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> | |
ac4c3f34 | 20 | #include <linux/sort.h> |
67324ea1 | 21 | #include <linux/rbtree.h> |
991d9fa0 JT |
22 | |
23 | #define DM_MSG_PREFIX "thin" | |
24 | ||
25 | /* | |
26 | * Tunable constants | |
27 | */ | |
7768ed33 | 28 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 29 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 30 | #define COMMIT_PERIOD HZ |
80c57893 MS |
31 | #define NO_SPACE_TIMEOUT_SECS 60 |
32 | ||
33 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 34 | |
df5d2e90 MP |
35 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
36 | "A percentage of time allocated for copy on write"); | |
37 | ||
991d9fa0 JT |
38 | /* |
39 | * The block size of the device holding pool data must be | |
40 | * between 64KB and 1GB. | |
41 | */ | |
42 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
43 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
44 | ||
991d9fa0 JT |
45 | /* |
46 | * Device id is restricted to 24 bits. | |
47 | */ | |
48 | #define MAX_DEV_ID ((1 << 24) - 1) | |
49 | ||
50 | /* | |
51 | * How do we handle breaking sharing of data blocks? | |
52 | * ================================================= | |
53 | * | |
54 | * We use a standard copy-on-write btree to store the mappings for the | |
55 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
56 | * the data). When you take an internal snapshot you clone the root node | |
57 | * of the origin btree. After this there is no concept of an origin or a | |
58 | * snapshot. They are just two device trees that happen to point to the | |
59 | * same data blocks. | |
60 | * | |
61 | * When we get a write in we decide if it's to a shared data block using | |
62 | * some timestamp magic. If it is, we have to break sharing. | |
63 | * | |
64 | * Let's say we write to a shared block in what was the origin. The | |
65 | * steps are: | |
66 | * | |
67 | * i) plug io further to this physical block. (see bio_prison code). | |
68 | * | |
69 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 70 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
71 | * |
72 | * iii) copy the data block to a newly allocate block. This step can be | |
73 | * missed out if the io covers the block. (schedule_copy). | |
74 | * | |
75 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 76 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
77 | * sharing of btree nodes between the two devices. Breaking sharing only |
78 | * effects the btree of that specific device. Btrees for the other | |
79 | * devices that share the block never change. The btree for the origin | |
80 | * device as it was after the last commit is untouched, ie. we're using | |
81 | * persistent data structures in the functional programming sense. | |
82 | * | |
83 | * v) unplug io to this physical block, including the io that triggered | |
84 | * the breaking of sharing. | |
85 | * | |
86 | * Steps (ii) and (iii) occur in parallel. | |
87 | * | |
88 | * The metadata _doesn't_ need to be committed before the io continues. We | |
89 | * get away with this because the io is always written to a _new_ block. | |
90 | * If there's a crash, then: | |
91 | * | |
92 | * - The origin mapping will point to the old origin block (the shared | |
93 | * one). This will contain the data as it was before the io that triggered | |
94 | * the breaking of sharing came in. | |
95 | * | |
96 | * - The snap mapping still points to the old block. As it would after | |
97 | * the commit. | |
98 | * | |
99 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
100 | * will continue to think that data block in the snapshot device is shared | |
101 | * even after the write to the origin has broken sharing. I suspect data | |
102 | * blocks will typically be shared by many different devices, so we're | |
103 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
104 | * devices that reference this data block. At the moment I think the | |
105 | * benefits far, far outweigh the disadvantages. | |
106 | */ | |
107 | ||
108 | /*----------------------------------------------------------------*/ | |
109 | ||
991d9fa0 JT |
110 | /* |
111 | * Key building. | |
112 | */ | |
113 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 114 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
115 | { |
116 | key->virtual = 0; | |
117 | key->dev = dm_thin_dev_id(td); | |
118 | key->block = b; | |
119 | } | |
120 | ||
121 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 122 | struct dm_cell_key *key) |
991d9fa0 JT |
123 | { |
124 | key->virtual = 1; | |
125 | key->dev = dm_thin_dev_id(td); | |
126 | key->block = b; | |
127 | } | |
128 | ||
129 | /*----------------------------------------------------------------*/ | |
130 | ||
7d327fe0 JT |
131 | #define THROTTLE_THRESHOLD (1 * HZ) |
132 | ||
133 | struct throttle { | |
134 | struct rw_semaphore lock; | |
135 | unsigned long threshold; | |
136 | bool throttle_applied; | |
137 | }; | |
138 | ||
139 | static void throttle_init(struct throttle *t) | |
140 | { | |
141 | init_rwsem(&t->lock); | |
142 | t->throttle_applied = false; | |
143 | } | |
144 | ||
145 | static void throttle_work_start(struct throttle *t) | |
146 | { | |
147 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
148 | } | |
149 | ||
150 | static void throttle_work_update(struct throttle *t) | |
151 | { | |
152 | if (!t->throttle_applied && jiffies > t->threshold) { | |
153 | down_write(&t->lock); | |
154 | t->throttle_applied = true; | |
155 | } | |
156 | } | |
157 | ||
158 | static void throttle_work_complete(struct throttle *t) | |
159 | { | |
160 | if (t->throttle_applied) { | |
161 | t->throttle_applied = false; | |
162 | up_write(&t->lock); | |
163 | } | |
164 | } | |
165 | ||
166 | static void throttle_lock(struct throttle *t) | |
167 | { | |
168 | down_read(&t->lock); | |
169 | } | |
170 | ||
171 | static void throttle_unlock(struct throttle *t) | |
172 | { | |
173 | up_read(&t->lock); | |
174 | } | |
175 | ||
176 | /*----------------------------------------------------------------*/ | |
177 | ||
991d9fa0 JT |
178 | /* |
179 | * A pool device ties together a metadata device and a data device. It | |
180 | * also provides the interface for creating and destroying internal | |
181 | * devices. | |
182 | */ | |
a24c2569 | 183 | struct dm_thin_new_mapping; |
67e2e2b2 | 184 | |
e49e5829 | 185 | /* |
3e1a0699 | 186 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
187 | */ |
188 | enum pool_mode { | |
189 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 190 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
191 | PM_READ_ONLY, /* metadata may not be changed */ |
192 | PM_FAIL, /* all I/O fails */ | |
193 | }; | |
194 | ||
67e2e2b2 | 195 | struct pool_features { |
e49e5829 JT |
196 | enum pool_mode mode; |
197 | ||
9bc142dd MS |
198 | bool zero_new_blocks:1; |
199 | bool discard_enabled:1; | |
200 | bool discard_passdown:1; | |
787a996c | 201 | bool error_if_no_space:1; |
67e2e2b2 JT |
202 | }; |
203 | ||
e49e5829 JT |
204 | struct thin_c; |
205 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 206 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
207 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
208 | ||
ac4c3f34 JT |
209 | #define CELL_SORT_ARRAY_SIZE 8192 |
210 | ||
991d9fa0 JT |
211 | struct pool { |
212 | struct list_head list; | |
213 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
214 | ||
215 | struct mapped_device *pool_md; | |
216 | struct block_device *md_dev; | |
217 | struct dm_pool_metadata *pmd; | |
218 | ||
991d9fa0 | 219 | dm_block_t low_water_blocks; |
55f2b8bd | 220 | uint32_t sectors_per_block; |
f9a8e0cd | 221 | int sectors_per_block_shift; |
991d9fa0 | 222 | |
67e2e2b2 | 223 | struct pool_features pf; |
88a6621b | 224 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 225 | |
44feb387 | 226 | struct dm_bio_prison *prison; |
991d9fa0 JT |
227 | struct dm_kcopyd_client *copier; |
228 | ||
229 | struct workqueue_struct *wq; | |
7d327fe0 | 230 | struct throttle throttle; |
991d9fa0 | 231 | struct work_struct worker; |
905e51b3 | 232 | struct delayed_work waker; |
85ad643b | 233 | struct delayed_work no_space_timeout; |
991d9fa0 | 234 | |
905e51b3 | 235 | unsigned long last_commit_jiffies; |
55f2b8bd | 236 | unsigned ref_count; |
991d9fa0 JT |
237 | |
238 | spinlock_t lock; | |
991d9fa0 JT |
239 | struct bio_list deferred_flush_bios; |
240 | struct list_head prepared_mappings; | |
104655fd | 241 | struct list_head prepared_discards; |
c140e1c4 | 242 | struct list_head active_thins; |
991d9fa0 | 243 | |
44feb387 MS |
244 | struct dm_deferred_set *shared_read_ds; |
245 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 246 | |
a24c2569 | 247 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 248 | mempool_t *mapping_pool; |
e49e5829 JT |
249 | |
250 | process_bio_fn process_bio; | |
251 | process_bio_fn process_discard; | |
252 | ||
a374bb21 JT |
253 | process_cell_fn process_cell; |
254 | process_cell_fn process_discard_cell; | |
255 | ||
e49e5829 JT |
256 | process_mapping_fn process_prepared_mapping; |
257 | process_mapping_fn process_prepared_discard; | |
ac4c3f34 JT |
258 | |
259 | struct dm_bio_prison_cell *cell_sort_array[CELL_SORT_ARRAY_SIZE]; | |
991d9fa0 JT |
260 | }; |
261 | ||
e49e5829 | 262 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 263 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 264 | |
991d9fa0 JT |
265 | /* |
266 | * Target context for a pool. | |
267 | */ | |
268 | struct pool_c { | |
269 | struct dm_target *ti; | |
270 | struct pool *pool; | |
271 | struct dm_dev *data_dev; | |
272 | struct dm_dev *metadata_dev; | |
273 | struct dm_target_callbacks callbacks; | |
274 | ||
275 | dm_block_t low_water_blocks; | |
0424caa1 MS |
276 | struct pool_features requested_pf; /* Features requested during table load */ |
277 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
278 | }; |
279 | ||
280 | /* | |
281 | * Target context for a thin. | |
282 | */ | |
283 | struct thin_c { | |
c140e1c4 | 284 | struct list_head list; |
991d9fa0 | 285 | struct dm_dev *pool_dev; |
2dd9c257 | 286 | struct dm_dev *origin_dev; |
e5aea7b4 | 287 | sector_t origin_size; |
991d9fa0 JT |
288 | dm_thin_id dev_id; |
289 | ||
290 | struct pool *pool; | |
291 | struct dm_thin_device *td; | |
738211f7 | 292 | bool requeue_mode:1; |
c140e1c4 | 293 | spinlock_t lock; |
a374bb21 | 294 | struct list_head deferred_cells; |
c140e1c4 MS |
295 | struct bio_list deferred_bio_list; |
296 | struct bio_list retry_on_resume_list; | |
67324ea1 | 297 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
298 | |
299 | /* | |
300 | * Ensures the thin is not destroyed until the worker has finished | |
301 | * iterating the active_thins list. | |
302 | */ | |
303 | atomic_t refcount; | |
304 | struct completion can_destroy; | |
991d9fa0 JT |
305 | }; |
306 | ||
307 | /*----------------------------------------------------------------*/ | |
308 | ||
025b9685 JT |
309 | /* |
310 | * wake_worker() is used when new work is queued and when pool_resume is | |
311 | * ready to continue deferred IO processing. | |
312 | */ | |
313 | static void wake_worker(struct pool *pool) | |
314 | { | |
315 | queue_work(pool->wq, &pool->worker); | |
316 | } | |
317 | ||
318 | /*----------------------------------------------------------------*/ | |
319 | ||
6beca5eb JT |
320 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
321 | struct dm_bio_prison_cell **cell_result) | |
322 | { | |
323 | int r; | |
324 | struct dm_bio_prison_cell *cell_prealloc; | |
325 | ||
326 | /* | |
327 | * Allocate a cell from the prison's mempool. | |
328 | * This might block but it can't fail. | |
329 | */ | |
330 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
331 | ||
332 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
333 | if (r) | |
334 | /* | |
335 | * We reused an old cell; we can get rid of | |
336 | * the new one. | |
337 | */ | |
338 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
339 | ||
340 | return r; | |
341 | } | |
342 | ||
343 | static void cell_release(struct pool *pool, | |
344 | struct dm_bio_prison_cell *cell, | |
345 | struct bio_list *bios) | |
346 | { | |
347 | dm_cell_release(pool->prison, cell, bios); | |
348 | dm_bio_prison_free_cell(pool->prison, cell); | |
349 | } | |
350 | ||
2d759a46 JT |
351 | static void cell_visit_release(struct pool *pool, |
352 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
353 | void *context, | |
354 | struct dm_bio_prison_cell *cell) | |
355 | { | |
356 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
357 | dm_bio_prison_free_cell(pool->prison, cell); | |
358 | } | |
359 | ||
6beca5eb JT |
360 | static void cell_release_no_holder(struct pool *pool, |
361 | struct dm_bio_prison_cell *cell, | |
362 | struct bio_list *bios) | |
363 | { | |
364 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
365 | dm_bio_prison_free_cell(pool->prison, cell); | |
366 | } | |
367 | ||
af91805a MS |
368 | static void cell_error_with_code(struct pool *pool, |
369 | struct dm_bio_prison_cell *cell, int error_code) | |
6beca5eb | 370 | { |
af91805a | 371 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
372 | dm_bio_prison_free_cell(pool->prison, cell); |
373 | } | |
374 | ||
af91805a MS |
375 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
376 | { | |
377 | cell_error_with_code(pool, cell, -EIO); | |
378 | } | |
379 | ||
a374bb21 JT |
380 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
381 | { | |
382 | cell_error_with_code(pool, cell, 0); | |
383 | } | |
384 | ||
385 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
386 | { | |
387 | cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE); | |
388 | } | |
389 | ||
6beca5eb JT |
390 | /*----------------------------------------------------------------*/ |
391 | ||
991d9fa0 JT |
392 | /* |
393 | * A global list of pools that uses a struct mapped_device as a key. | |
394 | */ | |
395 | static struct dm_thin_pool_table { | |
396 | struct mutex mutex; | |
397 | struct list_head pools; | |
398 | } dm_thin_pool_table; | |
399 | ||
400 | static void pool_table_init(void) | |
401 | { | |
402 | mutex_init(&dm_thin_pool_table.mutex); | |
403 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
404 | } | |
405 | ||
406 | static void __pool_table_insert(struct pool *pool) | |
407 | { | |
408 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
409 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
410 | } | |
411 | ||
412 | static void __pool_table_remove(struct pool *pool) | |
413 | { | |
414 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
415 | list_del(&pool->list); | |
416 | } | |
417 | ||
418 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
419 | { | |
420 | struct pool *pool = NULL, *tmp; | |
421 | ||
422 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
423 | ||
424 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
425 | if (tmp->pool_md == md) { | |
426 | pool = tmp; | |
427 | break; | |
428 | } | |
429 | } | |
430 | ||
431 | return pool; | |
432 | } | |
433 | ||
434 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
435 | { | |
436 | struct pool *pool = NULL, *tmp; | |
437 | ||
438 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
439 | ||
440 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
441 | if (tmp->md_dev == md_dev) { | |
442 | pool = tmp; | |
443 | break; | |
444 | } | |
445 | } | |
446 | ||
447 | return pool; | |
448 | } | |
449 | ||
450 | /*----------------------------------------------------------------*/ | |
451 | ||
a24c2569 | 452 | struct dm_thin_endio_hook { |
eb2aa48d | 453 | struct thin_c *tc; |
44feb387 MS |
454 | struct dm_deferred_entry *shared_read_entry; |
455 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 456 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 457 | struct rb_node rb_node; |
eb2aa48d JT |
458 | }; |
459 | ||
42d6a8ce MS |
460 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
461 | { | |
462 | bio_list_merge(bios, master); | |
463 | bio_list_init(master); | |
464 | } | |
465 | ||
466 | static void error_bio_list(struct bio_list *bios, int error) | |
991d9fa0 JT |
467 | { |
468 | struct bio *bio; | |
42d6a8ce MS |
469 | |
470 | while ((bio = bio_list_pop(bios))) | |
471 | bio_endio(bio, error); | |
472 | } | |
473 | ||
474 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error) | |
475 | { | |
991d9fa0 | 476 | struct bio_list bios; |
18adc577 | 477 | unsigned long flags; |
991d9fa0 JT |
478 | |
479 | bio_list_init(&bios); | |
18adc577 | 480 | |
c140e1c4 | 481 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce | 482 | __merge_bio_list(&bios, master); |
c140e1c4 | 483 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 484 | |
42d6a8ce | 485 | error_bio_list(&bios, error); |
991d9fa0 JT |
486 | } |
487 | ||
a374bb21 JT |
488 | static void requeue_deferred_cells(struct thin_c *tc) |
489 | { | |
490 | struct pool *pool = tc->pool; | |
491 | unsigned long flags; | |
492 | struct list_head cells; | |
493 | struct dm_bio_prison_cell *cell, *tmp; | |
494 | ||
495 | INIT_LIST_HEAD(&cells); | |
496 | ||
497 | spin_lock_irqsave(&tc->lock, flags); | |
498 | list_splice_init(&tc->deferred_cells, &cells); | |
499 | spin_unlock_irqrestore(&tc->lock, flags); | |
500 | ||
501 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
502 | cell_requeue(pool, cell); | |
503 | } | |
504 | ||
991d9fa0 JT |
505 | static void requeue_io(struct thin_c *tc) |
506 | { | |
3e1a0699 | 507 | struct bio_list bios; |
42d6a8ce | 508 | unsigned long flags; |
3e1a0699 JT |
509 | |
510 | bio_list_init(&bios); | |
511 | ||
c140e1c4 | 512 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce MS |
513 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
514 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
c140e1c4 | 515 | spin_unlock_irqrestore(&tc->lock, flags); |
3e1a0699 | 516 | |
42d6a8ce MS |
517 | error_bio_list(&bios, DM_ENDIO_REQUEUE); |
518 | requeue_deferred_cells(tc); | |
3e1a0699 JT |
519 | } |
520 | ||
c140e1c4 MS |
521 | static void error_retry_list(struct pool *pool) |
522 | { | |
523 | struct thin_c *tc; | |
524 | ||
525 | rcu_read_lock(); | |
526 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
42d6a8ce | 527 | error_thin_bio_list(tc, &tc->retry_on_resume_list, -EIO); |
c140e1c4 MS |
528 | rcu_read_unlock(); |
529 | } | |
530 | ||
991d9fa0 JT |
531 | /* |
532 | * This section of code contains the logic for processing a thin device's IO. | |
533 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
534 | * but most is exclusively called from the thin target rather than the thin-pool | |
535 | * target. | |
536 | */ | |
537 | ||
58f77a21 MS |
538 | static bool block_size_is_power_of_two(struct pool *pool) |
539 | { | |
540 | return pool->sectors_per_block_shift >= 0; | |
541 | } | |
542 | ||
991d9fa0 JT |
543 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
544 | { | |
58f77a21 | 545 | struct pool *pool = tc->pool; |
4f024f37 | 546 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 547 | |
58f77a21 MS |
548 | if (block_size_is_power_of_two(pool)) |
549 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 550 | else |
58f77a21 | 551 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
552 | |
553 | return block_nr; | |
991d9fa0 JT |
554 | } |
555 | ||
556 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
557 | { | |
558 | struct pool *pool = tc->pool; | |
4f024f37 | 559 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
560 | |
561 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 562 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
563 | bio->bi_iter.bi_sector = |
564 | (block << pool->sectors_per_block_shift) | | |
565 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 566 | else |
4f024f37 | 567 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 568 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
569 | } |
570 | ||
2dd9c257 JT |
571 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
572 | { | |
573 | bio->bi_bdev = tc->origin_dev->bdev; | |
574 | } | |
575 | ||
4afdd680 JT |
576 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
577 | { | |
578 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
579 | dm_thin_changed_this_transaction(tc->td); | |
580 | } | |
581 | ||
e8088073 JT |
582 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
583 | { | |
584 | struct dm_thin_endio_hook *h; | |
585 | ||
586 | if (bio->bi_rw & REQ_DISCARD) | |
587 | return; | |
588 | ||
59c3d2c6 | 589 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
590 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
591 | } | |
592 | ||
2dd9c257 | 593 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
594 | { |
595 | struct pool *pool = tc->pool; | |
596 | unsigned long flags; | |
597 | ||
e49e5829 JT |
598 | if (!bio_triggers_commit(tc, bio)) { |
599 | generic_make_request(bio); | |
600 | return; | |
601 | } | |
602 | ||
991d9fa0 | 603 | /* |
e49e5829 JT |
604 | * Complete bio with an error if earlier I/O caused changes to |
605 | * the metadata that can't be committed e.g, due to I/O errors | |
606 | * on the metadata device. | |
991d9fa0 | 607 | */ |
e49e5829 JT |
608 | if (dm_thin_aborted_changes(tc->td)) { |
609 | bio_io_error(bio); | |
610 | return; | |
611 | } | |
612 | ||
613 | /* | |
614 | * Batch together any bios that trigger commits and then issue a | |
615 | * single commit for them in process_deferred_bios(). | |
616 | */ | |
617 | spin_lock_irqsave(&pool->lock, flags); | |
618 | bio_list_add(&pool->deferred_flush_bios, bio); | |
619 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
620 | } |
621 | ||
2dd9c257 JT |
622 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
623 | { | |
624 | remap_to_origin(tc, bio); | |
625 | issue(tc, bio); | |
626 | } | |
627 | ||
628 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
629 | dm_block_t block) | |
630 | { | |
631 | remap(tc, bio, block); | |
632 | issue(tc, bio); | |
633 | } | |
634 | ||
991d9fa0 JT |
635 | /*----------------------------------------------------------------*/ |
636 | ||
637 | /* | |
638 | * Bio endio functions. | |
639 | */ | |
a24c2569 | 640 | struct dm_thin_new_mapping { |
991d9fa0 JT |
641 | struct list_head list; |
642 | ||
7f214665 MS |
643 | bool pass_discard:1; |
644 | bool definitely_not_shared:1; | |
991d9fa0 | 645 | |
50f3c3ef JT |
646 | /* |
647 | * Track quiescing, copying and zeroing preparation actions. When this | |
648 | * counter hits zero the block is prepared and can be inserted into the | |
649 | * btree. | |
650 | */ | |
651 | atomic_t prepare_actions; | |
652 | ||
7f214665 | 653 | int err; |
991d9fa0 JT |
654 | struct thin_c *tc; |
655 | dm_block_t virt_block; | |
656 | dm_block_t data_block; | |
a24c2569 | 657 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
658 | |
659 | /* | |
660 | * If the bio covers the whole area of a block then we can avoid | |
661 | * zeroing or copying. Instead this bio is hooked. The bio will | |
662 | * still be in the cell, so care has to be taken to avoid issuing | |
663 | * the bio twice. | |
664 | */ | |
665 | struct bio *bio; | |
666 | bio_end_io_t *saved_bi_end_io; | |
667 | }; | |
668 | ||
50f3c3ef | 669 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
670 | { |
671 | struct pool *pool = m->tc->pool; | |
672 | ||
50f3c3ef | 673 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 674 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
675 | wake_worker(pool); |
676 | } | |
677 | } | |
678 | ||
e5aea7b4 | 679 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
680 | { |
681 | unsigned long flags; | |
991d9fa0 JT |
682 | struct pool *pool = m->tc->pool; |
683 | ||
991d9fa0 | 684 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 685 | __complete_mapping_preparation(m); |
991d9fa0 JT |
686 | spin_unlock_irqrestore(&pool->lock, flags); |
687 | } | |
688 | ||
e5aea7b4 JT |
689 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
690 | { | |
691 | struct dm_thin_new_mapping *m = context; | |
692 | ||
693 | m->err = read_err || write_err ? -EIO : 0; | |
694 | complete_mapping_preparation(m); | |
695 | } | |
696 | ||
991d9fa0 JT |
697 | static void overwrite_endio(struct bio *bio, int err) |
698 | { | |
59c3d2c6 | 699 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 700 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
701 | |
702 | m->err = err; | |
e5aea7b4 | 703 | complete_mapping_preparation(m); |
991d9fa0 JT |
704 | } |
705 | ||
991d9fa0 JT |
706 | /*----------------------------------------------------------------*/ |
707 | ||
708 | /* | |
709 | * Workqueue. | |
710 | */ | |
711 | ||
712 | /* | |
713 | * Prepared mapping jobs. | |
714 | */ | |
715 | ||
716 | /* | |
2d759a46 JT |
717 | * This sends the bios in the cell, except the original holder, back |
718 | * to the deferred_bios list. | |
991d9fa0 | 719 | */ |
f286ba0e | 720 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 721 | { |
991d9fa0 JT |
722 | struct pool *pool = tc->pool; |
723 | unsigned long flags; | |
724 | ||
c140e1c4 MS |
725 | spin_lock_irqsave(&tc->lock, flags); |
726 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
727 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
728 | |
729 | wake_worker(pool); | |
730 | } | |
731 | ||
a374bb21 JT |
732 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
733 | ||
2d759a46 JT |
734 | struct remap_info { |
735 | struct thin_c *tc; | |
736 | struct bio_list defer_bios; | |
737 | struct bio_list issue_bios; | |
738 | }; | |
739 | ||
740 | static void __inc_remap_and_issue_cell(void *context, | |
741 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 742 | { |
2d759a46 | 743 | struct remap_info *info = context; |
a374bb21 | 744 | struct bio *bio; |
a374bb21 | 745 | |
2d759a46 | 746 | while ((bio = bio_list_pop(&cell->bios))) { |
a374bb21 | 747 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) |
2d759a46 | 748 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 749 | else { |
2d759a46 JT |
750 | inc_all_io_entry(info->tc->pool, bio); |
751 | ||
752 | /* | |
753 | * We can't issue the bios with the bio prison lock | |
754 | * held, so we add them to a list to issue on | |
755 | * return from this function. | |
756 | */ | |
757 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
758 | } |
759 | } | |
760 | } | |
761 | ||
2d759a46 JT |
762 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
763 | struct dm_bio_prison_cell *cell, | |
764 | dm_block_t block) | |
765 | { | |
766 | struct bio *bio; | |
767 | struct remap_info info; | |
768 | ||
769 | info.tc = tc; | |
770 | bio_list_init(&info.defer_bios); | |
771 | bio_list_init(&info.issue_bios); | |
772 | ||
773 | /* | |
774 | * We have to be careful to inc any bios we're about to issue | |
775 | * before the cell is released, and avoid a race with new bios | |
776 | * being added to the cell. | |
777 | */ | |
778 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
779 | &info, cell); | |
780 | ||
781 | while ((bio = bio_list_pop(&info.defer_bios))) | |
782 | thin_defer_bio(tc, bio); | |
783 | ||
784 | while ((bio = bio_list_pop(&info.issue_bios))) | |
785 | remap_and_issue(info.tc, bio, block); | |
786 | } | |
787 | ||
e49e5829 JT |
788 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
789 | { | |
196d38bc | 790 | if (m->bio) { |
e49e5829 | 791 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
792 | atomic_inc(&m->bio->bi_remaining); |
793 | } | |
6beca5eb | 794 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
795 | list_del(&m->list); |
796 | mempool_free(m, m->tc->pool->mapping_pool); | |
797 | } | |
025b9685 | 798 | |
a24c2569 | 799 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
800 | { |
801 | struct thin_c *tc = m->tc; | |
6beca5eb | 802 | struct pool *pool = tc->pool; |
991d9fa0 JT |
803 | struct bio *bio; |
804 | int r; | |
805 | ||
806 | bio = m->bio; | |
196d38bc | 807 | if (bio) { |
991d9fa0 | 808 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
809 | atomic_inc(&bio->bi_remaining); |
810 | } | |
991d9fa0 JT |
811 | |
812 | if (m->err) { | |
6beca5eb | 813 | cell_error(pool, m->cell); |
905386f8 | 814 | goto out; |
991d9fa0 JT |
815 | } |
816 | ||
817 | /* | |
818 | * Commit the prepared block into the mapping btree. | |
819 | * Any I/O for this block arriving after this point will get | |
820 | * remapped to it directly. | |
821 | */ | |
822 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
823 | if (r) { | |
b5330655 | 824 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 825 | cell_error(pool, m->cell); |
905386f8 | 826 | goto out; |
991d9fa0 JT |
827 | } |
828 | ||
829 | /* | |
830 | * Release any bios held while the block was being provisioned. | |
831 | * If we are processing a write bio that completely covers the block, | |
832 | * we already processed it so can ignore it now when processing | |
833 | * the bios in the cell. | |
834 | */ | |
835 | if (bio) { | |
2d759a46 | 836 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
991d9fa0 | 837 | bio_endio(bio, 0); |
2d759a46 JT |
838 | } else { |
839 | inc_all_io_entry(tc->pool, m->cell->holder); | |
840 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
841 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
842 | } | |
991d9fa0 | 843 | |
905386f8 | 844 | out: |
991d9fa0 | 845 | list_del(&m->list); |
6beca5eb | 846 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
847 | } |
848 | ||
e49e5829 | 849 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 850 | { |
104655fd JT |
851 | struct thin_c *tc = m->tc; |
852 | ||
e49e5829 | 853 | bio_io_error(m->bio); |
f286ba0e JT |
854 | cell_defer_no_holder(tc, m->cell); |
855 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
856 | mempool_free(m, tc->pool->mapping_pool); |
857 | } | |
858 | ||
859 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
860 | { | |
861 | struct thin_c *tc = m->tc; | |
104655fd | 862 | |
e8088073 | 863 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
864 | cell_defer_no_holder(tc, m->cell); |
865 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 866 | |
104655fd | 867 | if (m->pass_discard) |
19fa1a67 JT |
868 | if (m->definitely_not_shared) |
869 | remap_and_issue(tc, m->bio, m->data_block); | |
870 | else { | |
871 | bool used = false; | |
872 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
873 | bio_endio(m->bio, 0); | |
874 | else | |
875 | remap_and_issue(tc, m->bio, m->data_block); | |
876 | } | |
104655fd JT |
877 | else |
878 | bio_endio(m->bio, 0); | |
879 | ||
104655fd JT |
880 | mempool_free(m, tc->pool->mapping_pool); |
881 | } | |
882 | ||
e49e5829 JT |
883 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
884 | { | |
885 | int r; | |
886 | struct thin_c *tc = m->tc; | |
887 | ||
888 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
889 | if (r) | |
c397741c | 890 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
891 | |
892 | process_prepared_discard_passdown(m); | |
893 | } | |
894 | ||
104655fd | 895 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 896 | process_mapping_fn *fn) |
991d9fa0 JT |
897 | { |
898 | unsigned long flags; | |
899 | struct list_head maps; | |
a24c2569 | 900 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
901 | |
902 | INIT_LIST_HEAD(&maps); | |
903 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 904 | list_splice_init(head, &maps); |
991d9fa0 JT |
905 | spin_unlock_irqrestore(&pool->lock, flags); |
906 | ||
907 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 908 | (*fn)(m); |
991d9fa0 JT |
909 | } |
910 | ||
911 | /* | |
912 | * Deferred bio jobs. | |
913 | */ | |
104655fd | 914 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 915 | { |
4f024f37 KO |
916 | return bio->bi_iter.bi_size == |
917 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
918 | } |
919 | ||
920 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
921 | { | |
922 | return (bio_data_dir(bio) == WRITE) && | |
923 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
924 | } |
925 | ||
926 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
927 | bio_end_io_t *fn) | |
928 | { | |
929 | *save = bio->bi_end_io; | |
930 | bio->bi_end_io = fn; | |
931 | } | |
932 | ||
933 | static int ensure_next_mapping(struct pool *pool) | |
934 | { | |
935 | if (pool->next_mapping) | |
936 | return 0; | |
937 | ||
938 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
939 | ||
940 | return pool->next_mapping ? 0 : -ENOMEM; | |
941 | } | |
942 | ||
a24c2569 | 943 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 944 | { |
16961b04 | 945 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
946 | |
947 | BUG_ON(!pool->next_mapping); | |
948 | ||
16961b04 MS |
949 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
950 | INIT_LIST_HEAD(&m->list); | |
951 | m->bio = NULL; | |
952 | ||
991d9fa0 JT |
953 | pool->next_mapping = NULL; |
954 | ||
16961b04 | 955 | return m; |
991d9fa0 JT |
956 | } |
957 | ||
e5aea7b4 JT |
958 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
959 | sector_t begin, sector_t end) | |
960 | { | |
961 | int r; | |
962 | struct dm_io_region to; | |
963 | ||
964 | to.bdev = tc->pool_dev->bdev; | |
965 | to.sector = begin; | |
966 | to.count = end - begin; | |
967 | ||
968 | r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); | |
969 | if (r < 0) { | |
970 | DMERR_LIMIT("dm_kcopyd_zero() failed"); | |
971 | copy_complete(1, 1, m); | |
972 | } | |
973 | } | |
974 | ||
452d7a62 MS |
975 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
976 | dm_block_t data_block, | |
977 | struct dm_thin_new_mapping *m) | |
978 | { | |
979 | struct pool *pool = tc->pool; | |
980 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
981 | ||
982 | h->overwrite_mapping = m; | |
983 | m->bio = bio; | |
984 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
985 | inc_all_io_entry(pool, bio); | |
986 | remap_and_issue(tc, bio, data_block); | |
987 | } | |
988 | ||
e5aea7b4 JT |
989 | /* |
990 | * A partial copy also needs to zero the uncopied region. | |
991 | */ | |
991d9fa0 | 992 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
993 | struct dm_dev *origin, dm_block_t data_origin, |
994 | dm_block_t data_dest, | |
e5aea7b4 JT |
995 | struct dm_bio_prison_cell *cell, struct bio *bio, |
996 | sector_t len) | |
991d9fa0 JT |
997 | { |
998 | int r; | |
999 | struct pool *pool = tc->pool; | |
a24c2569 | 1000 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1001 | |
991d9fa0 JT |
1002 | m->tc = tc; |
1003 | m->virt_block = virt_block; | |
1004 | m->data_block = data_dest; | |
1005 | m->cell = cell; | |
991d9fa0 | 1006 | |
e5aea7b4 JT |
1007 | /* |
1008 | * quiesce action + copy action + an extra reference held for the | |
1009 | * duration of this function (we may need to inc later for a | |
1010 | * partial zero). | |
1011 | */ | |
1012 | atomic_set(&m->prepare_actions, 3); | |
1013 | ||
44feb387 | 1014 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1015 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1016 | |
1017 | /* | |
1018 | * IO to pool_dev remaps to the pool target's data_dev. | |
1019 | * | |
1020 | * If the whole block of data is being overwritten, we can issue the | |
1021 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1022 | */ | |
452d7a62 MS |
1023 | if (io_overwrites_block(pool, bio)) |
1024 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1025 | else { | |
991d9fa0 JT |
1026 | struct dm_io_region from, to; |
1027 | ||
2dd9c257 | 1028 | from.bdev = origin->bdev; |
991d9fa0 | 1029 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1030 | from.count = len; |
991d9fa0 JT |
1031 | |
1032 | to.bdev = tc->pool_dev->bdev; | |
1033 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1034 | to.count = len; |
991d9fa0 JT |
1035 | |
1036 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1037 | 0, copy_complete, m); | |
1038 | if (r < 0) { | |
c397741c | 1039 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
e5aea7b4 JT |
1040 | copy_complete(1, 1, m); |
1041 | ||
1042 | /* | |
1043 | * We allow the zero to be issued, to simplify the | |
1044 | * error path. Otherwise we'd need to start | |
1045 | * worrying about decrementing the prepare_actions | |
1046 | * counter. | |
1047 | */ | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Do we need to zero a tail region? | |
1052 | */ | |
1053 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1054 | atomic_inc(&m->prepare_actions); | |
1055 | ll_zero(tc, m, | |
1056 | data_dest * pool->sectors_per_block + len, | |
1057 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1058 | } |
1059 | } | |
e5aea7b4 JT |
1060 | |
1061 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1062 | } |
1063 | ||
2dd9c257 JT |
1064 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1065 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1066 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1067 | { |
1068 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1069 | data_origin, data_dest, cell, bio, |
1070 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1071 | } |
1072 | ||
991d9fa0 | 1073 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1074 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1075 | struct bio *bio) |
1076 | { | |
1077 | struct pool *pool = tc->pool; | |
a24c2569 | 1078 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1079 | |
50f3c3ef | 1080 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 JT |
1081 | m->tc = tc; |
1082 | m->virt_block = virt_block; | |
1083 | m->data_block = data_block; | |
1084 | m->cell = cell; | |
991d9fa0 JT |
1085 | |
1086 | /* | |
1087 | * If the whole block of data is being overwritten or we are not | |
1088 | * zeroing pre-existing data, we can issue the bio immediately. | |
1089 | * Otherwise we use kcopyd to zero the data first. | |
1090 | */ | |
67e2e2b2 | 1091 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
1092 | process_prepared_mapping(m); |
1093 | ||
452d7a62 MS |
1094 | else if (io_overwrites_block(pool, bio)) |
1095 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
991d9fa0 | 1096 | |
452d7a62 | 1097 | else |
e5aea7b4 JT |
1098 | ll_zero(tc, m, |
1099 | data_block * pool->sectors_per_block, | |
1100 | (data_block + 1) * pool->sectors_per_block); | |
1101 | } | |
991d9fa0 | 1102 | |
e5aea7b4 JT |
1103 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1104 | dm_block_t data_dest, | |
1105 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1106 | { | |
1107 | struct pool *pool = tc->pool; | |
1108 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1109 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1110 | ||
1111 | if (virt_block_end <= tc->origin_size) | |
1112 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1113 | virt_block, data_dest, cell, bio, | |
1114 | pool->sectors_per_block); | |
1115 | ||
1116 | else if (virt_block_begin < tc->origin_size) | |
1117 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1118 | virt_block, data_dest, cell, bio, | |
1119 | tc->origin_size - virt_block_begin); | |
1120 | ||
1121 | else | |
1122 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1123 | } |
1124 | ||
e49e5829 JT |
1125 | /* |
1126 | * A non-zero return indicates read_only or fail_io mode. | |
1127 | * Many callers don't care about the return value. | |
1128 | */ | |
020cc3b5 | 1129 | static int commit(struct pool *pool) |
e49e5829 JT |
1130 | { |
1131 | int r; | |
1132 | ||
8d07e8a5 | 1133 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
1134 | return -EINVAL; |
1135 | ||
020cc3b5 | 1136 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1137 | if (r) |
1138 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
1139 | |
1140 | return r; | |
1141 | } | |
1142 | ||
88a6621b JT |
1143 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1144 | { | |
1145 | unsigned long flags; | |
1146 | ||
1147 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1148 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1149 | dm_device_name(pool->pool_md)); | |
1150 | spin_lock_irqsave(&pool->lock, flags); | |
1151 | pool->low_water_triggered = true; | |
1152 | spin_unlock_irqrestore(&pool->lock, flags); | |
1153 | dm_table_event(pool->ti->table); | |
1154 | } | |
1155 | } | |
1156 | ||
3e1a0699 JT |
1157 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1158 | ||
991d9fa0 JT |
1159 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1160 | { | |
1161 | int r; | |
1162 | dm_block_t free_blocks; | |
991d9fa0 JT |
1163 | struct pool *pool = tc->pool; |
1164 | ||
3e1a0699 | 1165 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1166 | return -EINVAL; |
1167 | ||
991d9fa0 | 1168 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1169 | if (r) { |
1170 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1171 | return r; |
b5330655 | 1172 | } |
991d9fa0 | 1173 | |
88a6621b | 1174 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1175 | |
1176 | if (!free_blocks) { | |
94563bad MS |
1177 | /* |
1178 | * Try to commit to see if that will free up some | |
1179 | * more space. | |
1180 | */ | |
020cc3b5 JT |
1181 | r = commit(pool); |
1182 | if (r) | |
1183 | return r; | |
991d9fa0 | 1184 | |
94563bad | 1185 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1186 | if (r) { |
1187 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1188 | return r; |
b5330655 | 1189 | } |
991d9fa0 | 1190 | |
94563bad | 1191 | if (!free_blocks) { |
3e1a0699 | 1192 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1193 | return -ENOSPC; |
991d9fa0 JT |
1194 | } |
1195 | } | |
1196 | ||
1197 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1198 | if (r) { |
b5330655 | 1199 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1200 | return r; |
4a02b34e | 1201 | } |
991d9fa0 JT |
1202 | |
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * If we have run out of space, queue bios until the device is | |
1208 | * resumed, presumably after having been reloaded with more space. | |
1209 | */ | |
1210 | static void retry_on_resume(struct bio *bio) | |
1211 | { | |
59c3d2c6 | 1212 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1213 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1214 | unsigned long flags; |
1215 | ||
c140e1c4 MS |
1216 | spin_lock_irqsave(&tc->lock, flags); |
1217 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1218 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1219 | } |
1220 | ||
af91805a | 1221 | static int should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1222 | { |
3e1a0699 JT |
1223 | enum pool_mode m = get_pool_mode(pool); |
1224 | ||
1225 | switch (m) { | |
1226 | case PM_WRITE: | |
1227 | /* Shouldn't get here */ | |
1228 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
af91805a | 1229 | return -EIO; |
3e1a0699 JT |
1230 | |
1231 | case PM_OUT_OF_DATA_SPACE: | |
af91805a | 1232 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
3e1a0699 JT |
1233 | |
1234 | case PM_READ_ONLY: | |
1235 | case PM_FAIL: | |
af91805a | 1236 | return -EIO; |
3e1a0699 JT |
1237 | default: |
1238 | /* Shouldn't get here */ | |
1239 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
af91805a | 1240 | return -EIO; |
3e1a0699 JT |
1241 | } |
1242 | } | |
8c0f0e8c | 1243 | |
3e1a0699 JT |
1244 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1245 | { | |
af91805a MS |
1246 | int error = should_error_unserviceable_bio(pool); |
1247 | ||
1248 | if (error) | |
1249 | bio_endio(bio, error); | |
6d16202b MS |
1250 | else |
1251 | retry_on_resume(bio); | |
8c0f0e8c MS |
1252 | } |
1253 | ||
399caddf | 1254 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1255 | { |
1256 | struct bio *bio; | |
1257 | struct bio_list bios; | |
af91805a | 1258 | int error; |
991d9fa0 | 1259 | |
af91805a MS |
1260 | error = should_error_unserviceable_bio(pool); |
1261 | if (error) { | |
1262 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1263 | return; |
1264 | } | |
1265 | ||
991d9fa0 | 1266 | bio_list_init(&bios); |
6beca5eb | 1267 | cell_release(pool, cell, &bios); |
991d9fa0 | 1268 | |
9d094eeb MS |
1269 | while ((bio = bio_list_pop(&bios))) |
1270 | retry_on_resume(bio); | |
991d9fa0 JT |
1271 | } |
1272 | ||
a374bb21 | 1273 | static void process_discard_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
104655fd JT |
1274 | { |
1275 | int r; | |
a374bb21 | 1276 | struct bio *bio = cell->holder; |
104655fd | 1277 | struct pool *pool = tc->pool; |
a374bb21 JT |
1278 | struct dm_bio_prison_cell *cell2; |
1279 | struct dm_cell_key key2; | |
104655fd JT |
1280 | dm_block_t block = get_bio_block(tc, bio); |
1281 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1282 | struct dm_thin_new_mapping *m; |
104655fd | 1283 | |
a374bb21 JT |
1284 | if (tc->requeue_mode) { |
1285 | cell_requeue(pool, cell); | |
104655fd | 1286 | return; |
a374bb21 | 1287 | } |
104655fd JT |
1288 | |
1289 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1290 | switch (r) { | |
1291 | case 0: | |
1292 | /* | |
1293 | * Check nobody is fiddling with this pool block. This can | |
1294 | * happen if someone's in the process of breaking sharing | |
1295 | * on this block. | |
1296 | */ | |
1297 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1298 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1299 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1300 | break; |
1301 | } | |
1302 | ||
1303 | if (io_overlaps_block(pool, bio)) { | |
1304 | /* | |
1305 | * IO may still be going to the destination block. We must | |
1306 | * quiesce before we can do the removal. | |
1307 | */ | |
1308 | m = get_next_mapping(pool); | |
1309 | m->tc = tc; | |
19fa1a67 JT |
1310 | m->pass_discard = pool->pf.discard_passdown; |
1311 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1312 | m->virt_block = block; |
1313 | m->data_block = lookup_result.block; | |
1314 | m->cell = cell; | |
1315 | m->cell2 = cell2; | |
104655fd JT |
1316 | m->bio = bio; |
1317 | ||
7a7e97ca JT |
1318 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1319 | pool->process_prepared_discard(m); | |
1320 | ||
104655fd | 1321 | } else { |
e8088073 | 1322 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1323 | cell_defer_no_holder(tc, cell); |
1324 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1325 | |
104655fd | 1326 | /* |
49296309 MP |
1327 | * The DM core makes sure that the discard doesn't span |
1328 | * a block boundary. So we submit the discard of a | |
1329 | * partial block appropriately. | |
104655fd | 1330 | */ |
650d2a06 MP |
1331 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1332 | remap_and_issue(tc, bio, lookup_result.block); | |
1333 | else | |
1334 | bio_endio(bio, 0); | |
104655fd JT |
1335 | } |
1336 | break; | |
1337 | ||
1338 | case -ENODATA: | |
1339 | /* | |
1340 | * It isn't provisioned, just forget it. | |
1341 | */ | |
f286ba0e | 1342 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1343 | bio_endio(bio, 0); |
1344 | break; | |
1345 | ||
1346 | default: | |
c397741c MS |
1347 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1348 | __func__, r); | |
f286ba0e | 1349 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1350 | bio_io_error(bio); |
1351 | break; | |
1352 | } | |
1353 | } | |
1354 | ||
a374bb21 JT |
1355 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1356 | { | |
1357 | struct dm_bio_prison_cell *cell; | |
1358 | struct dm_cell_key key; | |
1359 | dm_block_t block = get_bio_block(tc, bio); | |
1360 | ||
1361 | build_virtual_key(tc->td, block, &key); | |
1362 | if (bio_detain(tc->pool, &key, bio, &cell)) | |
1363 | return; | |
1364 | ||
1365 | process_discard_cell(tc, cell); | |
1366 | } | |
1367 | ||
991d9fa0 | 1368 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1369 | struct dm_cell_key *key, |
991d9fa0 | 1370 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1371 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1372 | { |
1373 | int r; | |
1374 | dm_block_t data_block; | |
d6fc2042 | 1375 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1376 | |
1377 | r = alloc_data_block(tc, &data_block); | |
1378 | switch (r) { | |
1379 | case 0: | |
2dd9c257 JT |
1380 | schedule_internal_copy(tc, block, lookup_result->block, |
1381 | data_block, cell, bio); | |
991d9fa0 JT |
1382 | break; |
1383 | ||
1384 | case -ENOSPC: | |
399caddf | 1385 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1386 | break; |
1387 | ||
1388 | default: | |
c397741c MS |
1389 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1390 | __func__, r); | |
d6fc2042 | 1391 | cell_error(pool, cell); |
991d9fa0 JT |
1392 | break; |
1393 | } | |
1394 | } | |
1395 | ||
23ca2bb6 JT |
1396 | static void __remap_and_issue_shared_cell(void *context, |
1397 | struct dm_bio_prison_cell *cell) | |
1398 | { | |
1399 | struct remap_info *info = context; | |
1400 | struct bio *bio; | |
1401 | ||
1402 | while ((bio = bio_list_pop(&cell->bios))) { | |
1403 | if ((bio_data_dir(bio) == WRITE) || | |
1404 | (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))) | |
1405 | bio_list_add(&info->defer_bios, bio); | |
1406 | else { | |
1407 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));; | |
1408 | ||
1409 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1410 | inc_all_io_entry(info->tc->pool, bio); | |
1411 | bio_list_add(&info->issue_bios, bio); | |
1412 | } | |
1413 | } | |
1414 | } | |
1415 | ||
1416 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1417 | struct dm_bio_prison_cell *cell, | |
1418 | dm_block_t block) | |
1419 | { | |
1420 | struct bio *bio; | |
1421 | struct remap_info info; | |
1422 | ||
1423 | info.tc = tc; | |
1424 | bio_list_init(&info.defer_bios); | |
1425 | bio_list_init(&info.issue_bios); | |
1426 | ||
1427 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1428 | &info, cell); | |
1429 | ||
1430 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1431 | thin_defer_bio(tc, bio); | |
1432 | ||
1433 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1434 | remap_and_issue(tc, bio, block); | |
1435 | } | |
1436 | ||
991d9fa0 JT |
1437 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1438 | dm_block_t block, | |
23ca2bb6 JT |
1439 | struct dm_thin_lookup_result *lookup_result, |
1440 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1441 | { |
23ca2bb6 | 1442 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1443 | struct pool *pool = tc->pool; |
44feb387 | 1444 | struct dm_cell_key key; |
991d9fa0 JT |
1445 | |
1446 | /* | |
1447 | * If cell is already occupied, then sharing is already in the process | |
1448 | * of being broken so we have nothing further to do here. | |
1449 | */ | |
1450 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1451 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1452 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1453 | return; |
23ca2bb6 | 1454 | } |
991d9fa0 | 1455 | |
23ca2bb6 JT |
1456 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1457 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1458 | cell_defer_no_holder(tc, virt_cell); | |
1459 | } else { | |
59c3d2c6 | 1460 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1461 | |
44feb387 | 1462 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1463 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1464 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1465 | |
1466 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1467 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1468 | } |
1469 | } | |
1470 | ||
1471 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1472 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1473 | { |
1474 | int r; | |
1475 | dm_block_t data_block; | |
6beca5eb | 1476 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1477 | |
1478 | /* | |
1479 | * Remap empty bios (flushes) immediately, without provisioning. | |
1480 | */ | |
4f024f37 | 1481 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1482 | inc_all_io_entry(pool, bio); |
f286ba0e | 1483 | cell_defer_no_holder(tc, cell); |
e8088073 | 1484 | |
991d9fa0 JT |
1485 | remap_and_issue(tc, bio, 0); |
1486 | return; | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Fill read bios with zeroes and complete them immediately. | |
1491 | */ | |
1492 | if (bio_data_dir(bio) == READ) { | |
1493 | zero_fill_bio(bio); | |
f286ba0e | 1494 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1495 | bio_endio(bio, 0); |
1496 | return; | |
1497 | } | |
1498 | ||
1499 | r = alloc_data_block(tc, &data_block); | |
1500 | switch (r) { | |
1501 | case 0: | |
2dd9c257 JT |
1502 | if (tc->origin_dev) |
1503 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1504 | else | |
1505 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1506 | break; |
1507 | ||
1508 | case -ENOSPC: | |
399caddf | 1509 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1510 | break; |
1511 | ||
1512 | default: | |
c397741c MS |
1513 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1514 | __func__, r); | |
6beca5eb | 1515 | cell_error(pool, cell); |
991d9fa0 JT |
1516 | break; |
1517 | } | |
1518 | } | |
1519 | ||
a374bb21 | 1520 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1521 | { |
1522 | int r; | |
6beca5eb | 1523 | struct pool *pool = tc->pool; |
a374bb21 | 1524 | struct bio *bio = cell->holder; |
991d9fa0 | 1525 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1526 | struct dm_thin_lookup_result lookup_result; |
1527 | ||
a374bb21 JT |
1528 | if (tc->requeue_mode) { |
1529 | cell_requeue(pool, cell); | |
991d9fa0 | 1530 | return; |
a374bb21 | 1531 | } |
991d9fa0 JT |
1532 | |
1533 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1534 | switch (r) { | |
1535 | case 0: | |
23ca2bb6 JT |
1536 | if (lookup_result.shared) |
1537 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1538 | else { | |
6beca5eb | 1539 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1540 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1541 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1542 | } |
991d9fa0 JT |
1543 | break; |
1544 | ||
1545 | case -ENODATA: | |
2dd9c257 | 1546 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1547 | inc_all_io_entry(pool, bio); |
f286ba0e | 1548 | cell_defer_no_holder(tc, cell); |
e8088073 | 1549 | |
e5aea7b4 JT |
1550 | if (bio_end_sector(bio) <= tc->origin_size) |
1551 | remap_to_origin_and_issue(tc, bio); | |
1552 | ||
1553 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1554 | zero_fill_bio(bio); | |
1555 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1556 | remap_to_origin_and_issue(tc, bio); | |
1557 | ||
1558 | } else { | |
1559 | zero_fill_bio(bio); | |
1560 | bio_endio(bio, 0); | |
1561 | } | |
2dd9c257 JT |
1562 | } else |
1563 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1564 | break; |
1565 | ||
1566 | default: | |
c397741c MS |
1567 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1568 | __func__, r); | |
f286ba0e | 1569 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1570 | bio_io_error(bio); |
1571 | break; | |
1572 | } | |
1573 | } | |
1574 | ||
a374bb21 JT |
1575 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1576 | { | |
1577 | struct pool *pool = tc->pool; | |
1578 | dm_block_t block = get_bio_block(tc, bio); | |
1579 | struct dm_bio_prison_cell *cell; | |
1580 | struct dm_cell_key key; | |
1581 | ||
1582 | /* | |
1583 | * If cell is already occupied, then the block is already | |
1584 | * being provisioned so we have nothing further to do here. | |
1585 | */ | |
1586 | build_virtual_key(tc->td, block, &key); | |
1587 | if (bio_detain(pool, &key, bio, &cell)) | |
1588 | return; | |
1589 | ||
1590 | process_cell(tc, cell); | |
1591 | } | |
1592 | ||
1593 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
1594 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
1595 | { |
1596 | int r; | |
1597 | int rw = bio_data_dir(bio); | |
1598 | dm_block_t block = get_bio_block(tc, bio); | |
1599 | struct dm_thin_lookup_result lookup_result; | |
1600 | ||
1601 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1602 | switch (r) { | |
1603 | case 0: | |
a374bb21 | 1604 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 1605 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
1606 | if (cell) |
1607 | cell_defer_no_holder(tc, cell); | |
1608 | } else { | |
e8088073 | 1609 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 1610 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
1611 | if (cell) |
1612 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 1613 | } |
e49e5829 JT |
1614 | break; |
1615 | ||
1616 | case -ENODATA: | |
a374bb21 JT |
1617 | if (cell) |
1618 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 1619 | if (rw != READ) { |
8c0f0e8c | 1620 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1621 | break; |
1622 | } | |
1623 | ||
1624 | if (tc->origin_dev) { | |
e8088073 | 1625 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1626 | remap_to_origin_and_issue(tc, bio); |
1627 | break; | |
1628 | } | |
1629 | ||
1630 | zero_fill_bio(bio); | |
1631 | bio_endio(bio, 0); | |
1632 | break; | |
1633 | ||
1634 | default: | |
c397741c MS |
1635 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1636 | __func__, r); | |
a374bb21 JT |
1637 | if (cell) |
1638 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
1639 | bio_io_error(bio); |
1640 | break; | |
1641 | } | |
1642 | } | |
1643 | ||
a374bb21 JT |
1644 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1645 | { | |
1646 | __process_bio_read_only(tc, bio, NULL); | |
1647 | } | |
1648 | ||
1649 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1650 | { | |
1651 | __process_bio_read_only(tc, cell->holder, cell); | |
1652 | } | |
1653 | ||
3e1a0699 JT |
1654 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1655 | { | |
1656 | bio_endio(bio, 0); | |
1657 | } | |
1658 | ||
e49e5829 JT |
1659 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1660 | { | |
1661 | bio_io_error(bio); | |
1662 | } | |
1663 | ||
a374bb21 JT |
1664 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
1665 | { | |
1666 | cell_success(tc->pool, cell); | |
1667 | } | |
1668 | ||
1669 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1670 | { | |
1671 | cell_error(tc->pool, cell); | |
1672 | } | |
1673 | ||
ac8c3f3d JT |
1674 | /* |
1675 | * FIXME: should we also commit due to size of transaction, measured in | |
1676 | * metadata blocks? | |
1677 | */ | |
905e51b3 JT |
1678 | static int need_commit_due_to_time(struct pool *pool) |
1679 | { | |
1680 | return jiffies < pool->last_commit_jiffies || | |
1681 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1682 | } | |
1683 | ||
67324ea1 MS |
1684 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1685 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1686 | ||
1687 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1688 | { | |
1689 | struct rb_node **rbp, *parent; | |
1690 | struct dm_thin_endio_hook *pbd; | |
1691 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1692 | ||
1693 | rbp = &tc->sort_bio_list.rb_node; | |
1694 | parent = NULL; | |
1695 | while (*rbp) { | |
1696 | parent = *rbp; | |
1697 | pbd = thin_pbd(parent); | |
1698 | ||
1699 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1700 | rbp = &(*rbp)->rb_left; | |
1701 | else | |
1702 | rbp = &(*rbp)->rb_right; | |
1703 | } | |
1704 | ||
1705 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1706 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1707 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1708 | } | |
1709 | ||
1710 | static void __extract_sorted_bios(struct thin_c *tc) | |
1711 | { | |
1712 | struct rb_node *node; | |
1713 | struct dm_thin_endio_hook *pbd; | |
1714 | struct bio *bio; | |
1715 | ||
1716 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1717 | pbd = thin_pbd(node); | |
1718 | bio = thin_bio(pbd); | |
1719 | ||
1720 | bio_list_add(&tc->deferred_bio_list, bio); | |
1721 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1722 | } | |
1723 | ||
1724 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1725 | } | |
1726 | ||
1727 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1728 | { | |
1729 | struct bio *bio; | |
1730 | struct bio_list bios; | |
1731 | ||
1732 | bio_list_init(&bios); | |
1733 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1734 | bio_list_init(&tc->deferred_bio_list); | |
1735 | ||
1736 | /* Sort deferred_bio_list using rb-tree */ | |
1737 | while ((bio = bio_list_pop(&bios))) | |
1738 | __thin_bio_rb_add(tc, bio); | |
1739 | ||
1740 | /* | |
1741 | * Transfer the sorted bios in sort_bio_list back to | |
1742 | * deferred_bio_list to allow lockless submission of | |
1743 | * all bios. | |
1744 | */ | |
1745 | __extract_sorted_bios(tc); | |
1746 | } | |
1747 | ||
c140e1c4 | 1748 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1749 | { |
c140e1c4 | 1750 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1751 | unsigned long flags; |
1752 | struct bio *bio; | |
1753 | struct bio_list bios; | |
67324ea1 | 1754 | struct blk_plug plug; |
8a01a6af | 1755 | unsigned count = 0; |
991d9fa0 | 1756 | |
c140e1c4 | 1757 | if (tc->requeue_mode) { |
42d6a8ce | 1758 | error_thin_bio_list(tc, &tc->deferred_bio_list, DM_ENDIO_REQUEUE); |
c140e1c4 MS |
1759 | return; |
1760 | } | |
1761 | ||
991d9fa0 JT |
1762 | bio_list_init(&bios); |
1763 | ||
c140e1c4 | 1764 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1765 | |
1766 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1767 | spin_unlock_irqrestore(&tc->lock, flags); | |
1768 | return; | |
1769 | } | |
1770 | ||
1771 | __sort_thin_deferred_bios(tc); | |
1772 | ||
c140e1c4 MS |
1773 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1774 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1775 | |
c140e1c4 | 1776 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1777 | |
67324ea1 | 1778 | blk_start_plug(&plug); |
991d9fa0 | 1779 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1780 | /* |
1781 | * If we've got no free new_mapping structs, and processing | |
1782 | * this bio might require one, we pause until there are some | |
1783 | * prepared mappings to process. | |
1784 | */ | |
1785 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1786 | spin_lock_irqsave(&tc->lock, flags); |
1787 | bio_list_add(&tc->deferred_bio_list, bio); | |
1788 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1789 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1790 | break; |
1791 | } | |
104655fd JT |
1792 | |
1793 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1794 | pool->process_discard(tc, bio); |
104655fd | 1795 | else |
e49e5829 | 1796 | pool->process_bio(tc, bio); |
8a01a6af JT |
1797 | |
1798 | if ((count++ & 127) == 0) { | |
7d327fe0 | 1799 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
1800 | dm_pool_issue_prefetches(pool->pmd); |
1801 | } | |
991d9fa0 | 1802 | } |
67324ea1 | 1803 | blk_finish_plug(&plug); |
c140e1c4 MS |
1804 | } |
1805 | ||
ac4c3f34 JT |
1806 | static int cmp_cells(const void *lhs, const void *rhs) |
1807 | { | |
1808 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
1809 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
1810 | ||
1811 | BUG_ON(!lhs_cell->holder); | |
1812 | BUG_ON(!rhs_cell->holder); | |
1813 | ||
1814 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
1815 | return -1; | |
1816 | ||
1817 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
1818 | return 1; | |
1819 | ||
1820 | return 0; | |
1821 | } | |
1822 | ||
1823 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) | |
1824 | { | |
1825 | unsigned count = 0; | |
1826 | struct dm_bio_prison_cell *cell, *tmp; | |
1827 | ||
1828 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
1829 | if (count >= CELL_SORT_ARRAY_SIZE) | |
1830 | break; | |
1831 | ||
1832 | pool->cell_sort_array[count++] = cell; | |
1833 | list_del(&cell->user_list); | |
1834 | } | |
1835 | ||
1836 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
1837 | ||
1838 | return count; | |
1839 | } | |
1840 | ||
a374bb21 JT |
1841 | static void process_thin_deferred_cells(struct thin_c *tc) |
1842 | { | |
1843 | struct pool *pool = tc->pool; | |
1844 | unsigned long flags; | |
1845 | struct list_head cells; | |
ac4c3f34 JT |
1846 | struct dm_bio_prison_cell *cell; |
1847 | unsigned i, j, count; | |
a374bb21 JT |
1848 | |
1849 | INIT_LIST_HEAD(&cells); | |
1850 | ||
1851 | spin_lock_irqsave(&tc->lock, flags); | |
1852 | list_splice_init(&tc->deferred_cells, &cells); | |
1853 | spin_unlock_irqrestore(&tc->lock, flags); | |
1854 | ||
1855 | if (list_empty(&cells)) | |
1856 | return; | |
1857 | ||
ac4c3f34 JT |
1858 | do { |
1859 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 1860 | |
ac4c3f34 JT |
1861 | for (i = 0; i < count; i++) { |
1862 | cell = pool->cell_sort_array[i]; | |
1863 | BUG_ON(!cell->holder); | |
a374bb21 | 1864 | |
ac4c3f34 JT |
1865 | /* |
1866 | * If we've got no free new_mapping structs, and processing | |
1867 | * this bio might require one, we pause until there are some | |
1868 | * prepared mappings to process. | |
1869 | */ | |
1870 | if (ensure_next_mapping(pool)) { | |
1871 | for (j = i; j < count; j++) | |
1872 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
1873 | ||
1874 | spin_lock_irqsave(&tc->lock, flags); | |
1875 | list_splice(&cells, &tc->deferred_cells); | |
1876 | spin_unlock_irqrestore(&tc->lock, flags); | |
1877 | return; | |
1878 | } | |
1879 | ||
1880 | if (cell->holder->bi_rw & REQ_DISCARD) | |
1881 | pool->process_discard_cell(tc, cell); | |
1882 | else | |
1883 | pool->process_cell(tc, cell); | |
1884 | } | |
1885 | } while (!list_empty(&cells)); | |
a374bb21 JT |
1886 | } |
1887 | ||
b10ebd34 JT |
1888 | static void thin_get(struct thin_c *tc); |
1889 | static void thin_put(struct thin_c *tc); | |
1890 | ||
1891 | /* | |
1892 | * We can't hold rcu_read_lock() around code that can block. So we | |
1893 | * find a thin with the rcu lock held; bump a refcount; then drop | |
1894 | * the lock. | |
1895 | */ | |
1896 | static struct thin_c *get_first_thin(struct pool *pool) | |
1897 | { | |
1898 | struct thin_c *tc = NULL; | |
1899 | ||
1900 | rcu_read_lock(); | |
1901 | if (!list_empty(&pool->active_thins)) { | |
1902 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
1903 | thin_get(tc); | |
1904 | } | |
1905 | rcu_read_unlock(); | |
1906 | ||
1907 | return tc; | |
1908 | } | |
1909 | ||
1910 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
1911 | { | |
1912 | struct thin_c *old_tc = tc; | |
1913 | ||
1914 | rcu_read_lock(); | |
1915 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
1916 | thin_get(tc); | |
1917 | thin_put(old_tc); | |
1918 | rcu_read_unlock(); | |
1919 | return tc; | |
1920 | } | |
1921 | thin_put(old_tc); | |
1922 | rcu_read_unlock(); | |
1923 | ||
1924 | return NULL; | |
1925 | } | |
1926 | ||
c140e1c4 MS |
1927 | static void process_deferred_bios(struct pool *pool) |
1928 | { | |
1929 | unsigned long flags; | |
1930 | struct bio *bio; | |
1931 | struct bio_list bios; | |
1932 | struct thin_c *tc; | |
1933 | ||
b10ebd34 JT |
1934 | tc = get_first_thin(pool); |
1935 | while (tc) { | |
a374bb21 | 1936 | process_thin_deferred_cells(tc); |
c140e1c4 | 1937 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
1938 | tc = get_next_thin(pool, tc); |
1939 | } | |
991d9fa0 JT |
1940 | |
1941 | /* | |
1942 | * If there are any deferred flush bios, we must commit | |
1943 | * the metadata before issuing them. | |
1944 | */ | |
1945 | bio_list_init(&bios); | |
1946 | spin_lock_irqsave(&pool->lock, flags); | |
1947 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1948 | bio_list_init(&pool->deferred_flush_bios); | |
1949 | spin_unlock_irqrestore(&pool->lock, flags); | |
1950 | ||
4d1662a3 MS |
1951 | if (bio_list_empty(&bios) && |
1952 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1953 | return; |
1954 | ||
020cc3b5 | 1955 | if (commit(pool)) { |
991d9fa0 JT |
1956 | while ((bio = bio_list_pop(&bios))) |
1957 | bio_io_error(bio); | |
1958 | return; | |
1959 | } | |
905e51b3 | 1960 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1961 | |
1962 | while ((bio = bio_list_pop(&bios))) | |
1963 | generic_make_request(bio); | |
1964 | } | |
1965 | ||
1966 | static void do_worker(struct work_struct *ws) | |
1967 | { | |
1968 | struct pool *pool = container_of(ws, struct pool, worker); | |
1969 | ||
7d327fe0 | 1970 | throttle_work_start(&pool->throttle); |
8a01a6af | 1971 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 1972 | throttle_work_update(&pool->throttle); |
e49e5829 | 1973 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 1974 | throttle_work_update(&pool->throttle); |
e49e5829 | 1975 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 1976 | throttle_work_update(&pool->throttle); |
991d9fa0 | 1977 | process_deferred_bios(pool); |
7d327fe0 | 1978 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
1979 | } |
1980 | ||
905e51b3 JT |
1981 | /* |
1982 | * We want to commit periodically so that not too much | |
1983 | * unwritten data builds up. | |
1984 | */ | |
1985 | static void do_waker(struct work_struct *ws) | |
1986 | { | |
1987 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1988 | wake_worker(pool); | |
1989 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1990 | } | |
1991 | ||
85ad643b JT |
1992 | /* |
1993 | * We're holding onto IO to allow userland time to react. After the | |
1994 | * timeout either the pool will have been resized (and thus back in | |
1995 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
1996 | */ | |
1997 | static void do_no_space_timeout(struct work_struct *ws) | |
1998 | { | |
1999 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2000 | no_space_timeout); | |
2001 | ||
2002 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
2003 | set_pool_mode(pool, PM_READ_ONLY); | |
2004 | } | |
2005 | ||
991d9fa0 JT |
2006 | /*----------------------------------------------------------------*/ |
2007 | ||
e7a3e871 | 2008 | struct pool_work { |
738211f7 | 2009 | struct work_struct worker; |
e7a3e871 JT |
2010 | struct completion complete; |
2011 | }; | |
2012 | ||
2013 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2014 | { | |
2015 | return container_of(ws, struct pool_work, worker); | |
2016 | } | |
2017 | ||
2018 | static void pool_work_complete(struct pool_work *pw) | |
2019 | { | |
2020 | complete(&pw->complete); | |
2021 | } | |
738211f7 | 2022 | |
e7a3e871 JT |
2023 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2024 | void (*fn)(struct work_struct *)) | |
2025 | { | |
2026 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2027 | init_completion(&pw->complete); | |
2028 | queue_work(pool->wq, &pw->worker); | |
2029 | wait_for_completion(&pw->complete); | |
2030 | } | |
2031 | ||
2032 | /*----------------------------------------------------------------*/ | |
2033 | ||
2034 | struct noflush_work { | |
2035 | struct pool_work pw; | |
2036 | struct thin_c *tc; | |
738211f7 JT |
2037 | }; |
2038 | ||
e7a3e871 | 2039 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2040 | { |
e7a3e871 | 2041 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2042 | } |
2043 | ||
2044 | static void do_noflush_start(struct work_struct *ws) | |
2045 | { | |
e7a3e871 | 2046 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2047 | w->tc->requeue_mode = true; |
2048 | requeue_io(w->tc); | |
e7a3e871 | 2049 | pool_work_complete(&w->pw); |
738211f7 JT |
2050 | } |
2051 | ||
2052 | static void do_noflush_stop(struct work_struct *ws) | |
2053 | { | |
e7a3e871 | 2054 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2055 | w->tc->requeue_mode = false; |
e7a3e871 | 2056 | pool_work_complete(&w->pw); |
738211f7 JT |
2057 | } |
2058 | ||
2059 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2060 | { | |
2061 | struct noflush_work w; | |
2062 | ||
738211f7 | 2063 | w.tc = tc; |
e7a3e871 | 2064 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2065 | } |
2066 | ||
2067 | /*----------------------------------------------------------------*/ | |
2068 | ||
e49e5829 JT |
2069 | static enum pool_mode get_pool_mode(struct pool *pool) |
2070 | { | |
2071 | return pool->pf.mode; | |
2072 | } | |
2073 | ||
3e1a0699 JT |
2074 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
2075 | { | |
2076 | dm_table_event(pool->ti->table); | |
2077 | DMINFO("%s: switching pool to %s mode", | |
2078 | dm_device_name(pool->pool_md), new_mode); | |
2079 | } | |
2080 | ||
8b64e881 | 2081 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2082 | { |
cdc2b415 | 2083 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2084 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2085 | enum pool_mode old_mode = get_pool_mode(pool); | |
80c57893 | 2086 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2087 | |
2088 | /* | |
2089 | * Never allow the pool to transition to PM_WRITE mode if user | |
2090 | * intervention is required to verify metadata and data consistency. | |
2091 | */ | |
2092 | if (new_mode == PM_WRITE && needs_check) { | |
2093 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2094 | dm_device_name(pool->pool_md)); | |
2095 | if (old_mode != new_mode) | |
2096 | new_mode = old_mode; | |
2097 | else | |
2098 | new_mode = PM_READ_ONLY; | |
2099 | } | |
2100 | /* | |
2101 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2102 | * not going to recover without a thin_repair. So we never let the | |
2103 | * pool move out of the old mode. | |
2104 | */ | |
2105 | if (old_mode == PM_FAIL) | |
2106 | new_mode = old_mode; | |
e49e5829 | 2107 | |
8b64e881 | 2108 | switch (new_mode) { |
e49e5829 | 2109 | case PM_FAIL: |
8b64e881 | 2110 | if (old_mode != new_mode) |
3e1a0699 | 2111 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 2112 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2113 | pool->process_bio = process_bio_fail; |
2114 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2115 | pool->process_cell = process_cell_fail; |
2116 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2117 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2118 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2119 | |
2120 | error_retry_list(pool); | |
e49e5829 JT |
2121 | break; |
2122 | ||
2123 | case PM_READ_ONLY: | |
8b64e881 | 2124 | if (old_mode != new_mode) |
3e1a0699 JT |
2125 | notify_of_pool_mode_change(pool, "read-only"); |
2126 | dm_pool_metadata_read_only(pool->pmd); | |
2127 | pool->process_bio = process_bio_read_only; | |
2128 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2129 | pool->process_cell = process_cell_read_only; |
2130 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 JT |
2131 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2132 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
2133 | ||
2134 | error_retry_list(pool); | |
2135 | break; | |
2136 | ||
2137 | case PM_OUT_OF_DATA_SPACE: | |
2138 | /* | |
2139 | * Ideally we'd never hit this state; the low water mark | |
2140 | * would trigger userland to extend the pool before we | |
2141 | * completely run out of data space. However, many small | |
2142 | * IOs to unprovisioned space can consume data space at an | |
2143 | * alarming rate. Adjust your low water mark if you're | |
2144 | * frequently seeing this mode. | |
2145 | */ | |
2146 | if (old_mode != new_mode) | |
2147 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
2148 | pool->process_bio = process_bio_read_only; | |
a374bb21 JT |
2149 | pool->process_discard = process_discard_bio; |
2150 | pool->process_cell = process_cell_read_only; | |
2151 | pool->process_discard_cell = process_discard_cell; | |
3e1a0699 JT |
2152 | pool->process_prepared_mapping = process_prepared_mapping; |
2153 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
85ad643b | 2154 | |
80c57893 MS |
2155 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2156 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2157 | break; |
2158 | ||
2159 | case PM_WRITE: | |
8b64e881 | 2160 | if (old_mode != new_mode) |
3e1a0699 | 2161 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 2162 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2163 | pool->process_bio = process_bio; |
a374bb21 JT |
2164 | pool->process_discard = process_discard_bio; |
2165 | pool->process_cell = process_cell; | |
2166 | pool->process_discard_cell = process_discard_cell; | |
e49e5829 JT |
2167 | pool->process_prepared_mapping = process_prepared_mapping; |
2168 | pool->process_prepared_discard = process_prepared_discard; | |
2169 | break; | |
2170 | } | |
8b64e881 MS |
2171 | |
2172 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2173 | /* |
2174 | * The pool mode may have changed, sync it so bind_control_target() | |
2175 | * doesn't cause an unexpected mode transition on resume. | |
2176 | */ | |
2177 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
2178 | } |
2179 | ||
07f2b6e0 | 2180 | static void abort_transaction(struct pool *pool) |
b5330655 | 2181 | { |
07f2b6e0 MS |
2182 | const char *dev_name = dm_device_name(pool->pool_md); |
2183 | ||
2184 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2185 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2186 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2187 | set_pool_mode(pool, PM_FAIL); | |
2188 | } | |
2189 | ||
2190 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2191 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2192 | set_pool_mode(pool, PM_FAIL); | |
2193 | } | |
2194 | } | |
399caddf | 2195 | |
07f2b6e0 MS |
2196 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2197 | { | |
b5330655 JT |
2198 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2199 | dm_device_name(pool->pool_md), op, r); | |
2200 | ||
07f2b6e0 | 2201 | abort_transaction(pool); |
b5330655 JT |
2202 | set_pool_mode(pool, PM_READ_ONLY); |
2203 | } | |
2204 | ||
e49e5829 JT |
2205 | /*----------------------------------------------------------------*/ |
2206 | ||
991d9fa0 JT |
2207 | /* |
2208 | * Mapping functions. | |
2209 | */ | |
2210 | ||
2211 | /* | |
2212 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2213 | */ | |
2214 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2215 | { | |
2216 | unsigned long flags; | |
2217 | struct pool *pool = tc->pool; | |
2218 | ||
c140e1c4 MS |
2219 | spin_lock_irqsave(&tc->lock, flags); |
2220 | bio_list_add(&tc->deferred_bio_list, bio); | |
2221 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2222 | |
2223 | wake_worker(pool); | |
2224 | } | |
2225 | ||
7d327fe0 JT |
2226 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2227 | { | |
2228 | struct pool *pool = tc->pool; | |
2229 | ||
2230 | throttle_lock(&pool->throttle); | |
2231 | thin_defer_bio(tc, bio); | |
2232 | throttle_unlock(&pool->throttle); | |
2233 | } | |
2234 | ||
a374bb21 JT |
2235 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2236 | { | |
2237 | unsigned long flags; | |
2238 | struct pool *pool = tc->pool; | |
2239 | ||
2240 | throttle_lock(&pool->throttle); | |
2241 | spin_lock_irqsave(&tc->lock, flags); | |
2242 | list_add_tail(&cell->user_list, &tc->deferred_cells); | |
2243 | spin_unlock_irqrestore(&tc->lock, flags); | |
2244 | throttle_unlock(&pool->throttle); | |
2245 | ||
2246 | wake_worker(pool); | |
2247 | } | |
2248 | ||
59c3d2c6 | 2249 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2250 | { |
59c3d2c6 | 2251 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2252 | |
2253 | h->tc = tc; | |
2254 | h->shared_read_entry = NULL; | |
e8088073 | 2255 | h->all_io_entry = NULL; |
eb2aa48d | 2256 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
2257 | } |
2258 | ||
991d9fa0 JT |
2259 | /* |
2260 | * Non-blocking function called from the thin target's map function. | |
2261 | */ | |
7de3ee57 | 2262 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2263 | { |
2264 | int r; | |
2265 | struct thin_c *tc = ti->private; | |
2266 | dm_block_t block = get_bio_block(tc, bio); | |
2267 | struct dm_thin_device *td = tc->td; | |
2268 | struct dm_thin_lookup_result result; | |
a374bb21 | 2269 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2270 | struct dm_cell_key key; |
991d9fa0 | 2271 | |
59c3d2c6 | 2272 | thin_hook_bio(tc, bio); |
e49e5829 | 2273 | |
738211f7 JT |
2274 | if (tc->requeue_mode) { |
2275 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
2276 | return DM_MAPIO_SUBMITTED; | |
2277 | } | |
2278 | ||
e49e5829 JT |
2279 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2280 | bio_io_error(bio); | |
2281 | return DM_MAPIO_SUBMITTED; | |
2282 | } | |
2283 | ||
104655fd | 2284 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
7d327fe0 | 2285 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2286 | return DM_MAPIO_SUBMITTED; |
2287 | } | |
2288 | ||
c822ed96 JT |
2289 | /* |
2290 | * We must hold the virtual cell before doing the lookup, otherwise | |
2291 | * there's a race with discard. | |
2292 | */ | |
2293 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2294 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2295 | return DM_MAPIO_SUBMITTED; |
2296 | ||
991d9fa0 JT |
2297 | r = dm_thin_find_block(td, block, 0, &result); |
2298 | ||
2299 | /* | |
2300 | * Note that we defer readahead too. | |
2301 | */ | |
2302 | switch (r) { | |
2303 | case 0: | |
2304 | if (unlikely(result.shared)) { | |
2305 | /* | |
2306 | * We have a race condition here between the | |
2307 | * result.shared value returned by the lookup and | |
2308 | * snapshot creation, which may cause new | |
2309 | * sharing. | |
2310 | * | |
2311 | * To avoid this always quiesce the origin before | |
2312 | * taking the snap. You want to do this anyway to | |
2313 | * ensure a consistent application view | |
2314 | * (i.e. lockfs). | |
2315 | * | |
2316 | * More distant ancestors are irrelevant. The | |
2317 | * shared flag will be set in their case. | |
2318 | */ | |
a374bb21 | 2319 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2320 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2321 | } |
e8088073 | 2322 | |
e8088073 | 2323 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2324 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2325 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2326 | return DM_MAPIO_SUBMITTED; |
2327 | } | |
2328 | ||
2329 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2330 | cell_defer_no_holder(tc, data_cell); |
2331 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2332 | |
2333 | remap(tc, bio, result.block); | |
2334 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2335 | |
2336 | case -ENODATA: | |
e49e5829 JT |
2337 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
2338 | /* | |
2339 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 2340 | * of doing so. |
e49e5829 | 2341 | */ |
8c0f0e8c | 2342 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 | 2343 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2344 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2345 | } |
2346 | /* fall through */ | |
2347 | ||
2348 | case -EWOULDBLOCK: | |
a374bb21 | 2349 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2350 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2351 | |
2352 | default: | |
2353 | /* | |
2354 | * Must always call bio_io_error on failure. | |
2355 | * dm_thin_find_block can fail with -EINVAL if the | |
2356 | * pool is switched to fail-io mode. | |
2357 | */ | |
2358 | bio_io_error(bio); | |
a374bb21 | 2359 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2360 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2361 | } |
991d9fa0 JT |
2362 | } |
2363 | ||
2364 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2365 | { | |
991d9fa0 | 2366 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 2367 | struct request_queue *q; |
991d9fa0 | 2368 | |
760fe67e MS |
2369 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
2370 | return 1; | |
991d9fa0 | 2371 | |
760fe67e MS |
2372 | q = bdev_get_queue(pt->data_dev->bdev); |
2373 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
2374 | } |
2375 | ||
c140e1c4 | 2376 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2377 | { |
c140e1c4 MS |
2378 | unsigned long flags; |
2379 | struct thin_c *tc; | |
2380 | ||
2381 | rcu_read_lock(); | |
2382 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
2383 | spin_lock_irqsave(&tc->lock, flags); | |
2384 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
2385 | bio_list_init(&tc->retry_on_resume_list); | |
2386 | spin_unlock_irqrestore(&tc->lock, flags); | |
2387 | } | |
2388 | rcu_read_unlock(); | |
991d9fa0 JT |
2389 | } |
2390 | ||
2391 | /*---------------------------------------------------------------- | |
2392 | * Binding of control targets to a pool object | |
2393 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2394 | static bool data_dev_supports_discard(struct pool_c *pt) |
2395 | { | |
2396 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2397 | ||
2398 | return q && blk_queue_discard(q); | |
2399 | } | |
2400 | ||
58051b94 JT |
2401 | static bool is_factor(sector_t block_size, uint32_t n) |
2402 | { | |
2403 | return !sector_div(block_size, n); | |
2404 | } | |
2405 | ||
9bc142dd MS |
2406 | /* |
2407 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2408 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2409 | */ |
0424caa1 | 2410 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2411 | { |
0424caa1 MS |
2412 | struct pool *pool = pt->pool; |
2413 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2414 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
2415 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
2416 | const char *reason = NULL; | |
9bc142dd MS |
2417 | char buf[BDEVNAME_SIZE]; |
2418 | ||
0424caa1 | 2419 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2420 | return; |
2421 | ||
0424caa1 MS |
2422 | if (!data_dev_supports_discard(pt)) |
2423 | reason = "discard unsupported"; | |
2424 | ||
2425 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2426 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2427 | |
0424caa1 MS |
2428 | else if (data_limits->discard_granularity > block_size) |
2429 | reason = "discard granularity larger than a block"; | |
2430 | ||
58051b94 | 2431 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
2432 | reason = "discard granularity not a factor of block size"; |
2433 | ||
2434 | if (reason) { | |
2435 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2436 | pt->adjusted_pf.discard_passdown = false; | |
2437 | } | |
9bc142dd MS |
2438 | } |
2439 | ||
991d9fa0 JT |
2440 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2441 | { | |
2442 | struct pool_c *pt = ti->private; | |
2443 | ||
e49e5829 | 2444 | /* |
9b7aaa64 | 2445 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2446 | */ |
07f2b6e0 | 2447 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2448 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2449 | |
8b64e881 MS |
2450 | /* |
2451 | * Don't change the pool's mode until set_pool_mode() below. | |
2452 | * Otherwise the pool's process_* function pointers may | |
2453 | * not match the desired pool mode. | |
2454 | */ | |
2455 | pt->adjusted_pf.mode = old_mode; | |
2456 | ||
2457 | pool->ti = ti; | |
2458 | pool->pf = pt->adjusted_pf; | |
2459 | pool->low_water_blocks = pt->low_water_blocks; | |
2460 | ||
9bc142dd | 2461 | set_pool_mode(pool, new_mode); |
f402693d | 2462 | |
991d9fa0 JT |
2463 | return 0; |
2464 | } | |
2465 | ||
2466 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2467 | { | |
2468 | if (pool->ti == ti) | |
2469 | pool->ti = NULL; | |
2470 | } | |
2471 | ||
2472 | /*---------------------------------------------------------------- | |
2473 | * Pool creation | |
2474 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2475 | /* Initialize pool features. */ |
2476 | static void pool_features_init(struct pool_features *pf) | |
2477 | { | |
e49e5829 | 2478 | pf->mode = PM_WRITE; |
9bc142dd MS |
2479 | pf->zero_new_blocks = true; |
2480 | pf->discard_enabled = true; | |
2481 | pf->discard_passdown = true; | |
787a996c | 2482 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2483 | } |
2484 | ||
991d9fa0 JT |
2485 | static void __pool_destroy(struct pool *pool) |
2486 | { | |
2487 | __pool_table_remove(pool); | |
2488 | ||
2489 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2490 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2491 | ||
44feb387 | 2492 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2493 | dm_kcopyd_client_destroy(pool->copier); |
2494 | ||
2495 | if (pool->wq) | |
2496 | destroy_workqueue(pool->wq); | |
2497 | ||
2498 | if (pool->next_mapping) | |
2499 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2500 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2501 | dm_deferred_set_destroy(pool->shared_read_ds); |
2502 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2503 | kfree(pool); |
2504 | } | |
2505 | ||
a24c2569 | 2506 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2507 | |
991d9fa0 JT |
2508 | static struct pool *pool_create(struct mapped_device *pool_md, |
2509 | struct block_device *metadata_dev, | |
e49e5829 JT |
2510 | unsigned long block_size, |
2511 | int read_only, char **error) | |
991d9fa0 JT |
2512 | { |
2513 | int r; | |
2514 | void *err_p; | |
2515 | struct pool *pool; | |
2516 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2517 | bool format_device = read_only ? false : true; |
991d9fa0 | 2518 | |
e49e5829 | 2519 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2520 | if (IS_ERR(pmd)) { |
2521 | *error = "Error creating metadata object"; | |
2522 | return (struct pool *)pmd; | |
2523 | } | |
2524 | ||
2525 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2526 | if (!pool) { | |
2527 | *error = "Error allocating memory for pool"; | |
2528 | err_p = ERR_PTR(-ENOMEM); | |
2529 | goto bad_pool; | |
2530 | } | |
2531 | ||
2532 | pool->pmd = pmd; | |
2533 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2534 | if (block_size & (block_size - 1)) |
2535 | pool->sectors_per_block_shift = -1; | |
2536 | else | |
2537 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2538 | pool->low_water_blocks = 0; |
67e2e2b2 | 2539 | pool_features_init(&pool->pf); |
a195db2d | 2540 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2541 | if (!pool->prison) { |
2542 | *error = "Error creating pool's bio prison"; | |
2543 | err_p = ERR_PTR(-ENOMEM); | |
2544 | goto bad_prison; | |
2545 | } | |
2546 | ||
df5d2e90 | 2547 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2548 | if (IS_ERR(pool->copier)) { |
2549 | r = PTR_ERR(pool->copier); | |
2550 | *error = "Error creating pool's kcopyd client"; | |
2551 | err_p = ERR_PTR(r); | |
2552 | goto bad_kcopyd_client; | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * Create singlethreaded workqueue that will service all devices | |
2557 | * that use this metadata. | |
2558 | */ | |
2559 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2560 | if (!pool->wq) { | |
2561 | *error = "Error creating pool's workqueue"; | |
2562 | err_p = ERR_PTR(-ENOMEM); | |
2563 | goto bad_wq; | |
2564 | } | |
2565 | ||
7d327fe0 | 2566 | throttle_init(&pool->throttle); |
991d9fa0 | 2567 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 2568 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2569 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2570 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2571 | bio_list_init(&pool->deferred_flush_bios); |
2572 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2573 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2574 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2575 | pool->low_water_triggered = false; |
44feb387 MS |
2576 | |
2577 | pool->shared_read_ds = dm_deferred_set_create(); | |
2578 | if (!pool->shared_read_ds) { | |
2579 | *error = "Error creating pool's shared read deferred set"; | |
2580 | err_p = ERR_PTR(-ENOMEM); | |
2581 | goto bad_shared_read_ds; | |
2582 | } | |
2583 | ||
2584 | pool->all_io_ds = dm_deferred_set_create(); | |
2585 | if (!pool->all_io_ds) { | |
2586 | *error = "Error creating pool's all io deferred set"; | |
2587 | err_p = ERR_PTR(-ENOMEM); | |
2588 | goto bad_all_io_ds; | |
2589 | } | |
991d9fa0 JT |
2590 | |
2591 | pool->next_mapping = NULL; | |
a24c2569 MS |
2592 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2593 | _new_mapping_cache); | |
991d9fa0 JT |
2594 | if (!pool->mapping_pool) { |
2595 | *error = "Error creating pool's mapping mempool"; | |
2596 | err_p = ERR_PTR(-ENOMEM); | |
2597 | goto bad_mapping_pool; | |
2598 | } | |
2599 | ||
991d9fa0 | 2600 | pool->ref_count = 1; |
905e51b3 | 2601 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2602 | pool->pool_md = pool_md; |
2603 | pool->md_dev = metadata_dev; | |
2604 | __pool_table_insert(pool); | |
2605 | ||
2606 | return pool; | |
2607 | ||
991d9fa0 | 2608 | bad_mapping_pool: |
44feb387 MS |
2609 | dm_deferred_set_destroy(pool->all_io_ds); |
2610 | bad_all_io_ds: | |
2611 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2612 | bad_shared_read_ds: | |
991d9fa0 JT |
2613 | destroy_workqueue(pool->wq); |
2614 | bad_wq: | |
2615 | dm_kcopyd_client_destroy(pool->copier); | |
2616 | bad_kcopyd_client: | |
44feb387 | 2617 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2618 | bad_prison: |
2619 | kfree(pool); | |
2620 | bad_pool: | |
2621 | if (dm_pool_metadata_close(pmd)) | |
2622 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2623 | ||
2624 | return err_p; | |
2625 | } | |
2626 | ||
2627 | static void __pool_inc(struct pool *pool) | |
2628 | { | |
2629 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2630 | pool->ref_count++; | |
2631 | } | |
2632 | ||
2633 | static void __pool_dec(struct pool *pool) | |
2634 | { | |
2635 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2636 | BUG_ON(!pool->ref_count); | |
2637 | if (!--pool->ref_count) | |
2638 | __pool_destroy(pool); | |
2639 | } | |
2640 | ||
2641 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2642 | struct block_device *metadata_dev, | |
e49e5829 JT |
2643 | unsigned long block_size, int read_only, |
2644 | char **error, int *created) | |
991d9fa0 JT |
2645 | { |
2646 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2647 | ||
2648 | if (pool) { | |
f09996c9 MS |
2649 | if (pool->pool_md != pool_md) { |
2650 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2651 | return ERR_PTR(-EBUSY); |
f09996c9 | 2652 | } |
991d9fa0 JT |
2653 | __pool_inc(pool); |
2654 | ||
2655 | } else { | |
2656 | pool = __pool_table_lookup(pool_md); | |
2657 | if (pool) { | |
f09996c9 MS |
2658 | if (pool->md_dev != metadata_dev) { |
2659 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2660 | return ERR_PTR(-EINVAL); |
f09996c9 | 2661 | } |
991d9fa0 JT |
2662 | __pool_inc(pool); |
2663 | ||
67e2e2b2 | 2664 | } else { |
e49e5829 | 2665 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2666 | *created = 1; |
2667 | } | |
991d9fa0 JT |
2668 | } |
2669 | ||
2670 | return pool; | |
2671 | } | |
2672 | ||
2673 | /*---------------------------------------------------------------- | |
2674 | * Pool target methods | |
2675 | *--------------------------------------------------------------*/ | |
2676 | static void pool_dtr(struct dm_target *ti) | |
2677 | { | |
2678 | struct pool_c *pt = ti->private; | |
2679 | ||
2680 | mutex_lock(&dm_thin_pool_table.mutex); | |
2681 | ||
2682 | unbind_control_target(pt->pool, ti); | |
2683 | __pool_dec(pt->pool); | |
2684 | dm_put_device(ti, pt->metadata_dev); | |
2685 | dm_put_device(ti, pt->data_dev); | |
2686 | kfree(pt); | |
2687 | ||
2688 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2689 | } | |
2690 | ||
991d9fa0 JT |
2691 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2692 | struct dm_target *ti) | |
2693 | { | |
2694 | int r; | |
2695 | unsigned argc; | |
2696 | const char *arg_name; | |
2697 | ||
2698 | static struct dm_arg _args[] = { | |
74aa45c3 | 2699 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2700 | }; |
2701 | ||
2702 | /* | |
2703 | * No feature arguments supplied. | |
2704 | */ | |
2705 | if (!as->argc) | |
2706 | return 0; | |
2707 | ||
2708 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2709 | if (r) | |
2710 | return -EINVAL; | |
2711 | ||
2712 | while (argc && !r) { | |
2713 | arg_name = dm_shift_arg(as); | |
2714 | argc--; | |
2715 | ||
e49e5829 | 2716 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2717 | pf->zero_new_blocks = false; |
e49e5829 JT |
2718 | |
2719 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2720 | pf->discard_enabled = false; |
e49e5829 JT |
2721 | |
2722 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2723 | pf->discard_passdown = false; |
991d9fa0 | 2724 | |
e49e5829 JT |
2725 | else if (!strcasecmp(arg_name, "read_only")) |
2726 | pf->mode = PM_READ_ONLY; | |
2727 | ||
787a996c MS |
2728 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2729 | pf->error_if_no_space = true; | |
2730 | ||
e49e5829 JT |
2731 | else { |
2732 | ti->error = "Unrecognised pool feature requested"; | |
2733 | r = -EINVAL; | |
2734 | break; | |
2735 | } | |
991d9fa0 JT |
2736 | } |
2737 | ||
2738 | return r; | |
2739 | } | |
2740 | ||
ac8c3f3d JT |
2741 | static void metadata_low_callback(void *context) |
2742 | { | |
2743 | struct pool *pool = context; | |
2744 | ||
2745 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2746 | dm_device_name(pool->pool_md)); | |
2747 | ||
2748 | dm_table_event(pool->ti->table); | |
2749 | } | |
2750 | ||
7d48935e MS |
2751 | static sector_t get_dev_size(struct block_device *bdev) |
2752 | { | |
2753 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2754 | } | |
2755 | ||
2756 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2757 | { |
7d48935e | 2758 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2759 | char buffer[BDEVNAME_SIZE]; |
2760 | ||
7d48935e | 2761 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2762 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2763 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2764 | } |
2765 | ||
2766 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2767 | { | |
2768 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2769 | ||
2770 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2771 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2772 | |
2773 | return metadata_dev_size; | |
2774 | } | |
2775 | ||
24347e95 JT |
2776 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2777 | { | |
2778 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2779 | ||
7d48935e | 2780 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2781 | |
2782 | return metadata_dev_size; | |
2783 | } | |
2784 | ||
ac8c3f3d JT |
2785 | /* |
2786 | * When a metadata threshold is crossed a dm event is triggered, and | |
2787 | * userland should respond by growing the metadata device. We could let | |
2788 | * userland set the threshold, like we do with the data threshold, but I'm | |
2789 | * not sure they know enough to do this well. | |
2790 | */ | |
2791 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2792 | { | |
2793 | /* | |
2794 | * 4M is ample for all ops with the possible exception of thin | |
2795 | * device deletion which is harmless if it fails (just retry the | |
2796 | * delete after you've grown the device). | |
2797 | */ | |
2798 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2799 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2800 | } | |
2801 | ||
991d9fa0 JT |
2802 | /* |
2803 | * thin-pool <metadata dev> <data dev> | |
2804 | * <data block size (sectors)> | |
2805 | * <low water mark (blocks)> | |
2806 | * [<#feature args> [<arg>]*] | |
2807 | * | |
2808 | * Optional feature arguments are: | |
2809 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2810 | * ignore_discard: disable discard |
2811 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2812 | * read_only: Don't allow any changes to be made to the pool metadata. |
2813 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2814 | */ |
2815 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2816 | { | |
67e2e2b2 | 2817 | int r, pool_created = 0; |
991d9fa0 JT |
2818 | struct pool_c *pt; |
2819 | struct pool *pool; | |
2820 | struct pool_features pf; | |
2821 | struct dm_arg_set as; | |
2822 | struct dm_dev *data_dev; | |
2823 | unsigned long block_size; | |
2824 | dm_block_t low_water_blocks; | |
2825 | struct dm_dev *metadata_dev; | |
5d0db96d | 2826 | fmode_t metadata_mode; |
991d9fa0 JT |
2827 | |
2828 | /* | |
2829 | * FIXME Remove validation from scope of lock. | |
2830 | */ | |
2831 | mutex_lock(&dm_thin_pool_table.mutex); | |
2832 | ||
2833 | if (argc < 4) { | |
2834 | ti->error = "Invalid argument count"; | |
2835 | r = -EINVAL; | |
2836 | goto out_unlock; | |
2837 | } | |
5d0db96d | 2838 | |
991d9fa0 JT |
2839 | as.argc = argc; |
2840 | as.argv = argv; | |
2841 | ||
5d0db96d JT |
2842 | /* |
2843 | * Set default pool features. | |
2844 | */ | |
2845 | pool_features_init(&pf); | |
2846 | ||
2847 | dm_consume_args(&as, 4); | |
2848 | r = parse_pool_features(&as, &pf, ti); | |
2849 | if (r) | |
2850 | goto out_unlock; | |
2851 | ||
2852 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2853 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2854 | if (r) { |
2855 | ti->error = "Error opening metadata block device"; | |
2856 | goto out_unlock; | |
2857 | } | |
7d48935e | 2858 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2859 | |
2860 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2861 | if (r) { | |
2862 | ti->error = "Error getting data device"; | |
2863 | goto out_metadata; | |
2864 | } | |
2865 | ||
2866 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2867 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2868 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2869 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2870 | ti->error = "Invalid block size"; |
2871 | r = -EINVAL; | |
2872 | goto out; | |
2873 | } | |
2874 | ||
2875 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2876 | ti->error = "Invalid low water mark"; | |
2877 | r = -EINVAL; | |
2878 | goto out; | |
2879 | } | |
2880 | ||
991d9fa0 JT |
2881 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2882 | if (!pt) { | |
2883 | r = -ENOMEM; | |
2884 | goto out; | |
2885 | } | |
2886 | ||
2887 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2888 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2889 | if (IS_ERR(pool)) { |
2890 | r = PTR_ERR(pool); | |
2891 | goto out_free_pt; | |
2892 | } | |
2893 | ||
67e2e2b2 JT |
2894 | /* |
2895 | * 'pool_created' reflects whether this is the first table load. | |
2896 | * Top level discard support is not allowed to be changed after | |
2897 | * initial load. This would require a pool reload to trigger thin | |
2898 | * device changes. | |
2899 | */ | |
2900 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2901 | ti->error = "Discard support cannot be disabled once enabled"; | |
2902 | r = -EINVAL; | |
2903 | goto out_flags_changed; | |
2904 | } | |
2905 | ||
991d9fa0 JT |
2906 | pt->pool = pool; |
2907 | pt->ti = ti; | |
2908 | pt->metadata_dev = metadata_dev; | |
2909 | pt->data_dev = data_dev; | |
2910 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2911 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2912 | ti->num_flush_bios = 1; |
9bc142dd | 2913 | |
67e2e2b2 JT |
2914 | /* |
2915 | * Only need to enable discards if the pool should pass | |
2916 | * them down to the data device. The thin device's discard | |
2917 | * processing will cause mappings to be removed from the btree. | |
2918 | */ | |
b60ab990 | 2919 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2920 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2921 | ti->num_discard_bios = 1; |
9bc142dd | 2922 | |
67e2e2b2 JT |
2923 | /* |
2924 | * Setting 'discards_supported' circumvents the normal | |
2925 | * stacking of discard limits (this keeps the pool and | |
2926 | * thin devices' discard limits consistent). | |
2927 | */ | |
0ac55489 | 2928 | ti->discards_supported = true; |
67e2e2b2 | 2929 | } |
991d9fa0 JT |
2930 | ti->private = pt; |
2931 | ||
ac8c3f3d JT |
2932 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2933 | calc_metadata_threshold(pt), | |
2934 | metadata_low_callback, | |
2935 | pool); | |
2936 | if (r) | |
2937 | goto out_free_pt; | |
2938 | ||
991d9fa0 JT |
2939 | pt->callbacks.congested_fn = pool_is_congested; |
2940 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2941 | ||
2942 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2943 | ||
2944 | return 0; | |
2945 | ||
67e2e2b2 JT |
2946 | out_flags_changed: |
2947 | __pool_dec(pool); | |
991d9fa0 JT |
2948 | out_free_pt: |
2949 | kfree(pt); | |
2950 | out: | |
2951 | dm_put_device(ti, data_dev); | |
2952 | out_metadata: | |
2953 | dm_put_device(ti, metadata_dev); | |
2954 | out_unlock: | |
2955 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2956 | ||
2957 | return r; | |
2958 | } | |
2959 | ||
7de3ee57 | 2960 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2961 | { |
2962 | int r; | |
2963 | struct pool_c *pt = ti->private; | |
2964 | struct pool *pool = pt->pool; | |
2965 | unsigned long flags; | |
2966 | ||
2967 | /* | |
2968 | * As this is a singleton target, ti->begin is always zero. | |
2969 | */ | |
2970 | spin_lock_irqsave(&pool->lock, flags); | |
2971 | bio->bi_bdev = pt->data_dev->bdev; | |
2972 | r = DM_MAPIO_REMAPPED; | |
2973 | spin_unlock_irqrestore(&pool->lock, flags); | |
2974 | ||
2975 | return r; | |
2976 | } | |
2977 | ||
b17446df | 2978 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2979 | { |
2980 | int r; | |
2981 | struct pool_c *pt = ti->private; | |
2982 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2983 | sector_t data_size = ti->len; |
2984 | dm_block_t sb_data_size; | |
991d9fa0 | 2985 | |
b17446df | 2986 | *need_commit = false; |
991d9fa0 | 2987 | |
55f2b8bd MS |
2988 | (void) sector_div(data_size, pool->sectors_per_block); |
2989 | ||
991d9fa0 JT |
2990 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2991 | if (r) { | |
4fa5971a MS |
2992 | DMERR("%s: failed to retrieve data device size", |
2993 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2994 | return r; |
2995 | } | |
2996 | ||
2997 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2998 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2999 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3000 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3001 | return -EINVAL; |
3002 | ||
3003 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3004 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3005 | DMERR("%s: unable to grow the data device until repaired.", | |
3006 | dm_device_name(pool->pool_md)); | |
3007 | return 0; | |
3008 | } | |
3009 | ||
6f7f51d4 MS |
3010 | if (sb_data_size) |
3011 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3012 | dm_device_name(pool->pool_md), | |
3013 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3014 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3015 | if (r) { | |
b5330655 | 3016 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3017 | return r; |
3018 | } | |
3019 | ||
b17446df | 3020 | *need_commit = true; |
991d9fa0 JT |
3021 | } |
3022 | ||
3023 | return 0; | |
3024 | } | |
3025 | ||
24347e95 JT |
3026 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3027 | { | |
3028 | int r; | |
3029 | struct pool_c *pt = ti->private; | |
3030 | struct pool *pool = pt->pool; | |
3031 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3032 | ||
3033 | *need_commit = false; | |
3034 | ||
610bba8b | 3035 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3036 | |
3037 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3038 | if (r) { | |
4fa5971a MS |
3039 | DMERR("%s: failed to retrieve metadata device size", |
3040 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3041 | return r; |
3042 | } | |
3043 | ||
3044 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3045 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3046 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3047 | metadata_dev_size, sb_metadata_dev_size); |
3048 | return -EINVAL; | |
3049 | ||
3050 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3051 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3052 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3053 | dm_device_name(pool->pool_md)); | |
3054 | return 0; | |
3055 | } | |
3056 | ||
7d48935e | 3057 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3058 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3059 | dm_device_name(pool->pool_md), | |
3060 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
3061 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3062 | if (r) { | |
b5330655 | 3063 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3064 | return r; |
3065 | } | |
3066 | ||
3067 | *need_commit = true; | |
3068 | } | |
3069 | ||
3070 | return 0; | |
3071 | } | |
3072 | ||
b17446df JT |
3073 | /* |
3074 | * Retrieves the number of blocks of the data device from | |
3075 | * the superblock and compares it to the actual device size, | |
3076 | * thus resizing the data device in case it has grown. | |
3077 | * | |
3078 | * This both copes with opening preallocated data devices in the ctr | |
3079 | * being followed by a resume | |
3080 | * -and- | |
3081 | * calling the resume method individually after userspace has | |
3082 | * grown the data device in reaction to a table event. | |
3083 | */ | |
3084 | static int pool_preresume(struct dm_target *ti) | |
3085 | { | |
3086 | int r; | |
24347e95 | 3087 | bool need_commit1, need_commit2; |
b17446df JT |
3088 | struct pool_c *pt = ti->private; |
3089 | struct pool *pool = pt->pool; | |
3090 | ||
3091 | /* | |
3092 | * Take control of the pool object. | |
3093 | */ | |
3094 | r = bind_control_target(pool, ti); | |
3095 | if (r) | |
3096 | return r; | |
3097 | ||
3098 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3099 | if (r) | |
3100 | return r; | |
3101 | ||
24347e95 JT |
3102 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3103 | if (r) | |
3104 | return r; | |
3105 | ||
3106 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3107 | (void) commit(pool); |
b17446df JT |
3108 | |
3109 | return 0; | |
3110 | } | |
3111 | ||
991d9fa0 JT |
3112 | static void pool_resume(struct dm_target *ti) |
3113 | { | |
3114 | struct pool_c *pt = ti->private; | |
3115 | struct pool *pool = pt->pool; | |
3116 | unsigned long flags; | |
3117 | ||
3118 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 3119 | pool->low_water_triggered = false; |
991d9fa0 | 3120 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 3121 | requeue_bios(pool); |
991d9fa0 | 3122 | |
905e51b3 | 3123 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3124 | } |
3125 | ||
3126 | static void pool_postsuspend(struct dm_target *ti) | |
3127 | { | |
991d9fa0 JT |
3128 | struct pool_c *pt = ti->private; |
3129 | struct pool *pool = pt->pool; | |
3130 | ||
905e51b3 | 3131 | cancel_delayed_work(&pool->waker); |
85ad643b | 3132 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 3133 | flush_workqueue(pool->wq); |
020cc3b5 | 3134 | (void) commit(pool); |
991d9fa0 JT |
3135 | } |
3136 | ||
3137 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3138 | { | |
3139 | if (argc != args_required) { | |
3140 | DMWARN("Message received with %u arguments instead of %u.", | |
3141 | argc, args_required); | |
3142 | return -EINVAL; | |
3143 | } | |
3144 | ||
3145 | return 0; | |
3146 | } | |
3147 | ||
3148 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3149 | { | |
3150 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3151 | *dev_id <= MAX_DEV_ID) | |
3152 | return 0; | |
3153 | ||
3154 | if (warning) | |
3155 | DMWARN("Message received with invalid device id: %s", arg); | |
3156 | ||
3157 | return -EINVAL; | |
3158 | } | |
3159 | ||
3160 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3161 | { | |
3162 | dm_thin_id dev_id; | |
3163 | int r; | |
3164 | ||
3165 | r = check_arg_count(argc, 2); | |
3166 | if (r) | |
3167 | return r; | |
3168 | ||
3169 | r = read_dev_id(argv[1], &dev_id, 1); | |
3170 | if (r) | |
3171 | return r; | |
3172 | ||
3173 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3174 | if (r) { | |
3175 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3176 | argv[1]); | |
3177 | return r; | |
3178 | } | |
3179 | ||
3180 | return 0; | |
3181 | } | |
3182 | ||
3183 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3184 | { | |
3185 | dm_thin_id dev_id; | |
3186 | dm_thin_id origin_dev_id; | |
3187 | int r; | |
3188 | ||
3189 | r = check_arg_count(argc, 3); | |
3190 | if (r) | |
3191 | return r; | |
3192 | ||
3193 | r = read_dev_id(argv[1], &dev_id, 1); | |
3194 | if (r) | |
3195 | return r; | |
3196 | ||
3197 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3198 | if (r) | |
3199 | return r; | |
3200 | ||
3201 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3202 | if (r) { | |
3203 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3204 | argv[1], argv[2]); | |
3205 | return r; | |
3206 | } | |
3207 | ||
3208 | return 0; | |
3209 | } | |
3210 | ||
3211 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3212 | { | |
3213 | dm_thin_id dev_id; | |
3214 | int r; | |
3215 | ||
3216 | r = check_arg_count(argc, 2); | |
3217 | if (r) | |
3218 | return r; | |
3219 | ||
3220 | r = read_dev_id(argv[1], &dev_id, 1); | |
3221 | if (r) | |
3222 | return r; | |
3223 | ||
3224 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3225 | if (r) | |
3226 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3227 | ||
3228 | return r; | |
3229 | } | |
3230 | ||
3231 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3232 | { | |
3233 | dm_thin_id old_id, new_id; | |
3234 | int r; | |
3235 | ||
3236 | r = check_arg_count(argc, 3); | |
3237 | if (r) | |
3238 | return r; | |
3239 | ||
3240 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3241 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3242 | return -EINVAL; | |
3243 | } | |
3244 | ||
3245 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3246 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3247 | return -EINVAL; | |
3248 | } | |
3249 | ||
3250 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3251 | if (r) { | |
3252 | DMWARN("Failed to change transaction id from %s to %s.", | |
3253 | argv[1], argv[2]); | |
3254 | return r; | |
3255 | } | |
3256 | ||
3257 | return 0; | |
3258 | } | |
3259 | ||
cc8394d8 JT |
3260 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3261 | { | |
3262 | int r; | |
3263 | ||
3264 | r = check_arg_count(argc, 1); | |
3265 | if (r) | |
3266 | return r; | |
3267 | ||
020cc3b5 | 3268 | (void) commit(pool); |
0d200aef | 3269 | |
cc8394d8 JT |
3270 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3271 | if (r) | |
3272 | DMWARN("reserve_metadata_snap message failed."); | |
3273 | ||
3274 | return r; | |
3275 | } | |
3276 | ||
3277 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3278 | { | |
3279 | int r; | |
3280 | ||
3281 | r = check_arg_count(argc, 1); | |
3282 | if (r) | |
3283 | return r; | |
3284 | ||
3285 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3286 | if (r) | |
3287 | DMWARN("release_metadata_snap message failed."); | |
3288 | ||
3289 | return r; | |
3290 | } | |
3291 | ||
991d9fa0 JT |
3292 | /* |
3293 | * Messages supported: | |
3294 | * create_thin <dev_id> | |
3295 | * create_snap <dev_id> <origin_id> | |
3296 | * delete <dev_id> | |
3297 | * trim <dev_id> <new_size_in_sectors> | |
3298 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
3299 | * reserve_metadata_snap |
3300 | * release_metadata_snap | |
991d9fa0 JT |
3301 | */ |
3302 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
3303 | { | |
3304 | int r = -EINVAL; | |
3305 | struct pool_c *pt = ti->private; | |
3306 | struct pool *pool = pt->pool; | |
3307 | ||
3308 | if (!strcasecmp(argv[0], "create_thin")) | |
3309 | r = process_create_thin_mesg(argc, argv, pool); | |
3310 | ||
3311 | else if (!strcasecmp(argv[0], "create_snap")) | |
3312 | r = process_create_snap_mesg(argc, argv, pool); | |
3313 | ||
3314 | else if (!strcasecmp(argv[0], "delete")) | |
3315 | r = process_delete_mesg(argc, argv, pool); | |
3316 | ||
3317 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3318 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3319 | ||
cc8394d8 JT |
3320 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3321 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3322 | ||
3323 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3324 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3325 | ||
991d9fa0 JT |
3326 | else |
3327 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3328 | ||
e49e5829 | 3329 | if (!r) |
020cc3b5 | 3330 | (void) commit(pool); |
991d9fa0 JT |
3331 | |
3332 | return r; | |
3333 | } | |
3334 | ||
e49e5829 JT |
3335 | static void emit_flags(struct pool_features *pf, char *result, |
3336 | unsigned sz, unsigned maxlen) | |
3337 | { | |
3338 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3339 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3340 | pf->error_if_no_space; | |
e49e5829 JT |
3341 | DMEMIT("%u ", count); |
3342 | ||
3343 | if (!pf->zero_new_blocks) | |
3344 | DMEMIT("skip_block_zeroing "); | |
3345 | ||
3346 | if (!pf->discard_enabled) | |
3347 | DMEMIT("ignore_discard "); | |
3348 | ||
3349 | if (!pf->discard_passdown) | |
3350 | DMEMIT("no_discard_passdown "); | |
3351 | ||
3352 | if (pf->mode == PM_READ_ONLY) | |
3353 | DMEMIT("read_only "); | |
787a996c MS |
3354 | |
3355 | if (pf->error_if_no_space) | |
3356 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3357 | } |
3358 | ||
991d9fa0 JT |
3359 | /* |
3360 | * Status line is: | |
3361 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3362 | * <used data sectors>/<total data sectors> <held metadata root> | |
3363 | */ | |
fd7c092e MP |
3364 | static void pool_status(struct dm_target *ti, status_type_t type, |
3365 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3366 | { |
e49e5829 | 3367 | int r; |
991d9fa0 JT |
3368 | unsigned sz = 0; |
3369 | uint64_t transaction_id; | |
3370 | dm_block_t nr_free_blocks_data; | |
3371 | dm_block_t nr_free_blocks_metadata; | |
3372 | dm_block_t nr_blocks_data; | |
3373 | dm_block_t nr_blocks_metadata; | |
3374 | dm_block_t held_root; | |
3375 | char buf[BDEVNAME_SIZE]; | |
3376 | char buf2[BDEVNAME_SIZE]; | |
3377 | struct pool_c *pt = ti->private; | |
3378 | struct pool *pool = pt->pool; | |
3379 | ||
3380 | switch (type) { | |
3381 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3382 | if (get_pool_mode(pool) == PM_FAIL) { |
3383 | DMEMIT("Fail"); | |
3384 | break; | |
3385 | } | |
3386 | ||
1f4e0ff0 AK |
3387 | /* Commit to ensure statistics aren't out-of-date */ |
3388 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3389 | (void) commit(pool); |
1f4e0ff0 | 3390 | |
fd7c092e MP |
3391 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3392 | if (r) { | |
4fa5971a MS |
3393 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3394 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3395 | goto err; |
3396 | } | |
991d9fa0 | 3397 | |
fd7c092e MP |
3398 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3399 | if (r) { | |
4fa5971a MS |
3400 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3401 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3402 | goto err; |
3403 | } | |
991d9fa0 JT |
3404 | |
3405 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3406 | if (r) { |
4fa5971a MS |
3407 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3408 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3409 | goto err; |
3410 | } | |
991d9fa0 | 3411 | |
fd7c092e MP |
3412 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3413 | if (r) { | |
4fa5971a MS |
3414 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3415 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3416 | goto err; |
3417 | } | |
991d9fa0 JT |
3418 | |
3419 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3420 | if (r) { |
4fa5971a MS |
3421 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3422 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3423 | goto err; |
3424 | } | |
991d9fa0 | 3425 | |
cc8394d8 | 3426 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3427 | if (r) { |
4fa5971a MS |
3428 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3429 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3430 | goto err; |
3431 | } | |
991d9fa0 JT |
3432 | |
3433 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3434 | (unsigned long long)transaction_id, | |
3435 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3436 | (unsigned long long)nr_blocks_metadata, | |
3437 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3438 | (unsigned long long)nr_blocks_data); | |
3439 | ||
3440 | if (held_root) | |
e49e5829 JT |
3441 | DMEMIT("%llu ", held_root); |
3442 | else | |
3443 | DMEMIT("- "); | |
3444 | ||
3e1a0699 JT |
3445 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3446 | DMEMIT("out_of_data_space "); | |
3447 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3448 | DMEMIT("ro "); |
991d9fa0 | 3449 | else |
e49e5829 JT |
3450 | DMEMIT("rw "); |
3451 | ||
018debea | 3452 | if (!pool->pf.discard_enabled) |
787a996c | 3453 | DMEMIT("ignore_discard "); |
018debea | 3454 | else if (pool->pf.discard_passdown) |
787a996c MS |
3455 | DMEMIT("discard_passdown "); |
3456 | else | |
3457 | DMEMIT("no_discard_passdown "); | |
3458 | ||
3459 | if (pool->pf.error_if_no_space) | |
3460 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3461 | else |
787a996c | 3462 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3463 | |
3464 | break; | |
3465 | ||
3466 | case STATUSTYPE_TABLE: | |
3467 | DMEMIT("%s %s %lu %llu ", | |
3468 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3469 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3470 | (unsigned long)pool->sectors_per_block, | |
3471 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3472 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3473 | break; |
3474 | } | |
fd7c092e | 3475 | return; |
991d9fa0 | 3476 | |
fd7c092e MP |
3477 | err: |
3478 | DMEMIT("Error"); | |
991d9fa0 JT |
3479 | } |
3480 | ||
3481 | static int pool_iterate_devices(struct dm_target *ti, | |
3482 | iterate_devices_callout_fn fn, void *data) | |
3483 | { | |
3484 | struct pool_c *pt = ti->private; | |
3485 | ||
3486 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3487 | } | |
3488 | ||
3489 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3490 | struct bio_vec *biovec, int max_size) | |
3491 | { | |
3492 | struct pool_c *pt = ti->private; | |
3493 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3494 | ||
3495 | if (!q->merge_bvec_fn) | |
3496 | return max_size; | |
3497 | ||
3498 | bvm->bi_bdev = pt->data_dev->bdev; | |
3499 | ||
3500 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3501 | } | |
3502 | ||
0424caa1 | 3503 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 3504 | { |
0424caa1 MS |
3505 | struct pool *pool = pt->pool; |
3506 | struct queue_limits *data_limits; | |
3507 | ||
104655fd JT |
3508 | limits->max_discard_sectors = pool->sectors_per_block; |
3509 | ||
3510 | /* | |
0424caa1 | 3511 | * discard_granularity is just a hint, and not enforced. |
104655fd | 3512 | */ |
0424caa1 MS |
3513 | if (pt->adjusted_pf.discard_passdown) { |
3514 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
09869de5 LC |
3515 | limits->discard_granularity = max(data_limits->discard_granularity, |
3516 | pool->sectors_per_block << SECTOR_SHIFT); | |
f13945d7 | 3517 | } else |
0424caa1 | 3518 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3519 | } |
3520 | ||
991d9fa0 JT |
3521 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3522 | { | |
3523 | struct pool_c *pt = ti->private; | |
3524 | struct pool *pool = pt->pool; | |
604ea906 MS |
3525 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
3526 | ||
3527 | /* | |
3528 | * Adjust max_sectors_kb to highest possible power-of-2 | |
3529 | * factor of pool->sectors_per_block. | |
3530 | */ | |
3531 | if (limits->max_hw_sectors & (limits->max_hw_sectors - 1)) | |
3532 | limits->max_sectors = rounddown_pow_of_two(limits->max_hw_sectors); | |
3533 | else | |
3534 | limits->max_sectors = limits->max_hw_sectors; | |
3535 | ||
3536 | if (limits->max_sectors < pool->sectors_per_block) { | |
3537 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
3538 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3539 | limits->max_sectors--; | |
3540 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3541 | } | |
3542 | } else if (block_size_is_power_of_two(pool)) { | |
3543 | /* max_sectors_kb is >= power-of-2 thinp blocksize */ | |
3544 | while (!is_factor(limits->max_sectors, pool->sectors_per_block)) { | |
3545 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3546 | limits->max_sectors--; | |
3547 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3548 | } | |
3549 | } | |
991d9fa0 | 3550 | |
0cc67cd9 MS |
3551 | /* |
3552 | * If the system-determined stacked limits are compatible with the | |
3553 | * pool's blocksize (io_opt is a factor) do not override them. | |
3554 | */ | |
3555 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
3556 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
3557 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
3558 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
3559 | else | |
3560 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
3561 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
3562 | } | |
0424caa1 MS |
3563 | |
3564 | /* | |
3565 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3566 | * They get transferred to the live pool in bind_control_target() | |
3567 | * called from pool_preresume(). | |
3568 | */ | |
b60ab990 MS |
3569 | if (!pt->adjusted_pf.discard_enabled) { |
3570 | /* | |
3571 | * Must explicitly disallow stacking discard limits otherwise the | |
3572 | * block layer will stack them if pool's data device has support. | |
3573 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3574 | * user to see that, so make sure to set all discard limits to 0. | |
3575 | */ | |
3576 | limits->discard_granularity = 0; | |
0424caa1 | 3577 | return; |
b60ab990 | 3578 | } |
0424caa1 MS |
3579 | |
3580 | disable_passdown_if_not_supported(pt); | |
3581 | ||
3582 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3583 | } |
3584 | ||
3585 | static struct target_type pool_target = { | |
3586 | .name = "thin-pool", | |
3587 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3588 | DM_TARGET_IMMUTABLE, | |
36f12aeb | 3589 | .version = {1, 14, 0}, |
991d9fa0 JT |
3590 | .module = THIS_MODULE, |
3591 | .ctr = pool_ctr, | |
3592 | .dtr = pool_dtr, | |
3593 | .map = pool_map, | |
3594 | .postsuspend = pool_postsuspend, | |
3595 | .preresume = pool_preresume, | |
3596 | .resume = pool_resume, | |
3597 | .message = pool_message, | |
3598 | .status = pool_status, | |
3599 | .merge = pool_merge, | |
3600 | .iterate_devices = pool_iterate_devices, | |
3601 | .io_hints = pool_io_hints, | |
3602 | }; | |
3603 | ||
3604 | /*---------------------------------------------------------------- | |
3605 | * Thin target methods | |
3606 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3607 | static void thin_get(struct thin_c *tc) |
3608 | { | |
3609 | atomic_inc(&tc->refcount); | |
3610 | } | |
3611 | ||
3612 | static void thin_put(struct thin_c *tc) | |
3613 | { | |
3614 | if (atomic_dec_and_test(&tc->refcount)) | |
3615 | complete(&tc->can_destroy); | |
3616 | } | |
3617 | ||
991d9fa0 JT |
3618 | static void thin_dtr(struct dm_target *ti) |
3619 | { | |
3620 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3621 | unsigned long flags; |
3622 | ||
b10ebd34 JT |
3623 | thin_put(tc); |
3624 | wait_for_completion(&tc->can_destroy); | |
3625 | ||
c140e1c4 MS |
3626 | spin_lock_irqsave(&tc->pool->lock, flags); |
3627 | list_del_rcu(&tc->list); | |
3628 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3629 | synchronize_rcu(); | |
991d9fa0 JT |
3630 | |
3631 | mutex_lock(&dm_thin_pool_table.mutex); | |
3632 | ||
3633 | __pool_dec(tc->pool); | |
3634 | dm_pool_close_thin_device(tc->td); | |
3635 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3636 | if (tc->origin_dev) |
3637 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3638 | kfree(tc); |
3639 | ||
3640 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3641 | } | |
3642 | ||
3643 | /* | |
3644 | * Thin target parameters: | |
3645 | * | |
2dd9c257 | 3646 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3647 | * |
3648 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3649 | * dev_id: the internal device identifier | |
2dd9c257 | 3650 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3651 | * |
3652 | * If the pool device has discards disabled, they get disabled for the thin | |
3653 | * device as well. | |
991d9fa0 JT |
3654 | */ |
3655 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3656 | { | |
3657 | int r; | |
3658 | struct thin_c *tc; | |
2dd9c257 | 3659 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3660 | struct mapped_device *pool_md; |
5e3283e2 | 3661 | unsigned long flags; |
991d9fa0 JT |
3662 | |
3663 | mutex_lock(&dm_thin_pool_table.mutex); | |
3664 | ||
2dd9c257 | 3665 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3666 | ti->error = "Invalid argument count"; |
3667 | r = -EINVAL; | |
3668 | goto out_unlock; | |
3669 | } | |
3670 | ||
3671 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3672 | if (!tc) { | |
3673 | ti->error = "Out of memory"; | |
3674 | r = -ENOMEM; | |
3675 | goto out_unlock; | |
3676 | } | |
c140e1c4 | 3677 | spin_lock_init(&tc->lock); |
a374bb21 | 3678 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
3679 | bio_list_init(&tc->deferred_bio_list); |
3680 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3681 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3682 | |
2dd9c257 JT |
3683 | if (argc == 3) { |
3684 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3685 | if (r) { | |
3686 | ti->error = "Error opening origin device"; | |
3687 | goto bad_origin_dev; | |
3688 | } | |
3689 | tc->origin_dev = origin_dev; | |
3690 | } | |
3691 | ||
991d9fa0 JT |
3692 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3693 | if (r) { | |
3694 | ti->error = "Error opening pool device"; | |
3695 | goto bad_pool_dev; | |
3696 | } | |
3697 | tc->pool_dev = pool_dev; | |
3698 | ||
3699 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3700 | ti->error = "Invalid device id"; | |
3701 | r = -EINVAL; | |
3702 | goto bad_common; | |
3703 | } | |
3704 | ||
3705 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3706 | if (!pool_md) { | |
3707 | ti->error = "Couldn't get pool mapped device"; | |
3708 | r = -EINVAL; | |
3709 | goto bad_common; | |
3710 | } | |
3711 | ||
3712 | tc->pool = __pool_table_lookup(pool_md); | |
3713 | if (!tc->pool) { | |
3714 | ti->error = "Couldn't find pool object"; | |
3715 | r = -EINVAL; | |
3716 | goto bad_pool_lookup; | |
3717 | } | |
3718 | __pool_inc(tc->pool); | |
3719 | ||
e49e5829 JT |
3720 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3721 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3722 | r = -EINVAL; |
e49e5829 JT |
3723 | goto bad_thin_open; |
3724 | } | |
3725 | ||
991d9fa0 JT |
3726 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3727 | if (r) { | |
3728 | ti->error = "Couldn't open thin internal device"; | |
3729 | goto bad_thin_open; | |
3730 | } | |
3731 | ||
542f9038 MS |
3732 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3733 | if (r) | |
1acacc07 | 3734 | goto bad_target_max_io_len; |
542f9038 | 3735 | |
55a62eef | 3736 | ti->num_flush_bios = 1; |
16ad3d10 | 3737 | ti->flush_supported = true; |
59c3d2c6 | 3738 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3739 | |
3740 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3741 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3742 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3743 | ti->discards_supported = true; |
55a62eef | 3744 | ti->num_discard_bios = 1; |
55a62eef AK |
3745 | /* Discard bios must be split on a block boundary */ |
3746 | ti->split_discard_bios = true; | |
67e2e2b2 | 3747 | } |
991d9fa0 JT |
3748 | |
3749 | dm_put(pool_md); | |
3750 | ||
3751 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3752 | ||
b10ebd34 JT |
3753 | atomic_set(&tc->refcount, 1); |
3754 | init_completion(&tc->can_destroy); | |
3755 | ||
5e3283e2 | 3756 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3757 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3758 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3759 | /* |
3760 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3761 | * wake_worker() call finding no bios to process (because the newly | |
3762 | * added tc isn't yet visible). So this reduces latency since we | |
3763 | * aren't then dependent on the periodic commit to wake_worker(). | |
3764 | */ | |
3765 | synchronize_rcu(); | |
3766 | ||
991d9fa0 JT |
3767 | return 0; |
3768 | ||
1acacc07 MS |
3769 | bad_target_max_io_len: |
3770 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3771 | bad_thin_open: |
3772 | __pool_dec(tc->pool); | |
3773 | bad_pool_lookup: | |
3774 | dm_put(pool_md); | |
3775 | bad_common: | |
3776 | dm_put_device(ti, tc->pool_dev); | |
3777 | bad_pool_dev: | |
2dd9c257 JT |
3778 | if (tc->origin_dev) |
3779 | dm_put_device(ti, tc->origin_dev); | |
3780 | bad_origin_dev: | |
991d9fa0 JT |
3781 | kfree(tc); |
3782 | out_unlock: | |
3783 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3784 | ||
3785 | return r; | |
3786 | } | |
3787 | ||
7de3ee57 | 3788 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3789 | { |
4f024f37 | 3790 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3791 | |
7de3ee57 | 3792 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3793 | } |
3794 | ||
7de3ee57 | 3795 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3796 | { |
3797 | unsigned long flags; | |
59c3d2c6 | 3798 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3799 | struct list_head work; |
a24c2569 | 3800 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3801 | struct pool *pool = h->tc->pool; |
3802 | ||
3803 | if (h->shared_read_entry) { | |
3804 | INIT_LIST_HEAD(&work); | |
44feb387 | 3805 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3806 | |
3807 | spin_lock_irqsave(&pool->lock, flags); | |
3808 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3809 | list_del(&m->list); | |
50f3c3ef | 3810 | __complete_mapping_preparation(m); |
eb2aa48d JT |
3811 | } |
3812 | spin_unlock_irqrestore(&pool->lock, flags); | |
3813 | } | |
3814 | ||
104655fd JT |
3815 | if (h->all_io_entry) { |
3816 | INIT_LIST_HEAD(&work); | |
44feb387 | 3817 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3818 | if (!list_empty(&work)) { |
3819 | spin_lock_irqsave(&pool->lock, flags); | |
3820 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3821 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3822 | spin_unlock_irqrestore(&pool->lock, flags); |
3823 | wake_worker(pool); | |
3824 | } | |
104655fd JT |
3825 | } |
3826 | ||
eb2aa48d JT |
3827 | return 0; |
3828 | } | |
3829 | ||
738211f7 | 3830 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3831 | { |
738211f7 JT |
3832 | struct thin_c *tc = ti->private; |
3833 | ||
991d9fa0 | 3834 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3835 | noflush_work(tc, do_noflush_start); |
3836 | } | |
3837 | ||
3838 | static void thin_postsuspend(struct dm_target *ti) | |
3839 | { | |
3840 | struct thin_c *tc = ti->private; | |
3841 | ||
3842 | /* | |
3843 | * The dm_noflush_suspending flag has been cleared by now, so | |
3844 | * unfortunately we must always run this. | |
3845 | */ | |
3846 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3847 | } |
3848 | ||
e5aea7b4 JT |
3849 | static int thin_preresume(struct dm_target *ti) |
3850 | { | |
3851 | struct thin_c *tc = ti->private; | |
3852 | ||
3853 | if (tc->origin_dev) | |
3854 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
3855 | ||
3856 | return 0; | |
3857 | } | |
3858 | ||
991d9fa0 JT |
3859 | /* |
3860 | * <nr mapped sectors> <highest mapped sector> | |
3861 | */ | |
fd7c092e MP |
3862 | static void thin_status(struct dm_target *ti, status_type_t type, |
3863 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3864 | { |
3865 | int r; | |
3866 | ssize_t sz = 0; | |
3867 | dm_block_t mapped, highest; | |
3868 | char buf[BDEVNAME_SIZE]; | |
3869 | struct thin_c *tc = ti->private; | |
3870 | ||
e49e5829 JT |
3871 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3872 | DMEMIT("Fail"); | |
fd7c092e | 3873 | return; |
e49e5829 JT |
3874 | } |
3875 | ||
991d9fa0 JT |
3876 | if (!tc->td) |
3877 | DMEMIT("-"); | |
3878 | else { | |
3879 | switch (type) { | |
3880 | case STATUSTYPE_INFO: | |
3881 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3882 | if (r) { |
3883 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3884 | goto err; | |
3885 | } | |
991d9fa0 JT |
3886 | |
3887 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3888 | if (r < 0) { |
3889 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3890 | goto err; | |
3891 | } | |
991d9fa0 JT |
3892 | |
3893 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3894 | if (r) | |
3895 | DMEMIT("%llu", ((highest + 1) * | |
3896 | tc->pool->sectors_per_block) - 1); | |
3897 | else | |
3898 | DMEMIT("-"); | |
3899 | break; | |
3900 | ||
3901 | case STATUSTYPE_TABLE: | |
3902 | DMEMIT("%s %lu", | |
3903 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3904 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3905 | if (tc->origin_dev) |
3906 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3907 | break; |
3908 | } | |
3909 | } | |
3910 | ||
fd7c092e MP |
3911 | return; |
3912 | ||
3913 | err: | |
3914 | DMEMIT("Error"); | |
991d9fa0 JT |
3915 | } |
3916 | ||
36f12aeb MS |
3917 | static int thin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
3918 | struct bio_vec *biovec, int max_size) | |
3919 | { | |
3920 | struct thin_c *tc = ti->private; | |
3921 | struct request_queue *q = bdev_get_queue(tc->pool_dev->bdev); | |
3922 | ||
3923 | if (!q->merge_bvec_fn) | |
3924 | return max_size; | |
3925 | ||
3926 | bvm->bi_bdev = tc->pool_dev->bdev; | |
3927 | bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector); | |
3928 | ||
3929 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3930 | } | |
3931 | ||
991d9fa0 JT |
3932 | static int thin_iterate_devices(struct dm_target *ti, |
3933 | iterate_devices_callout_fn fn, void *data) | |
3934 | { | |
55f2b8bd | 3935 | sector_t blocks; |
991d9fa0 | 3936 | struct thin_c *tc = ti->private; |
55f2b8bd | 3937 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3938 | |
3939 | /* | |
3940 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3941 | * we follow a more convoluted path through to the pool's target. | |
3942 | */ | |
55f2b8bd | 3943 | if (!pool->ti) |
991d9fa0 JT |
3944 | return 0; /* nothing is bound */ |
3945 | ||
55f2b8bd MS |
3946 | blocks = pool->ti->len; |
3947 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3948 | if (blocks) |
55f2b8bd | 3949 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3950 | |
3951 | return 0; | |
3952 | } | |
3953 | ||
991d9fa0 JT |
3954 | static struct target_type thin_target = { |
3955 | .name = "thin", | |
36f12aeb | 3956 | .version = {1, 14, 0}, |
991d9fa0 JT |
3957 | .module = THIS_MODULE, |
3958 | .ctr = thin_ctr, | |
3959 | .dtr = thin_dtr, | |
3960 | .map = thin_map, | |
eb2aa48d | 3961 | .end_io = thin_endio, |
e5aea7b4 | 3962 | .preresume = thin_preresume, |
738211f7 | 3963 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3964 | .postsuspend = thin_postsuspend, |
3965 | .status = thin_status, | |
36f12aeb | 3966 | .merge = thin_merge, |
991d9fa0 | 3967 | .iterate_devices = thin_iterate_devices, |
991d9fa0 JT |
3968 | }; |
3969 | ||
3970 | /*----------------------------------------------------------------*/ | |
3971 | ||
3972 | static int __init dm_thin_init(void) | |
3973 | { | |
3974 | int r; | |
3975 | ||
3976 | pool_table_init(); | |
3977 | ||
3978 | r = dm_register_target(&thin_target); | |
3979 | if (r) | |
3980 | return r; | |
3981 | ||
3982 | r = dm_register_target(&pool_target); | |
3983 | if (r) | |
a24c2569 MS |
3984 | goto bad_pool_target; |
3985 | ||
3986 | r = -ENOMEM; | |
3987 | ||
a24c2569 MS |
3988 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3989 | if (!_new_mapping_cache) | |
3990 | goto bad_new_mapping_cache; | |
3991 | ||
a24c2569 MS |
3992 | return 0; |
3993 | ||
a24c2569 | 3994 | bad_new_mapping_cache: |
a24c2569 MS |
3995 | dm_unregister_target(&pool_target); |
3996 | bad_pool_target: | |
3997 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3998 | |
3999 | return r; | |
4000 | } | |
4001 | ||
4002 | static void dm_thin_exit(void) | |
4003 | { | |
4004 | dm_unregister_target(&thin_target); | |
4005 | dm_unregister_target(&pool_target); | |
a24c2569 | 4006 | |
a24c2569 | 4007 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
4008 | } |
4009 | ||
4010 | module_init(dm_thin_init); | |
4011 | module_exit(dm_thin_exit); | |
4012 | ||
80c57893 MS |
4013 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
4014 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
4015 | ||
7cab8bf1 | 4016 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4017 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4018 | MODULE_LICENSE("GPL"); |