96c54 MS |
3618 | } |
3619 | ||
3620 | static void pool_presuspend_undo(struct dm_target *ti) | |
3621 | { | |
3622 | struct pool_c *pt = ti->private; | |
3623 | struct pool *pool = pt->pool; | |
80e96c54 | 3624 | |
583024d2 MS |
3625 | pool_resume_active_thins(pool); |
3626 | ||
8e0c9dac | 3627 | spin_lock_irq(&pool->lock); |
80e96c54 | 3628 | pool->suspended = false; |
8e0c9dac | 3629 | spin_unlock_irq(&pool->lock); |
80e96c54 MS |
3630 | } |
3631 | ||
991d9fa0 JT |
3632 | static void pool_postsuspend(struct dm_target *ti) |
3633 | { | |
991d9fa0 JT |
3634 | struct pool_c *pt = ti->private; |
3635 | struct pool *pool = pt->pool; | |
3636 | ||
18d03e8c NB |
3637 | cancel_delayed_work_sync(&pool->waker); |
3638 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
991d9fa0 | 3639 | flush_workqueue(pool->wq); |
020cc3b5 | 3640 | (void) commit(pool); |
991d9fa0 JT |
3641 | } |
3642 | ||
3643 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3644 | { | |
3645 | if (argc != args_required) { | |
3646 | DMWARN("Message received with %u arguments instead of %u.", | |
3647 | argc, args_required); | |
3648 | return -EINVAL; | |
3649 | } | |
3650 | ||
3651 | return 0; | |
3652 | } | |
3653 | ||
3654 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3655 | { | |
3656 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3657 | *dev_id <= MAX_DEV_ID) | |
3658 | return 0; | |
3659 | ||
3660 | if (warning) | |
3661 | DMWARN("Message received with invalid device id: %s", arg); | |
3662 | ||
3663 | return -EINVAL; | |
3664 | } | |
3665 | ||
3666 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3667 | { | |
3668 | dm_thin_id dev_id; | |
3669 | int r; | |
3670 | ||
3671 | r = check_arg_count(argc, 2); | |
3672 | if (r) | |
3673 | return r; | |
3674 | ||
3675 | r = read_dev_id(argv[1], &dev_id, 1); | |
3676 | if (r) | |
3677 | return r; | |
3678 | ||
3679 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3680 | if (r) { | |
3681 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3682 | argv[1]); | |
3683 | return r; | |
3684 | } | |
3685 | ||
3686 | return 0; | |
3687 | } | |
3688 | ||
3689 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3690 | { | |
3691 | dm_thin_id dev_id; | |
3692 | dm_thin_id origin_dev_id; | |
3693 | int r; | |
3694 | ||
3695 | r = check_arg_count(argc, 3); | |
3696 | if (r) | |
3697 | return r; | |
3698 | ||
3699 | r = read_dev_id(argv[1], &dev_id, 1); | |
3700 | if (r) | |
3701 | return r; | |
3702 | ||
3703 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3704 | if (r) | |
3705 | return r; | |
3706 | ||
3707 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3708 | if (r) { | |
3709 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3710 | argv[1], argv[2]); | |
3711 | return r; | |
3712 | } | |
3713 | ||
3714 | return 0; | |
3715 | } | |
3716 | ||
3717 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3718 | { | |
3719 | dm_thin_id dev_id; | |
3720 | int r; | |
3721 | ||
3722 | r = check_arg_count(argc, 2); | |
3723 | if (r) | |
3724 | return r; | |
3725 | ||
3726 | r = read_dev_id(argv[1], &dev_id, 1); | |
3727 | if (r) | |
3728 | return r; | |
3729 | ||
3730 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3731 | if (r) | |
3732 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3733 | ||
3734 | return r; | |
3735 | } | |
3736 | ||
3737 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3738 | { | |
3739 | dm_thin_id old_id, new_id; | |
3740 | int r; | |
3741 | ||
3742 | r = check_arg_count(argc, 3); | |
3743 | if (r) | |
3744 | return r; | |
3745 | ||
3746 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3747 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3748 | return -EINVAL; | |
3749 | } | |
3750 | ||
3751 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3752 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3753 | return -EINVAL; | |
3754 | } | |
3755 | ||
3756 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3757 | if (r) { | |
3758 | DMWARN("Failed to change transaction id from %s to %s.", | |
3759 | argv[1], argv[2]); | |
3760 | return r; | |
3761 | } | |
3762 | ||
3763 | return 0; | |
3764 | } | |
3765 | ||
cc8394d8 JT |
3766 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3767 | { | |
3768 | int r; | |
3769 | ||
3770 | r = check_arg_count(argc, 1); | |
3771 | if (r) | |
3772 | return r; | |
3773 | ||
020cc3b5 | 3774 | (void) commit(pool); |
0d200aef | 3775 | |
cc8394d8 JT |
3776 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3777 | if (r) | |
3778 | DMWARN("reserve_metadata_snap message failed."); | |
3779 | ||
3780 | return r; | |
3781 | } | |
3782 | ||
3783 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3784 | { | |
3785 | int r; | |
3786 | ||
3787 | r = check_arg_count(argc, 1); | |
3788 | if (r) | |
3789 | return r; | |
3790 | ||
3791 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3792 | if (r) | |
3793 | DMWARN("release_metadata_snap message failed."); | |
3794 | ||
3795 | return r; | |
3796 | } | |
3797 | ||
991d9fa0 JT |
3798 | /* |
3799 | * Messages supported: | |
3800 | * create_thin <dev_id> | |
3801 | * create_snap <dev_id> <origin_id> | |
3802 | * delete <dev_id> | |
991d9fa0 | 3803 | * set_transaction_id <current_trans_id> <new_trans_id> |
cc8394d8 JT |
3804 | * reserve_metadata_snap |
3805 | * release_metadata_snap | |
991d9fa0 | 3806 | */ |
1eb5fa84 MS |
3807 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv, |
3808 | char *result, unsigned maxlen) | |
991d9fa0 JT |
3809 | { |
3810 | int r = -EINVAL; | |
3811 | struct pool_c *pt = ti->private; | |
3812 | struct pool *pool = pt->pool; | |
3813 | ||
3ab91828 | 3814 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) { |
2a7eaea0 JT |
3815 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", |
3816 | dm_device_name(pool->pool_md)); | |
fd467696 | 3817 | return -EOPNOTSUPP; |
2a7eaea0 JT |
3818 | } |
3819 | ||
991d9fa0 JT |
3820 | if (!strcasecmp(argv[0], "create_thin")) |
3821 | r = process_create_thin_mesg(argc, argv, pool); | |
3822 | ||
3823 | else if (!strcasecmp(argv[0], "create_snap")) | |
3824 | r = process_create_snap_mesg(argc, argv, pool); | |
3825 | ||
3826 | else if (!strcasecmp(argv[0], "delete")) | |
3827 | r = process_delete_mesg(argc, argv, pool); | |
3828 | ||
3829 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3830 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3831 | ||
cc8394d8 JT |
3832 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3833 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3834 | ||
3835 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3836 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3837 | ||
991d9fa0 JT |
3838 | else |
3839 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3840 | ||
e49e5829 | 3841 | if (!r) |
020cc3b5 | 3842 | (void) commit(pool); |
991d9fa0 JT |
3843 | |
3844 | return r; | |
3845 | } | |
3846 | ||
e49e5829 JT |
3847 | static void emit_flags(struct pool_features *pf, char *result, |
3848 | unsigned sz, unsigned maxlen) | |
3849 | { | |
3850 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3851 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3852 | pf->error_if_no_space; | |
e49e5829 JT |
3853 | DMEMIT("%u ", count); |
3854 | ||
3855 | if (!pf->zero_new_blocks) | |
3856 | DMEMIT("skip_block_zeroing "); | |
3857 | ||
3858 | if (!pf->discard_enabled) | |
3859 | DMEMIT("ignore_discard "); | |
3860 | ||
3861 | if (!pf->discard_passdown) | |
3862 | DMEMIT("no_discard_passdown "); | |
3863 | ||
3864 | if (pf->mode == PM_READ_ONLY) | |
3865 | DMEMIT("read_only "); | |
787a996c MS |
3866 | |
3867 | if (pf->error_if_no_space) | |
3868 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3869 | } |
3870 | ||
991d9fa0 JT |
3871 | /* |
3872 | * Status line is: | |
3873 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3874 | * <used data sectors>/<total data sectors> <held metadata root> | |
e4c78e21 | 3875 | * <pool mode> <discard config> <no space config> <needs_check> |
991d9fa0 | 3876 | */ |
fd7c092e MP |
3877 | static void pool_status(struct dm_target *ti, status_type_t type, |
3878 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3879 | { |
e49e5829 | 3880 | int r; |
991d9fa0 JT |
3881 | unsigned sz = 0; |
3882 | uint64_t transaction_id; | |
3883 | dm_block_t nr_free_blocks_data; | |
3884 | dm_block_t nr_free_blocks_metadata; | |
3885 | dm_block_t nr_blocks_data; | |
3886 | dm_block_t nr_blocks_metadata; | |
3887 | dm_block_t held_root; | |
3ab91828 | 3888 | enum pool_mode mode; |
991d9fa0 JT |
3889 | char buf[BDEVNAME_SIZE]; |
3890 | char buf2[BDEVNAME_SIZE]; | |
3891 | struct pool_c *pt = ti->private; | |
3892 | struct pool *pool = pt->pool; | |
3893 | ||
3894 | switch (type) { | |
3895 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3896 | if (get_pool_mode(pool) == PM_FAIL) { |
3897 | DMEMIT("Fail"); | |
3898 | break; | |
3899 | } | |
3900 | ||
1f4e0ff0 AK |
3901 | /* Commit to ensure statistics aren't out-of-date */ |
3902 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3903 | (void) commit(pool); |
1f4e0ff0 | 3904 | |
fd7c092e MP |
3905 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3906 | if (r) { | |
4fa5971a MS |
3907 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3908 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3909 | goto err; |
3910 | } | |
991d9fa0 | 3911 | |
fd7c092e MP |
3912 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3913 | if (r) { | |
4fa5971a MS |
3914 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3915 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3916 | goto err; |
3917 | } | |
991d9fa0 JT |
3918 | |
3919 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3920 | if (r) { |
4fa5971a MS |
3921 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3922 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3923 | goto err; |
3924 | } | |
991d9fa0 | 3925 | |
fd7c092e MP |
3926 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3927 | if (r) { | |
4fa5971a MS |
3928 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3929 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3930 | goto err; |
3931 | } | |
991d9fa0 JT |
3932 | |
3933 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3934 | if (r) { |
4fa5971a MS |
3935 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3936 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3937 | goto err; |
3938 | } | |
991d9fa0 | 3939 | |
cc8394d8 | 3940 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3941 | if (r) { |
4fa5971a MS |
3942 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3943 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3944 | goto err; |
3945 | } | |
991d9fa0 JT |
3946 | |
3947 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3948 | (unsigned long long)transaction_id, | |
3949 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3950 | (unsigned long long)nr_blocks_metadata, | |
3951 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3952 | (unsigned long long)nr_blocks_data); | |
3953 | ||
3954 | if (held_root) | |
e49e5829 JT |
3955 | DMEMIT("%llu ", held_root); |
3956 | else | |
3957 | DMEMIT("- "); | |
3958 | ||
3ab91828 JT |
3959 | mode = get_pool_mode(pool); |
3960 | if (mode == PM_OUT_OF_DATA_SPACE) | |
3e1a0699 | 3961 | DMEMIT("out_of_data_space "); |
3ab91828 | 3962 | else if (is_read_only_pool_mode(mode)) |
e49e5829 | 3963 | DMEMIT("ro "); |
991d9fa0 | 3964 | else |
e49e5829 JT |
3965 | DMEMIT("rw "); |
3966 | ||
018debea | 3967 | if (!pool->pf.discard_enabled) |
787a996c | 3968 | DMEMIT("ignore_discard "); |
018debea | 3969 | else if (pool->pf.discard_passdown) |
787a996c MS |
3970 | DMEMIT("discard_passdown "); |
3971 | else | |
3972 | DMEMIT("no_discard_passdown "); | |
3973 | ||
3974 | if (pool->pf.error_if_no_space) | |
3975 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3976 | else |
787a996c | 3977 | DMEMIT("queue_if_no_space "); |
991d9fa0 | 3978 | |
e4c78e21 MS |
3979 | if (dm_pool_metadata_needs_check(pool->pmd)) |
3980 | DMEMIT("needs_check "); | |
3981 | else | |
3982 | DMEMIT("- "); | |
3983 | ||
63c8ecb6 AG |
3984 | DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt)); |
3985 | ||
991d9fa0 JT |
3986 | break; |
3987 | ||
3988 | case STATUSTYPE_TABLE: | |
3989 | DMEMIT("%s %s %lu %llu ", | |
3990 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3991 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3992 | (unsigned long)pool->sectors_per_block, | |
3993 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3994 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 | 3995 | break; |
8ec45662 TS |
3996 | |
3997 | case STATUSTYPE_IMA: | |
3998 | *result = '\0'; | |
3999 | break; | |
991d9fa0 | 4000 | } |
fd7c092e | 4001 | return; |
991d9fa0 | 4002 | |
fd7c092e MP |
4003 | err: |
4004 | DMEMIT("Error"); | |
991d9fa0 JT |
4005 | } |
4006 | ||
4007 | static int pool_iterate_devices(struct dm_target *ti, | |
4008 | iterate_devices_callout_fn fn, void *data) | |
4009 | { | |
4010 | struct pool_c *pt = ti->private; | |
4011 | ||
4012 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
4013 | } | |
4014 | ||
991d9fa0 JT |
4015 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4016 | { | |
4017 | struct pool_c *pt = ti->private; | |
4018 | struct pool *pool = pt->pool; | |
604ea906 MS |
4019 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
4020 | ||
4021 | /* | |
d200c30e MS |
4022 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
4023 | * to the highest possible power-of-2 factor of pool->sectors_per_block. | |
4024 | * This is especially beneficial when the pool's data device is a RAID | |
4025 | * device that has a full stripe width that matches pool->sectors_per_block | |
4026 | * -- because even though partial RAID stripe-sized IOs will be issued to a | |
4027 | * single RAID stripe; when aggregated they will end on a full RAID stripe | |
4028 | * boundary.. which avoids additional partial RAID stripe writes cascading | |
604ea906 | 4029 | */ |
604ea906 MS |
4030 | if (limits->max_sectors < pool->sectors_per_block) { |
4031 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
4032 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
4033 | limits->max_sectors--; | |
4034 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
4035 | } | |
604ea906 | 4036 | } |
991d9fa0 | 4037 | |
0cc67cd9 MS |
4038 | /* |
4039 | * If the system-determined stacked limits are compatible with the | |
4040 | * pool's blocksize (io_opt is a factor) do not override them. | |
4041 | */ | |
4042 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
4043 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
4044 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
4045 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
4046 | else | |
4047 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
4048 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
4049 | } | |
0424caa1 MS |
4050 | |
4051 | /* | |
4052 | * pt->adjusted_pf is a staging area for the actual features to use. | |
4053 | * They get transferred to the live pool in bind_control_target() | |
4054 | * called from pool_preresume(). | |
4055 | */ | |
b60ab990 MS |
4056 | if (!pt->adjusted_pf.discard_enabled) { |
4057 | /* | |
4058 | * Must explicitly disallow stacking discard limits otherwise the | |
4059 | * block layer will stack them if pool's data device has support. | |
4060 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
4061 | * user to see that, so make sure to set all discard limits to 0. | |
4062 | */ | |
4063 | limits->discard_granularity = 0; | |
0424caa1 | 4064 | return; |
b60ab990 | 4065 | } |
0424caa1 MS |
4066 | |
4067 | disable_passdown_if_not_supported(pt); | |
4068 | ||
34fbcf62 JT |
4069 | /* |
4070 | * The pool uses the same discard limits as the underlying data | |
4071 | * device. DM core has already set this up. | |
4072 | */ | |
991d9fa0 JT |
4073 | } |
4074 | ||
4075 | static struct target_type pool_target = { | |
4076 | .name = "thin-pool", | |
4077 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
4078 | DM_TARGET_IMMUTABLE, | |
873937e7 | 4079 | .version = {1, 22, 0}, |
991d9fa0 JT |
4080 | .module = THIS_MODULE, |
4081 | .ctr = pool_ctr, | |
4082 | .dtr = pool_dtr, | |
4083 | .map = pool_map, | |
80e96c54 MS |
4084 | .presuspend = pool_presuspend, |
4085 | .presuspend_undo = pool_presuspend_undo, | |
991d9fa0 JT |
4086 | .postsuspend = pool_postsuspend, |
4087 | .preresume = pool_preresume, | |
4088 | .resume = pool_resume, | |
4089 | .message = pool_message, | |
4090 | .status = pool_status, | |
991d9fa0 JT |
4091 | .iterate_devices = pool_iterate_devices, |
4092 | .io_hints = pool_io_hints, | |
4093 | }; | |
4094 | ||
4095 | /*---------------------------------------------------------------- | |
4096 | * Thin target methods | |
4097 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
4098 | static void thin_get(struct thin_c *tc) |
4099 | { | |
22d4c291 | 4100 | refcount_inc(&tc->refcount); |
b10ebd34 JT |
4101 | } |
4102 | ||
4103 | static void thin_put(struct thin_c *tc) | |
4104 | { | |
22d4c291 | 4105 | if (refcount_dec_and_test(&tc->refcount)) |
b10ebd34 JT |
4106 | complete(&tc->can_destroy); |
4107 | } | |
4108 | ||
991d9fa0 JT |
4109 | static void thin_dtr(struct dm_target *ti) |
4110 | { | |
4111 | struct thin_c *tc = ti->private; | |
c140e1c4 | 4112 | |
8e0c9dac | 4113 | spin_lock_irq(&tc->pool->lock); |
c140e1c4 | 4114 | list_del_rcu(&tc->list); |
8e0c9dac | 4115 | spin_unlock_irq(&tc->pool->lock); |
c140e1c4 | 4116 | synchronize_rcu(); |
991d9fa0 | 4117 | |
17181fb7 MP |
4118 | thin_put(tc); |
4119 | wait_for_completion(&tc->can_destroy); | |
4120 | ||
991d9fa0 JT |
4121 | mutex_lock(&dm_thin_pool_table.mutex); |
4122 | ||
4123 | __pool_dec(tc->pool); | |
4124 | dm_pool_close_thin_device(tc->td); | |
4125 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
4126 | if (tc->origin_dev) |
4127 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
4128 | kfree(tc); |
4129 | ||
4130 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4131 | } | |
4132 | ||
4133 | /* | |
4134 | * Thin target parameters: | |
4135 | * | |
2dd9c257 | 4136 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
4137 | * |
4138 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
4139 | * dev_id: the internal device identifier | |
2dd9c257 | 4140 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
4141 | * |
4142 | * If the pool device has discards disabled, they get disabled for the thin | |
4143 | * device as well. | |
991d9fa0 JT |
4144 | */ |
4145 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
4146 | { | |
4147 | int r; | |
4148 | struct thin_c *tc; | |
2dd9c257 | 4149 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
4150 | struct mapped_device *pool_md; |
4151 | ||
4152 | mutex_lock(&dm_thin_pool_table.mutex); | |
4153 | ||
2dd9c257 | 4154 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
4155 | ti->error = "Invalid argument count"; |
4156 | r = -EINVAL; | |
4157 | goto out_unlock; | |
4158 | } | |
4159 | ||
4160 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
4161 | if (!tc) { | |
4162 | ti->error = "Out of memory"; | |
4163 | r = -ENOMEM; | |
4164 | goto out_unlock; | |
4165 | } | |
583024d2 | 4166 | tc->thin_md = dm_table_get_md(ti->table); |
c140e1c4 | 4167 | spin_lock_init(&tc->lock); |
a374bb21 | 4168 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
4169 | bio_list_init(&tc->deferred_bio_list); |
4170 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 4171 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 4172 | |
2dd9c257 | 4173 | if (argc == 3) { |
70de2cbd JCXF |
4174 | if (!strcmp(argv[0], argv[2])) { |
4175 | ti->error = "Error setting origin device"; | |
4176 | r = -EINVAL; | |
4177 | goto bad_origin_dev; | |
4178 | } | |
4179 | ||
2dd9c257 JT |
4180 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
4181 | if (r) { | |
4182 | ti->error = "Error opening origin device"; | |
4183 | goto bad_origin_dev; | |
4184 | } | |
4185 | tc->origin_dev = origin_dev; | |
4186 | } | |
4187 | ||
991d9fa0 JT |
4188 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
4189 | if (r) { | |
4190 | ti->error = "Error opening pool device"; | |
4191 | goto bad_pool_dev; | |
4192 | } | |
4193 | tc->pool_dev = pool_dev; | |
4194 | ||
4195 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
4196 | ti->error = "Invalid device id"; | |
4197 | r = -EINVAL; | |
4198 | goto bad_common; | |
4199 | } | |
4200 | ||
4201 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
4202 | if (!pool_md) { | |
4203 | ti->error = "Couldn't get pool mapped device"; | |
4204 | r = -EINVAL; | |
4205 | goto bad_common; | |
4206 | } | |
4207 | ||
4208 | tc->pool = __pool_table_lookup(pool_md); | |
4209 | if (!tc->pool) { | |
4210 | ti->error = "Couldn't find pool object"; | |
4211 | r = -EINVAL; | |
4212 | goto bad_pool_lookup; | |
4213 | } | |
4214 | __pool_inc(tc->pool); | |
4215 | ||
e49e5829 JT |
4216 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4217 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 4218 | r = -EINVAL; |
80e96c54 | 4219 | goto bad_pool; |
e49e5829 JT |
4220 | } |
4221 | ||
991d9fa0 JT |
4222 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
4223 | if (r) { | |
4224 | ti->error = "Couldn't open thin internal device"; | |
80e96c54 | 4225 | goto bad_pool; |
991d9fa0 JT |
4226 | } |
4227 | ||
542f9038 MS |
4228 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
4229 | if (r) | |
80e96c54 | 4230 | goto bad; |
542f9038 | 4231 | |
55a62eef | 4232 | ti->num_flush_bios = 1; |
16ad3d10 | 4233 | ti->flush_supported = true; |
30187e1d | 4234 | ti->per_io_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
4235 | |
4236 | /* In case the pool supports discards, pass them on. */ | |
4237 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 4238 | ti->discards_supported = true; |
55a62eef | 4239 | ti->num_discard_bios = 1; |
67e2e2b2 | 4240 | } |
991d9fa0 | 4241 | |
991d9fa0 JT |
4242 | mutex_unlock(&dm_thin_pool_table.mutex); |
4243 | ||
8e0c9dac | 4244 | spin_lock_irq(&tc->pool->lock); |
80e96c54 | 4245 | if (tc->pool->suspended) { |
8e0c9dac | 4246 | spin_unlock_irq(&tc->pool->lock); |
80e96c54 MS |
4247 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ |
4248 | ti->error = "Unable to activate thin device while pool is suspended"; | |
4249 | r = -EINVAL; | |
4250 | goto bad; | |
4251 | } | |
22d4c291 | 4252 | refcount_set(&tc->refcount, 1); |
2b94e896 | 4253 | init_completion(&tc->can_destroy); |
c140e1c4 | 4254 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
8e0c9dac | 4255 | spin_unlock_irq(&tc->pool->lock); |
c140e1c4 MS |
4256 | /* |
4257 | * This synchronize_rcu() call is needed here otherwise we risk a | |
4258 | * wake_worker() call finding no bios to process (because the newly | |
4259 | * added tc isn't yet visible). So this reduces latency since we | |
4260 | * aren't then dependent on the periodic commit to wake_worker(). | |
4261 | */ | |
4262 | synchronize_rcu(); | |
4263 | ||
80e96c54 MS |
4264 | dm_put(pool_md); |
4265 | ||
991d9fa0 JT |
4266 | return 0; |
4267 | ||
80e96c54 | 4268 | bad: |
1acacc07 | 4269 | dm_pool_close_thin_device(tc->td); |
80e96c54 | 4270 | bad_pool: |
991d9fa0 JT |
4271 | __pool_dec(tc->pool); |
4272 | bad_pool_lookup: | |
4273 | dm_put(pool_md); | |
4274 | bad_common: | |
4275 | dm_put_device(ti, tc->pool_dev); | |
4276 | bad_pool_dev: | |
2dd9c257 JT |
4277 | if (tc->origin_dev) |
4278 | dm_put_device(ti, tc->origin_dev); | |
4279 | bad_origin_dev: | |
991d9fa0 JT |
4280 | kfree(tc); |
4281 | out_unlock: | |
4282 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4283 | ||
4284 | return r; | |
4285 | } | |
4286 | ||
7de3ee57 | 4287 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 4288 | { |
4f024f37 | 4289 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 4290 | |
7de3ee57 | 4291 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
4292 | } |
4293 | ||
4e4cbee9 CH |
4294 | static int thin_endio(struct dm_target *ti, struct bio *bio, |
4295 | blk_status_t *err) | |
eb2aa48d JT |
4296 | { |
4297 | unsigned long flags; | |
59c3d2c6 | 4298 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 4299 | struct list_head work; |
a24c2569 | 4300 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
4301 | struct pool *pool = h->tc->pool; |
4302 | ||
4303 | if (h->shared_read_entry) { | |
4304 | INIT_LIST_HEAD(&work); | |
44feb387 | 4305 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
4306 | |
4307 | spin_lock_irqsave(&pool->lock, flags); | |
4308 | list_for_each_entry_safe(m, tmp, &work, list) { | |
4309 | list_del(&m->list); | |
50f3c3ef | 4310 | __complete_mapping_preparation(m); |
eb2aa48d JT |
4311 | } |
4312 | spin_unlock_irqrestore(&pool->lock, flags); | |
4313 | } | |
4314 | ||
104655fd JT |
4315 | if (h->all_io_entry) { |
4316 | INIT_LIST_HEAD(&work); | |
44feb387 | 4317 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
4318 | if (!list_empty(&work)) { |
4319 | spin_lock_irqsave(&pool->lock, flags); | |
4320 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 4321 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
4322 | spin_unlock_irqrestore(&pool->lock, flags); |
4323 | wake_worker(pool); | |
4324 | } | |
104655fd JT |
4325 | } |
4326 | ||
34fbcf62 JT |
4327 | if (h->cell) |
4328 | cell_defer_no_holder(h->tc, h->cell); | |
4329 | ||
1be56909 | 4330 | return DM_ENDIO_DONE; |
eb2aa48d JT |
4331 | } |
4332 | ||
738211f7 | 4333 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 4334 | { |
738211f7 JT |
4335 | struct thin_c *tc = ti->private; |
4336 | ||
991d9fa0 | 4337 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
4338 | noflush_work(tc, do_noflush_start); |
4339 | } | |
4340 | ||
4341 | static void thin_postsuspend(struct dm_target *ti) | |
4342 | { | |
4343 | struct thin_c *tc = ti->private; | |
4344 | ||
4345 | /* | |
4346 | * The dm_noflush_suspending flag has been cleared by now, so | |
4347 | * unfortunately we must always run this. | |
4348 | */ | |
4349 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
4350 | } |
4351 | ||
e5aea7b4 JT |
4352 | static int thin_preresume(struct dm_target *ti) |
4353 | { | |
4354 | struct thin_c *tc = ti->private; | |
4355 | ||
4356 | if (tc->origin_dev) | |
4357 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
4358 | ||
4359 | return 0; | |
4360 | } | |
4361 | ||
991d9fa0 JT |
4362 | /* |
4363 | * <nr mapped sectors> <highest mapped sector> | |
4364 | */ | |
fd7c092e MP |
4365 | static void thin_status(struct dm_target *ti, status_type_t type, |
4366 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
4367 | { |
4368 | int r; | |
4369 | ssize_t sz = 0; | |
4370 | dm_block_t mapped, highest; | |
4371 | char buf[BDEVNAME_SIZE]; | |
4372 | struct thin_c *tc = ti->private; | |
4373 | ||
e49e5829 JT |
4374 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4375 | DMEMIT("Fail"); | |
fd7c092e | 4376 | return; |
e49e5829 JT |
4377 | } |
4378 | ||
991d9fa0 JT |
4379 | if (!tc->td) |
4380 | DMEMIT("-"); | |
4381 | else { | |
4382 | switch (type) { | |
4383 | case STATUSTYPE_INFO: | |
4384 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
4385 | if (r) { |
4386 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
4387 | goto err; | |
4388 | } | |
991d9fa0 JT |
4389 | |
4390 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
4391 | if (r < 0) { |
4392 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
4393 | goto err; | |
4394 | } | |
991d9fa0 JT |
4395 | |
4396 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
4397 | if (r) | |
4398 | DMEMIT("%llu", ((highest + 1) * | |
4399 | tc->pool->sectors_per_block) - 1); | |
4400 | else | |
4401 | DMEMIT("-"); | |
4402 | break; | |
4403 | ||
4404 | case STATUSTYPE_TABLE: | |
4405 | DMEMIT("%s %lu", | |
4406 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
4407 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
4408 | if (tc->origin_dev) |
4409 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 | 4410 | break; |
8ec45662 TS |
4411 | |
4412 | case STATUSTYPE_IMA: | |
4413 | *result = '\0'; | |
4414 | break; | |
991d9fa0 JT |
4415 | } |
4416 | } | |
4417 | ||
fd7c092e MP |
4418 | return; |
4419 | ||
4420 | err: | |
4421 | DMEMIT("Error"); | |
991d9fa0 JT |
4422 | } |
4423 | ||
4424 | static int thin_iterate_devices(struct dm_target *ti, | |
4425 | iterate_devices_callout_fn fn, void *data) | |
4426 | { | |
55f2b8bd | 4427 | sector_t blocks; |
991d9fa0 | 4428 | struct thin_c *tc = ti->private; |
55f2b8bd | 4429 | struct pool *pool = tc->pool; |
991d9fa0 JT |
4430 | |
4431 | /* | |
4432 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
4433 | * we follow a more convoluted path through to the pool's target. | |
4434 | */ | |
55f2b8bd | 4435 | if (!pool->ti) |
991d9fa0 JT |
4436 | return 0; /* nothing is bound */ |
4437 | ||
55f2b8bd MS |
4438 | blocks = pool->ti->len; |
4439 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 4440 | if (blocks) |
55f2b8bd | 4441 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
4442 | |
4443 | return 0; | |
4444 | } | |
4445 | ||
34fbcf62 JT |
4446 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4447 | { | |
4448 | struct thin_c *tc = ti->private; | |
4449 | struct pool *pool = tc->pool; | |
21607670 | 4450 | |
0fcb04d5 MS |
4451 | if (!pool->pf.discard_enabled) |
4452 | return; | |
34fbcf62 JT |
4453 | |
4454 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
4455 | limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ | |
4456 | } | |
4457 | ||
991d9fa0 JT |
4458 | static struct target_type thin_target = { |
4459 | .name = "thin", | |
873937e7 | 4460 | .version = {1, 22, 0}, |
991d9fa0 JT |
4461 | .module = THIS_MODULE, |
4462 | .ctr = thin_ctr, | |
4463 | .dtr = thin_dtr, | |
4464 | .map = thin_map, | |
eb2aa48d | 4465 | .end_io = thin_endio, |
e5aea7b4 | 4466 | .preresume = thin_preresume, |
738211f7 | 4467 | .presuspend = thin_presuspend, |
991d9fa0 JT |
4468 | .postsuspend = thin_postsuspend, |
4469 | .status = thin_status, | |
4470 | .iterate_devices = thin_iterate_devices, | |
34fbcf62 | 4471 | .io_hints = thin_io_hints, |
991d9fa0 JT |
4472 | }; |
4473 | ||
4474 | /*----------------------------------------------------------------*/ | |
4475 | ||
4476 | static int __init dm_thin_init(void) | |
4477 | { | |
7e6358d2 | 4478 | int r = -ENOMEM; |
991d9fa0 JT |
4479 | |
4480 | pool_table_init(); | |
4481 | ||
7e6358d2 | 4482 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
4483 | if (!_new_mapping_cache) | |
4484 | return r; | |
4485 | ||
991d9fa0 JT |
4486 | r = dm_register_target(&thin_target); |
4487 | if (r) | |
7e6358d2 | 4488 | goto bad_new_mapping_cache; |
991d9fa0 JT |
4489 | |
4490 | r = dm_register_target(&pool_target); | |
4491 | if (r) | |
7e6358d2 | 4492 | goto bad_thin_target; |
a24c2569 | 4493 | |
a24c2569 MS |
4494 | return 0; |
4495 | ||
7e6358d2 | 4496 | bad_thin_target: |
a24c2569 | 4497 | dm_unregister_target(&thin_target); |
7e6358d2 | 4498 | bad_new_mapping_cache: |
4499 | kmem_cache_destroy(_new_mapping_cache); | |
991d9fa0 JT |
4500 | |
4501 | return r; | |
4502 | } | |
4503 | ||
4504 | static void dm_thin_exit(void) | |
4505 | { | |
4506 | dm_unregister_target(&thin_target); | |
4507 | dm_unregister_target(&pool_target); | |
a24c2569 | 4508 | |
a24c2569 | 4509 | kmem_cache_destroy(_new_mapping_cache); |
d5ffebdd MS |
4510 | |
4511 | pool_table_exit(); | |
991d9fa0 JT |
4512 | } |
4513 | ||
4514 | module_init(dm_thin_init); | |
4515 | module_exit(dm_thin_exit); | |
4516 | ||
80c57893 MS |
4517 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
4518 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
4519 | ||
7cab8bf1 | 4520 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4521 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4522 | MODULE_LICENSE("GPL"); |