]>
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" | |
4f81a417 | 8 | #include "dm-bio-prison.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> | |
604ea906 | 14 | #include <linux/log2.h> |
991d9fa0 | 15 | #include <linux/list.h> |
c140e1c4 | 16 | #include <linux/rculist.h> |
991d9fa0 JT |
17 | #include <linux/init.h> |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
67324ea1 | 20 | #include <linux/rbtree.h> |
991d9fa0 JT |
21 | |
22 | #define DM_MSG_PREFIX "thin" | |
23 | ||
24 | /* | |
25 | * Tunable constants | |
26 | */ | |
7768ed33 | 27 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 28 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 29 | #define COMMIT_PERIOD HZ |
80c57893 MS |
30 | #define NO_SPACE_TIMEOUT_SECS 60 |
31 | ||
32 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 33 | |
df5d2e90 MP |
34 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
35 | "A percentage of time allocated for copy on write"); | |
36 | ||
991d9fa0 JT |
37 | /* |
38 | * The block size of the device holding pool data must be | |
39 | * between 64KB and 1GB. | |
40 | */ | |
41 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
42 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
43 | ||
991d9fa0 JT |
44 | /* |
45 | * Device id is restricted to 24 bits. | |
46 | */ | |
47 | #define MAX_DEV_ID ((1 << 24) - 1) | |
48 | ||
49 | /* | |
50 | * How do we handle breaking sharing of data blocks? | |
51 | * ================================================= | |
52 | * | |
53 | * We use a standard copy-on-write btree to store the mappings for the | |
54 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
55 | * the data). When you take an internal snapshot you clone the root node | |
56 | * of the origin btree. After this there is no concept of an origin or a | |
57 | * snapshot. They are just two device trees that happen to point to the | |
58 | * same data blocks. | |
59 | * | |
60 | * When we get a write in we decide if it's to a shared data block using | |
61 | * some timestamp magic. If it is, we have to break sharing. | |
62 | * | |
63 | * Let's say we write to a shared block in what was the origin. The | |
64 | * steps are: | |
65 | * | |
66 | * i) plug io further to this physical block. (see bio_prison code). | |
67 | * | |
68 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 69 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
70 | * |
71 | * iii) copy the data block to a newly allocate block. This step can be | |
72 | * missed out if the io covers the block. (schedule_copy). | |
73 | * | |
74 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 75 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
76 | * sharing of btree nodes between the two devices. Breaking sharing only |
77 | * effects the btree of that specific device. Btrees for the other | |
78 | * devices that share the block never change. The btree for the origin | |
79 | * device as it was after the last commit is untouched, ie. we're using | |
80 | * persistent data structures in the functional programming sense. | |
81 | * | |
82 | * v) unplug io to this physical block, including the io that triggered | |
83 | * the breaking of sharing. | |
84 | * | |
85 | * Steps (ii) and (iii) occur in parallel. | |
86 | * | |
87 | * The metadata _doesn't_ need to be committed before the io continues. We | |
88 | * get away with this because the io is always written to a _new_ block. | |
89 | * If there's a crash, then: | |
90 | * | |
91 | * - The origin mapping will point to the old origin block (the shared | |
92 | * one). This will contain the data as it was before the io that triggered | |
93 | * the breaking of sharing came in. | |
94 | * | |
95 | * - The snap mapping still points to the old block. As it would after | |
96 | * the commit. | |
97 | * | |
98 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
99 | * will continue to think that data block in the snapshot device is shared | |
100 | * even after the write to the origin has broken sharing. I suspect data | |
101 | * blocks will typically be shared by many different devices, so we're | |
102 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
103 | * devices that reference this data block. At the moment I think the | |
104 | * benefits far, far outweigh the disadvantages. | |
105 | */ | |
106 | ||
107 | /*----------------------------------------------------------------*/ | |
108 | ||
991d9fa0 JT |
109 | /* |
110 | * Key building. | |
111 | */ | |
112 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 113 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
114 | { |
115 | key->virtual = 0; | |
116 | key->dev = dm_thin_dev_id(td); | |
117 | key->block = b; | |
118 | } | |
119 | ||
120 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 121 | struct dm_cell_key *key) |
991d9fa0 JT |
122 | { |
123 | key->virtual = 1; | |
124 | key->dev = dm_thin_dev_id(td); | |
125 | key->block = b; | |
126 | } | |
127 | ||
128 | /*----------------------------------------------------------------*/ | |
129 | ||
7d327fe0 JT |
130 | #define THROTTLE_THRESHOLD (1 * HZ) |
131 | ||
132 | struct throttle { | |
133 | struct rw_semaphore lock; | |
134 | unsigned long threshold; | |
135 | bool throttle_applied; | |
136 | }; | |
137 | ||
138 | static void throttle_init(struct throttle *t) | |
139 | { | |
140 | init_rwsem(&t->lock); | |
141 | t->throttle_applied = false; | |
142 | } | |
143 | ||
144 | static void throttle_work_start(struct throttle *t) | |
145 | { | |
146 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
147 | } | |
148 | ||
149 | static void throttle_work_update(struct throttle *t) | |
150 | { | |
151 | if (!t->throttle_applied && jiffies > t->threshold) { | |
152 | down_write(&t->lock); | |
153 | t->throttle_applied = true; | |
154 | } | |
155 | } | |
156 | ||
157 | static void throttle_work_complete(struct throttle *t) | |
158 | { | |
159 | if (t->throttle_applied) { | |
160 | t->throttle_applied = false; | |
161 | up_write(&t->lock); | |
162 | } | |
163 | } | |
164 | ||
165 | static void throttle_lock(struct throttle *t) | |
166 | { | |
167 | down_read(&t->lock); | |
168 | } | |
169 | ||
170 | static void throttle_unlock(struct throttle *t) | |
171 | { | |
172 | up_read(&t->lock); | |
173 | } | |
174 | ||
175 | /*----------------------------------------------------------------*/ | |
176 | ||
991d9fa0 JT |
177 | /* |
178 | * A pool device ties together a metadata device and a data device. It | |
179 | * also provides the interface for creating and destroying internal | |
180 | * devices. | |
181 | */ | |
a24c2569 | 182 | struct dm_thin_new_mapping; |
67e2e2b2 | 183 | |
e49e5829 | 184 | /* |
3e1a0699 | 185 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
186 | */ |
187 | enum pool_mode { | |
188 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 189 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
190 | PM_READ_ONLY, /* metadata may not be changed */ |
191 | PM_FAIL, /* all I/O fails */ | |
192 | }; | |
193 | ||
67e2e2b2 | 194 | struct pool_features { |
e49e5829 JT |
195 | enum pool_mode mode; |
196 | ||
9bc142dd MS |
197 | bool zero_new_blocks:1; |
198 | bool discard_enabled:1; | |
199 | bool discard_passdown:1; | |
787a996c | 200 | bool error_if_no_space:1; |
67e2e2b2 JT |
201 | }; |
202 | ||
e49e5829 JT |
203 | struct thin_c; |
204 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 205 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
206 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
207 | ||
991d9fa0 JT |
208 | struct pool { |
209 | struct list_head list; | |
210 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
211 | ||
212 | struct mapped_device *pool_md; | |
213 | struct block_device *md_dev; | |
214 | struct dm_pool_metadata *pmd; | |
215 | ||
991d9fa0 | 216 | dm_block_t low_water_blocks; |
55f2b8bd | 217 | uint32_t sectors_per_block; |
f9a8e0cd | 218 | int sectors_per_block_shift; |
991d9fa0 | 219 | |
67e2e2b2 | 220 | struct pool_features pf; |
88a6621b | 221 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 222 | |
44feb387 | 223 | struct dm_bio_prison *prison; |
991d9fa0 JT |
224 | struct dm_kcopyd_client *copier; |
225 | ||
226 | struct workqueue_struct *wq; | |
7d327fe0 | 227 | struct throttle throttle; |
991d9fa0 | 228 | struct work_struct worker; |
905e51b3 | 229 | struct delayed_work waker; |
85ad643b | 230 | struct delayed_work no_space_timeout; |
991d9fa0 | 231 | |
905e51b3 | 232 | unsigned long last_commit_jiffies; |
55f2b8bd | 233 | unsigned ref_count; |
991d9fa0 JT |
234 | |
235 | spinlock_t lock; | |
991d9fa0 JT |
236 | struct bio_list deferred_flush_bios; |
237 | struct list_head prepared_mappings; | |
104655fd | 238 | struct list_head prepared_discards; |
c140e1c4 | 239 | struct list_head active_thins; |
991d9fa0 | 240 | |
44feb387 MS |
241 | struct dm_deferred_set *shared_read_ds; |
242 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 243 | |
a24c2569 | 244 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 245 | mempool_t *mapping_pool; |
e49e5829 JT |
246 | |
247 | process_bio_fn process_bio; | |
248 | process_bio_fn process_discard; | |
249 | ||
a374bb21 JT |
250 | process_cell_fn process_cell; |
251 | process_cell_fn process_discard_cell; | |
252 | ||
e49e5829 JT |
253 | process_mapping_fn process_prepared_mapping; |
254 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
255 | }; |
256 | ||
e49e5829 | 257 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 258 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 259 | |
991d9fa0 JT |
260 | /* |
261 | * Target context for a pool. | |
262 | */ | |
263 | struct pool_c { | |
264 | struct dm_target *ti; | |
265 | struct pool *pool; | |
266 | struct dm_dev *data_dev; | |
267 | struct dm_dev *metadata_dev; | |
268 | struct dm_target_callbacks callbacks; | |
269 | ||
270 | dm_block_t low_water_blocks; | |
0424caa1 MS |
271 | struct pool_features requested_pf; /* Features requested during table load */ |
272 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
273 | }; |
274 | ||
275 | /* | |
276 | * Target context for a thin. | |
277 | */ | |
278 | struct thin_c { | |
c140e1c4 | 279 | struct list_head list; |
991d9fa0 | 280 | struct dm_dev *pool_dev; |
2dd9c257 | 281 | struct dm_dev *origin_dev; |
e5aea7b4 | 282 | sector_t origin_size; |
991d9fa0 JT |
283 | dm_thin_id dev_id; |
284 | ||
285 | struct pool *pool; | |
286 | struct dm_thin_device *td; | |
738211f7 | 287 | bool requeue_mode:1; |
c140e1c4 | 288 | spinlock_t lock; |
a374bb21 | 289 | struct list_head deferred_cells; |
c140e1c4 MS |
290 | struct bio_list deferred_bio_list; |
291 | struct bio_list retry_on_resume_list; | |
67324ea1 | 292 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
293 | |
294 | /* | |
295 | * Ensures the thin is not destroyed until the worker has finished | |
296 | * iterating the active_thins list. | |
297 | */ | |
298 | atomic_t refcount; | |
299 | struct completion can_destroy; | |
991d9fa0 JT |
300 | }; |
301 | ||
302 | /*----------------------------------------------------------------*/ | |
303 | ||
025b9685 JT |
304 | /* |
305 | * wake_worker() is used when new work is queued and when pool_resume is | |
306 | * ready to continue deferred IO processing. | |
307 | */ | |
308 | static void wake_worker(struct pool *pool) | |
309 | { | |
310 | queue_work(pool->wq, &pool->worker); | |
311 | } | |
312 | ||
313 | /*----------------------------------------------------------------*/ | |
314 | ||
6beca5eb JT |
315 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
316 | struct dm_bio_prison_cell **cell_result) | |
317 | { | |
318 | int r; | |
319 | struct dm_bio_prison_cell *cell_prealloc; | |
320 | ||
321 | /* | |
322 | * Allocate a cell from the prison's mempool. | |
323 | * This might block but it can't fail. | |
324 | */ | |
325 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
326 | ||
327 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
328 | if (r) | |
329 | /* | |
330 | * We reused an old cell; we can get rid of | |
331 | * the new one. | |
332 | */ | |
333 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
334 | ||
335 | return r; | |
336 | } | |
337 | ||
338 | static void cell_release(struct pool *pool, | |
339 | struct dm_bio_prison_cell *cell, | |
340 | struct bio_list *bios) | |
341 | { | |
342 | dm_cell_release(pool->prison, cell, bios); | |
343 | dm_bio_prison_free_cell(pool->prison, cell); | |
344 | } | |
345 | ||
2d759a46 JT |
346 | static void cell_visit_release(struct pool *pool, |
347 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
348 | void *context, | |
349 | struct dm_bio_prison_cell *cell) | |
350 | { | |
351 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
352 | dm_bio_prison_free_cell(pool->prison, cell); | |
353 | } | |
354 | ||
6beca5eb JT |
355 | static void cell_release_no_holder(struct pool *pool, |
356 | struct dm_bio_prison_cell *cell, | |
357 | struct bio_list *bios) | |
358 | { | |
359 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
360 | dm_bio_prison_free_cell(pool->prison, cell); | |
361 | } | |
362 | ||
af91805a MS |
363 | static void cell_error_with_code(struct pool *pool, |
364 | struct dm_bio_prison_cell *cell, int error_code) | |
6beca5eb | 365 | { |
af91805a | 366 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
367 | dm_bio_prison_free_cell(pool->prison, cell); |
368 | } | |
369 | ||
af91805a MS |
370 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
371 | { | |
372 | cell_error_with_code(pool, cell, -EIO); | |
373 | } | |
374 | ||
a374bb21 JT |
375 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
376 | { | |
377 | cell_error_with_code(pool, cell, 0); | |
378 | } | |
379 | ||
380 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
381 | { | |
382 | cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE); | |
383 | } | |
384 | ||
6beca5eb JT |
385 | /*----------------------------------------------------------------*/ |
386 | ||
991d9fa0 JT |
387 | /* |
388 | * A global list of pools that uses a struct mapped_device as a key. | |
389 | */ | |
390 | static struct dm_thin_pool_table { | |
391 | struct mutex mutex; | |
392 | struct list_head pools; | |
393 | } dm_thin_pool_table; | |
394 | ||
395 | static void pool_table_init(void) | |
396 | { | |
397 | mutex_init(&dm_thin_pool_table.mutex); | |
398 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
399 | } | |
400 | ||
401 | static void __pool_table_insert(struct pool *pool) | |
402 | { | |
403 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
404 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
405 | } | |
406 | ||
407 | static void __pool_table_remove(struct pool *pool) | |
408 | { | |
409 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
410 | list_del(&pool->list); | |
411 | } | |
412 | ||
413 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
414 | { | |
415 | struct pool *pool = NULL, *tmp; | |
416 | ||
417 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
418 | ||
419 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
420 | if (tmp->pool_md == md) { | |
421 | pool = tmp; | |
422 | break; | |
423 | } | |
424 | } | |
425 | ||
426 | return pool; | |
427 | } | |
428 | ||
429 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
430 | { | |
431 | struct pool *pool = NULL, *tmp; | |
432 | ||
433 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
434 | ||
435 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
436 | if (tmp->md_dev == md_dev) { | |
437 | pool = tmp; | |
438 | break; | |
439 | } | |
440 | } | |
441 | ||
442 | return pool; | |
443 | } | |
444 | ||
445 | /*----------------------------------------------------------------*/ | |
446 | ||
a24c2569 | 447 | struct dm_thin_endio_hook { |
eb2aa48d | 448 | struct thin_c *tc; |
44feb387 MS |
449 | struct dm_deferred_entry *shared_read_entry; |
450 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 451 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 452 | struct rb_node rb_node; |
eb2aa48d JT |
453 | }; |
454 | ||
18adc577 | 455 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
456 | { |
457 | struct bio *bio; | |
458 | struct bio_list bios; | |
18adc577 | 459 | unsigned long flags; |
991d9fa0 JT |
460 | |
461 | bio_list_init(&bios); | |
18adc577 | 462 | |
c140e1c4 | 463 | spin_lock_irqsave(&tc->lock, flags); |
991d9fa0 JT |
464 | bio_list_merge(&bios, master); |
465 | bio_list_init(master); | |
c140e1c4 | 466 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 467 | |
c140e1c4 MS |
468 | while ((bio = bio_list_pop(&bios))) |
469 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
991d9fa0 JT |
470 | } |
471 | ||
a374bb21 JT |
472 | static void requeue_deferred_cells(struct thin_c *tc) |
473 | { | |
474 | struct pool *pool = tc->pool; | |
475 | unsigned long flags; | |
476 | struct list_head cells; | |
477 | struct dm_bio_prison_cell *cell, *tmp; | |
478 | ||
479 | INIT_LIST_HEAD(&cells); | |
480 | ||
481 | spin_lock_irqsave(&tc->lock, flags); | |
482 | list_splice_init(&tc->deferred_cells, &cells); | |
483 | spin_unlock_irqrestore(&tc->lock, flags); | |
484 | ||
485 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
486 | cell_requeue(pool, cell); | |
487 | } | |
488 | ||
991d9fa0 JT |
489 | static void requeue_io(struct thin_c *tc) |
490 | { | |
c140e1c4 MS |
491 | requeue_bio_list(tc, &tc->deferred_bio_list); |
492 | requeue_bio_list(tc, &tc->retry_on_resume_list); | |
a374bb21 | 493 | requeue_deferred_cells(tc); |
991d9fa0 JT |
494 | } |
495 | ||
c140e1c4 | 496 | static void error_thin_retry_list(struct thin_c *tc) |
3e1a0699 JT |
497 | { |
498 | struct bio *bio; | |
499 | unsigned long flags; | |
500 | struct bio_list bios; | |
501 | ||
502 | bio_list_init(&bios); | |
503 | ||
c140e1c4 MS |
504 | spin_lock_irqsave(&tc->lock, flags); |
505 | bio_list_merge(&bios, &tc->retry_on_resume_list); | |
506 | bio_list_init(&tc->retry_on_resume_list); | |
507 | spin_unlock_irqrestore(&tc->lock, flags); | |
3e1a0699 JT |
508 | |
509 | while ((bio = bio_list_pop(&bios))) | |
510 | bio_io_error(bio); | |
511 | } | |
512 | ||
c140e1c4 MS |
513 | static void error_retry_list(struct pool *pool) |
514 | { | |
515 | struct thin_c *tc; | |
516 | ||
517 | rcu_read_lock(); | |
518 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
519 | error_thin_retry_list(tc); | |
520 | rcu_read_unlock(); | |
521 | } | |
522 | ||
991d9fa0 JT |
523 | /* |
524 | * This section of code contains the logic for processing a thin device's IO. | |
525 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
526 | * but most is exclusively called from the thin target rather than the thin-pool | |
527 | * target. | |
528 | */ | |
529 | ||
58f77a21 MS |
530 | static bool block_size_is_power_of_two(struct pool *pool) |
531 | { | |
532 | return pool->sectors_per_block_shift >= 0; | |
533 | } | |
534 | ||
991d9fa0 JT |
535 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
536 | { | |
58f77a21 | 537 | struct pool *pool = tc->pool; |
4f024f37 | 538 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 539 | |
58f77a21 MS |
540 | if (block_size_is_power_of_two(pool)) |
541 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 542 | else |
58f77a21 | 543 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
544 | |
545 | return block_nr; | |
991d9fa0 JT |
546 | } |
547 | ||
548 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
549 | { | |
550 | struct pool *pool = tc->pool; | |
4f024f37 | 551 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
552 | |
553 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 554 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
555 | bio->bi_iter.bi_sector = |
556 | (block << pool->sectors_per_block_shift) | | |
557 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 558 | else |
4f024f37 | 559 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 560 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
561 | } |
562 | ||
2dd9c257 JT |
563 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
564 | { | |
565 | bio->bi_bdev = tc->origin_dev->bdev; | |
566 | } | |
567 | ||
4afdd680 JT |
568 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
569 | { | |
570 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
571 | dm_thin_changed_this_transaction(tc->td); | |
572 | } | |
573 | ||
e8088073 JT |
574 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
575 | { | |
576 | struct dm_thin_endio_hook *h; | |
577 | ||
578 | if (bio->bi_rw & REQ_DISCARD) | |
579 | return; | |
580 | ||
59c3d2c6 | 581 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
582 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
583 | } | |
584 | ||
2dd9c257 | 585 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
586 | { |
587 | struct pool *pool = tc->pool; | |
588 | unsigned long flags; | |
589 | ||
e49e5829 JT |
590 | if (!bio_triggers_commit(tc, bio)) { |
591 | generic_make_request(bio); | |
592 | return; | |
593 | } | |
594 | ||
991d9fa0 | 595 | /* |
e49e5829 JT |
596 | * Complete bio with an error if earlier I/O caused changes to |
597 | * the metadata that can't be committed e.g, due to I/O errors | |
598 | * on the metadata device. | |
991d9fa0 | 599 | */ |
e49e5829 JT |
600 | if (dm_thin_aborted_changes(tc->td)) { |
601 | bio_io_error(bio); | |
602 | return; | |
603 | } | |
604 | ||
605 | /* | |
606 | * Batch together any bios that trigger commits and then issue a | |
607 | * single commit for them in process_deferred_bios(). | |
608 | */ | |
609 | spin_lock_irqsave(&pool->lock, flags); | |
610 | bio_list_add(&pool->deferred_flush_bios, bio); | |
611 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
612 | } |
613 | ||
2dd9c257 JT |
614 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
615 | { | |
616 | remap_to_origin(tc, bio); | |
617 | issue(tc, bio); | |
618 | } | |
619 | ||
620 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
621 | dm_block_t block) | |
622 | { | |
623 | remap(tc, bio, block); | |
624 | issue(tc, bio); | |
625 | } | |
626 | ||
991d9fa0 JT |
627 | /*----------------------------------------------------------------*/ |
628 | ||
629 | /* | |
630 | * Bio endio functions. | |
631 | */ | |
a24c2569 | 632 | struct dm_thin_new_mapping { |
991d9fa0 JT |
633 | struct list_head list; |
634 | ||
7f214665 MS |
635 | bool pass_discard:1; |
636 | bool definitely_not_shared:1; | |
991d9fa0 | 637 | |
50f3c3ef JT |
638 | /* |
639 | * Track quiescing, copying and zeroing preparation actions. When this | |
640 | * counter hits zero the block is prepared and can be inserted into the | |
641 | * btree. | |
642 | */ | |
643 | atomic_t prepare_actions; | |
644 | ||
7f214665 | 645 | int err; |
991d9fa0 JT |
646 | struct thin_c *tc; |
647 | dm_block_t virt_block; | |
648 | dm_block_t data_block; | |
a24c2569 | 649 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
650 | |
651 | /* | |
652 | * If the bio covers the whole area of a block then we can avoid | |
653 | * zeroing or copying. Instead this bio is hooked. The bio will | |
654 | * still be in the cell, so care has to be taken to avoid issuing | |
655 | * the bio twice. | |
656 | */ | |
657 | struct bio *bio; | |
658 | bio_end_io_t *saved_bi_end_io; | |
659 | }; | |
660 | ||
50f3c3ef | 661 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
662 | { |
663 | struct pool *pool = m->tc->pool; | |
664 | ||
50f3c3ef | 665 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 666 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
667 | wake_worker(pool); |
668 | } | |
669 | } | |
670 | ||
e5aea7b4 | 671 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
672 | { |
673 | unsigned long flags; | |
991d9fa0 JT |
674 | struct pool *pool = m->tc->pool; |
675 | ||
991d9fa0 | 676 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 677 | __complete_mapping_preparation(m); |
991d9fa0 JT |
678 | spin_unlock_irqrestore(&pool->lock, flags); |
679 | } | |
680 | ||
e5aea7b4 JT |
681 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
682 | { | |
683 | struct dm_thin_new_mapping *m = context; | |
684 | ||
685 | m->err = read_err || write_err ? -EIO : 0; | |
686 | complete_mapping_preparation(m); | |
687 | } | |
688 | ||
991d9fa0 JT |
689 | static void overwrite_endio(struct bio *bio, int err) |
690 | { | |
59c3d2c6 | 691 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 692 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
693 | |
694 | m->err = err; | |
e5aea7b4 | 695 | complete_mapping_preparation(m); |
991d9fa0 JT |
696 | } |
697 | ||
991d9fa0 JT |
698 | /*----------------------------------------------------------------*/ |
699 | ||
700 | /* | |
701 | * Workqueue. | |
702 | */ | |
703 | ||
704 | /* | |
705 | * Prepared mapping jobs. | |
706 | */ | |
707 | ||
708 | /* | |
2d759a46 JT |
709 | * This sends the bios in the cell, except the original holder, back |
710 | * to the deferred_bios list. | |
991d9fa0 | 711 | */ |
f286ba0e | 712 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 713 | { |
991d9fa0 JT |
714 | struct pool *pool = tc->pool; |
715 | unsigned long flags; | |
716 | ||
c140e1c4 MS |
717 | spin_lock_irqsave(&tc->lock, flags); |
718 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
719 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
720 | |
721 | wake_worker(pool); | |
722 | } | |
723 | ||
a374bb21 JT |
724 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
725 | ||
2d759a46 JT |
726 | struct remap_info { |
727 | struct thin_c *tc; | |
728 | struct bio_list defer_bios; | |
729 | struct bio_list issue_bios; | |
730 | }; | |
731 | ||
732 | static void __inc_remap_and_issue_cell(void *context, | |
733 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 734 | { |
2d759a46 | 735 | struct remap_info *info = context; |
a374bb21 | 736 | struct bio *bio; |
a374bb21 | 737 | |
2d759a46 | 738 | while ((bio = bio_list_pop(&cell->bios))) { |
a374bb21 | 739 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) |
2d759a46 | 740 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 741 | else { |
2d759a46 JT |
742 | inc_all_io_entry(info->tc->pool, bio); |
743 | ||
744 | /* | |
745 | * We can't issue the bios with the bio prison lock | |
746 | * held, so we add them to a list to issue on | |
747 | * return from this function. | |
748 | */ | |
749 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
750 | } |
751 | } | |
752 | } | |
753 | ||
2d759a46 JT |
754 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
755 | struct dm_bio_prison_cell *cell, | |
756 | dm_block_t block) | |
757 | { | |
758 | struct bio *bio; | |
759 | struct remap_info info; | |
760 | ||
761 | info.tc = tc; | |
762 | bio_list_init(&info.defer_bios); | |
763 | bio_list_init(&info.issue_bios); | |
764 | ||
765 | /* | |
766 | * We have to be careful to inc any bios we're about to issue | |
767 | * before the cell is released, and avoid a race with new bios | |
768 | * being added to the cell. | |
769 | */ | |
770 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
771 | &info, cell); | |
772 | ||
773 | while ((bio = bio_list_pop(&info.defer_bios))) | |
774 | thin_defer_bio(tc, bio); | |
775 | ||
776 | while ((bio = bio_list_pop(&info.issue_bios))) | |
777 | remap_and_issue(info.tc, bio, block); | |
778 | } | |
779 | ||
e49e5829 JT |
780 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
781 | { | |
196d38bc | 782 | if (m->bio) { |
e49e5829 | 783 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
784 | atomic_inc(&m->bio->bi_remaining); |
785 | } | |
6beca5eb | 786 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
787 | list_del(&m->list); |
788 | mempool_free(m, m->tc->pool->mapping_pool); | |
789 | } | |
025b9685 | 790 | |
a24c2569 | 791 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
792 | { |
793 | struct thin_c *tc = m->tc; | |
6beca5eb | 794 | struct pool *pool = tc->pool; |
991d9fa0 JT |
795 | struct bio *bio; |
796 | int r; | |
797 | ||
798 | bio = m->bio; | |
196d38bc | 799 | if (bio) { |
991d9fa0 | 800 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
801 | atomic_inc(&bio->bi_remaining); |
802 | } | |
991d9fa0 JT |
803 | |
804 | if (m->err) { | |
6beca5eb | 805 | cell_error(pool, m->cell); |
905386f8 | 806 | goto out; |
991d9fa0 JT |
807 | } |
808 | ||
809 | /* | |
810 | * Commit the prepared block into the mapping btree. | |
811 | * Any I/O for this block arriving after this point will get | |
812 | * remapped to it directly. | |
813 | */ | |
814 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
815 | if (r) { | |
b5330655 | 816 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 817 | cell_error(pool, m->cell); |
905386f8 | 818 | goto out; |
991d9fa0 JT |
819 | } |
820 | ||
821 | /* | |
822 | * Release any bios held while the block was being provisioned. | |
823 | * If we are processing a write bio that completely covers the block, | |
824 | * we already processed it so can ignore it now when processing | |
825 | * the bios in the cell. | |
826 | */ | |
827 | if (bio) { | |
2d759a46 | 828 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
991d9fa0 | 829 | bio_endio(bio, 0); |
2d759a46 JT |
830 | } else { |
831 | inc_all_io_entry(tc->pool, m->cell->holder); | |
832 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
833 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
834 | } | |
991d9fa0 | 835 | |
905386f8 | 836 | out: |
991d9fa0 | 837 | list_del(&m->list); |
6beca5eb | 838 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
839 | } |
840 | ||
e49e5829 | 841 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 842 | { |
104655fd JT |
843 | struct thin_c *tc = m->tc; |
844 | ||
e49e5829 | 845 | bio_io_error(m->bio); |
f286ba0e JT |
846 | cell_defer_no_holder(tc, m->cell); |
847 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
848 | mempool_free(m, tc->pool->mapping_pool); |
849 | } | |
850 | ||
851 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
852 | { | |
853 | struct thin_c *tc = m->tc; | |
104655fd | 854 | |
e8088073 | 855 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
856 | cell_defer_no_holder(tc, m->cell); |
857 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 858 | |
104655fd | 859 | if (m->pass_discard) |
19fa1a67 JT |
860 | if (m->definitely_not_shared) |
861 | remap_and_issue(tc, m->bio, m->data_block); | |
862 | else { | |
863 | bool used = false; | |
864 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
865 | bio_endio(m->bio, 0); | |
866 | else | |
867 | remap_and_issue(tc, m->bio, m->data_block); | |
868 | } | |
104655fd JT |
869 | else |
870 | bio_endio(m->bio, 0); | |
871 | ||
104655fd JT |
872 | mempool_free(m, tc->pool->mapping_pool); |
873 | } | |
874 | ||
e49e5829 JT |
875 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
876 | { | |
877 | int r; | |
878 | struct thin_c *tc = m->tc; | |
879 | ||
880 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
881 | if (r) | |
c397741c | 882 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
883 | |
884 | process_prepared_discard_passdown(m); | |
885 | } | |
886 | ||
104655fd | 887 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 888 | process_mapping_fn *fn) |
991d9fa0 JT |
889 | { |
890 | unsigned long flags; | |
891 | struct list_head maps; | |
a24c2569 | 892 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
893 | |
894 | INIT_LIST_HEAD(&maps); | |
895 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 896 | list_splice_init(head, &maps); |
991d9fa0 JT |
897 | spin_unlock_irqrestore(&pool->lock, flags); |
898 | ||
899 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 900 | (*fn)(m); |
991d9fa0 JT |
901 | } |
902 | ||
903 | /* | |
904 | * Deferred bio jobs. | |
905 | */ | |
104655fd | 906 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 907 | { |
4f024f37 KO |
908 | return bio->bi_iter.bi_size == |
909 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
910 | } |
911 | ||
912 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
913 | { | |
914 | return (bio_data_dir(bio) == WRITE) && | |
915 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
916 | } |
917 | ||
918 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
919 | bio_end_io_t *fn) | |
920 | { | |
921 | *save = bio->bi_end_io; | |
922 | bio->bi_end_io = fn; | |
923 | } | |
924 | ||
925 | static int ensure_next_mapping(struct pool *pool) | |
926 | { | |
927 | if (pool->next_mapping) | |
928 | return 0; | |
929 | ||
930 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
931 | ||
932 | return pool->next_mapping ? 0 : -ENOMEM; | |
933 | } | |
934 | ||
a24c2569 | 935 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 936 | { |
16961b04 | 937 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
938 | |
939 | BUG_ON(!pool->next_mapping); | |
940 | ||
16961b04 MS |
941 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
942 | INIT_LIST_HEAD(&m->list); | |
943 | m->bio = NULL; | |
944 | ||
991d9fa0 JT |
945 | pool->next_mapping = NULL; |
946 | ||
16961b04 | 947 | return m; |
991d9fa0 JT |
948 | } |
949 | ||
e5aea7b4 JT |
950 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
951 | sector_t begin, sector_t end) | |
952 | { | |
953 | int r; | |
954 | struct dm_io_region to; | |
955 | ||
956 | to.bdev = tc->pool_dev->bdev; | |
957 | to.sector = begin; | |
958 | to.count = end - begin; | |
959 | ||
960 | r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); | |
961 | if (r < 0) { | |
962 | DMERR_LIMIT("dm_kcopyd_zero() failed"); | |
963 | copy_complete(1, 1, m); | |
964 | } | |
965 | } | |
966 | ||
452d7a62 MS |
967 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
968 | dm_block_t data_block, | |
969 | struct dm_thin_new_mapping *m) | |
970 | { | |
971 | struct pool *pool = tc->pool; | |
972 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
973 | ||
974 | h->overwrite_mapping = m; | |
975 | m->bio = bio; | |
976 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
977 | inc_all_io_entry(pool, bio); | |
978 | remap_and_issue(tc, bio, data_block); | |
979 | } | |
980 | ||
e5aea7b4 JT |
981 | /* |
982 | * A partial copy also needs to zero the uncopied region. | |
983 | */ | |
991d9fa0 | 984 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
985 | struct dm_dev *origin, dm_block_t data_origin, |
986 | dm_block_t data_dest, | |
e5aea7b4 JT |
987 | struct dm_bio_prison_cell *cell, struct bio *bio, |
988 | sector_t len) | |
991d9fa0 JT |
989 | { |
990 | int r; | |
991 | struct pool *pool = tc->pool; | |
a24c2569 | 992 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 993 | |
991d9fa0 JT |
994 | m->tc = tc; |
995 | m->virt_block = virt_block; | |
996 | m->data_block = data_dest; | |
997 | m->cell = cell; | |
991d9fa0 | 998 | |
e5aea7b4 JT |
999 | /* |
1000 | * quiesce action + copy action + an extra reference held for the | |
1001 | * duration of this function (we may need to inc later for a | |
1002 | * partial zero). | |
1003 | */ | |
1004 | atomic_set(&m->prepare_actions, 3); | |
1005 | ||
44feb387 | 1006 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1007 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1008 | |
1009 | /* | |
1010 | * IO to pool_dev remaps to the pool target's data_dev. | |
1011 | * | |
1012 | * If the whole block of data is being overwritten, we can issue the | |
1013 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1014 | */ | |
452d7a62 MS |
1015 | if (io_overwrites_block(pool, bio)) |
1016 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1017 | else { | |
991d9fa0 JT |
1018 | struct dm_io_region from, to; |
1019 | ||
2dd9c257 | 1020 | from.bdev = origin->bdev; |
991d9fa0 | 1021 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1022 | from.count = len; |
991d9fa0 JT |
1023 | |
1024 | to.bdev = tc->pool_dev->bdev; | |
1025 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1026 | to.count = len; |
991d9fa0 JT |
1027 | |
1028 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1029 | 0, copy_complete, m); | |
1030 | if (r < 0) { | |
c397741c | 1031 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
e5aea7b4 JT |
1032 | copy_complete(1, 1, m); |
1033 | ||
1034 | /* | |
1035 | * We allow the zero to be issued, to simplify the | |
1036 | * error path. Otherwise we'd need to start | |
1037 | * worrying about decrementing the prepare_actions | |
1038 | * counter. | |
1039 | */ | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Do we need to zero a tail region? | |
1044 | */ | |
1045 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1046 | atomic_inc(&m->prepare_actions); | |
1047 | ll_zero(tc, m, | |
1048 | data_dest * pool->sectors_per_block + len, | |
1049 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1050 | } |
1051 | } | |
e5aea7b4 JT |
1052 | |
1053 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1054 | } |
1055 | ||
2dd9c257 JT |
1056 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1057 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1058 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1059 | { |
1060 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1061 | data_origin, data_dest, cell, bio, |
1062 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1063 | } |
1064 | ||
991d9fa0 | 1065 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1066 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1067 | struct bio *bio) |
1068 | { | |
1069 | struct pool *pool = tc->pool; | |
a24c2569 | 1070 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1071 | |
50f3c3ef | 1072 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 JT |
1073 | m->tc = tc; |
1074 | m->virt_block = virt_block; | |
1075 | m->data_block = data_block; | |
1076 | m->cell = cell; | |
991d9fa0 JT |
1077 | |
1078 | /* | |
1079 | * If the whole block of data is being overwritten or we are not | |
1080 | * zeroing pre-existing data, we can issue the bio immediately. | |
1081 | * Otherwise we use kcopyd to zero the data first. | |
1082 | */ | |
67e2e2b2 | 1083 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
1084 | process_prepared_mapping(m); |
1085 | ||
452d7a62 MS |
1086 | else if (io_overwrites_block(pool, bio)) |
1087 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
991d9fa0 | 1088 | |
452d7a62 | 1089 | else |
e5aea7b4 JT |
1090 | ll_zero(tc, m, |
1091 | data_block * pool->sectors_per_block, | |
1092 | (data_block + 1) * pool->sectors_per_block); | |
1093 | } | |
991d9fa0 | 1094 | |
e5aea7b4 JT |
1095 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1096 | dm_block_t data_dest, | |
1097 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1098 | { | |
1099 | struct pool *pool = tc->pool; | |
1100 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1101 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1102 | ||
1103 | if (virt_block_end <= tc->origin_size) | |
1104 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1105 | virt_block, data_dest, cell, bio, | |
1106 | pool->sectors_per_block); | |
1107 | ||
1108 | else if (virt_block_begin < tc->origin_size) | |
1109 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1110 | virt_block, data_dest, cell, bio, | |
1111 | tc->origin_size - virt_block_begin); | |
1112 | ||
1113 | else | |
1114 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1115 | } |
1116 | ||
e49e5829 JT |
1117 | /* |
1118 | * A non-zero return indicates read_only or fail_io mode. | |
1119 | * Many callers don't care about the return value. | |
1120 | */ | |
020cc3b5 | 1121 | static int commit(struct pool *pool) |
e49e5829 JT |
1122 | { |
1123 | int r; | |
1124 | ||
8d07e8a5 | 1125 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
1126 | return -EINVAL; |
1127 | ||
020cc3b5 | 1128 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1129 | if (r) |
1130 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
1131 | |
1132 | return r; | |
1133 | } | |
1134 | ||
88a6621b JT |
1135 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1136 | { | |
1137 | unsigned long flags; | |
1138 | ||
1139 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1140 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1141 | dm_device_name(pool->pool_md)); | |
1142 | spin_lock_irqsave(&pool->lock, flags); | |
1143 | pool->low_water_triggered = true; | |
1144 | spin_unlock_irqrestore(&pool->lock, flags); | |
1145 | dm_table_event(pool->ti->table); | |
1146 | } | |
1147 | } | |
1148 | ||
3e1a0699 JT |
1149 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1150 | ||
991d9fa0 JT |
1151 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1152 | { | |
1153 | int r; | |
1154 | dm_block_t free_blocks; | |
991d9fa0 JT |
1155 | struct pool *pool = tc->pool; |
1156 | ||
3e1a0699 | 1157 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1158 | return -EINVAL; |
1159 | ||
991d9fa0 | 1160 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1161 | if (r) { |
1162 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1163 | return r; |
b5330655 | 1164 | } |
991d9fa0 | 1165 | |
88a6621b | 1166 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1167 | |
1168 | if (!free_blocks) { | |
94563bad MS |
1169 | /* |
1170 | * Try to commit to see if that will free up some | |
1171 | * more space. | |
1172 | */ | |
020cc3b5 JT |
1173 | r = commit(pool); |
1174 | if (r) | |
1175 | return r; | |
991d9fa0 | 1176 | |
94563bad | 1177 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1178 | if (r) { |
1179 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1180 | return r; |
b5330655 | 1181 | } |
991d9fa0 | 1182 | |
94563bad | 1183 | if (!free_blocks) { |
3e1a0699 | 1184 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1185 | return -ENOSPC; |
991d9fa0 JT |
1186 | } |
1187 | } | |
1188 | ||
1189 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1190 | if (r) { |
b5330655 | 1191 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1192 | return r; |
4a02b34e | 1193 | } |
991d9fa0 JT |
1194 | |
1195 | return 0; | |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * If we have run out of space, queue bios until the device is | |
1200 | * resumed, presumably after having been reloaded with more space. | |
1201 | */ | |
1202 | static void retry_on_resume(struct bio *bio) | |
1203 | { | |
59c3d2c6 | 1204 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1205 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1206 | unsigned long flags; |
1207 | ||
c140e1c4 MS |
1208 | spin_lock_irqsave(&tc->lock, flags); |
1209 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1210 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1211 | } |
1212 | ||
af91805a | 1213 | static int should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1214 | { |
3e1a0699 JT |
1215 | enum pool_mode m = get_pool_mode(pool); |
1216 | ||
1217 | switch (m) { | |
1218 | case PM_WRITE: | |
1219 | /* Shouldn't get here */ | |
1220 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
af91805a | 1221 | return -EIO; |
3e1a0699 JT |
1222 | |
1223 | case PM_OUT_OF_DATA_SPACE: | |
af91805a | 1224 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
3e1a0699 JT |
1225 | |
1226 | case PM_READ_ONLY: | |
1227 | case PM_FAIL: | |
af91805a | 1228 | return -EIO; |
3e1a0699 JT |
1229 | default: |
1230 | /* Shouldn't get here */ | |
1231 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
af91805a | 1232 | return -EIO; |
3e1a0699 JT |
1233 | } |
1234 | } | |
8c0f0e8c | 1235 | |
3e1a0699 JT |
1236 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1237 | { | |
af91805a MS |
1238 | int error = should_error_unserviceable_bio(pool); |
1239 | ||
1240 | if (error) | |
1241 | bio_endio(bio, error); | |
6d16202b MS |
1242 | else |
1243 | retry_on_resume(bio); | |
8c0f0e8c MS |
1244 | } |
1245 | ||
399caddf | 1246 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1247 | { |
1248 | struct bio *bio; | |
1249 | struct bio_list bios; | |
af91805a | 1250 | int error; |
991d9fa0 | 1251 | |
af91805a MS |
1252 | error = should_error_unserviceable_bio(pool); |
1253 | if (error) { | |
1254 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1255 | return; |
1256 | } | |
1257 | ||
991d9fa0 | 1258 | bio_list_init(&bios); |
6beca5eb | 1259 | cell_release(pool, cell, &bios); |
991d9fa0 | 1260 | |
af91805a MS |
1261 | error = should_error_unserviceable_bio(pool); |
1262 | if (error) | |
3e1a0699 | 1263 | while ((bio = bio_list_pop(&bios))) |
af91805a | 1264 | bio_endio(bio, error); |
3e1a0699 JT |
1265 | else |
1266 | while ((bio = bio_list_pop(&bios))) | |
1267 | retry_on_resume(bio); | |
991d9fa0 JT |
1268 | } |
1269 | ||
a374bb21 | 1270 | static void process_discard_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
104655fd JT |
1271 | { |
1272 | int r; | |
a374bb21 | 1273 | struct bio *bio = cell->holder; |
104655fd | 1274 | struct pool *pool = tc->pool; |
a374bb21 JT |
1275 | struct dm_bio_prison_cell *cell2; |
1276 | struct dm_cell_key key2; | |
104655fd JT |
1277 | dm_block_t block = get_bio_block(tc, bio); |
1278 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1279 | struct dm_thin_new_mapping *m; |
104655fd | 1280 | |
a374bb21 JT |
1281 | if (tc->requeue_mode) { |
1282 | cell_requeue(pool, cell); | |
104655fd | 1283 | return; |
a374bb21 | 1284 | } |
104655fd JT |
1285 | |
1286 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1287 | switch (r) { | |
1288 | case 0: | |
1289 | /* | |
1290 | * Check nobody is fiddling with this pool block. This can | |
1291 | * happen if someone's in the process of breaking sharing | |
1292 | * on this block. | |
1293 | */ | |
1294 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1295 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1296 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1297 | break; |
1298 | } | |
1299 | ||
1300 | if (io_overlaps_block(pool, bio)) { | |
1301 | /* | |
1302 | * IO may still be going to the destination block. We must | |
1303 | * quiesce before we can do the removal. | |
1304 | */ | |
1305 | m = get_next_mapping(pool); | |
1306 | m->tc = tc; | |
19fa1a67 JT |
1307 | m->pass_discard = pool->pf.discard_passdown; |
1308 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1309 | m->virt_block = block; |
1310 | m->data_block = lookup_result.block; | |
1311 | m->cell = cell; | |
1312 | m->cell2 = cell2; | |
104655fd JT |
1313 | m->bio = bio; |
1314 | ||
7a7e97ca JT |
1315 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1316 | pool->process_prepared_discard(m); | |
1317 | ||
104655fd | 1318 | } else { |
e8088073 | 1319 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1320 | cell_defer_no_holder(tc, cell); |
1321 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1322 | |
104655fd | 1323 | /* |
49296309 MP |
1324 | * The DM core makes sure that the discard doesn't span |
1325 | * a block boundary. So we submit the discard of a | |
1326 | * partial block appropriately. | |
104655fd | 1327 | */ |
650d2a06 MP |
1328 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1329 | remap_and_issue(tc, bio, lookup_result.block); | |
1330 | else | |
1331 | bio_endio(bio, 0); | |
104655fd JT |
1332 | } |
1333 | break; | |
1334 | ||
1335 | case -ENODATA: | |
1336 | /* | |
1337 | * It isn't provisioned, just forget it. | |
1338 | */ | |
f286ba0e | 1339 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1340 | bio_endio(bio, 0); |
1341 | break; | |
1342 | ||
1343 | default: | |
c397741c MS |
1344 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1345 | __func__, r); | |
f286ba0e | 1346 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1347 | bio_io_error(bio); |
1348 | break; | |
1349 | } | |
1350 | } | |
1351 | ||
a374bb21 JT |
1352 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1353 | { | |
1354 | struct dm_bio_prison_cell *cell; | |
1355 | struct dm_cell_key key; | |
1356 | dm_block_t block = get_bio_block(tc, bio); | |
1357 | ||
1358 | build_virtual_key(tc->td, block, &key); | |
1359 | if (bio_detain(tc->pool, &key, bio, &cell)) | |
1360 | return; | |
1361 | ||
1362 | process_discard_cell(tc, cell); | |
1363 | } | |
1364 | ||
991d9fa0 | 1365 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1366 | struct dm_cell_key *key, |
991d9fa0 | 1367 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1368 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1369 | { |
1370 | int r; | |
1371 | dm_block_t data_block; | |
d6fc2042 | 1372 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1373 | |
1374 | r = alloc_data_block(tc, &data_block); | |
1375 | switch (r) { | |
1376 | case 0: | |
2dd9c257 JT |
1377 | schedule_internal_copy(tc, block, lookup_result->block, |
1378 | data_block, cell, bio); | |
991d9fa0 JT |
1379 | break; |
1380 | ||
1381 | case -ENOSPC: | |
399caddf | 1382 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1383 | break; |
1384 | ||
1385 | default: | |
c397741c MS |
1386 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1387 | __func__, r); | |
d6fc2042 | 1388 | cell_error(pool, cell); |
991d9fa0 JT |
1389 | break; |
1390 | } | |
1391 | } | |
1392 | ||
23ca2bb6 JT |
1393 | static void __remap_and_issue_shared_cell(void *context, |
1394 | struct dm_bio_prison_cell *cell) | |
1395 | { | |
1396 | struct remap_info *info = context; | |
1397 | struct bio *bio; | |
1398 | ||
1399 | while ((bio = bio_list_pop(&cell->bios))) { | |
1400 | if ((bio_data_dir(bio) == WRITE) || | |
1401 | (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))) | |
1402 | bio_list_add(&info->defer_bios, bio); | |
1403 | else { | |
1404 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));; | |
1405 | ||
1406 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1407 | inc_all_io_entry(info->tc->pool, bio); | |
1408 | bio_list_add(&info->issue_bios, bio); | |
1409 | } | |
1410 | } | |
1411 | } | |
1412 | ||
1413 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1414 | struct dm_bio_prison_cell *cell, | |
1415 | dm_block_t block) | |
1416 | { | |
1417 | struct bio *bio; | |
1418 | struct remap_info info; | |
1419 | ||
1420 | info.tc = tc; | |
1421 | bio_list_init(&info.defer_bios); | |
1422 | bio_list_init(&info.issue_bios); | |
1423 | ||
1424 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1425 | &info, cell); | |
1426 | ||
1427 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1428 | thin_defer_bio(tc, bio); | |
1429 | ||
1430 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1431 | remap_and_issue(tc, bio, block); | |
1432 | } | |
1433 | ||
991d9fa0 JT |
1434 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1435 | dm_block_t block, | |
23ca2bb6 JT |
1436 | struct dm_thin_lookup_result *lookup_result, |
1437 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1438 | { |
23ca2bb6 | 1439 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1440 | struct pool *pool = tc->pool; |
44feb387 | 1441 | struct dm_cell_key key; |
991d9fa0 JT |
1442 | |
1443 | /* | |
1444 | * If cell is already occupied, then sharing is already in the process | |
1445 | * of being broken so we have nothing further to do here. | |
1446 | */ | |
1447 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1448 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1449 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1450 | return; |
23ca2bb6 | 1451 | } |
991d9fa0 | 1452 | |
23ca2bb6 JT |
1453 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1454 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1455 | cell_defer_no_holder(tc, virt_cell); | |
1456 | } else { | |
59c3d2c6 | 1457 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1458 | |
44feb387 | 1459 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1460 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1461 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1462 | |
1463 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1464 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1465 | } |
1466 | } | |
1467 | ||
1468 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1469 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1470 | { |
1471 | int r; | |
1472 | dm_block_t data_block; | |
6beca5eb | 1473 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1474 | |
1475 | /* | |
1476 | * Remap empty bios (flushes) immediately, without provisioning. | |
1477 | */ | |
4f024f37 | 1478 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1479 | inc_all_io_entry(pool, bio); |
f286ba0e | 1480 | cell_defer_no_holder(tc, cell); |
e8088073 | 1481 | |
991d9fa0 JT |
1482 | remap_and_issue(tc, bio, 0); |
1483 | return; | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * Fill read bios with zeroes and complete them immediately. | |
1488 | */ | |
1489 | if (bio_data_dir(bio) == READ) { | |
1490 | zero_fill_bio(bio); | |
f286ba0e | 1491 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1492 | bio_endio(bio, 0); |
1493 | return; | |
1494 | } | |
1495 | ||
1496 | r = alloc_data_block(tc, &data_block); | |
1497 | switch (r) { | |
1498 | case 0: | |
2dd9c257 JT |
1499 | if (tc->origin_dev) |
1500 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1501 | else | |
1502 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1503 | break; |
1504 | ||
1505 | case -ENOSPC: | |
399caddf | 1506 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1507 | break; |
1508 | ||
1509 | default: | |
c397741c MS |
1510 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1511 | __func__, r); | |
6beca5eb | 1512 | cell_error(pool, cell); |
991d9fa0 JT |
1513 | break; |
1514 | } | |
1515 | } | |
1516 | ||
a374bb21 | 1517 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1518 | { |
1519 | int r; | |
6beca5eb | 1520 | struct pool *pool = tc->pool; |
a374bb21 | 1521 | struct bio *bio = cell->holder; |
991d9fa0 | 1522 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1523 | struct dm_thin_lookup_result lookup_result; |
1524 | ||
a374bb21 JT |
1525 | if (tc->requeue_mode) { |
1526 | cell_requeue(pool, cell); | |
991d9fa0 | 1527 | return; |
a374bb21 | 1528 | } |
991d9fa0 JT |
1529 | |
1530 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1531 | switch (r) { | |
1532 | case 0: | |
23ca2bb6 JT |
1533 | if (lookup_result.shared) |
1534 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1535 | else { | |
6beca5eb | 1536 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1537 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1538 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1539 | } |
991d9fa0 JT |
1540 | break; |
1541 | ||
1542 | case -ENODATA: | |
2dd9c257 | 1543 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1544 | inc_all_io_entry(pool, bio); |
f286ba0e | 1545 | cell_defer_no_holder(tc, cell); |
e8088073 | 1546 | |
e5aea7b4 JT |
1547 | if (bio_end_sector(bio) <= tc->origin_size) |
1548 | remap_to_origin_and_issue(tc, bio); | |
1549 | ||
1550 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1551 | zero_fill_bio(bio); | |
1552 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1553 | remap_to_origin_and_issue(tc, bio); | |
1554 | ||
1555 | } else { | |
1556 | zero_fill_bio(bio); | |
1557 | bio_endio(bio, 0); | |
1558 | } | |
2dd9c257 JT |
1559 | } else |
1560 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1561 | break; |
1562 | ||
1563 | default: | |
c397741c MS |
1564 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1565 | __func__, r); | |
f286ba0e | 1566 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1567 | bio_io_error(bio); |
1568 | break; | |
1569 | } | |
1570 | } | |
1571 | ||
a374bb21 JT |
1572 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1573 | { | |
1574 | struct pool *pool = tc->pool; | |
1575 | dm_block_t block = get_bio_block(tc, bio); | |
1576 | struct dm_bio_prison_cell *cell; | |
1577 | struct dm_cell_key key; | |
1578 | ||
1579 | /* | |
1580 | * If cell is already occupied, then the block is already | |
1581 | * being provisioned so we have nothing further to do here. | |
1582 | */ | |
1583 | build_virtual_key(tc->td, block, &key); | |
1584 | if (bio_detain(pool, &key, bio, &cell)) | |
1585 | return; | |
1586 | ||
1587 | process_cell(tc, cell); | |
1588 | } | |
1589 | ||
1590 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
1591 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
1592 | { |
1593 | int r; | |
1594 | int rw = bio_data_dir(bio); | |
1595 | dm_block_t block = get_bio_block(tc, bio); | |
1596 | struct dm_thin_lookup_result lookup_result; | |
1597 | ||
1598 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1599 | switch (r) { | |
1600 | case 0: | |
a374bb21 | 1601 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 1602 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
1603 | if (cell) |
1604 | cell_defer_no_holder(tc, cell); | |
1605 | } else { | |
e8088073 | 1606 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 1607 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
1608 | if (cell) |
1609 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 1610 | } |
e49e5829 JT |
1611 | break; |
1612 | ||
1613 | case -ENODATA: | |
a374bb21 JT |
1614 | if (cell) |
1615 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 1616 | if (rw != READ) { |
8c0f0e8c | 1617 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1618 | break; |
1619 | } | |
1620 | ||
1621 | if (tc->origin_dev) { | |
e8088073 | 1622 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1623 | remap_to_origin_and_issue(tc, bio); |
1624 | break; | |
1625 | } | |
1626 | ||
1627 | zero_fill_bio(bio); | |
1628 | bio_endio(bio, 0); | |
1629 | break; | |
1630 | ||
1631 | default: | |
c397741c MS |
1632 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1633 | __func__, r); | |
a374bb21 JT |
1634 | if (cell) |
1635 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
1636 | bio_io_error(bio); |
1637 | break; | |
1638 | } | |
1639 | } | |
1640 | ||
a374bb21 JT |
1641 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1642 | { | |
1643 | __process_bio_read_only(tc, bio, NULL); | |
1644 | } | |
1645 | ||
1646 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1647 | { | |
1648 | __process_bio_read_only(tc, cell->holder, cell); | |
1649 | } | |
1650 | ||
3e1a0699 JT |
1651 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1652 | { | |
1653 | bio_endio(bio, 0); | |
1654 | } | |
1655 | ||
e49e5829 JT |
1656 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1657 | { | |
1658 | bio_io_error(bio); | |
1659 | } | |
1660 | ||
a374bb21 JT |
1661 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
1662 | { | |
1663 | cell_success(tc->pool, cell); | |
1664 | } | |
1665 | ||
1666 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1667 | { | |
1668 | cell_error(tc->pool, cell); | |
1669 | } | |
1670 | ||
ac8c3f3d JT |
1671 | /* |
1672 | * FIXME: should we also commit due to size of transaction, measured in | |
1673 | * metadata blocks? | |
1674 | */ | |
905e51b3 JT |
1675 | static int need_commit_due_to_time(struct pool *pool) |
1676 | { | |
1677 | return jiffies < pool->last_commit_jiffies || | |
1678 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1679 | } | |
1680 | ||
67324ea1 MS |
1681 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1682 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1683 | ||
1684 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1685 | { | |
1686 | struct rb_node **rbp, *parent; | |
1687 | struct dm_thin_endio_hook *pbd; | |
1688 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1689 | ||
1690 | rbp = &tc->sort_bio_list.rb_node; | |
1691 | parent = NULL; | |
1692 | while (*rbp) { | |
1693 | parent = *rbp; | |
1694 | pbd = thin_pbd(parent); | |
1695 | ||
1696 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1697 | rbp = &(*rbp)->rb_left; | |
1698 | else | |
1699 | rbp = &(*rbp)->rb_right; | |
1700 | } | |
1701 | ||
1702 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1703 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1704 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1705 | } | |
1706 | ||
1707 | static void __extract_sorted_bios(struct thin_c *tc) | |
1708 | { | |
1709 | struct rb_node *node; | |
1710 | struct dm_thin_endio_hook *pbd; | |
1711 | struct bio *bio; | |
1712 | ||
1713 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1714 | pbd = thin_pbd(node); | |
1715 | bio = thin_bio(pbd); | |
1716 | ||
1717 | bio_list_add(&tc->deferred_bio_list, bio); | |
1718 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1719 | } | |
1720 | ||
1721 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1722 | } | |
1723 | ||
1724 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1725 | { | |
1726 | struct bio *bio; | |
1727 | struct bio_list bios; | |
1728 | ||
1729 | bio_list_init(&bios); | |
1730 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1731 | bio_list_init(&tc->deferred_bio_list); | |
1732 | ||
1733 | /* Sort deferred_bio_list using rb-tree */ | |
1734 | while ((bio = bio_list_pop(&bios))) | |
1735 | __thin_bio_rb_add(tc, bio); | |
1736 | ||
1737 | /* | |
1738 | * Transfer the sorted bios in sort_bio_list back to | |
1739 | * deferred_bio_list to allow lockless submission of | |
1740 | * all bios. | |
1741 | */ | |
1742 | __extract_sorted_bios(tc); | |
1743 | } | |
1744 | ||
c140e1c4 | 1745 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1746 | { |
c140e1c4 | 1747 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1748 | unsigned long flags; |
1749 | struct bio *bio; | |
1750 | struct bio_list bios; | |
67324ea1 | 1751 | struct blk_plug plug; |
8a01a6af | 1752 | unsigned count = 0; |
991d9fa0 | 1753 | |
c140e1c4 MS |
1754 | if (tc->requeue_mode) { |
1755 | requeue_bio_list(tc, &tc->deferred_bio_list); | |
1756 | return; | |
1757 | } | |
1758 | ||
991d9fa0 JT |
1759 | bio_list_init(&bios); |
1760 | ||
c140e1c4 | 1761 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1762 | |
1763 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1764 | spin_unlock_irqrestore(&tc->lock, flags); | |
1765 | return; | |
1766 | } | |
1767 | ||
1768 | __sort_thin_deferred_bios(tc); | |
1769 | ||
c140e1c4 MS |
1770 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1771 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1772 | |
c140e1c4 | 1773 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1774 | |
67324ea1 | 1775 | blk_start_plug(&plug); |
991d9fa0 | 1776 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1777 | /* |
1778 | * If we've got no free new_mapping structs, and processing | |
1779 | * this bio might require one, we pause until there are some | |
1780 | * prepared mappings to process. | |
1781 | */ | |
1782 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1783 | spin_lock_irqsave(&tc->lock, flags); |
1784 | bio_list_add(&tc->deferred_bio_list, bio); | |
1785 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1786 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1787 | break; |
1788 | } | |
104655fd JT |
1789 | |
1790 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1791 | pool->process_discard(tc, bio); |
104655fd | 1792 | else |
e49e5829 | 1793 | pool->process_bio(tc, bio); |
8a01a6af JT |
1794 | |
1795 | if ((count++ & 127) == 0) { | |
7d327fe0 | 1796 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
1797 | dm_pool_issue_prefetches(pool->pmd); |
1798 | } | |
991d9fa0 | 1799 | } |
67324ea1 | 1800 | blk_finish_plug(&plug); |
c140e1c4 MS |
1801 | } |
1802 | ||
a374bb21 JT |
1803 | static void process_thin_deferred_cells(struct thin_c *tc) |
1804 | { | |
1805 | struct pool *pool = tc->pool; | |
1806 | unsigned long flags; | |
1807 | struct list_head cells; | |
1808 | struct dm_bio_prison_cell *cell, *tmp; | |
1809 | ||
1810 | INIT_LIST_HEAD(&cells); | |
1811 | ||
1812 | spin_lock_irqsave(&tc->lock, flags); | |
1813 | list_splice_init(&tc->deferred_cells, &cells); | |
1814 | spin_unlock_irqrestore(&tc->lock, flags); | |
1815 | ||
1816 | if (list_empty(&cells)) | |
1817 | return; | |
1818 | ||
1819 | list_for_each_entry_safe(cell, tmp, &cells, user_list) { | |
1820 | BUG_ON(!cell->holder); | |
1821 | ||
1822 | /* | |
1823 | * If we've got no free new_mapping structs, and processing | |
1824 | * this bio might require one, we pause until there are some | |
1825 | * prepared mappings to process. | |
1826 | */ | |
1827 | if (ensure_next_mapping(pool)) { | |
1828 | spin_lock_irqsave(&tc->lock, flags); | |
1829 | list_add(&cell->user_list, &tc->deferred_cells); | |
1830 | list_splice(&cells, &tc->deferred_cells); | |
1831 | spin_unlock_irqrestore(&tc->lock, flags); | |
1832 | break; | |
1833 | } | |
1834 | ||
1835 | if (cell->holder->bi_rw & REQ_DISCARD) | |
1836 | pool->process_discard_cell(tc, cell); | |
1837 | else | |
1838 | pool->process_cell(tc, cell); | |
1839 | } | |
1840 | } | |
1841 | ||
b10ebd34 JT |
1842 | static void thin_get(struct thin_c *tc); |
1843 | static void thin_put(struct thin_c *tc); | |
1844 | ||
1845 | /* | |
1846 | * We can't hold rcu_read_lock() around code that can block. So we | |
1847 | * find a thin with the rcu lock held; bump a refcount; then drop | |
1848 | * the lock. | |
1849 | */ | |
1850 | static struct thin_c *get_first_thin(struct pool *pool) | |
1851 | { | |
1852 | struct thin_c *tc = NULL; | |
1853 | ||
1854 | rcu_read_lock(); | |
1855 | if (!list_empty(&pool->active_thins)) { | |
1856 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
1857 | thin_get(tc); | |
1858 | } | |
1859 | rcu_read_unlock(); | |
1860 | ||
1861 | return tc; | |
1862 | } | |
1863 | ||
1864 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
1865 | { | |
1866 | struct thin_c *old_tc = tc; | |
1867 | ||
1868 | rcu_read_lock(); | |
1869 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
1870 | thin_get(tc); | |
1871 | thin_put(old_tc); | |
1872 | rcu_read_unlock(); | |
1873 | return tc; | |
1874 | } | |
1875 | thin_put(old_tc); | |
1876 | rcu_read_unlock(); | |
1877 | ||
1878 | return NULL; | |
1879 | } | |
1880 | ||
c140e1c4 MS |
1881 | static void process_deferred_bios(struct pool *pool) |
1882 | { | |
1883 | unsigned long flags; | |
1884 | struct bio *bio; | |
1885 | struct bio_list bios; | |
1886 | struct thin_c *tc; | |
1887 | ||
b10ebd34 JT |
1888 | tc = get_first_thin(pool); |
1889 | while (tc) { | |
a374bb21 | 1890 | process_thin_deferred_cells(tc); |
c140e1c4 | 1891 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
1892 | tc = get_next_thin(pool, tc); |
1893 | } | |
991d9fa0 JT |
1894 | |
1895 | /* | |
1896 | * If there are any deferred flush bios, we must commit | |
1897 | * the metadata before issuing them. | |
1898 | */ | |
1899 | bio_list_init(&bios); | |
1900 | spin_lock_irqsave(&pool->lock, flags); | |
1901 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1902 | bio_list_init(&pool->deferred_flush_bios); | |
1903 | spin_unlock_irqrestore(&pool->lock, flags); | |
1904 | ||
4d1662a3 MS |
1905 | if (bio_list_empty(&bios) && |
1906 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1907 | return; |
1908 | ||
020cc3b5 | 1909 | if (commit(pool)) { |
991d9fa0 JT |
1910 | while ((bio = bio_list_pop(&bios))) |
1911 | bio_io_error(bio); | |
1912 | return; | |
1913 | } | |
905e51b3 | 1914 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1915 | |
1916 | while ((bio = bio_list_pop(&bios))) | |
1917 | generic_make_request(bio); | |
1918 | } | |
1919 | ||
1920 | static void do_worker(struct work_struct *ws) | |
1921 | { | |
1922 | struct pool *pool = container_of(ws, struct pool, worker); | |
1923 | ||
7d327fe0 | 1924 | throttle_work_start(&pool->throttle); |
8a01a6af | 1925 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 1926 | throttle_work_update(&pool->throttle); |
e49e5829 | 1927 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 1928 | throttle_work_update(&pool->throttle); |
e49e5829 | 1929 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 1930 | throttle_work_update(&pool->throttle); |
991d9fa0 | 1931 | process_deferred_bios(pool); |
7d327fe0 | 1932 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
1933 | } |
1934 | ||
905e51b3 JT |
1935 | /* |
1936 | * We want to commit periodically so that not too much | |
1937 | * unwritten data builds up. | |
1938 | */ | |
1939 | static void do_waker(struct work_struct *ws) | |
1940 | { | |
1941 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1942 | wake_worker(pool); | |
1943 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1944 | } | |
1945 | ||
85ad643b JT |
1946 | /* |
1947 | * We're holding onto IO to allow userland time to react. After the | |
1948 | * timeout either the pool will have been resized (and thus back in | |
1949 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
1950 | */ | |
1951 | static void do_no_space_timeout(struct work_struct *ws) | |
1952 | { | |
1953 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
1954 | no_space_timeout); | |
1955 | ||
1956 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
1957 | set_pool_mode(pool, PM_READ_ONLY); | |
1958 | } | |
1959 | ||
991d9fa0 JT |
1960 | /*----------------------------------------------------------------*/ |
1961 | ||
e7a3e871 | 1962 | struct pool_work { |
738211f7 | 1963 | struct work_struct worker; |
e7a3e871 JT |
1964 | struct completion complete; |
1965 | }; | |
1966 | ||
1967 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
1968 | { | |
1969 | return container_of(ws, struct pool_work, worker); | |
1970 | } | |
1971 | ||
1972 | static void pool_work_complete(struct pool_work *pw) | |
1973 | { | |
1974 | complete(&pw->complete); | |
1975 | } | |
738211f7 | 1976 | |
e7a3e871 JT |
1977 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
1978 | void (*fn)(struct work_struct *)) | |
1979 | { | |
1980 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
1981 | init_completion(&pw->complete); | |
1982 | queue_work(pool->wq, &pw->worker); | |
1983 | wait_for_completion(&pw->complete); | |
1984 | } | |
1985 | ||
1986 | /*----------------------------------------------------------------*/ | |
1987 | ||
1988 | struct noflush_work { | |
1989 | struct pool_work pw; | |
1990 | struct thin_c *tc; | |
738211f7 JT |
1991 | }; |
1992 | ||
e7a3e871 | 1993 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 1994 | { |
e7a3e871 | 1995 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
1996 | } |
1997 | ||
1998 | static void do_noflush_start(struct work_struct *ws) | |
1999 | { | |
e7a3e871 | 2000 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2001 | w->tc->requeue_mode = true; |
2002 | requeue_io(w->tc); | |
e7a3e871 | 2003 | pool_work_complete(&w->pw); |
738211f7 JT |
2004 | } |
2005 | ||
2006 | static void do_noflush_stop(struct work_struct *ws) | |
2007 | { | |
e7a3e871 | 2008 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2009 | w->tc->requeue_mode = false; |
e7a3e871 | 2010 | pool_work_complete(&w->pw); |
738211f7 JT |
2011 | } |
2012 | ||
2013 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2014 | { | |
2015 | struct noflush_work w; | |
2016 | ||
738211f7 | 2017 | w.tc = tc; |
e7a3e871 | 2018 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2019 | } |
2020 | ||
2021 | /*----------------------------------------------------------------*/ | |
2022 | ||
e49e5829 JT |
2023 | static enum pool_mode get_pool_mode(struct pool *pool) |
2024 | { | |
2025 | return pool->pf.mode; | |
2026 | } | |
2027 | ||
3e1a0699 JT |
2028 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
2029 | { | |
2030 | dm_table_event(pool->ti->table); | |
2031 | DMINFO("%s: switching pool to %s mode", | |
2032 | dm_device_name(pool->pool_md), new_mode); | |
2033 | } | |
2034 | ||
8b64e881 | 2035 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2036 | { |
cdc2b415 | 2037 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2038 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2039 | enum pool_mode old_mode = get_pool_mode(pool); | |
80c57893 | 2040 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2041 | |
2042 | /* | |
2043 | * Never allow the pool to transition to PM_WRITE mode if user | |
2044 | * intervention is required to verify metadata and data consistency. | |
2045 | */ | |
2046 | if (new_mode == PM_WRITE && needs_check) { | |
2047 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2048 | dm_device_name(pool->pool_md)); | |
2049 | if (old_mode != new_mode) | |
2050 | new_mode = old_mode; | |
2051 | else | |
2052 | new_mode = PM_READ_ONLY; | |
2053 | } | |
2054 | /* | |
2055 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2056 | * not going to recover without a thin_repair. So we never let the | |
2057 | * pool move out of the old mode. | |
2058 | */ | |
2059 | if (old_mode == PM_FAIL) | |
2060 | new_mode = old_mode; | |
e49e5829 | 2061 | |
8b64e881 | 2062 | switch (new_mode) { |
e49e5829 | 2063 | case PM_FAIL: |
8b64e881 | 2064 | if (old_mode != new_mode) |
3e1a0699 | 2065 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 2066 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2067 | pool->process_bio = process_bio_fail; |
2068 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2069 | pool->process_cell = process_cell_fail; |
2070 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2071 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2072 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2073 | |
2074 | error_retry_list(pool); | |
e49e5829 JT |
2075 | break; |
2076 | ||
2077 | case PM_READ_ONLY: | |
8b64e881 | 2078 | if (old_mode != new_mode) |
3e1a0699 JT |
2079 | notify_of_pool_mode_change(pool, "read-only"); |
2080 | dm_pool_metadata_read_only(pool->pmd); | |
2081 | pool->process_bio = process_bio_read_only; | |
2082 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2083 | pool->process_cell = process_cell_read_only; |
2084 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 JT |
2085 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2086 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
2087 | ||
2088 | error_retry_list(pool); | |
2089 | break; | |
2090 | ||
2091 | case PM_OUT_OF_DATA_SPACE: | |
2092 | /* | |
2093 | * Ideally we'd never hit this state; the low water mark | |
2094 | * would trigger userland to extend the pool before we | |
2095 | * completely run out of data space. However, many small | |
2096 | * IOs to unprovisioned space can consume data space at an | |
2097 | * alarming rate. Adjust your low water mark if you're | |
2098 | * frequently seeing this mode. | |
2099 | */ | |
2100 | if (old_mode != new_mode) | |
2101 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
2102 | pool->process_bio = process_bio_read_only; | |
a374bb21 JT |
2103 | pool->process_discard = process_discard_bio; |
2104 | pool->process_cell = process_cell_read_only; | |
2105 | pool->process_discard_cell = process_discard_cell; | |
3e1a0699 JT |
2106 | pool->process_prepared_mapping = process_prepared_mapping; |
2107 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
85ad643b | 2108 | |
80c57893 MS |
2109 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2110 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2111 | break; |
2112 | ||
2113 | case PM_WRITE: | |
8b64e881 | 2114 | if (old_mode != new_mode) |
3e1a0699 | 2115 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 2116 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2117 | pool->process_bio = process_bio; |
a374bb21 JT |
2118 | pool->process_discard = process_discard_bio; |
2119 | pool->process_cell = process_cell; | |
2120 | pool->process_discard_cell = process_discard_cell; | |
e49e5829 JT |
2121 | pool->process_prepared_mapping = process_prepared_mapping; |
2122 | pool->process_prepared_discard = process_prepared_discard; | |
2123 | break; | |
2124 | } | |
8b64e881 MS |
2125 | |
2126 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2127 | /* |
2128 | * The pool mode may have changed, sync it so bind_control_target() | |
2129 | * doesn't cause an unexpected mode transition on resume. | |
2130 | */ | |
2131 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
2132 | } |
2133 | ||
07f2b6e0 | 2134 | static void abort_transaction(struct pool *pool) |
b5330655 | 2135 | { |
07f2b6e0 MS |
2136 | const char *dev_name = dm_device_name(pool->pool_md); |
2137 | ||
2138 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2139 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2140 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2141 | set_pool_mode(pool, PM_FAIL); | |
2142 | } | |
2143 | ||
2144 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2145 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2146 | set_pool_mode(pool, PM_FAIL); | |
2147 | } | |
2148 | } | |
399caddf | 2149 | |
07f2b6e0 MS |
2150 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2151 | { | |
b5330655 JT |
2152 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2153 | dm_device_name(pool->pool_md), op, r); | |
2154 | ||
07f2b6e0 | 2155 | abort_transaction(pool); |
b5330655 JT |
2156 | set_pool_mode(pool, PM_READ_ONLY); |
2157 | } | |
2158 | ||
e49e5829 JT |
2159 | /*----------------------------------------------------------------*/ |
2160 | ||
991d9fa0 JT |
2161 | /* |
2162 | * Mapping functions. | |
2163 | */ | |
2164 | ||
2165 | /* | |
2166 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2167 | */ | |
2168 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2169 | { | |
2170 | unsigned long flags; | |
2171 | struct pool *pool = tc->pool; | |
2172 | ||
c140e1c4 MS |
2173 | spin_lock_irqsave(&tc->lock, flags); |
2174 | bio_list_add(&tc->deferred_bio_list, bio); | |
2175 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2176 | |
2177 | wake_worker(pool); | |
2178 | } | |
2179 | ||
7d327fe0 JT |
2180 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2181 | { | |
2182 | struct pool *pool = tc->pool; | |
2183 | ||
2184 | throttle_lock(&pool->throttle); | |
2185 | thin_defer_bio(tc, bio); | |
2186 | throttle_unlock(&pool->throttle); | |
2187 | } | |
2188 | ||
a374bb21 JT |
2189 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2190 | { | |
2191 | unsigned long flags; | |
2192 | struct pool *pool = tc->pool; | |
2193 | ||
2194 | throttle_lock(&pool->throttle); | |
2195 | spin_lock_irqsave(&tc->lock, flags); | |
2196 | list_add_tail(&cell->user_list, &tc->deferred_cells); | |
2197 | spin_unlock_irqrestore(&tc->lock, flags); | |
2198 | throttle_unlock(&pool->throttle); | |
2199 | ||
2200 | wake_worker(pool); | |
2201 | } | |
2202 | ||
59c3d2c6 | 2203 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2204 | { |
59c3d2c6 | 2205 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2206 | |
2207 | h->tc = tc; | |
2208 | h->shared_read_entry = NULL; | |
e8088073 | 2209 | h->all_io_entry = NULL; |
eb2aa48d | 2210 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
2211 | } |
2212 | ||
991d9fa0 JT |
2213 | /* |
2214 | * Non-blocking function called from the thin target's map function. | |
2215 | */ | |
7de3ee57 | 2216 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2217 | { |
2218 | int r; | |
2219 | struct thin_c *tc = ti->private; | |
2220 | dm_block_t block = get_bio_block(tc, bio); | |
2221 | struct dm_thin_device *td = tc->td; | |
2222 | struct dm_thin_lookup_result result; | |
a374bb21 | 2223 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2224 | struct dm_cell_key key; |
991d9fa0 | 2225 | |
59c3d2c6 | 2226 | thin_hook_bio(tc, bio); |
e49e5829 | 2227 | |
738211f7 JT |
2228 | if (tc->requeue_mode) { |
2229 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
2230 | return DM_MAPIO_SUBMITTED; | |
2231 | } | |
2232 | ||
e49e5829 JT |
2233 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2234 | bio_io_error(bio); | |
2235 | return DM_MAPIO_SUBMITTED; | |
2236 | } | |
2237 | ||
104655fd | 2238 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
7d327fe0 | 2239 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2240 | return DM_MAPIO_SUBMITTED; |
2241 | } | |
2242 | ||
c822ed96 JT |
2243 | /* |
2244 | * We must hold the virtual cell before doing the lookup, otherwise | |
2245 | * there's a race with discard. | |
2246 | */ | |
2247 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2248 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2249 | return DM_MAPIO_SUBMITTED; |
2250 | ||
991d9fa0 JT |
2251 | r = dm_thin_find_block(td, block, 0, &result); |
2252 | ||
2253 | /* | |
2254 | * Note that we defer readahead too. | |
2255 | */ | |
2256 | switch (r) { | |
2257 | case 0: | |
2258 | if (unlikely(result.shared)) { | |
2259 | /* | |
2260 | * We have a race condition here between the | |
2261 | * result.shared value returned by the lookup and | |
2262 | * snapshot creation, which may cause new | |
2263 | * sharing. | |
2264 | * | |
2265 | * To avoid this always quiesce the origin before | |
2266 | * taking the snap. You want to do this anyway to | |
2267 | * ensure a consistent application view | |
2268 | * (i.e. lockfs). | |
2269 | * | |
2270 | * More distant ancestors are irrelevant. The | |
2271 | * shared flag will be set in their case. | |
2272 | */ | |
a374bb21 | 2273 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2274 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2275 | } |
e8088073 | 2276 | |
e8088073 | 2277 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2278 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2279 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2280 | return DM_MAPIO_SUBMITTED; |
2281 | } | |
2282 | ||
2283 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2284 | cell_defer_no_holder(tc, data_cell); |
2285 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2286 | |
2287 | remap(tc, bio, result.block); | |
2288 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2289 | |
2290 | case -ENODATA: | |
e49e5829 JT |
2291 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
2292 | /* | |
2293 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 2294 | * of doing so. |
e49e5829 | 2295 | */ |
8c0f0e8c | 2296 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 | 2297 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2298 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2299 | } |
2300 | /* fall through */ | |
2301 | ||
2302 | case -EWOULDBLOCK: | |
a374bb21 | 2303 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2304 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2305 | |
2306 | default: | |
2307 | /* | |
2308 | * Must always call bio_io_error on failure. | |
2309 | * dm_thin_find_block can fail with -EINVAL if the | |
2310 | * pool is switched to fail-io mode. | |
2311 | */ | |
2312 | bio_io_error(bio); | |
a374bb21 | 2313 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2314 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2315 | } |
991d9fa0 JT |
2316 | } |
2317 | ||
2318 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2319 | { | |
991d9fa0 | 2320 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 2321 | struct request_queue *q; |
991d9fa0 | 2322 | |
760fe67e MS |
2323 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
2324 | return 1; | |
991d9fa0 | 2325 | |
760fe67e MS |
2326 | q = bdev_get_queue(pt->data_dev->bdev); |
2327 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
2328 | } |
2329 | ||
c140e1c4 | 2330 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2331 | { |
c140e1c4 MS |
2332 | unsigned long flags; |
2333 | struct thin_c *tc; | |
2334 | ||
2335 | rcu_read_lock(); | |
2336 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
2337 | spin_lock_irqsave(&tc->lock, flags); | |
2338 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
2339 | bio_list_init(&tc->retry_on_resume_list); | |
2340 | spin_unlock_irqrestore(&tc->lock, flags); | |
2341 | } | |
2342 | rcu_read_unlock(); | |
991d9fa0 JT |
2343 | } |
2344 | ||
2345 | /*---------------------------------------------------------------- | |
2346 | * Binding of control targets to a pool object | |
2347 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2348 | static bool data_dev_supports_discard(struct pool_c *pt) |
2349 | { | |
2350 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2351 | ||
2352 | return q && blk_queue_discard(q); | |
2353 | } | |
2354 | ||
58051b94 JT |
2355 | static bool is_factor(sector_t block_size, uint32_t n) |
2356 | { | |
2357 | return !sector_div(block_size, n); | |
2358 | } | |
2359 | ||
9bc142dd MS |
2360 | /* |
2361 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2362 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2363 | */ |
0424caa1 | 2364 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2365 | { |
0424caa1 MS |
2366 | struct pool *pool = pt->pool; |
2367 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2368 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
2369 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
2370 | const char *reason = NULL; | |
9bc142dd MS |
2371 | char buf[BDEVNAME_SIZE]; |
2372 | ||
0424caa1 | 2373 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2374 | return; |
2375 | ||
0424caa1 MS |
2376 | if (!data_dev_supports_discard(pt)) |
2377 | reason = "discard unsupported"; | |
2378 | ||
2379 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2380 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2381 | |
0424caa1 MS |
2382 | else if (data_limits->discard_granularity > block_size) |
2383 | reason = "discard granularity larger than a block"; | |
2384 | ||
58051b94 | 2385 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
2386 | reason = "discard granularity not a factor of block size"; |
2387 | ||
2388 | if (reason) { | |
2389 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2390 | pt->adjusted_pf.discard_passdown = false; | |
2391 | } | |
9bc142dd MS |
2392 | } |
2393 | ||
991d9fa0 JT |
2394 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2395 | { | |
2396 | struct pool_c *pt = ti->private; | |
2397 | ||
e49e5829 | 2398 | /* |
9b7aaa64 | 2399 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2400 | */ |
07f2b6e0 | 2401 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2402 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2403 | |
8b64e881 MS |
2404 | /* |
2405 | * Don't change the pool's mode until set_pool_mode() below. | |
2406 | * Otherwise the pool's process_* function pointers may | |
2407 | * not match the desired pool mode. | |
2408 | */ | |
2409 | pt->adjusted_pf.mode = old_mode; | |
2410 | ||
2411 | pool->ti = ti; | |
2412 | pool->pf = pt->adjusted_pf; | |
2413 | pool->low_water_blocks = pt->low_water_blocks; | |
2414 | ||
9bc142dd | 2415 | set_pool_mode(pool, new_mode); |
f402693d | 2416 | |
991d9fa0 JT |
2417 | return 0; |
2418 | } | |
2419 | ||
2420 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2421 | { | |
2422 | if (pool->ti == ti) | |
2423 | pool->ti = NULL; | |
2424 | } | |
2425 | ||
2426 | /*---------------------------------------------------------------- | |
2427 | * Pool creation | |
2428 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2429 | /* Initialize pool features. */ |
2430 | static void pool_features_init(struct pool_features *pf) | |
2431 | { | |
e49e5829 | 2432 | pf->mode = PM_WRITE; |
9bc142dd MS |
2433 | pf->zero_new_blocks = true; |
2434 | pf->discard_enabled = true; | |
2435 | pf->discard_passdown = true; | |
787a996c | 2436 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2437 | } |
2438 | ||
991d9fa0 JT |
2439 | static void __pool_destroy(struct pool *pool) |
2440 | { | |
2441 | __pool_table_remove(pool); | |
2442 | ||
2443 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2444 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2445 | ||
44feb387 | 2446 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2447 | dm_kcopyd_client_destroy(pool->copier); |
2448 | ||
2449 | if (pool->wq) | |
2450 | destroy_workqueue(pool->wq); | |
2451 | ||
2452 | if (pool->next_mapping) | |
2453 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2454 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2455 | dm_deferred_set_destroy(pool->shared_read_ds); |
2456 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2457 | kfree(pool); |
2458 | } | |
2459 | ||
a24c2569 | 2460 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2461 | |
991d9fa0 JT |
2462 | static struct pool *pool_create(struct mapped_device *pool_md, |
2463 | struct block_device *metadata_dev, | |
e49e5829 JT |
2464 | unsigned long block_size, |
2465 | int read_only, char **error) | |
991d9fa0 JT |
2466 | { |
2467 | int r; | |
2468 | void *err_p; | |
2469 | struct pool *pool; | |
2470 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2471 | bool format_device = read_only ? false : true; |
991d9fa0 | 2472 | |
e49e5829 | 2473 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2474 | if (IS_ERR(pmd)) { |
2475 | *error = "Error creating metadata object"; | |
2476 | return (struct pool *)pmd; | |
2477 | } | |
2478 | ||
2479 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2480 | if (!pool) { | |
2481 | *error = "Error allocating memory for pool"; | |
2482 | err_p = ERR_PTR(-ENOMEM); | |
2483 | goto bad_pool; | |
2484 | } | |
2485 | ||
2486 | pool->pmd = pmd; | |
2487 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2488 | if (block_size & (block_size - 1)) |
2489 | pool->sectors_per_block_shift = -1; | |
2490 | else | |
2491 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2492 | pool->low_water_blocks = 0; |
67e2e2b2 | 2493 | pool_features_init(&pool->pf); |
a195db2d | 2494 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2495 | if (!pool->prison) { |
2496 | *error = "Error creating pool's bio prison"; | |
2497 | err_p = ERR_PTR(-ENOMEM); | |
2498 | goto bad_prison; | |
2499 | } | |
2500 | ||
df5d2e90 | 2501 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2502 | if (IS_ERR(pool->copier)) { |
2503 | r = PTR_ERR(pool->copier); | |
2504 | *error = "Error creating pool's kcopyd client"; | |
2505 | err_p = ERR_PTR(r); | |
2506 | goto bad_kcopyd_client; | |
2507 | } | |
2508 | ||
2509 | /* | |
2510 | * Create singlethreaded workqueue that will service all devices | |
2511 | * that use this metadata. | |
2512 | */ | |
2513 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2514 | if (!pool->wq) { | |
2515 | *error = "Error creating pool's workqueue"; | |
2516 | err_p = ERR_PTR(-ENOMEM); | |
2517 | goto bad_wq; | |
2518 | } | |
2519 | ||
7d327fe0 | 2520 | throttle_init(&pool->throttle); |
991d9fa0 | 2521 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 2522 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2523 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2524 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2525 | bio_list_init(&pool->deferred_flush_bios); |
2526 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2527 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2528 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2529 | pool->low_water_triggered = false; |
44feb387 MS |
2530 | |
2531 | pool->shared_read_ds = dm_deferred_set_create(); | |
2532 | if (!pool->shared_read_ds) { | |
2533 | *error = "Error creating pool's shared read deferred set"; | |
2534 | err_p = ERR_PTR(-ENOMEM); | |
2535 | goto bad_shared_read_ds; | |
2536 | } | |
2537 | ||
2538 | pool->all_io_ds = dm_deferred_set_create(); | |
2539 | if (!pool->all_io_ds) { | |
2540 | *error = "Error creating pool's all io deferred set"; | |
2541 | err_p = ERR_PTR(-ENOMEM); | |
2542 | goto bad_all_io_ds; | |
2543 | } | |
991d9fa0 JT |
2544 | |
2545 | pool->next_mapping = NULL; | |
a24c2569 MS |
2546 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2547 | _new_mapping_cache); | |
991d9fa0 JT |
2548 | if (!pool->mapping_pool) { |
2549 | *error = "Error creating pool's mapping mempool"; | |
2550 | err_p = ERR_PTR(-ENOMEM); | |
2551 | goto bad_mapping_pool; | |
2552 | } | |
2553 | ||
991d9fa0 | 2554 | pool->ref_count = 1; |
905e51b3 | 2555 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2556 | pool->pool_md = pool_md; |
2557 | pool->md_dev = metadata_dev; | |
2558 | __pool_table_insert(pool); | |
2559 | ||
2560 | return pool; | |
2561 | ||
991d9fa0 | 2562 | bad_mapping_pool: |
44feb387 MS |
2563 | dm_deferred_set_destroy(pool->all_io_ds); |
2564 | bad_all_io_ds: | |
2565 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2566 | bad_shared_read_ds: | |
991d9fa0 JT |
2567 | destroy_workqueue(pool->wq); |
2568 | bad_wq: | |
2569 | dm_kcopyd_client_destroy(pool->copier); | |
2570 | bad_kcopyd_client: | |
44feb387 | 2571 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2572 | bad_prison: |
2573 | kfree(pool); | |
2574 | bad_pool: | |
2575 | if (dm_pool_metadata_close(pmd)) | |
2576 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2577 | ||
2578 | return err_p; | |
2579 | } | |
2580 | ||
2581 | static void __pool_inc(struct pool *pool) | |
2582 | { | |
2583 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2584 | pool->ref_count++; | |
2585 | } | |
2586 | ||
2587 | static void __pool_dec(struct pool *pool) | |
2588 | { | |
2589 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2590 | BUG_ON(!pool->ref_count); | |
2591 | if (!--pool->ref_count) | |
2592 | __pool_destroy(pool); | |
2593 | } | |
2594 | ||
2595 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2596 | struct block_device *metadata_dev, | |
e49e5829 JT |
2597 | unsigned long block_size, int read_only, |
2598 | char **error, int *created) | |
991d9fa0 JT |
2599 | { |
2600 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2601 | ||
2602 | if (pool) { | |
f09996c9 MS |
2603 | if (pool->pool_md != pool_md) { |
2604 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2605 | return ERR_PTR(-EBUSY); |
f09996c9 | 2606 | } |
991d9fa0 JT |
2607 | __pool_inc(pool); |
2608 | ||
2609 | } else { | |
2610 | pool = __pool_table_lookup(pool_md); | |
2611 | if (pool) { | |
f09996c9 MS |
2612 | if (pool->md_dev != metadata_dev) { |
2613 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2614 | return ERR_PTR(-EINVAL); |
f09996c9 | 2615 | } |
991d9fa0 JT |
2616 | __pool_inc(pool); |
2617 | ||
67e2e2b2 | 2618 | } else { |
e49e5829 | 2619 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2620 | *created = 1; |
2621 | } | |
991d9fa0 JT |
2622 | } |
2623 | ||
2624 | return pool; | |
2625 | } | |
2626 | ||
2627 | /*---------------------------------------------------------------- | |
2628 | * Pool target methods | |
2629 | *--------------------------------------------------------------*/ | |
2630 | static void pool_dtr(struct dm_target *ti) | |
2631 | { | |
2632 | struct pool_c *pt = ti->private; | |
2633 | ||
2634 | mutex_lock(&dm_thin_pool_table.mutex); | |
2635 | ||
2636 | unbind_control_target(pt->pool, ti); | |
2637 | __pool_dec(pt->pool); | |
2638 | dm_put_device(ti, pt->metadata_dev); | |
2639 | dm_put_device(ti, pt->data_dev); | |
2640 | kfree(pt); | |
2641 | ||
2642 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2643 | } | |
2644 | ||
991d9fa0 JT |
2645 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2646 | struct dm_target *ti) | |
2647 | { | |
2648 | int r; | |
2649 | unsigned argc; | |
2650 | const char *arg_name; | |
2651 | ||
2652 | static struct dm_arg _args[] = { | |
74aa45c3 | 2653 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2654 | }; |
2655 | ||
2656 | /* | |
2657 | * No feature arguments supplied. | |
2658 | */ | |
2659 | if (!as->argc) | |
2660 | return 0; | |
2661 | ||
2662 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2663 | if (r) | |
2664 | return -EINVAL; | |
2665 | ||
2666 | while (argc && !r) { | |
2667 | arg_name = dm_shift_arg(as); | |
2668 | argc--; | |
2669 | ||
e49e5829 | 2670 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2671 | pf->zero_new_blocks = false; |
e49e5829 JT |
2672 | |
2673 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2674 | pf->discard_enabled = false; |
e49e5829 JT |
2675 | |
2676 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2677 | pf->discard_passdown = false; |
991d9fa0 | 2678 | |
e49e5829 JT |
2679 | else if (!strcasecmp(arg_name, "read_only")) |
2680 | pf->mode = PM_READ_ONLY; | |
2681 | ||
787a996c MS |
2682 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2683 | pf->error_if_no_space = true; | |
2684 | ||
e49e5829 JT |
2685 | else { |
2686 | ti->error = "Unrecognised pool feature requested"; | |
2687 | r = -EINVAL; | |
2688 | break; | |
2689 | } | |
991d9fa0 JT |
2690 | } |
2691 | ||
2692 | return r; | |
2693 | } | |
2694 | ||
ac8c3f3d JT |
2695 | static void metadata_low_callback(void *context) |
2696 | { | |
2697 | struct pool *pool = context; | |
2698 | ||
2699 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2700 | dm_device_name(pool->pool_md)); | |
2701 | ||
2702 | dm_table_event(pool->ti->table); | |
2703 | } | |
2704 | ||
7d48935e MS |
2705 | static sector_t get_dev_size(struct block_device *bdev) |
2706 | { | |
2707 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2708 | } | |
2709 | ||
2710 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2711 | { |
7d48935e | 2712 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2713 | char buffer[BDEVNAME_SIZE]; |
2714 | ||
7d48935e | 2715 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2716 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2717 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2718 | } |
2719 | ||
2720 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2721 | { | |
2722 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2723 | ||
2724 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2725 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2726 | |
2727 | return metadata_dev_size; | |
2728 | } | |
2729 | ||
24347e95 JT |
2730 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2731 | { | |
2732 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2733 | ||
7d48935e | 2734 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2735 | |
2736 | return metadata_dev_size; | |
2737 | } | |
2738 | ||
ac8c3f3d JT |
2739 | /* |
2740 | * When a metadata threshold is crossed a dm event is triggered, and | |
2741 | * userland should respond by growing the metadata device. We could let | |
2742 | * userland set the threshold, like we do with the data threshold, but I'm | |
2743 | * not sure they know enough to do this well. | |
2744 | */ | |
2745 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2746 | { | |
2747 | /* | |
2748 | * 4M is ample for all ops with the possible exception of thin | |
2749 | * device deletion which is harmless if it fails (just retry the | |
2750 | * delete after you've grown the device). | |
2751 | */ | |
2752 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2753 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2754 | } | |
2755 | ||
991d9fa0 JT |
2756 | /* |
2757 | * thin-pool <metadata dev> <data dev> | |
2758 | * <data block size (sectors)> | |
2759 | * <low water mark (blocks)> | |
2760 | * [<#feature args> [<arg>]*] | |
2761 | * | |
2762 | * Optional feature arguments are: | |
2763 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2764 | * ignore_discard: disable discard |
2765 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2766 | * read_only: Don't allow any changes to be made to the pool metadata. |
2767 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2768 | */ |
2769 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2770 | { | |
67e2e2b2 | 2771 | int r, pool_created = 0; |
991d9fa0 JT |
2772 | struct pool_c *pt; |
2773 | struct pool *pool; | |
2774 | struct pool_features pf; | |
2775 | struct dm_arg_set as; | |
2776 | struct dm_dev *data_dev; | |
2777 | unsigned long block_size; | |
2778 | dm_block_t low_water_blocks; | |
2779 | struct dm_dev *metadata_dev; | |
5d0db96d | 2780 | fmode_t metadata_mode; |
991d9fa0 JT |
2781 | |
2782 | /* | |
2783 | * FIXME Remove validation from scope of lock. | |
2784 | */ | |
2785 | mutex_lock(&dm_thin_pool_table.mutex); | |
2786 | ||
2787 | if (argc < 4) { | |
2788 | ti->error = "Invalid argument count"; | |
2789 | r = -EINVAL; | |
2790 | goto out_unlock; | |
2791 | } | |
5d0db96d | 2792 | |
991d9fa0 JT |
2793 | as.argc = argc; |
2794 | as.argv = argv; | |
2795 | ||
5d0db96d JT |
2796 | /* |
2797 | * Set default pool features. | |
2798 | */ | |
2799 | pool_features_init(&pf); | |
2800 | ||
2801 | dm_consume_args(&as, 4); | |
2802 | r = parse_pool_features(&as, &pf, ti); | |
2803 | if (r) | |
2804 | goto out_unlock; | |
2805 | ||
2806 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2807 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2808 | if (r) { |
2809 | ti->error = "Error opening metadata block device"; | |
2810 | goto out_unlock; | |
2811 | } | |
7d48935e | 2812 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2813 | |
2814 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2815 | if (r) { | |
2816 | ti->error = "Error getting data device"; | |
2817 | goto out_metadata; | |
2818 | } | |
2819 | ||
2820 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2821 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2822 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2823 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2824 | ti->error = "Invalid block size"; |
2825 | r = -EINVAL; | |
2826 | goto out; | |
2827 | } | |
2828 | ||
2829 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2830 | ti->error = "Invalid low water mark"; | |
2831 | r = -EINVAL; | |
2832 | goto out; | |
2833 | } | |
2834 | ||
991d9fa0 JT |
2835 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2836 | if (!pt) { | |
2837 | r = -ENOMEM; | |
2838 | goto out; | |
2839 | } | |
2840 | ||
2841 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2842 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2843 | if (IS_ERR(pool)) { |
2844 | r = PTR_ERR(pool); | |
2845 | goto out_free_pt; | |
2846 | } | |
2847 | ||
67e2e2b2 JT |
2848 | /* |
2849 | * 'pool_created' reflects whether this is the first table load. | |
2850 | * Top level discard support is not allowed to be changed after | |
2851 | * initial load. This would require a pool reload to trigger thin | |
2852 | * device changes. | |
2853 | */ | |
2854 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2855 | ti->error = "Discard support cannot be disabled once enabled"; | |
2856 | r = -EINVAL; | |
2857 | goto out_flags_changed; | |
2858 | } | |
2859 | ||
991d9fa0 JT |
2860 | pt->pool = pool; |
2861 | pt->ti = ti; | |
2862 | pt->metadata_dev = metadata_dev; | |
2863 | pt->data_dev = data_dev; | |
2864 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2865 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2866 | ti->num_flush_bios = 1; |
9bc142dd | 2867 | |
67e2e2b2 JT |
2868 | /* |
2869 | * Only need to enable discards if the pool should pass | |
2870 | * them down to the data device. The thin device's discard | |
2871 | * processing will cause mappings to be removed from the btree. | |
2872 | */ | |
b60ab990 | 2873 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2874 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2875 | ti->num_discard_bios = 1; |
9bc142dd | 2876 | |
67e2e2b2 JT |
2877 | /* |
2878 | * Setting 'discards_supported' circumvents the normal | |
2879 | * stacking of discard limits (this keeps the pool and | |
2880 | * thin devices' discard limits consistent). | |
2881 | */ | |
0ac55489 | 2882 | ti->discards_supported = true; |
67e2e2b2 | 2883 | } |
991d9fa0 JT |
2884 | ti->private = pt; |
2885 | ||
ac8c3f3d JT |
2886 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2887 | calc_metadata_threshold(pt), | |
2888 | metadata_low_callback, | |
2889 | pool); | |
2890 | if (r) | |
2891 | goto out_free_pt; | |
2892 | ||
991d9fa0 JT |
2893 | pt->callbacks.congested_fn = pool_is_congested; |
2894 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2895 | ||
2896 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2897 | ||
2898 | return 0; | |
2899 | ||
67e2e2b2 JT |
2900 | out_flags_changed: |
2901 | __pool_dec(pool); | |
991d9fa0 JT |
2902 | out_free_pt: |
2903 | kfree(pt); | |
2904 | out: | |
2905 | dm_put_device(ti, data_dev); | |
2906 | out_metadata: | |
2907 | dm_put_device(ti, metadata_dev); | |
2908 | out_unlock: | |
2909 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2910 | ||
2911 | return r; | |
2912 | } | |
2913 | ||
7de3ee57 | 2914 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2915 | { |
2916 | int r; | |
2917 | struct pool_c *pt = ti->private; | |
2918 | struct pool *pool = pt->pool; | |
2919 | unsigned long flags; | |
2920 | ||
2921 | /* | |
2922 | * As this is a singleton target, ti->begin is always zero. | |
2923 | */ | |
2924 | spin_lock_irqsave(&pool->lock, flags); | |
2925 | bio->bi_bdev = pt->data_dev->bdev; | |
2926 | r = DM_MAPIO_REMAPPED; | |
2927 | spin_unlock_irqrestore(&pool->lock, flags); | |
2928 | ||
2929 | return r; | |
2930 | } | |
2931 | ||
b17446df | 2932 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2933 | { |
2934 | int r; | |
2935 | struct pool_c *pt = ti->private; | |
2936 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2937 | sector_t data_size = ti->len; |
2938 | dm_block_t sb_data_size; | |
991d9fa0 | 2939 | |
b17446df | 2940 | *need_commit = false; |
991d9fa0 | 2941 | |
55f2b8bd MS |
2942 | (void) sector_div(data_size, pool->sectors_per_block); |
2943 | ||
991d9fa0 JT |
2944 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2945 | if (r) { | |
4fa5971a MS |
2946 | DMERR("%s: failed to retrieve data device size", |
2947 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2948 | return r; |
2949 | } | |
2950 | ||
2951 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2952 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2953 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2954 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2955 | return -EINVAL; |
2956 | ||
2957 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2958 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2959 | DMERR("%s: unable to grow the data device until repaired.", | |
2960 | dm_device_name(pool->pool_md)); | |
2961 | return 0; | |
2962 | } | |
2963 | ||
6f7f51d4 MS |
2964 | if (sb_data_size) |
2965 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2966 | dm_device_name(pool->pool_md), | |
2967 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2968 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2969 | if (r) { | |
b5330655 | 2970 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2971 | return r; |
2972 | } | |
2973 | ||
b17446df | 2974 | *need_commit = true; |
991d9fa0 JT |
2975 | } |
2976 | ||
2977 | return 0; | |
2978 | } | |
2979 | ||
24347e95 JT |
2980 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2981 | { | |
2982 | int r; | |
2983 | struct pool_c *pt = ti->private; | |
2984 | struct pool *pool = pt->pool; | |
2985 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2986 | ||
2987 | *need_commit = false; | |
2988 | ||
610bba8b | 2989 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2990 | |
2991 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2992 | if (r) { | |
4fa5971a MS |
2993 | DMERR("%s: failed to retrieve metadata device size", |
2994 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2995 | return r; |
2996 | } | |
2997 | ||
2998 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2999 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3000 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3001 | metadata_dev_size, sb_metadata_dev_size); |
3002 | return -EINVAL; | |
3003 | ||
3004 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3005 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3006 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3007 | dm_device_name(pool->pool_md)); | |
3008 | return 0; | |
3009 | } | |
3010 | ||
7d48935e | 3011 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3012 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3013 | dm_device_name(pool->pool_md), | |
3014 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
3015 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3016 | if (r) { | |
b5330655 | 3017 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3018 | return r; |
3019 | } | |
3020 | ||
3021 | *need_commit = true; | |
3022 | } | |
3023 | ||
3024 | return 0; | |
3025 | } | |
3026 | ||
b17446df JT |
3027 | /* |
3028 | * Retrieves the number of blocks of the data device from | |
3029 | * the superblock and compares it to the actual device size, | |
3030 | * thus resizing the data device in case it has grown. | |
3031 | * | |
3032 | * This both copes with opening preallocated data devices in the ctr | |
3033 | * being followed by a resume | |
3034 | * -and- | |
3035 | * calling the resume method individually after userspace has | |
3036 | * grown the data device in reaction to a table event. | |
3037 | */ | |
3038 | static int pool_preresume(struct dm_target *ti) | |
3039 | { | |
3040 | int r; | |
24347e95 | 3041 | bool need_commit1, need_commit2; |
b17446df JT |
3042 | struct pool_c *pt = ti->private; |
3043 | struct pool *pool = pt->pool; | |
3044 | ||
3045 | /* | |
3046 | * Take control of the pool object. | |
3047 | */ | |
3048 | r = bind_control_target(pool, ti); | |
3049 | if (r) | |
3050 | return r; | |
3051 | ||
3052 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3053 | if (r) | |
3054 | return r; | |
3055 | ||
24347e95 JT |
3056 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3057 | if (r) | |
3058 | return r; | |
3059 | ||
3060 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3061 | (void) commit(pool); |
b17446df JT |
3062 | |
3063 | return 0; | |
3064 | } | |
3065 | ||
991d9fa0 JT |
3066 | static void pool_resume(struct dm_target *ti) |
3067 | { | |
3068 | struct pool_c *pt = ti->private; | |
3069 | struct pool *pool = pt->pool; | |
3070 | unsigned long flags; | |
3071 | ||
3072 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 3073 | pool->low_water_triggered = false; |
991d9fa0 | 3074 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 3075 | requeue_bios(pool); |
991d9fa0 | 3076 | |
905e51b3 | 3077 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3078 | } |
3079 | ||
3080 | static void pool_postsuspend(struct dm_target *ti) | |
3081 | { | |
991d9fa0 JT |
3082 | struct pool_c *pt = ti->private; |
3083 | struct pool *pool = pt->pool; | |
3084 | ||
905e51b3 | 3085 | cancel_delayed_work(&pool->waker); |
85ad643b | 3086 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 3087 | flush_workqueue(pool->wq); |
020cc3b5 | 3088 | (void) commit(pool); |
991d9fa0 JT |
3089 | } |
3090 | ||
3091 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3092 | { | |
3093 | if (argc != args_required) { | |
3094 | DMWARN("Message received with %u arguments instead of %u.", | |
3095 | argc, args_required); | |
3096 | return -EINVAL; | |
3097 | } | |
3098 | ||
3099 | return 0; | |
3100 | } | |
3101 | ||
3102 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3103 | { | |
3104 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3105 | *dev_id <= MAX_DEV_ID) | |
3106 | return 0; | |
3107 | ||
3108 | if (warning) | |
3109 | DMWARN("Message received with invalid device id: %s", arg); | |
3110 | ||
3111 | return -EINVAL; | |
3112 | } | |
3113 | ||
3114 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3115 | { | |
3116 | dm_thin_id dev_id; | |
3117 | int r; | |
3118 | ||
3119 | r = check_arg_count(argc, 2); | |
3120 | if (r) | |
3121 | return r; | |
3122 | ||
3123 | r = read_dev_id(argv[1], &dev_id, 1); | |
3124 | if (r) | |
3125 | return r; | |
3126 | ||
3127 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3128 | if (r) { | |
3129 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3130 | argv[1]); | |
3131 | return r; | |
3132 | } | |
3133 | ||
3134 | return 0; | |
3135 | } | |
3136 | ||
3137 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3138 | { | |
3139 | dm_thin_id dev_id; | |
3140 | dm_thin_id origin_dev_id; | |
3141 | int r; | |
3142 | ||
3143 | r = check_arg_count(argc, 3); | |
3144 | if (r) | |
3145 | return r; | |
3146 | ||
3147 | r = read_dev_id(argv[1], &dev_id, 1); | |
3148 | if (r) | |
3149 | return r; | |
3150 | ||
3151 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3152 | if (r) | |
3153 | return r; | |
3154 | ||
3155 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3156 | if (r) { | |
3157 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3158 | argv[1], argv[2]); | |
3159 | return r; | |
3160 | } | |
3161 | ||
3162 | return 0; | |
3163 | } | |
3164 | ||
3165 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3166 | { | |
3167 | dm_thin_id dev_id; | |
3168 | int r; | |
3169 | ||
3170 | r = check_arg_count(argc, 2); | |
3171 | if (r) | |
3172 | return r; | |
3173 | ||
3174 | r = read_dev_id(argv[1], &dev_id, 1); | |
3175 | if (r) | |
3176 | return r; | |
3177 | ||
3178 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3179 | if (r) | |
3180 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3181 | ||
3182 | return r; | |
3183 | } | |
3184 | ||
3185 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3186 | { | |
3187 | dm_thin_id old_id, new_id; | |
3188 | int r; | |
3189 | ||
3190 | r = check_arg_count(argc, 3); | |
3191 | if (r) | |
3192 | return r; | |
3193 | ||
3194 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3195 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3196 | return -EINVAL; | |
3197 | } | |
3198 | ||
3199 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3200 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3201 | return -EINVAL; | |
3202 | } | |
3203 | ||
3204 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3205 | if (r) { | |
3206 | DMWARN("Failed to change transaction id from %s to %s.", | |
3207 | argv[1], argv[2]); | |
3208 | return r; | |
3209 | } | |
3210 | ||
3211 | return 0; | |
3212 | } | |
3213 | ||
cc8394d8 JT |
3214 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3215 | { | |
3216 | int r; | |
3217 | ||
3218 | r = check_arg_count(argc, 1); | |
3219 | if (r) | |
3220 | return r; | |
3221 | ||
020cc3b5 | 3222 | (void) commit(pool); |
0d200aef | 3223 | |
cc8394d8 JT |
3224 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3225 | if (r) | |
3226 | DMWARN("reserve_metadata_snap message failed."); | |
3227 | ||
3228 | return r; | |
3229 | } | |
3230 | ||
3231 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3232 | { | |
3233 | int r; | |
3234 | ||
3235 | r = check_arg_count(argc, 1); | |
3236 | if (r) | |
3237 | return r; | |
3238 | ||
3239 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3240 | if (r) | |
3241 | DMWARN("release_metadata_snap message failed."); | |
3242 | ||
3243 | return r; | |
3244 | } | |
3245 | ||
991d9fa0 JT |
3246 | /* |
3247 | * Messages supported: | |
3248 | * create_thin <dev_id> | |
3249 | * create_snap <dev_id> <origin_id> | |
3250 | * delete <dev_id> | |
3251 | * trim <dev_id> <new_size_in_sectors> | |
3252 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
3253 | * reserve_metadata_snap |
3254 | * release_metadata_snap | |
991d9fa0 JT |
3255 | */ |
3256 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
3257 | { | |
3258 | int r = -EINVAL; | |
3259 | struct pool_c *pt = ti->private; | |
3260 | struct pool *pool = pt->pool; | |
3261 | ||
3262 | if (!strcasecmp(argv[0], "create_thin")) | |
3263 | r = process_create_thin_mesg(argc, argv, pool); | |
3264 | ||
3265 | else if (!strcasecmp(argv[0], "create_snap")) | |
3266 | r = process_create_snap_mesg(argc, argv, pool); | |
3267 | ||
3268 | else if (!strcasecmp(argv[0], "delete")) | |
3269 | r = process_delete_mesg(argc, argv, pool); | |
3270 | ||
3271 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3272 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3273 | ||
cc8394d8 JT |
3274 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3275 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3276 | ||
3277 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3278 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3279 | ||
991d9fa0 JT |
3280 | else |
3281 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3282 | ||
e49e5829 | 3283 | if (!r) |
020cc3b5 | 3284 | (void) commit(pool); |
991d9fa0 JT |
3285 | |
3286 | return r; | |
3287 | } | |
3288 | ||
e49e5829 JT |
3289 | static void emit_flags(struct pool_features *pf, char *result, |
3290 | unsigned sz, unsigned maxlen) | |
3291 | { | |
3292 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3293 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3294 | pf->error_if_no_space; | |
e49e5829 JT |
3295 | DMEMIT("%u ", count); |
3296 | ||
3297 | if (!pf->zero_new_blocks) | |
3298 | DMEMIT("skip_block_zeroing "); | |
3299 | ||
3300 | if (!pf->discard_enabled) | |
3301 | DMEMIT("ignore_discard "); | |
3302 | ||
3303 | if (!pf->discard_passdown) | |
3304 | DMEMIT("no_discard_passdown "); | |
3305 | ||
3306 | if (pf->mode == PM_READ_ONLY) | |
3307 | DMEMIT("read_only "); | |
787a996c MS |
3308 | |
3309 | if (pf->error_if_no_space) | |
3310 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3311 | } |
3312 | ||
991d9fa0 JT |
3313 | /* |
3314 | * Status line is: | |
3315 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3316 | * <used data sectors>/<total data sectors> <held metadata root> | |
3317 | */ | |
fd7c092e MP |
3318 | static void pool_status(struct dm_target *ti, status_type_t type, |
3319 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3320 | { |
e49e5829 | 3321 | int r; |
991d9fa0 JT |
3322 | unsigned sz = 0; |
3323 | uint64_t transaction_id; | |
3324 | dm_block_t nr_free_blocks_data; | |
3325 | dm_block_t nr_free_blocks_metadata; | |
3326 | dm_block_t nr_blocks_data; | |
3327 | dm_block_t nr_blocks_metadata; | |
3328 | dm_block_t held_root; | |
3329 | char buf[BDEVNAME_SIZE]; | |
3330 | char buf2[BDEVNAME_SIZE]; | |
3331 | struct pool_c *pt = ti->private; | |
3332 | struct pool *pool = pt->pool; | |
3333 | ||
3334 | switch (type) { | |
3335 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3336 | if (get_pool_mode(pool) == PM_FAIL) { |
3337 | DMEMIT("Fail"); | |
3338 | break; | |
3339 | } | |
3340 | ||
1f4e0ff0 AK |
3341 | /* Commit to ensure statistics aren't out-of-date */ |
3342 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3343 | (void) commit(pool); |
1f4e0ff0 | 3344 | |
fd7c092e MP |
3345 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3346 | if (r) { | |
4fa5971a MS |
3347 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3348 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3349 | goto err; |
3350 | } | |
991d9fa0 | 3351 | |
fd7c092e MP |
3352 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3353 | if (r) { | |
4fa5971a MS |
3354 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3355 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3356 | goto err; |
3357 | } | |
991d9fa0 JT |
3358 | |
3359 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3360 | if (r) { |
4fa5971a MS |
3361 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3362 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3363 | goto err; |
3364 | } | |
991d9fa0 | 3365 | |
fd7c092e MP |
3366 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3367 | if (r) { | |
4fa5971a MS |
3368 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3369 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3370 | goto err; |
3371 | } | |
991d9fa0 JT |
3372 | |
3373 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3374 | if (r) { |
4fa5971a MS |
3375 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3376 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3377 | goto err; |
3378 | } | |
991d9fa0 | 3379 | |
cc8394d8 | 3380 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3381 | if (r) { |
4fa5971a MS |
3382 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3383 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3384 | goto err; |
3385 | } | |
991d9fa0 JT |
3386 | |
3387 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3388 | (unsigned long long)transaction_id, | |
3389 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3390 | (unsigned long long)nr_blocks_metadata, | |
3391 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3392 | (unsigned long long)nr_blocks_data); | |
3393 | ||
3394 | if (held_root) | |
e49e5829 JT |
3395 | DMEMIT("%llu ", held_root); |
3396 | else | |
3397 | DMEMIT("- "); | |
3398 | ||
3e1a0699 JT |
3399 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3400 | DMEMIT("out_of_data_space "); | |
3401 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3402 | DMEMIT("ro "); |
991d9fa0 | 3403 | else |
e49e5829 JT |
3404 | DMEMIT("rw "); |
3405 | ||
018debea | 3406 | if (!pool->pf.discard_enabled) |
787a996c | 3407 | DMEMIT("ignore_discard "); |
018debea | 3408 | else if (pool->pf.discard_passdown) |
787a996c MS |
3409 | DMEMIT("discard_passdown "); |
3410 | else | |
3411 | DMEMIT("no_discard_passdown "); | |
3412 | ||
3413 | if (pool->pf.error_if_no_space) | |
3414 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3415 | else |
787a996c | 3416 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3417 | |
3418 | break; | |
3419 | ||
3420 | case STATUSTYPE_TABLE: | |
3421 | DMEMIT("%s %s %lu %llu ", | |
3422 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3423 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3424 | (unsigned long)pool->sectors_per_block, | |
3425 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3426 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3427 | break; |
3428 | } | |
fd7c092e | 3429 | return; |
991d9fa0 | 3430 | |
fd7c092e MP |
3431 | err: |
3432 | DMEMIT("Error"); | |
991d9fa0 JT |
3433 | } |
3434 | ||
3435 | static int pool_iterate_devices(struct dm_target *ti, | |
3436 | iterate_devices_callout_fn fn, void *data) | |
3437 | { | |
3438 | struct pool_c *pt = ti->private; | |
3439 | ||
3440 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3441 | } | |
3442 | ||
3443 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3444 | struct bio_vec *biovec, int max_size) | |
3445 | { | |
3446 | struct pool_c *pt = ti->private; | |
3447 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3448 | ||
3449 | if (!q->merge_bvec_fn) | |
3450 | return max_size; | |
3451 | ||
3452 | bvm->bi_bdev = pt->data_dev->bdev; | |
3453 | ||
3454 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3455 | } | |
3456 | ||
0424caa1 | 3457 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 3458 | { |
0424caa1 MS |
3459 | struct pool *pool = pt->pool; |
3460 | struct queue_limits *data_limits; | |
3461 | ||
104655fd JT |
3462 | limits->max_discard_sectors = pool->sectors_per_block; |
3463 | ||
3464 | /* | |
0424caa1 | 3465 | * discard_granularity is just a hint, and not enforced. |
104655fd | 3466 | */ |
0424caa1 MS |
3467 | if (pt->adjusted_pf.discard_passdown) { |
3468 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
09869de5 LC |
3469 | limits->discard_granularity = max(data_limits->discard_granularity, |
3470 | pool->sectors_per_block << SECTOR_SHIFT); | |
f13945d7 | 3471 | } else |
0424caa1 | 3472 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3473 | } |
3474 | ||
991d9fa0 JT |
3475 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3476 | { | |
3477 | struct pool_c *pt = ti->private; | |
3478 | struct pool *pool = pt->pool; | |
604ea906 MS |
3479 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
3480 | ||
3481 | /* | |
3482 | * Adjust max_sectors_kb to highest possible power-of-2 | |
3483 | * factor of pool->sectors_per_block. | |
3484 | */ | |
3485 | if (limits->max_hw_sectors & (limits->max_hw_sectors - 1)) | |
3486 | limits->max_sectors = rounddown_pow_of_two(limits->max_hw_sectors); | |
3487 | else | |
3488 | limits->max_sectors = limits->max_hw_sectors; | |
3489 | ||
3490 | if (limits->max_sectors < pool->sectors_per_block) { | |
3491 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
3492 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3493 | limits->max_sectors--; | |
3494 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3495 | } | |
3496 | } else if (block_size_is_power_of_two(pool)) { | |
3497 | /* max_sectors_kb is >= power-of-2 thinp blocksize */ | |
3498 | while (!is_factor(limits->max_sectors, pool->sectors_per_block)) { | |
3499 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3500 | limits->max_sectors--; | |
3501 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3502 | } | |
3503 | } | |
991d9fa0 | 3504 | |
0cc67cd9 MS |
3505 | /* |
3506 | * If the system-determined stacked limits are compatible with the | |
3507 | * pool's blocksize (io_opt is a factor) do not override them. | |
3508 | */ | |
3509 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
3510 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
3511 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
3512 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
3513 | else | |
3514 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
3515 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
3516 | } | |
0424caa1 MS |
3517 | |
3518 | /* | |
3519 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3520 | * They get transferred to the live pool in bind_control_target() | |
3521 | * called from pool_preresume(). | |
3522 | */ | |
b60ab990 MS |
3523 | if (!pt->adjusted_pf.discard_enabled) { |
3524 | /* | |
3525 | * Must explicitly disallow stacking discard limits otherwise the | |
3526 | * block layer will stack them if pool's data device has support. | |
3527 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3528 | * user to see that, so make sure to set all discard limits to 0. | |
3529 | */ | |
3530 | limits->discard_granularity = 0; | |
0424caa1 | 3531 | return; |
b60ab990 | 3532 | } |
0424caa1 MS |
3533 | |
3534 | disable_passdown_if_not_supported(pt); | |
3535 | ||
3536 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3537 | } |
3538 | ||
3539 | static struct target_type pool_target = { | |
3540 | .name = "thin-pool", | |
3541 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3542 | DM_TARGET_IMMUTABLE, | |
36f12aeb | 3543 | .version = {1, 14, 0}, |
991d9fa0 JT |
3544 | .module = THIS_MODULE, |
3545 | .ctr = pool_ctr, | |
3546 | .dtr = pool_dtr, | |
3547 | .map = pool_map, | |
3548 | .postsuspend = pool_postsuspend, | |
3549 | .preresume = pool_preresume, | |
3550 | .resume = pool_resume, | |
3551 | .message = pool_message, | |
3552 | .status = pool_status, | |
3553 | .merge = pool_merge, | |
3554 | .iterate_devices = pool_iterate_devices, | |
3555 | .io_hints = pool_io_hints, | |
3556 | }; | |
3557 | ||
3558 | /*---------------------------------------------------------------- | |
3559 | * Thin target methods | |
3560 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3561 | static void thin_get(struct thin_c *tc) |
3562 | { | |
3563 | atomic_inc(&tc->refcount); | |
3564 | } | |
3565 | ||
3566 | static void thin_put(struct thin_c *tc) | |
3567 | { | |
3568 | if (atomic_dec_and_test(&tc->refcount)) | |
3569 | complete(&tc->can_destroy); | |
3570 | } | |
3571 | ||
991d9fa0 JT |
3572 | static void thin_dtr(struct dm_target *ti) |
3573 | { | |
3574 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3575 | unsigned long flags; |
3576 | ||
b10ebd34 JT |
3577 | thin_put(tc); |
3578 | wait_for_completion(&tc->can_destroy); | |
3579 | ||
c140e1c4 MS |
3580 | spin_lock_irqsave(&tc->pool->lock, flags); |
3581 | list_del_rcu(&tc->list); | |
3582 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3583 | synchronize_rcu(); | |
991d9fa0 JT |
3584 | |
3585 | mutex_lock(&dm_thin_pool_table.mutex); | |
3586 | ||
3587 | __pool_dec(tc->pool); | |
3588 | dm_pool_close_thin_device(tc->td); | |
3589 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3590 | if (tc->origin_dev) |
3591 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3592 | kfree(tc); |
3593 | ||
3594 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3595 | } | |
3596 | ||
3597 | /* | |
3598 | * Thin target parameters: | |
3599 | * | |
2dd9c257 | 3600 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3601 | * |
3602 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3603 | * dev_id: the internal device identifier | |
2dd9c257 | 3604 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3605 | * |
3606 | * If the pool device has discards disabled, they get disabled for the thin | |
3607 | * device as well. | |
991d9fa0 JT |
3608 | */ |
3609 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3610 | { | |
3611 | int r; | |
3612 | struct thin_c *tc; | |
2dd9c257 | 3613 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3614 | struct mapped_device *pool_md; |
5e3283e2 | 3615 | unsigned long flags; |
991d9fa0 JT |
3616 | |
3617 | mutex_lock(&dm_thin_pool_table.mutex); | |
3618 | ||
2dd9c257 | 3619 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3620 | ti->error = "Invalid argument count"; |
3621 | r = -EINVAL; | |
3622 | goto out_unlock; | |
3623 | } | |
3624 | ||
3625 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3626 | if (!tc) { | |
3627 | ti->error = "Out of memory"; | |
3628 | r = -ENOMEM; | |
3629 | goto out_unlock; | |
3630 | } | |
c140e1c4 | 3631 | spin_lock_init(&tc->lock); |
a374bb21 | 3632 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
3633 | bio_list_init(&tc->deferred_bio_list); |
3634 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3635 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3636 | |
2dd9c257 JT |
3637 | if (argc == 3) { |
3638 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3639 | if (r) { | |
3640 | ti->error = "Error opening origin device"; | |
3641 | goto bad_origin_dev; | |
3642 | } | |
3643 | tc->origin_dev = origin_dev; | |
3644 | } | |
3645 | ||
991d9fa0 JT |
3646 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3647 | if (r) { | |
3648 | ti->error = "Error opening pool device"; | |
3649 | goto bad_pool_dev; | |
3650 | } | |
3651 | tc->pool_dev = pool_dev; | |
3652 | ||
3653 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3654 | ti->error = "Invalid device id"; | |
3655 | r = -EINVAL; | |
3656 | goto bad_common; | |
3657 | } | |
3658 | ||
3659 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3660 | if (!pool_md) { | |
3661 | ti->error = "Couldn't get pool mapped device"; | |
3662 | r = -EINVAL; | |
3663 | goto bad_common; | |
3664 | } | |
3665 | ||
3666 | tc->pool = __pool_table_lookup(pool_md); | |
3667 | if (!tc->pool) { | |
3668 | ti->error = "Couldn't find pool object"; | |
3669 | r = -EINVAL; | |
3670 | goto bad_pool_lookup; | |
3671 | } | |
3672 | __pool_inc(tc->pool); | |
3673 | ||
e49e5829 JT |
3674 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3675 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3676 | r = -EINVAL; |
e49e5829 JT |
3677 | goto bad_thin_open; |
3678 | } | |
3679 | ||
991d9fa0 JT |
3680 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3681 | if (r) { | |
3682 | ti->error = "Couldn't open thin internal device"; | |
3683 | goto bad_thin_open; | |
3684 | } | |
3685 | ||
542f9038 MS |
3686 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3687 | if (r) | |
1acacc07 | 3688 | goto bad_target_max_io_len; |
542f9038 | 3689 | |
55a62eef | 3690 | ti->num_flush_bios = 1; |
16ad3d10 | 3691 | ti->flush_supported = true; |
59c3d2c6 | 3692 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3693 | |
3694 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3695 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3696 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3697 | ti->discards_supported = true; |
55a62eef | 3698 | ti->num_discard_bios = 1; |
55a62eef AK |
3699 | /* Discard bios must be split on a block boundary */ |
3700 | ti->split_discard_bios = true; | |
67e2e2b2 | 3701 | } |
991d9fa0 JT |
3702 | |
3703 | dm_put(pool_md); | |
3704 | ||
3705 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3706 | ||
b10ebd34 JT |
3707 | atomic_set(&tc->refcount, 1); |
3708 | init_completion(&tc->can_destroy); | |
3709 | ||
5e3283e2 | 3710 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3711 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3712 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3713 | /* |
3714 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3715 | * wake_worker() call finding no bios to process (because the newly | |
3716 | * added tc isn't yet visible). So this reduces latency since we | |
3717 | * aren't then dependent on the periodic commit to wake_worker(). | |
3718 | */ | |
3719 | synchronize_rcu(); | |
3720 | ||
991d9fa0 JT |
3721 | return 0; |
3722 | ||
1acacc07 MS |
3723 | bad_target_max_io_len: |
3724 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3725 | bad_thin_open: |
3726 | __pool_dec(tc->pool); | |
3727 | bad_pool_lookup: | |
3728 | dm_put(pool_md); | |
3729 | bad_common: | |
3730 | dm_put_device(ti, tc->pool_dev); | |
3731 | bad_pool_dev: | |
2dd9c257 JT |
3732 | if (tc->origin_dev) |
3733 | dm_put_device(ti, tc->origin_dev); | |
3734 | bad_origin_dev: | |
991d9fa0 JT |
3735 | kfree(tc); |
3736 | out_unlock: | |
3737 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3738 | ||
3739 | return r; | |
3740 | } | |
3741 | ||
7de3ee57 | 3742 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3743 | { |
4f024f37 | 3744 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3745 | |
7de3ee57 | 3746 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3747 | } |
3748 | ||
7de3ee57 | 3749 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3750 | { |
3751 | unsigned long flags; | |
59c3d2c6 | 3752 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3753 | struct list_head work; |
a24c2569 | 3754 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3755 | struct pool *pool = h->tc->pool; |
3756 | ||
3757 | if (h->shared_read_entry) { | |
3758 | INIT_LIST_HEAD(&work); | |
44feb387 | 3759 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3760 | |
3761 | spin_lock_irqsave(&pool->lock, flags); | |
3762 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3763 | list_del(&m->list); | |
50f3c3ef | 3764 | __complete_mapping_preparation(m); |
eb2aa48d JT |
3765 | } |
3766 | spin_unlock_irqrestore(&pool->lock, flags); | |
3767 | } | |
3768 | ||
104655fd JT |
3769 | if (h->all_io_entry) { |
3770 | INIT_LIST_HEAD(&work); | |
44feb387 | 3771 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3772 | if (!list_empty(&work)) { |
3773 | spin_lock_irqsave(&pool->lock, flags); | |
3774 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3775 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3776 | spin_unlock_irqrestore(&pool->lock, flags); |
3777 | wake_worker(pool); | |
3778 | } | |
104655fd JT |
3779 | } |
3780 | ||
eb2aa48d JT |
3781 | return 0; |
3782 | } | |
3783 | ||
738211f7 | 3784 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3785 | { |
738211f7 JT |
3786 | struct thin_c *tc = ti->private; |
3787 | ||
991d9fa0 | 3788 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3789 | noflush_work(tc, do_noflush_start); |
3790 | } | |
3791 | ||
3792 | static void thin_postsuspend(struct dm_target *ti) | |
3793 | { | |
3794 | struct thin_c *tc = ti->private; | |
3795 | ||
3796 | /* | |
3797 | * The dm_noflush_suspending flag has been cleared by now, so | |
3798 | * unfortunately we must always run this. | |
3799 | */ | |
3800 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3801 | } |
3802 | ||
e5aea7b4 JT |
3803 | static int thin_preresume(struct dm_target *ti) |
3804 | { | |
3805 | struct thin_c *tc = ti->private; | |
3806 | ||
3807 | if (tc->origin_dev) | |
3808 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
3809 | ||
3810 | return 0; | |
3811 | } | |
3812 | ||
991d9fa0 JT |
3813 | /* |
3814 | * <nr mapped sectors> <highest mapped sector> | |
3815 | */ | |
fd7c092e MP |
3816 | static void thin_status(struct dm_target *ti, status_type_t type, |
3817 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3818 | { |
3819 | int r; | |
3820 | ssize_t sz = 0; | |
3821 | dm_block_t mapped, highest; | |
3822 | char buf[BDEVNAME_SIZE]; | |
3823 | struct thin_c *tc = ti->private; | |
3824 | ||
e49e5829 JT |
3825 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3826 | DMEMIT("Fail"); | |
fd7c092e | 3827 | return; |
e49e5829 JT |
3828 | } |
3829 | ||
991d9fa0 JT |
3830 | if (!tc->td) |
3831 | DMEMIT("-"); | |
3832 | else { | |
3833 | switch (type) { | |
3834 | case STATUSTYPE_INFO: | |
3835 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3836 | if (r) { |
3837 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3838 | goto err; | |
3839 | } | |
991d9fa0 JT |
3840 | |
3841 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3842 | if (r < 0) { |
3843 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3844 | goto err; | |
3845 | } | |
991d9fa0 JT |
3846 | |
3847 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3848 | if (r) | |
3849 | DMEMIT("%llu", ((highest + 1) * | |
3850 | tc->pool->sectors_per_block) - 1); | |
3851 | else | |
3852 | DMEMIT("-"); | |
3853 | break; | |
3854 | ||
3855 | case STATUSTYPE_TABLE: | |
3856 | DMEMIT("%s %lu", | |
3857 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3858 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3859 | if (tc->origin_dev) |
3860 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3861 | break; |
3862 | } | |
3863 | } | |
3864 | ||
fd7c092e MP |
3865 | return; |
3866 | ||
3867 | err: | |
3868 | DMEMIT("Error"); | |
991d9fa0 JT |
3869 | } |
3870 | ||
36f12aeb MS |
3871 | static int thin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
3872 | struct bio_vec *biovec, int max_size) | |
3873 | { | |
3874 | struct thin_c *tc = ti->private; | |
3875 | struct request_queue *q = bdev_get_queue(tc->pool_dev->bdev); | |
3876 | ||
3877 | if (!q->merge_bvec_fn) | |
3878 | return max_size; | |
3879 | ||
3880 | bvm->bi_bdev = tc->pool_dev->bdev; | |
3881 | bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector); | |
3882 | ||
3883 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3884 | } | |
3885 | ||
991d9fa0 JT |
3886 | static int thin_iterate_devices(struct dm_target *ti, |
3887 | iterate_devices_callout_fn fn, void *data) | |
3888 | { | |
55f2b8bd | 3889 | sector_t blocks; |
991d9fa0 | 3890 | struct thin_c *tc = ti->private; |
55f2b8bd | 3891 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3892 | |
3893 | /* | |
3894 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3895 | * we follow a more convoluted path through to the pool's target. | |
3896 | */ | |
55f2b8bd | 3897 | if (!pool->ti) |
991d9fa0 JT |
3898 | return 0; /* nothing is bound */ |
3899 | ||
55f2b8bd MS |
3900 | blocks = pool->ti->len; |
3901 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3902 | if (blocks) |
55f2b8bd | 3903 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3904 | |
3905 | return 0; | |
3906 | } | |
3907 | ||
991d9fa0 JT |
3908 | static struct target_type thin_target = { |
3909 | .name = "thin", | |
36f12aeb | 3910 | .version = {1, 14, 0}, |
991d9fa0 JT |
3911 | .module = THIS_MODULE, |
3912 | .ctr = thin_ctr, | |
3913 | .dtr = thin_dtr, | |
3914 | .map = thin_map, | |
eb2aa48d | 3915 | .end_io = thin_endio, |
e5aea7b4 | 3916 | .preresume = thin_preresume, |
738211f7 | 3917 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3918 | .postsuspend = thin_postsuspend, |
3919 | .status = thin_status, | |
36f12aeb | 3920 | .merge = thin_merge, |
991d9fa0 | 3921 | .iterate_devices = thin_iterate_devices, |
991d9fa0 JT |
3922 | }; |
3923 | ||
3924 | /*----------------------------------------------------------------*/ | |
3925 | ||
3926 | static int __init dm_thin_init(void) | |
3927 | { | |
3928 | int r; | |
3929 | ||
3930 | pool_table_init(); | |
3931 | ||
3932 | r = dm_register_target(&thin_target); | |
3933 | if (r) | |
3934 | return r; | |
3935 | ||
3936 | r = dm_register_target(&pool_target); | |
3937 | if (r) | |
a24c2569 MS |
3938 | goto bad_pool_target; |
3939 | ||
3940 | r = -ENOMEM; | |
3941 | ||
a24c2569 MS |
3942 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3943 | if (!_new_mapping_cache) | |
3944 | goto bad_new_mapping_cache; | |
3945 | ||
a24c2569 MS |
3946 | return 0; |
3947 | ||
a24c2569 | 3948 | bad_new_mapping_cache: |
a24c2569 MS |
3949 | dm_unregister_target(&pool_target); |
3950 | bad_pool_target: | |
3951 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3952 | |
3953 | return r; | |
3954 | } | |
3955 | ||
3956 | static void dm_thin_exit(void) | |
3957 | { | |
3958 | dm_unregister_target(&thin_target); | |
3959 | dm_unregister_target(&pool_target); | |
a24c2569 | 3960 | |
a24c2569 | 3961 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3962 | } |
3963 | ||
3964 | module_init(dm_thin_init); | |
3965 | module_exit(dm_thin_exit); | |
3966 | ||
80c57893 MS |
3967 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
3968 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
3969 | ||
7cab8bf1 | 3970 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3971 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3972 | MODULE_LICENSE("GPL"); |