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> | |
14 | #include <linux/list.h> | |
c140e1c4 | 15 | #include <linux/rculist.h> |
991d9fa0 JT |
16 | #include <linux/init.h> |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
67324ea1 | 19 | #include <linux/rbtree.h> |
991d9fa0 JT |
20 | |
21 | #define DM_MSG_PREFIX "thin" | |
22 | ||
23 | /* | |
24 | * Tunable constants | |
25 | */ | |
7768ed33 | 26 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
27 | #define MAPPING_POOL_SIZE 1024 |
28 | #define PRISON_CELLS 1024 | |
905e51b3 | 29 | #define COMMIT_PERIOD HZ |
85ad643b | 30 | #define NO_SPACE_TIMEOUT (HZ * 60) |
991d9fa0 | 31 | |
df5d2e90 MP |
32 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
33 | "A percentage of time allocated for copy on write"); | |
34 | ||
991d9fa0 JT |
35 | /* |
36 | * The block size of the device holding pool data must be | |
37 | * between 64KB and 1GB. | |
38 | */ | |
39 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
40 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
41 | ||
991d9fa0 JT |
42 | /* |
43 | * Device id is restricted to 24 bits. | |
44 | */ | |
45 | #define MAX_DEV_ID ((1 << 24) - 1) | |
46 | ||
47 | /* | |
48 | * How do we handle breaking sharing of data blocks? | |
49 | * ================================================= | |
50 | * | |
51 | * We use a standard copy-on-write btree to store the mappings for the | |
52 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
53 | * the data). When you take an internal snapshot you clone the root node | |
54 | * of the origin btree. After this there is no concept of an origin or a | |
55 | * snapshot. They are just two device trees that happen to point to the | |
56 | * same data blocks. | |
57 | * | |
58 | * When we get a write in we decide if it's to a shared data block using | |
59 | * some timestamp magic. If it is, we have to break sharing. | |
60 | * | |
61 | * Let's say we write to a shared block in what was the origin. The | |
62 | * steps are: | |
63 | * | |
64 | * i) plug io further to this physical block. (see bio_prison code). | |
65 | * | |
66 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 67 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
68 | * |
69 | * iii) copy the data block to a newly allocate block. This step can be | |
70 | * missed out if the io covers the block. (schedule_copy). | |
71 | * | |
72 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 73 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
74 | * sharing of btree nodes between the two devices. Breaking sharing only |
75 | * effects the btree of that specific device. Btrees for the other | |
76 | * devices that share the block never change. The btree for the origin | |
77 | * device as it was after the last commit is untouched, ie. we're using | |
78 | * persistent data structures in the functional programming sense. | |
79 | * | |
80 | * v) unplug io to this physical block, including the io that triggered | |
81 | * the breaking of sharing. | |
82 | * | |
83 | * Steps (ii) and (iii) occur in parallel. | |
84 | * | |
85 | * The metadata _doesn't_ need to be committed before the io continues. We | |
86 | * get away with this because the io is always written to a _new_ block. | |
87 | * If there's a crash, then: | |
88 | * | |
89 | * - The origin mapping will point to the old origin block (the shared | |
90 | * one). This will contain the data as it was before the io that triggered | |
91 | * the breaking of sharing came in. | |
92 | * | |
93 | * - The snap mapping still points to the old block. As it would after | |
94 | * the commit. | |
95 | * | |
96 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
97 | * will continue to think that data block in the snapshot device is shared | |
98 | * even after the write to the origin has broken sharing. I suspect data | |
99 | * blocks will typically be shared by many different devices, so we're | |
100 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
101 | * devices that reference this data block. At the moment I think the | |
102 | * benefits far, far outweigh the disadvantages. | |
103 | */ | |
104 | ||
105 | /*----------------------------------------------------------------*/ | |
106 | ||
991d9fa0 JT |
107 | /* |
108 | * Key building. | |
109 | */ | |
110 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 111 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
112 | { |
113 | key->virtual = 0; | |
114 | key->dev = dm_thin_dev_id(td); | |
115 | key->block = b; | |
116 | } | |
117 | ||
118 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 119 | struct dm_cell_key *key) |
991d9fa0 JT |
120 | { |
121 | key->virtual = 1; | |
122 | key->dev = dm_thin_dev_id(td); | |
123 | key->block = b; | |
124 | } | |
125 | ||
126 | /*----------------------------------------------------------------*/ | |
127 | ||
128 | /* | |
129 | * A pool device ties together a metadata device and a data device. It | |
130 | * also provides the interface for creating and destroying internal | |
131 | * devices. | |
132 | */ | |
a24c2569 | 133 | struct dm_thin_new_mapping; |
67e2e2b2 | 134 | |
e49e5829 | 135 | /* |
3e1a0699 | 136 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
137 | */ |
138 | enum pool_mode { | |
139 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 140 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
141 | PM_READ_ONLY, /* metadata may not be changed */ |
142 | PM_FAIL, /* all I/O fails */ | |
143 | }; | |
144 | ||
67e2e2b2 | 145 | struct pool_features { |
e49e5829 JT |
146 | enum pool_mode mode; |
147 | ||
9bc142dd MS |
148 | bool zero_new_blocks:1; |
149 | bool discard_enabled:1; | |
150 | bool discard_passdown:1; | |
787a996c | 151 | bool error_if_no_space:1; |
67e2e2b2 JT |
152 | }; |
153 | ||
e49e5829 JT |
154 | struct thin_c; |
155 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
156 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
157 | ||
991d9fa0 JT |
158 | struct pool { |
159 | struct list_head list; | |
160 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
161 | ||
162 | struct mapped_device *pool_md; | |
163 | struct block_device *md_dev; | |
164 | struct dm_pool_metadata *pmd; | |
165 | ||
991d9fa0 | 166 | dm_block_t low_water_blocks; |
55f2b8bd | 167 | uint32_t sectors_per_block; |
f9a8e0cd | 168 | int sectors_per_block_shift; |
991d9fa0 | 169 | |
67e2e2b2 | 170 | struct pool_features pf; |
88a6621b | 171 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 172 | |
44feb387 | 173 | struct dm_bio_prison *prison; |
991d9fa0 JT |
174 | struct dm_kcopyd_client *copier; |
175 | ||
176 | struct workqueue_struct *wq; | |
177 | struct work_struct worker; | |
905e51b3 | 178 | struct delayed_work waker; |
85ad643b | 179 | struct delayed_work no_space_timeout; |
991d9fa0 | 180 | |
905e51b3 | 181 | unsigned long last_commit_jiffies; |
55f2b8bd | 182 | unsigned ref_count; |
991d9fa0 JT |
183 | |
184 | spinlock_t lock; | |
991d9fa0 JT |
185 | struct bio_list deferred_flush_bios; |
186 | struct list_head prepared_mappings; | |
104655fd | 187 | struct list_head prepared_discards; |
c140e1c4 | 188 | struct list_head active_thins; |
991d9fa0 | 189 | |
44feb387 MS |
190 | struct dm_deferred_set *shared_read_ds; |
191 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 192 | |
a24c2569 | 193 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 194 | mempool_t *mapping_pool; |
e49e5829 JT |
195 | |
196 | process_bio_fn process_bio; | |
197 | process_bio_fn process_discard; | |
198 | ||
199 | process_mapping_fn process_prepared_mapping; | |
200 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
201 | }; |
202 | ||
e49e5829 | 203 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 204 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 205 | |
991d9fa0 JT |
206 | /* |
207 | * Target context for a pool. | |
208 | */ | |
209 | struct pool_c { | |
210 | struct dm_target *ti; | |
211 | struct pool *pool; | |
212 | struct dm_dev *data_dev; | |
213 | struct dm_dev *metadata_dev; | |
214 | struct dm_target_callbacks callbacks; | |
215 | ||
216 | dm_block_t low_water_blocks; | |
0424caa1 MS |
217 | struct pool_features requested_pf; /* Features requested during table load */ |
218 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
219 | }; |
220 | ||
221 | /* | |
222 | * Target context for a thin. | |
223 | */ | |
224 | struct thin_c { | |
c140e1c4 | 225 | struct list_head list; |
991d9fa0 | 226 | struct dm_dev *pool_dev; |
2dd9c257 | 227 | struct dm_dev *origin_dev; |
991d9fa0 JT |
228 | dm_thin_id dev_id; |
229 | ||
230 | struct pool *pool; | |
231 | struct dm_thin_device *td; | |
738211f7 | 232 | bool requeue_mode:1; |
c140e1c4 MS |
233 | spinlock_t lock; |
234 | struct bio_list deferred_bio_list; | |
235 | struct bio_list retry_on_resume_list; | |
67324ea1 | 236 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
237 | |
238 | /* | |
239 | * Ensures the thin is not destroyed until the worker has finished | |
240 | * iterating the active_thins list. | |
241 | */ | |
242 | atomic_t refcount; | |
243 | struct completion can_destroy; | |
991d9fa0 JT |
244 | }; |
245 | ||
246 | /*----------------------------------------------------------------*/ | |
247 | ||
025b9685 JT |
248 | /* |
249 | * wake_worker() is used when new work is queued and when pool_resume is | |
250 | * ready to continue deferred IO processing. | |
251 | */ | |
252 | static void wake_worker(struct pool *pool) | |
253 | { | |
254 | queue_work(pool->wq, &pool->worker); | |
255 | } | |
256 | ||
257 | /*----------------------------------------------------------------*/ | |
258 | ||
6beca5eb JT |
259 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
260 | struct dm_bio_prison_cell **cell_result) | |
261 | { | |
262 | int r; | |
263 | struct dm_bio_prison_cell *cell_prealloc; | |
264 | ||
265 | /* | |
266 | * Allocate a cell from the prison's mempool. | |
267 | * This might block but it can't fail. | |
268 | */ | |
269 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
270 | ||
271 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
272 | if (r) | |
273 | /* | |
274 | * We reused an old cell; we can get rid of | |
275 | * the new one. | |
276 | */ | |
277 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
278 | ||
279 | return r; | |
280 | } | |
281 | ||
282 | static void cell_release(struct pool *pool, | |
283 | struct dm_bio_prison_cell *cell, | |
284 | struct bio_list *bios) | |
285 | { | |
286 | dm_cell_release(pool->prison, cell, bios); | |
287 | dm_bio_prison_free_cell(pool->prison, cell); | |
288 | } | |
289 | ||
290 | static void cell_release_no_holder(struct pool *pool, | |
291 | struct dm_bio_prison_cell *cell, | |
292 | struct bio_list *bios) | |
293 | { | |
294 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
295 | dm_bio_prison_free_cell(pool->prison, cell); | |
296 | } | |
297 | ||
025b9685 JT |
298 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
299 | struct dm_bio_prison_cell *cell) | |
300 | { | |
301 | struct pool *pool = tc->pool; | |
302 | unsigned long flags; | |
303 | ||
c140e1c4 MS |
304 | spin_lock_irqsave(&tc->lock, flags); |
305 | dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list); | |
306 | spin_unlock_irqrestore(&tc->lock, flags); | |
025b9685 JT |
307 | |
308 | wake_worker(pool); | |
309 | } | |
310 | ||
6beca5eb JT |
311 | static void cell_error(struct pool *pool, |
312 | struct dm_bio_prison_cell *cell) | |
313 | { | |
314 | dm_cell_error(pool->prison, cell); | |
315 | dm_bio_prison_free_cell(pool->prison, cell); | |
316 | } | |
317 | ||
318 | /*----------------------------------------------------------------*/ | |
319 | ||
991d9fa0 JT |
320 | /* |
321 | * A global list of pools that uses a struct mapped_device as a key. | |
322 | */ | |
323 | static struct dm_thin_pool_table { | |
324 | struct mutex mutex; | |
325 | struct list_head pools; | |
326 | } dm_thin_pool_table; | |
327 | ||
328 | static void pool_table_init(void) | |
329 | { | |
330 | mutex_init(&dm_thin_pool_table.mutex); | |
331 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
332 | } | |
333 | ||
334 | static void __pool_table_insert(struct pool *pool) | |
335 | { | |
336 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
337 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
338 | } | |
339 | ||
340 | static void __pool_table_remove(struct pool *pool) | |
341 | { | |
342 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
343 | list_del(&pool->list); | |
344 | } | |
345 | ||
346 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
347 | { | |
348 | struct pool *pool = NULL, *tmp; | |
349 | ||
350 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
351 | ||
352 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
353 | if (tmp->pool_md == md) { | |
354 | pool = tmp; | |
355 | break; | |
356 | } | |
357 | } | |
358 | ||
359 | return pool; | |
360 | } | |
361 | ||
362 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
363 | { | |
364 | struct pool *pool = NULL, *tmp; | |
365 | ||
366 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
367 | ||
368 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
369 | if (tmp->md_dev == md_dev) { | |
370 | pool = tmp; | |
371 | break; | |
372 | } | |
373 | } | |
374 | ||
375 | return pool; | |
376 | } | |
377 | ||
378 | /*----------------------------------------------------------------*/ | |
379 | ||
a24c2569 | 380 | struct dm_thin_endio_hook { |
eb2aa48d | 381 | struct thin_c *tc; |
44feb387 MS |
382 | struct dm_deferred_entry *shared_read_entry; |
383 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 384 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 385 | struct rb_node rb_node; |
eb2aa48d JT |
386 | }; |
387 | ||
18adc577 | 388 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
389 | { |
390 | struct bio *bio; | |
391 | struct bio_list bios; | |
18adc577 | 392 | unsigned long flags; |
991d9fa0 JT |
393 | |
394 | bio_list_init(&bios); | |
18adc577 | 395 | |
c140e1c4 | 396 | spin_lock_irqsave(&tc->lock, flags); |
991d9fa0 JT |
397 | bio_list_merge(&bios, master); |
398 | bio_list_init(master); | |
c140e1c4 | 399 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 400 | |
c140e1c4 MS |
401 | while ((bio = bio_list_pop(&bios))) |
402 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
991d9fa0 JT |
403 | } |
404 | ||
405 | static void requeue_io(struct thin_c *tc) | |
406 | { | |
c140e1c4 MS |
407 | requeue_bio_list(tc, &tc->deferred_bio_list); |
408 | requeue_bio_list(tc, &tc->retry_on_resume_list); | |
991d9fa0 JT |
409 | } |
410 | ||
c140e1c4 | 411 | static void error_thin_retry_list(struct thin_c *tc) |
3e1a0699 JT |
412 | { |
413 | struct bio *bio; | |
414 | unsigned long flags; | |
415 | struct bio_list bios; | |
416 | ||
417 | bio_list_init(&bios); | |
418 | ||
c140e1c4 MS |
419 | spin_lock_irqsave(&tc->lock, flags); |
420 | bio_list_merge(&bios, &tc->retry_on_resume_list); | |
421 | bio_list_init(&tc->retry_on_resume_list); | |
422 | spin_unlock_irqrestore(&tc->lock, flags); | |
3e1a0699 JT |
423 | |
424 | while ((bio = bio_list_pop(&bios))) | |
425 | bio_io_error(bio); | |
426 | } | |
427 | ||
c140e1c4 MS |
428 | static void error_retry_list(struct pool *pool) |
429 | { | |
430 | struct thin_c *tc; | |
431 | ||
432 | rcu_read_lock(); | |
433 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
434 | error_thin_retry_list(tc); | |
435 | rcu_read_unlock(); | |
436 | } | |
437 | ||
991d9fa0 JT |
438 | /* |
439 | * This section of code contains the logic for processing a thin device's IO. | |
440 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
441 | * but most is exclusively called from the thin target rather than the thin-pool | |
442 | * target. | |
443 | */ | |
444 | ||
58f77a21 MS |
445 | static bool block_size_is_power_of_two(struct pool *pool) |
446 | { | |
447 | return pool->sectors_per_block_shift >= 0; | |
448 | } | |
449 | ||
991d9fa0 JT |
450 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
451 | { | |
58f77a21 | 452 | struct pool *pool = tc->pool; |
4f024f37 | 453 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 454 | |
58f77a21 MS |
455 | if (block_size_is_power_of_two(pool)) |
456 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 457 | else |
58f77a21 | 458 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
459 | |
460 | return block_nr; | |
991d9fa0 JT |
461 | } |
462 | ||
463 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
464 | { | |
465 | struct pool *pool = tc->pool; | |
4f024f37 | 466 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
467 | |
468 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 469 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
470 | bio->bi_iter.bi_sector = |
471 | (block << pool->sectors_per_block_shift) | | |
472 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 473 | else |
4f024f37 | 474 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 475 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
476 | } |
477 | ||
2dd9c257 JT |
478 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
479 | { | |
480 | bio->bi_bdev = tc->origin_dev->bdev; | |
481 | } | |
482 | ||
4afdd680 JT |
483 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
484 | { | |
485 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
486 | dm_thin_changed_this_transaction(tc->td); | |
487 | } | |
488 | ||
e8088073 JT |
489 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
490 | { | |
491 | struct dm_thin_endio_hook *h; | |
492 | ||
493 | if (bio->bi_rw & REQ_DISCARD) | |
494 | return; | |
495 | ||
59c3d2c6 | 496 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
497 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
498 | } | |
499 | ||
2dd9c257 | 500 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
501 | { |
502 | struct pool *pool = tc->pool; | |
503 | unsigned long flags; | |
504 | ||
e49e5829 JT |
505 | if (!bio_triggers_commit(tc, bio)) { |
506 | generic_make_request(bio); | |
507 | return; | |
508 | } | |
509 | ||
991d9fa0 | 510 | /* |
e49e5829 JT |
511 | * Complete bio with an error if earlier I/O caused changes to |
512 | * the metadata that can't be committed e.g, due to I/O errors | |
513 | * on the metadata device. | |
991d9fa0 | 514 | */ |
e49e5829 JT |
515 | if (dm_thin_aborted_changes(tc->td)) { |
516 | bio_io_error(bio); | |
517 | return; | |
518 | } | |
519 | ||
520 | /* | |
521 | * Batch together any bios that trigger commits and then issue a | |
522 | * single commit for them in process_deferred_bios(). | |
523 | */ | |
524 | spin_lock_irqsave(&pool->lock, flags); | |
525 | bio_list_add(&pool->deferred_flush_bios, bio); | |
526 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
527 | } |
528 | ||
2dd9c257 JT |
529 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
530 | { | |
531 | remap_to_origin(tc, bio); | |
532 | issue(tc, bio); | |
533 | } | |
534 | ||
535 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
536 | dm_block_t block) | |
537 | { | |
538 | remap(tc, bio, block); | |
539 | issue(tc, bio); | |
540 | } | |
541 | ||
991d9fa0 JT |
542 | /*----------------------------------------------------------------*/ |
543 | ||
544 | /* | |
545 | * Bio endio functions. | |
546 | */ | |
a24c2569 | 547 | struct dm_thin_new_mapping { |
991d9fa0 JT |
548 | struct list_head list; |
549 | ||
7f214665 MS |
550 | bool quiesced:1; |
551 | bool prepared:1; | |
552 | bool pass_discard:1; | |
553 | bool definitely_not_shared:1; | |
991d9fa0 | 554 | |
7f214665 | 555 | int err; |
991d9fa0 JT |
556 | struct thin_c *tc; |
557 | dm_block_t virt_block; | |
558 | dm_block_t data_block; | |
a24c2569 | 559 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
560 | |
561 | /* | |
562 | * If the bio covers the whole area of a block then we can avoid | |
563 | * zeroing or copying. Instead this bio is hooked. The bio will | |
564 | * still be in the cell, so care has to be taken to avoid issuing | |
565 | * the bio twice. | |
566 | */ | |
567 | struct bio *bio; | |
568 | bio_end_io_t *saved_bi_end_io; | |
569 | }; | |
570 | ||
a24c2569 | 571 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
572 | { |
573 | struct pool *pool = m->tc->pool; | |
574 | ||
eb2aa48d | 575 | if (m->quiesced && m->prepared) { |
daec338b | 576 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
577 | wake_worker(pool); |
578 | } | |
579 | } | |
580 | ||
581 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
582 | { | |
583 | unsigned long flags; | |
a24c2569 | 584 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
585 | struct pool *pool = m->tc->pool; |
586 | ||
587 | m->err = read_err || write_err ? -EIO : 0; | |
588 | ||
589 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 590 | m->prepared = true; |
991d9fa0 JT |
591 | __maybe_add_mapping(m); |
592 | spin_unlock_irqrestore(&pool->lock, flags); | |
593 | } | |
594 | ||
595 | static void overwrite_endio(struct bio *bio, int err) | |
596 | { | |
597 | unsigned long flags; | |
59c3d2c6 | 598 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 599 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
600 | struct pool *pool = m->tc->pool; |
601 | ||
602 | m->err = err; | |
603 | ||
604 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 605 | m->prepared = true; |
991d9fa0 JT |
606 | __maybe_add_mapping(m); |
607 | spin_unlock_irqrestore(&pool->lock, flags); | |
608 | } | |
609 | ||
991d9fa0 JT |
610 | /*----------------------------------------------------------------*/ |
611 | ||
612 | /* | |
613 | * Workqueue. | |
614 | */ | |
615 | ||
616 | /* | |
617 | * Prepared mapping jobs. | |
618 | */ | |
619 | ||
620 | /* | |
621 | * This sends the bios in the cell back to the deferred_bios list. | |
622 | */ | |
2aab3850 | 623 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
624 | { |
625 | struct pool *pool = tc->pool; | |
626 | unsigned long flags; | |
627 | ||
c140e1c4 MS |
628 | spin_lock_irqsave(&tc->lock, flags); |
629 | cell_release(pool, cell, &tc->deferred_bio_list); | |
630 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
631 | |
632 | wake_worker(pool); | |
633 | } | |
634 | ||
635 | /* | |
6beca5eb | 636 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 637 | */ |
f286ba0e | 638 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 639 | { |
991d9fa0 JT |
640 | struct pool *pool = tc->pool; |
641 | unsigned long flags; | |
642 | ||
c140e1c4 MS |
643 | spin_lock_irqsave(&tc->lock, flags); |
644 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
645 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
646 | |
647 | wake_worker(pool); | |
648 | } | |
649 | ||
e49e5829 JT |
650 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
651 | { | |
196d38bc | 652 | if (m->bio) { |
e49e5829 | 653 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
654 | atomic_inc(&m->bio->bi_remaining); |
655 | } | |
6beca5eb | 656 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
657 | list_del(&m->list); |
658 | mempool_free(m, m->tc->pool->mapping_pool); | |
659 | } | |
025b9685 | 660 | |
a24c2569 | 661 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
662 | { |
663 | struct thin_c *tc = m->tc; | |
6beca5eb | 664 | struct pool *pool = tc->pool; |
991d9fa0 JT |
665 | struct bio *bio; |
666 | int r; | |
667 | ||
668 | bio = m->bio; | |
196d38bc | 669 | if (bio) { |
991d9fa0 | 670 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
671 | atomic_inc(&bio->bi_remaining); |
672 | } | |
991d9fa0 JT |
673 | |
674 | if (m->err) { | |
6beca5eb | 675 | cell_error(pool, m->cell); |
905386f8 | 676 | goto out; |
991d9fa0 JT |
677 | } |
678 | ||
679 | /* | |
680 | * Commit the prepared block into the mapping btree. | |
681 | * Any I/O for this block arriving after this point will get | |
682 | * remapped to it directly. | |
683 | */ | |
684 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
685 | if (r) { | |
b5330655 | 686 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 687 | cell_error(pool, m->cell); |
905386f8 | 688 | goto out; |
991d9fa0 JT |
689 | } |
690 | ||
691 | /* | |
692 | * Release any bios held while the block was being provisioned. | |
693 | * If we are processing a write bio that completely covers the block, | |
694 | * we already processed it so can ignore it now when processing | |
695 | * the bios in the cell. | |
696 | */ | |
697 | if (bio) { | |
f286ba0e | 698 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
699 | bio_endio(bio, 0); |
700 | } else | |
2aab3850 | 701 | cell_defer(tc, m->cell); |
991d9fa0 | 702 | |
905386f8 | 703 | out: |
991d9fa0 | 704 | list_del(&m->list); |
6beca5eb | 705 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
706 | } |
707 | ||
e49e5829 | 708 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 709 | { |
104655fd JT |
710 | struct thin_c *tc = m->tc; |
711 | ||
e49e5829 | 712 | bio_io_error(m->bio); |
f286ba0e JT |
713 | cell_defer_no_holder(tc, m->cell); |
714 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
715 | mempool_free(m, tc->pool->mapping_pool); |
716 | } | |
717 | ||
718 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
719 | { | |
720 | struct thin_c *tc = m->tc; | |
104655fd | 721 | |
e8088073 | 722 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
723 | cell_defer_no_holder(tc, m->cell); |
724 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 725 | |
104655fd | 726 | if (m->pass_discard) |
19fa1a67 JT |
727 | if (m->definitely_not_shared) |
728 | remap_and_issue(tc, m->bio, m->data_block); | |
729 | else { | |
730 | bool used = false; | |
731 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
732 | bio_endio(m->bio, 0); | |
733 | else | |
734 | remap_and_issue(tc, m->bio, m->data_block); | |
735 | } | |
104655fd JT |
736 | else |
737 | bio_endio(m->bio, 0); | |
738 | ||
104655fd JT |
739 | mempool_free(m, tc->pool->mapping_pool); |
740 | } | |
741 | ||
e49e5829 JT |
742 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
743 | { | |
744 | int r; | |
745 | struct thin_c *tc = m->tc; | |
746 | ||
747 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
748 | if (r) | |
c397741c | 749 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
750 | |
751 | process_prepared_discard_passdown(m); | |
752 | } | |
753 | ||
104655fd | 754 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 755 | process_mapping_fn *fn) |
991d9fa0 JT |
756 | { |
757 | unsigned long flags; | |
758 | struct list_head maps; | |
a24c2569 | 759 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
760 | |
761 | INIT_LIST_HEAD(&maps); | |
762 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 763 | list_splice_init(head, &maps); |
991d9fa0 JT |
764 | spin_unlock_irqrestore(&pool->lock, flags); |
765 | ||
766 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 767 | (*fn)(m); |
991d9fa0 JT |
768 | } |
769 | ||
770 | /* | |
771 | * Deferred bio jobs. | |
772 | */ | |
104655fd | 773 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 774 | { |
4f024f37 KO |
775 | return bio->bi_iter.bi_size == |
776 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
777 | } |
778 | ||
779 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
780 | { | |
781 | return (bio_data_dir(bio) == WRITE) && | |
782 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
783 | } |
784 | ||
785 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
786 | bio_end_io_t *fn) | |
787 | { | |
788 | *save = bio->bi_end_io; | |
789 | bio->bi_end_io = fn; | |
790 | } | |
791 | ||
792 | static int ensure_next_mapping(struct pool *pool) | |
793 | { | |
794 | if (pool->next_mapping) | |
795 | return 0; | |
796 | ||
797 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
798 | ||
799 | return pool->next_mapping ? 0 : -ENOMEM; | |
800 | } | |
801 | ||
a24c2569 | 802 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 803 | { |
16961b04 | 804 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
805 | |
806 | BUG_ON(!pool->next_mapping); | |
807 | ||
16961b04 MS |
808 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
809 | INIT_LIST_HEAD(&m->list); | |
810 | m->bio = NULL; | |
811 | ||
991d9fa0 JT |
812 | pool->next_mapping = NULL; |
813 | ||
16961b04 | 814 | return m; |
991d9fa0 JT |
815 | } |
816 | ||
817 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
818 | struct dm_dev *origin, dm_block_t data_origin, |
819 | dm_block_t data_dest, | |
a24c2569 | 820 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
821 | { |
822 | int r; | |
823 | struct pool *pool = tc->pool; | |
a24c2569 | 824 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 825 | |
991d9fa0 JT |
826 | m->tc = tc; |
827 | m->virt_block = virt_block; | |
828 | m->data_block = data_dest; | |
829 | m->cell = cell; | |
991d9fa0 | 830 | |
44feb387 | 831 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 832 | m->quiesced = true; |
991d9fa0 JT |
833 | |
834 | /* | |
835 | * IO to pool_dev remaps to the pool target's data_dev. | |
836 | * | |
837 | * If the whole block of data is being overwritten, we can issue the | |
838 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
839 | */ | |
840 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 841 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 842 | |
eb2aa48d | 843 | h->overwrite_mapping = m; |
991d9fa0 JT |
844 | m->bio = bio; |
845 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 846 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
847 | remap_and_issue(tc, bio, data_dest); |
848 | } else { | |
849 | struct dm_io_region from, to; | |
850 | ||
2dd9c257 | 851 | from.bdev = origin->bdev; |
991d9fa0 JT |
852 | from.sector = data_origin * pool->sectors_per_block; |
853 | from.count = pool->sectors_per_block; | |
854 | ||
855 | to.bdev = tc->pool_dev->bdev; | |
856 | to.sector = data_dest * pool->sectors_per_block; | |
857 | to.count = pool->sectors_per_block; | |
858 | ||
859 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
860 | 0, copy_complete, m); | |
861 | if (r < 0) { | |
862 | mempool_free(m, pool->mapping_pool); | |
c397741c | 863 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 864 | cell_error(pool, cell); |
991d9fa0 JT |
865 | } |
866 | } | |
867 | } | |
868 | ||
2dd9c257 JT |
869 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
870 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 871 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
872 | { |
873 | schedule_copy(tc, virt_block, tc->pool_dev, | |
874 | data_origin, data_dest, cell, bio); | |
875 | } | |
876 | ||
877 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
878 | dm_block_t data_dest, | |
a24c2569 | 879 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
880 | { |
881 | schedule_copy(tc, virt_block, tc->origin_dev, | |
882 | virt_block, data_dest, cell, bio); | |
883 | } | |
884 | ||
991d9fa0 | 885 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 886 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
887 | struct bio *bio) |
888 | { | |
889 | struct pool *pool = tc->pool; | |
a24c2569 | 890 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 891 | |
7f214665 MS |
892 | m->quiesced = true; |
893 | m->prepared = false; | |
991d9fa0 JT |
894 | m->tc = tc; |
895 | m->virt_block = virt_block; | |
896 | m->data_block = data_block; | |
897 | m->cell = cell; | |
991d9fa0 JT |
898 | |
899 | /* | |
900 | * If the whole block of data is being overwritten or we are not | |
901 | * zeroing pre-existing data, we can issue the bio immediately. | |
902 | * Otherwise we use kcopyd to zero the data first. | |
903 | */ | |
67e2e2b2 | 904 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
905 | process_prepared_mapping(m); |
906 | ||
907 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 908 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 909 | |
eb2aa48d | 910 | h->overwrite_mapping = m; |
991d9fa0 JT |
911 | m->bio = bio; |
912 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 913 | inc_all_io_entry(pool, bio); |
991d9fa0 | 914 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
915 | } else { |
916 | int r; | |
917 | struct dm_io_region to; | |
918 | ||
919 | to.bdev = tc->pool_dev->bdev; | |
920 | to.sector = data_block * pool->sectors_per_block; | |
921 | to.count = pool->sectors_per_block; | |
922 | ||
923 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
924 | if (r < 0) { | |
925 | mempool_free(m, pool->mapping_pool); | |
c397741c | 926 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 927 | cell_error(pool, cell); |
991d9fa0 JT |
928 | } |
929 | } | |
930 | } | |
931 | ||
e49e5829 JT |
932 | /* |
933 | * A non-zero return indicates read_only or fail_io mode. | |
934 | * Many callers don't care about the return value. | |
935 | */ | |
020cc3b5 | 936 | static int commit(struct pool *pool) |
e49e5829 JT |
937 | { |
938 | int r; | |
939 | ||
8d07e8a5 | 940 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
941 | return -EINVAL; |
942 | ||
020cc3b5 | 943 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
944 | if (r) |
945 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
946 | |
947 | return r; | |
948 | } | |
949 | ||
88a6621b JT |
950 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
951 | { | |
952 | unsigned long flags; | |
953 | ||
954 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
955 | DMWARN("%s: reached low water mark for data device: sending event.", | |
956 | dm_device_name(pool->pool_md)); | |
957 | spin_lock_irqsave(&pool->lock, flags); | |
958 | pool->low_water_triggered = true; | |
959 | spin_unlock_irqrestore(&pool->lock, flags); | |
960 | dm_table_event(pool->ti->table); | |
961 | } | |
962 | } | |
963 | ||
3e1a0699 JT |
964 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
965 | ||
991d9fa0 JT |
966 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
967 | { | |
968 | int r; | |
969 | dm_block_t free_blocks; | |
991d9fa0 JT |
970 | struct pool *pool = tc->pool; |
971 | ||
3e1a0699 | 972 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
973 | return -EINVAL; |
974 | ||
991d9fa0 | 975 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
976 | if (r) { |
977 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 978 | return r; |
b5330655 | 979 | } |
991d9fa0 | 980 | |
88a6621b | 981 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
982 | |
983 | if (!free_blocks) { | |
94563bad MS |
984 | /* |
985 | * Try to commit to see if that will free up some | |
986 | * more space. | |
987 | */ | |
020cc3b5 JT |
988 | r = commit(pool); |
989 | if (r) | |
990 | return r; | |
991d9fa0 | 991 | |
94563bad | 992 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
993 | if (r) { |
994 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 995 | return r; |
b5330655 | 996 | } |
991d9fa0 | 997 | |
94563bad | 998 | if (!free_blocks) { |
3e1a0699 | 999 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1000 | return -ENOSPC; |
991d9fa0 JT |
1001 | } |
1002 | } | |
1003 | ||
1004 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1005 | if (r) { |
b5330655 | 1006 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1007 | return r; |
4a02b34e | 1008 | } |
991d9fa0 JT |
1009 | |
1010 | return 0; | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * If we have run out of space, queue bios until the device is | |
1015 | * resumed, presumably after having been reloaded with more space. | |
1016 | */ | |
1017 | static void retry_on_resume(struct bio *bio) | |
1018 | { | |
59c3d2c6 | 1019 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1020 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1021 | unsigned long flags; |
1022 | ||
c140e1c4 MS |
1023 | spin_lock_irqsave(&tc->lock, flags); |
1024 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1025 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1026 | } |
1027 | ||
3e1a0699 | 1028 | static bool should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1029 | { |
3e1a0699 JT |
1030 | enum pool_mode m = get_pool_mode(pool); |
1031 | ||
1032 | switch (m) { | |
1033 | case PM_WRITE: | |
1034 | /* Shouldn't get here */ | |
1035 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
1036 | return true; | |
1037 | ||
1038 | case PM_OUT_OF_DATA_SPACE: | |
1039 | return pool->pf.error_if_no_space; | |
1040 | ||
1041 | case PM_READ_ONLY: | |
1042 | case PM_FAIL: | |
1043 | return true; | |
1044 | default: | |
1045 | /* Shouldn't get here */ | |
1046 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
1047 | return true; | |
1048 | } | |
1049 | } | |
8c0f0e8c | 1050 | |
3e1a0699 JT |
1051 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1052 | { | |
1053 | if (should_error_unserviceable_bio(pool)) | |
8c0f0e8c | 1054 | bio_io_error(bio); |
6d16202b MS |
1055 | else |
1056 | retry_on_resume(bio); | |
8c0f0e8c MS |
1057 | } |
1058 | ||
399caddf | 1059 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1060 | { |
1061 | struct bio *bio; | |
1062 | struct bio_list bios; | |
1063 | ||
3e1a0699 JT |
1064 | if (should_error_unserviceable_bio(pool)) { |
1065 | cell_error(pool, cell); | |
1066 | return; | |
1067 | } | |
1068 | ||
991d9fa0 | 1069 | bio_list_init(&bios); |
6beca5eb | 1070 | cell_release(pool, cell, &bios); |
991d9fa0 | 1071 | |
3e1a0699 JT |
1072 | if (should_error_unserviceable_bio(pool)) |
1073 | while ((bio = bio_list_pop(&bios))) | |
1074 | bio_io_error(bio); | |
1075 | else | |
1076 | while ((bio = bio_list_pop(&bios))) | |
1077 | retry_on_resume(bio); | |
991d9fa0 JT |
1078 | } |
1079 | ||
104655fd JT |
1080 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1081 | { | |
1082 | int r; | |
c3a0ce2e | 1083 | unsigned long flags; |
104655fd | 1084 | struct pool *pool = tc->pool; |
a24c2569 | 1085 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1086 | struct dm_cell_key key, key2; |
104655fd JT |
1087 | dm_block_t block = get_bio_block(tc, bio); |
1088 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1089 | struct dm_thin_new_mapping *m; |
104655fd JT |
1090 | |
1091 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1092 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1093 | return; |
1094 | ||
1095 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1096 | switch (r) { | |
1097 | case 0: | |
1098 | /* | |
1099 | * Check nobody is fiddling with this pool block. This can | |
1100 | * happen if someone's in the process of breaking sharing | |
1101 | * on this block. | |
1102 | */ | |
1103 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1104 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1105 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1106 | break; |
1107 | } | |
1108 | ||
1109 | if (io_overlaps_block(pool, bio)) { | |
1110 | /* | |
1111 | * IO may still be going to the destination block. We must | |
1112 | * quiesce before we can do the removal. | |
1113 | */ | |
1114 | m = get_next_mapping(pool); | |
1115 | m->tc = tc; | |
19fa1a67 JT |
1116 | m->pass_discard = pool->pf.discard_passdown; |
1117 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1118 | m->virt_block = block; |
1119 | m->data_block = lookup_result.block; | |
1120 | m->cell = cell; | |
1121 | m->cell2 = cell2; | |
104655fd JT |
1122 | m->bio = bio; |
1123 | ||
44feb387 | 1124 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1125 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1126 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1127 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1128 | wake_worker(pool); |
1129 | } | |
1130 | } else { | |
e8088073 | 1131 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1132 | cell_defer_no_holder(tc, cell); |
1133 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1134 | |
104655fd | 1135 | /* |
49296309 MP |
1136 | * The DM core makes sure that the discard doesn't span |
1137 | * a block boundary. So we submit the discard of a | |
1138 | * partial block appropriately. | |
104655fd | 1139 | */ |
650d2a06 MP |
1140 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1141 | remap_and_issue(tc, bio, lookup_result.block); | |
1142 | else | |
1143 | bio_endio(bio, 0); | |
104655fd JT |
1144 | } |
1145 | break; | |
1146 | ||
1147 | case -ENODATA: | |
1148 | /* | |
1149 | * It isn't provisioned, just forget it. | |
1150 | */ | |
f286ba0e | 1151 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1152 | bio_endio(bio, 0); |
1153 | break; | |
1154 | ||
1155 | default: | |
c397741c MS |
1156 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1157 | __func__, r); | |
f286ba0e | 1158 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1159 | bio_io_error(bio); |
1160 | break; | |
1161 | } | |
1162 | } | |
1163 | ||
991d9fa0 | 1164 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1165 | struct dm_cell_key *key, |
991d9fa0 | 1166 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1167 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1168 | { |
1169 | int r; | |
1170 | dm_block_t data_block; | |
d6fc2042 | 1171 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1172 | |
1173 | r = alloc_data_block(tc, &data_block); | |
1174 | switch (r) { | |
1175 | case 0: | |
2dd9c257 JT |
1176 | schedule_internal_copy(tc, block, lookup_result->block, |
1177 | data_block, cell, bio); | |
991d9fa0 JT |
1178 | break; |
1179 | ||
1180 | case -ENOSPC: | |
399caddf | 1181 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1182 | break; |
1183 | ||
1184 | default: | |
c397741c MS |
1185 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1186 | __func__, r); | |
d6fc2042 | 1187 | cell_error(pool, cell); |
991d9fa0 JT |
1188 | break; |
1189 | } | |
1190 | } | |
1191 | ||
1192 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1193 | dm_block_t block, | |
1194 | struct dm_thin_lookup_result *lookup_result) | |
1195 | { | |
a24c2569 | 1196 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1197 | struct pool *pool = tc->pool; |
44feb387 | 1198 | struct dm_cell_key key; |
991d9fa0 JT |
1199 | |
1200 | /* | |
1201 | * If cell is already occupied, then sharing is already in the process | |
1202 | * of being broken so we have nothing further to do here. | |
1203 | */ | |
1204 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1205 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1206 | return; |
1207 | ||
4f024f37 | 1208 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
991d9fa0 JT |
1209 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1210 | else { | |
59c3d2c6 | 1211 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1212 | |
44feb387 | 1213 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1214 | inc_all_io_entry(pool, bio); |
f286ba0e | 1215 | cell_defer_no_holder(tc, cell); |
e8088073 | 1216 | |
991d9fa0 JT |
1217 | remap_and_issue(tc, bio, lookup_result->block); |
1218 | } | |
1219 | } | |
1220 | ||
1221 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1222 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1223 | { |
1224 | int r; | |
1225 | dm_block_t data_block; | |
6beca5eb | 1226 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1227 | |
1228 | /* | |
1229 | * Remap empty bios (flushes) immediately, without provisioning. | |
1230 | */ | |
4f024f37 | 1231 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1232 | inc_all_io_entry(pool, bio); |
f286ba0e | 1233 | cell_defer_no_holder(tc, cell); |
e8088073 | 1234 | |
991d9fa0 JT |
1235 | remap_and_issue(tc, bio, 0); |
1236 | return; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Fill read bios with zeroes and complete them immediately. | |
1241 | */ | |
1242 | if (bio_data_dir(bio) == READ) { | |
1243 | zero_fill_bio(bio); | |
f286ba0e | 1244 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1245 | bio_endio(bio, 0); |
1246 | return; | |
1247 | } | |
1248 | ||
1249 | r = alloc_data_block(tc, &data_block); | |
1250 | switch (r) { | |
1251 | case 0: | |
2dd9c257 JT |
1252 | if (tc->origin_dev) |
1253 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1254 | else | |
1255 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1256 | break; |
1257 | ||
1258 | case -ENOSPC: | |
399caddf | 1259 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1260 | break; |
1261 | ||
1262 | default: | |
c397741c MS |
1263 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1264 | __func__, r); | |
6beca5eb | 1265 | cell_error(pool, cell); |
991d9fa0 JT |
1266 | break; |
1267 | } | |
1268 | } | |
1269 | ||
1270 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1271 | { | |
1272 | int r; | |
6beca5eb | 1273 | struct pool *pool = tc->pool; |
991d9fa0 | 1274 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1275 | struct dm_bio_prison_cell *cell; |
44feb387 | 1276 | struct dm_cell_key key; |
991d9fa0 JT |
1277 | struct dm_thin_lookup_result lookup_result; |
1278 | ||
1279 | /* | |
1280 | * If cell is already occupied, then the block is already | |
1281 | * being provisioned so we have nothing further to do here. | |
1282 | */ | |
1283 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1284 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1285 | return; |
1286 | ||
1287 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1288 | switch (r) { | |
1289 | case 0: | |
e8088073 | 1290 | if (lookup_result.shared) { |
991d9fa0 | 1291 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1292 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1293 | } else { |
6beca5eb | 1294 | inc_all_io_entry(pool, bio); |
f286ba0e | 1295 | cell_defer_no_holder(tc, cell); |
e8088073 | 1296 | |
991d9fa0 | 1297 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1298 | } |
991d9fa0 JT |
1299 | break; |
1300 | ||
1301 | case -ENODATA: | |
2dd9c257 | 1302 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1303 | inc_all_io_entry(pool, bio); |
f286ba0e | 1304 | cell_defer_no_holder(tc, cell); |
e8088073 | 1305 | |
2dd9c257 JT |
1306 | remap_to_origin_and_issue(tc, bio); |
1307 | } else | |
1308 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1309 | break; |
1310 | ||
1311 | default: | |
c397741c MS |
1312 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1313 | __func__, r); | |
f286ba0e | 1314 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1315 | bio_io_error(bio); |
1316 | break; | |
1317 | } | |
1318 | } | |
1319 | ||
e49e5829 JT |
1320 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1321 | { | |
1322 | int r; | |
1323 | int rw = bio_data_dir(bio); | |
1324 | dm_block_t block = get_bio_block(tc, bio); | |
1325 | struct dm_thin_lookup_result lookup_result; | |
1326 | ||
1327 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1328 | switch (r) { | |
1329 | case 0: | |
4f024f37 | 1330 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
8c0f0e8c | 1331 | handle_unserviceable_bio(tc->pool, bio); |
e8088073 JT |
1332 | else { |
1333 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1334 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1335 | } |
e49e5829 JT |
1336 | break; |
1337 | ||
1338 | case -ENODATA: | |
1339 | if (rw != READ) { | |
8c0f0e8c | 1340 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1341 | break; |
1342 | } | |
1343 | ||
1344 | if (tc->origin_dev) { | |
e8088073 | 1345 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1346 | remap_to_origin_and_issue(tc, bio); |
1347 | break; | |
1348 | } | |
1349 | ||
1350 | zero_fill_bio(bio); | |
1351 | bio_endio(bio, 0); | |
1352 | break; | |
1353 | ||
1354 | default: | |
c397741c MS |
1355 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1356 | __func__, r); | |
e49e5829 JT |
1357 | bio_io_error(bio); |
1358 | break; | |
1359 | } | |
1360 | } | |
1361 | ||
3e1a0699 JT |
1362 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1363 | { | |
1364 | bio_endio(bio, 0); | |
1365 | } | |
1366 | ||
e49e5829 JT |
1367 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1368 | { | |
1369 | bio_io_error(bio); | |
1370 | } | |
1371 | ||
ac8c3f3d JT |
1372 | /* |
1373 | * FIXME: should we also commit due to size of transaction, measured in | |
1374 | * metadata blocks? | |
1375 | */ | |
905e51b3 JT |
1376 | static int need_commit_due_to_time(struct pool *pool) |
1377 | { | |
1378 | return jiffies < pool->last_commit_jiffies || | |
1379 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1380 | } | |
1381 | ||
67324ea1 MS |
1382 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1383 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1384 | ||
1385 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1386 | { | |
1387 | struct rb_node **rbp, *parent; | |
1388 | struct dm_thin_endio_hook *pbd; | |
1389 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1390 | ||
1391 | rbp = &tc->sort_bio_list.rb_node; | |
1392 | parent = NULL; | |
1393 | while (*rbp) { | |
1394 | parent = *rbp; | |
1395 | pbd = thin_pbd(parent); | |
1396 | ||
1397 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1398 | rbp = &(*rbp)->rb_left; | |
1399 | else | |
1400 | rbp = &(*rbp)->rb_right; | |
1401 | } | |
1402 | ||
1403 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1404 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1405 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1406 | } | |
1407 | ||
1408 | static void __extract_sorted_bios(struct thin_c *tc) | |
1409 | { | |
1410 | struct rb_node *node; | |
1411 | struct dm_thin_endio_hook *pbd; | |
1412 | struct bio *bio; | |
1413 | ||
1414 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1415 | pbd = thin_pbd(node); | |
1416 | bio = thin_bio(pbd); | |
1417 | ||
1418 | bio_list_add(&tc->deferred_bio_list, bio); | |
1419 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1420 | } | |
1421 | ||
1422 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1423 | } | |
1424 | ||
1425 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1426 | { | |
1427 | struct bio *bio; | |
1428 | struct bio_list bios; | |
1429 | ||
1430 | bio_list_init(&bios); | |
1431 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1432 | bio_list_init(&tc->deferred_bio_list); | |
1433 | ||
1434 | /* Sort deferred_bio_list using rb-tree */ | |
1435 | while ((bio = bio_list_pop(&bios))) | |
1436 | __thin_bio_rb_add(tc, bio); | |
1437 | ||
1438 | /* | |
1439 | * Transfer the sorted bios in sort_bio_list back to | |
1440 | * deferred_bio_list to allow lockless submission of | |
1441 | * all bios. | |
1442 | */ | |
1443 | __extract_sorted_bios(tc); | |
1444 | } | |
1445 | ||
c140e1c4 | 1446 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1447 | { |
c140e1c4 | 1448 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1449 | unsigned long flags; |
1450 | struct bio *bio; | |
1451 | struct bio_list bios; | |
67324ea1 | 1452 | struct blk_plug plug; |
991d9fa0 | 1453 | |
c140e1c4 MS |
1454 | if (tc->requeue_mode) { |
1455 | requeue_bio_list(tc, &tc->deferred_bio_list); | |
1456 | return; | |
1457 | } | |
1458 | ||
991d9fa0 JT |
1459 | bio_list_init(&bios); |
1460 | ||
c140e1c4 | 1461 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1462 | |
1463 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1464 | spin_unlock_irqrestore(&tc->lock, flags); | |
1465 | return; | |
1466 | } | |
1467 | ||
1468 | __sort_thin_deferred_bios(tc); | |
1469 | ||
c140e1c4 MS |
1470 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1471 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1472 | |
c140e1c4 | 1473 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1474 | |
67324ea1 | 1475 | blk_start_plug(&plug); |
991d9fa0 | 1476 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1477 | /* |
1478 | * If we've got no free new_mapping structs, and processing | |
1479 | * this bio might require one, we pause until there are some | |
1480 | * prepared mappings to process. | |
1481 | */ | |
1482 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1483 | spin_lock_irqsave(&tc->lock, flags); |
1484 | bio_list_add(&tc->deferred_bio_list, bio); | |
1485 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1486 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1487 | break; |
1488 | } | |
104655fd JT |
1489 | |
1490 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1491 | pool->process_discard(tc, bio); |
104655fd | 1492 | else |
e49e5829 | 1493 | pool->process_bio(tc, bio); |
991d9fa0 | 1494 | } |
67324ea1 | 1495 | blk_finish_plug(&plug); |
c140e1c4 MS |
1496 | } |
1497 | ||
b10ebd34 JT |
1498 | static void thin_get(struct thin_c *tc); |
1499 | static void thin_put(struct thin_c *tc); | |
1500 | ||
1501 | /* | |
1502 | * We can't hold rcu_read_lock() around code that can block. So we | |
1503 | * find a thin with the rcu lock held; bump a refcount; then drop | |
1504 | * the lock. | |
1505 | */ | |
1506 | static struct thin_c *get_first_thin(struct pool *pool) | |
1507 | { | |
1508 | struct thin_c *tc = NULL; | |
1509 | ||
1510 | rcu_read_lock(); | |
1511 | if (!list_empty(&pool->active_thins)) { | |
1512 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
1513 | thin_get(tc); | |
1514 | } | |
1515 | rcu_read_unlock(); | |
1516 | ||
1517 | return tc; | |
1518 | } | |
1519 | ||
1520 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
1521 | { | |
1522 | struct thin_c *old_tc = tc; | |
1523 | ||
1524 | rcu_read_lock(); | |
1525 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
1526 | thin_get(tc); | |
1527 | thin_put(old_tc); | |
1528 | rcu_read_unlock(); | |
1529 | return tc; | |
1530 | } | |
1531 | thin_put(old_tc); | |
1532 | rcu_read_unlock(); | |
1533 | ||
1534 | return NULL; | |
1535 | } | |
1536 | ||
c140e1c4 MS |
1537 | static void process_deferred_bios(struct pool *pool) |
1538 | { | |
1539 | unsigned long flags; | |
1540 | struct bio *bio; | |
1541 | struct bio_list bios; | |
1542 | struct thin_c *tc; | |
1543 | ||
b10ebd34 JT |
1544 | tc = get_first_thin(pool); |
1545 | while (tc) { | |
c140e1c4 | 1546 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
1547 | tc = get_next_thin(pool, tc); |
1548 | } | |
991d9fa0 JT |
1549 | |
1550 | /* | |
1551 | * If there are any deferred flush bios, we must commit | |
1552 | * the metadata before issuing them. | |
1553 | */ | |
1554 | bio_list_init(&bios); | |
1555 | spin_lock_irqsave(&pool->lock, flags); | |
1556 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1557 | bio_list_init(&pool->deferred_flush_bios); | |
1558 | spin_unlock_irqrestore(&pool->lock, flags); | |
1559 | ||
4d1662a3 MS |
1560 | if (bio_list_empty(&bios) && |
1561 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1562 | return; |
1563 | ||
020cc3b5 | 1564 | if (commit(pool)) { |
991d9fa0 JT |
1565 | while ((bio = bio_list_pop(&bios))) |
1566 | bio_io_error(bio); | |
1567 | return; | |
1568 | } | |
905e51b3 | 1569 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1570 | |
1571 | while ((bio = bio_list_pop(&bios))) | |
1572 | generic_make_request(bio); | |
1573 | } | |
1574 | ||
1575 | static void do_worker(struct work_struct *ws) | |
1576 | { | |
1577 | struct pool *pool = container_of(ws, struct pool, worker); | |
1578 | ||
e49e5829 JT |
1579 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1580 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1581 | process_deferred_bios(pool); |
1582 | } | |
1583 | ||
905e51b3 JT |
1584 | /* |
1585 | * We want to commit periodically so that not too much | |
1586 | * unwritten data builds up. | |
1587 | */ | |
1588 | static void do_waker(struct work_struct *ws) | |
1589 | { | |
1590 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1591 | wake_worker(pool); | |
1592 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1593 | } | |
1594 | ||
85ad643b JT |
1595 | /* |
1596 | * We're holding onto IO to allow userland time to react. After the | |
1597 | * timeout either the pool will have been resized (and thus back in | |
1598 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
1599 | */ | |
1600 | static void do_no_space_timeout(struct work_struct *ws) | |
1601 | { | |
1602 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
1603 | no_space_timeout); | |
1604 | ||
1605 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
1606 | set_pool_mode(pool, PM_READ_ONLY); | |
1607 | } | |
1608 | ||
991d9fa0 JT |
1609 | /*----------------------------------------------------------------*/ |
1610 | ||
738211f7 JT |
1611 | struct noflush_work { |
1612 | struct work_struct worker; | |
1613 | struct thin_c *tc; | |
1614 | ||
1615 | atomic_t complete; | |
1616 | wait_queue_head_t wait; | |
1617 | }; | |
1618 | ||
1619 | static void complete_noflush_work(struct noflush_work *w) | |
1620 | { | |
1621 | atomic_set(&w->complete, 1); | |
1622 | wake_up(&w->wait); | |
1623 | } | |
1624 | ||
1625 | static void do_noflush_start(struct work_struct *ws) | |
1626 | { | |
1627 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1628 | w->tc->requeue_mode = true; | |
1629 | requeue_io(w->tc); | |
1630 | complete_noflush_work(w); | |
1631 | } | |
1632 | ||
1633 | static void do_noflush_stop(struct work_struct *ws) | |
1634 | { | |
1635 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1636 | w->tc->requeue_mode = false; | |
1637 | complete_noflush_work(w); | |
1638 | } | |
1639 | ||
1640 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
1641 | { | |
1642 | struct noflush_work w; | |
1643 | ||
fbcde3d8 | 1644 | INIT_WORK_ONSTACK(&w.worker, fn); |
738211f7 JT |
1645 | w.tc = tc; |
1646 | atomic_set(&w.complete, 0); | |
1647 | init_waitqueue_head(&w.wait); | |
1648 | ||
1649 | queue_work(tc->pool->wq, &w.worker); | |
1650 | ||
1651 | wait_event(w.wait, atomic_read(&w.complete)); | |
1652 | } | |
1653 | ||
1654 | /*----------------------------------------------------------------*/ | |
1655 | ||
e49e5829 JT |
1656 | static enum pool_mode get_pool_mode(struct pool *pool) |
1657 | { | |
1658 | return pool->pf.mode; | |
1659 | } | |
1660 | ||
3e1a0699 JT |
1661 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
1662 | { | |
1663 | dm_table_event(pool->ti->table); | |
1664 | DMINFO("%s: switching pool to %s mode", | |
1665 | dm_device_name(pool->pool_md), new_mode); | |
1666 | } | |
1667 | ||
8b64e881 | 1668 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 1669 | { |
cdc2b415 | 1670 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
1671 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
1672 | enum pool_mode old_mode = get_pool_mode(pool); | |
1673 | ||
1674 | /* | |
1675 | * Never allow the pool to transition to PM_WRITE mode if user | |
1676 | * intervention is required to verify metadata and data consistency. | |
1677 | */ | |
1678 | if (new_mode == PM_WRITE && needs_check) { | |
1679 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
1680 | dm_device_name(pool->pool_md)); | |
1681 | if (old_mode != new_mode) | |
1682 | new_mode = old_mode; | |
1683 | else | |
1684 | new_mode = PM_READ_ONLY; | |
1685 | } | |
1686 | /* | |
1687 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1688 | * not going to recover without a thin_repair. So we never let the | |
1689 | * pool move out of the old mode. | |
1690 | */ | |
1691 | if (old_mode == PM_FAIL) | |
1692 | new_mode = old_mode; | |
e49e5829 | 1693 | |
8b64e881 | 1694 | switch (new_mode) { |
e49e5829 | 1695 | case PM_FAIL: |
8b64e881 | 1696 | if (old_mode != new_mode) |
3e1a0699 | 1697 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 1698 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1699 | pool->process_bio = process_bio_fail; |
1700 | pool->process_discard = process_bio_fail; | |
1701 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1702 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
1703 | |
1704 | error_retry_list(pool); | |
e49e5829 JT |
1705 | break; |
1706 | ||
1707 | case PM_READ_ONLY: | |
8b64e881 | 1708 | if (old_mode != new_mode) |
3e1a0699 JT |
1709 | notify_of_pool_mode_change(pool, "read-only"); |
1710 | dm_pool_metadata_read_only(pool->pmd); | |
1711 | pool->process_bio = process_bio_read_only; | |
1712 | pool->process_discard = process_bio_success; | |
1713 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1714 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1715 | ||
1716 | error_retry_list(pool); | |
1717 | break; | |
1718 | ||
1719 | case PM_OUT_OF_DATA_SPACE: | |
1720 | /* | |
1721 | * Ideally we'd never hit this state; the low water mark | |
1722 | * would trigger userland to extend the pool before we | |
1723 | * completely run out of data space. However, many small | |
1724 | * IOs to unprovisioned space can consume data space at an | |
1725 | * alarming rate. Adjust your low water mark if you're | |
1726 | * frequently seeing this mode. | |
1727 | */ | |
1728 | if (old_mode != new_mode) | |
1729 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
1730 | pool->process_bio = process_bio_read_only; | |
1731 | pool->process_discard = process_discard; | |
1732 | pool->process_prepared_mapping = process_prepared_mapping; | |
1733 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
85ad643b JT |
1734 | |
1735 | if (!pool->pf.error_if_no_space) | |
1736 | queue_delayed_work(pool->wq, &pool->no_space_timeout, NO_SPACE_TIMEOUT); | |
e49e5829 JT |
1737 | break; |
1738 | ||
1739 | case PM_WRITE: | |
8b64e881 | 1740 | if (old_mode != new_mode) |
3e1a0699 | 1741 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 1742 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1743 | pool->process_bio = process_bio; |
1744 | pool->process_discard = process_discard; | |
1745 | pool->process_prepared_mapping = process_prepared_mapping; | |
1746 | pool->process_prepared_discard = process_prepared_discard; | |
1747 | break; | |
1748 | } | |
8b64e881 MS |
1749 | |
1750 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
1751 | /* |
1752 | * The pool mode may have changed, sync it so bind_control_target() | |
1753 | * doesn't cause an unexpected mode transition on resume. | |
1754 | */ | |
1755 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
1756 | } |
1757 | ||
07f2b6e0 | 1758 | static void abort_transaction(struct pool *pool) |
b5330655 | 1759 | { |
07f2b6e0 MS |
1760 | const char *dev_name = dm_device_name(pool->pool_md); |
1761 | ||
1762 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
1763 | if (dm_pool_abort_metadata(pool->pmd)) { | |
1764 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
1765 | set_pool_mode(pool, PM_FAIL); | |
1766 | } | |
1767 | ||
1768 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
1769 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
1770 | set_pool_mode(pool, PM_FAIL); | |
1771 | } | |
1772 | } | |
399caddf | 1773 | |
07f2b6e0 MS |
1774 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1775 | { | |
b5330655 JT |
1776 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1777 | dm_device_name(pool->pool_md), op, r); | |
1778 | ||
07f2b6e0 | 1779 | abort_transaction(pool); |
b5330655 JT |
1780 | set_pool_mode(pool, PM_READ_ONLY); |
1781 | } | |
1782 | ||
e49e5829 JT |
1783 | /*----------------------------------------------------------------*/ |
1784 | ||
991d9fa0 JT |
1785 | /* |
1786 | * Mapping functions. | |
1787 | */ | |
1788 | ||
1789 | /* | |
1790 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1791 | */ | |
1792 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1793 | { | |
1794 | unsigned long flags; | |
1795 | struct pool *pool = tc->pool; | |
1796 | ||
c140e1c4 MS |
1797 | spin_lock_irqsave(&tc->lock, flags); |
1798 | bio_list_add(&tc->deferred_bio_list, bio); | |
1799 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1800 | |
1801 | wake_worker(pool); | |
1802 | } | |
1803 | ||
59c3d2c6 | 1804 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1805 | { |
59c3d2c6 | 1806 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1807 | |
1808 | h->tc = tc; | |
1809 | h->shared_read_entry = NULL; | |
e8088073 | 1810 | h->all_io_entry = NULL; |
eb2aa48d | 1811 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1812 | } |
1813 | ||
991d9fa0 JT |
1814 | /* |
1815 | * Non-blocking function called from the thin target's map function. | |
1816 | */ | |
7de3ee57 | 1817 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1818 | { |
1819 | int r; | |
1820 | struct thin_c *tc = ti->private; | |
1821 | dm_block_t block = get_bio_block(tc, bio); | |
1822 | struct dm_thin_device *td = tc->td; | |
1823 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1824 | struct dm_bio_prison_cell cell1, cell2; |
1825 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1826 | struct dm_cell_key key; |
991d9fa0 | 1827 | |
59c3d2c6 | 1828 | thin_hook_bio(tc, bio); |
e49e5829 | 1829 | |
738211f7 JT |
1830 | if (tc->requeue_mode) { |
1831 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1832 | return DM_MAPIO_SUBMITTED; | |
1833 | } | |
1834 | ||
e49e5829 JT |
1835 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
1836 | bio_io_error(bio); | |
1837 | return DM_MAPIO_SUBMITTED; | |
1838 | } | |
1839 | ||
104655fd | 1840 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1841 | thin_defer_bio(tc, bio); |
1842 | return DM_MAPIO_SUBMITTED; | |
1843 | } | |
1844 | ||
1845 | r = dm_thin_find_block(td, block, 0, &result); | |
1846 | ||
1847 | /* | |
1848 | * Note that we defer readahead too. | |
1849 | */ | |
1850 | switch (r) { | |
1851 | case 0: | |
1852 | if (unlikely(result.shared)) { | |
1853 | /* | |
1854 | * We have a race condition here between the | |
1855 | * result.shared value returned by the lookup and | |
1856 | * snapshot creation, which may cause new | |
1857 | * sharing. | |
1858 | * | |
1859 | * To avoid this always quiesce the origin before | |
1860 | * taking the snap. You want to do this anyway to | |
1861 | * ensure a consistent application view | |
1862 | * (i.e. lockfs). | |
1863 | * | |
1864 | * More distant ancestors are irrelevant. The | |
1865 | * shared flag will be set in their case. | |
1866 | */ | |
1867 | thin_defer_bio(tc, bio); | |
e8088073 | 1868 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1869 | } |
e8088073 JT |
1870 | |
1871 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1872 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1873 | return DM_MAPIO_SUBMITTED; |
1874 | ||
1875 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1876 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1877 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1878 | return DM_MAPIO_SUBMITTED; |
1879 | } | |
1880 | ||
1881 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1882 | cell_defer_no_holder_no_free(tc, &cell2); |
1883 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1884 | |
1885 | remap(tc, bio, result.block); | |
1886 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1887 | |
1888 | case -ENODATA: | |
e49e5829 JT |
1889 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1890 | /* | |
1891 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 1892 | * of doing so. |
e49e5829 | 1893 | */ |
8c0f0e8c | 1894 | handle_unserviceable_bio(tc->pool, bio); |
2aab3850 | 1895 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1896 | } |
1897 | /* fall through */ | |
1898 | ||
1899 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1900 | /* |
1901 | * In future, the failed dm_thin_find_block above could | |
1902 | * provide the hint to load the metadata into cache. | |
1903 | */ | |
991d9fa0 | 1904 | thin_defer_bio(tc, bio); |
2aab3850 | 1905 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1906 | |
1907 | default: | |
1908 | /* | |
1909 | * Must always call bio_io_error on failure. | |
1910 | * dm_thin_find_block can fail with -EINVAL if the | |
1911 | * pool is switched to fail-io mode. | |
1912 | */ | |
1913 | bio_io_error(bio); | |
2aab3850 | 1914 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1915 | } |
991d9fa0 JT |
1916 | } |
1917 | ||
1918 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1919 | { | |
991d9fa0 | 1920 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 1921 | struct request_queue *q; |
991d9fa0 | 1922 | |
760fe67e MS |
1923 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
1924 | return 1; | |
991d9fa0 | 1925 | |
760fe67e MS |
1926 | q = bdev_get_queue(pt->data_dev->bdev); |
1927 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
1928 | } |
1929 | ||
c140e1c4 | 1930 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 1931 | { |
c140e1c4 MS |
1932 | unsigned long flags; |
1933 | struct thin_c *tc; | |
1934 | ||
1935 | rcu_read_lock(); | |
1936 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
1937 | spin_lock_irqsave(&tc->lock, flags); | |
1938 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
1939 | bio_list_init(&tc->retry_on_resume_list); | |
1940 | spin_unlock_irqrestore(&tc->lock, flags); | |
1941 | } | |
1942 | rcu_read_unlock(); | |
991d9fa0 JT |
1943 | } |
1944 | ||
1945 | /*---------------------------------------------------------------- | |
1946 | * Binding of control targets to a pool object | |
1947 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1948 | static bool data_dev_supports_discard(struct pool_c *pt) |
1949 | { | |
1950 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1951 | ||
1952 | return q && blk_queue_discard(q); | |
1953 | } | |
1954 | ||
58051b94 JT |
1955 | static bool is_factor(sector_t block_size, uint32_t n) |
1956 | { | |
1957 | return !sector_div(block_size, n); | |
1958 | } | |
1959 | ||
9bc142dd MS |
1960 | /* |
1961 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1962 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1963 | */ |
0424caa1 | 1964 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1965 | { |
0424caa1 MS |
1966 | struct pool *pool = pt->pool; |
1967 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1968 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1969 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1970 | const char *reason = NULL; | |
9bc142dd MS |
1971 | char buf[BDEVNAME_SIZE]; |
1972 | ||
0424caa1 | 1973 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1974 | return; |
1975 | ||
0424caa1 MS |
1976 | if (!data_dev_supports_discard(pt)) |
1977 | reason = "discard unsupported"; | |
1978 | ||
1979 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1980 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1981 | |
0424caa1 MS |
1982 | else if (data_limits->discard_granularity > block_size) |
1983 | reason = "discard granularity larger than a block"; | |
1984 | ||
58051b94 | 1985 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1986 | reason = "discard granularity not a factor of block size"; |
1987 | ||
1988 | if (reason) { | |
1989 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1990 | pt->adjusted_pf.discard_passdown = false; | |
1991 | } | |
9bc142dd MS |
1992 | } |
1993 | ||
991d9fa0 JT |
1994 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1995 | { | |
1996 | struct pool_c *pt = ti->private; | |
1997 | ||
e49e5829 | 1998 | /* |
9b7aaa64 | 1999 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2000 | */ |
07f2b6e0 | 2001 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2002 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2003 | |
8b64e881 MS |
2004 | /* |
2005 | * Don't change the pool's mode until set_pool_mode() below. | |
2006 | * Otherwise the pool's process_* function pointers may | |
2007 | * not match the desired pool mode. | |
2008 | */ | |
2009 | pt->adjusted_pf.mode = old_mode; | |
2010 | ||
2011 | pool->ti = ti; | |
2012 | pool->pf = pt->adjusted_pf; | |
2013 | pool->low_water_blocks = pt->low_water_blocks; | |
2014 | ||
9bc142dd | 2015 | set_pool_mode(pool, new_mode); |
f402693d | 2016 | |
991d9fa0 JT |
2017 | return 0; |
2018 | } | |
2019 | ||
2020 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2021 | { | |
2022 | if (pool->ti == ti) | |
2023 | pool->ti = NULL; | |
2024 | } | |
2025 | ||
2026 | /*---------------------------------------------------------------- | |
2027 | * Pool creation | |
2028 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2029 | /* Initialize pool features. */ |
2030 | static void pool_features_init(struct pool_features *pf) | |
2031 | { | |
e49e5829 | 2032 | pf->mode = PM_WRITE; |
9bc142dd MS |
2033 | pf->zero_new_blocks = true; |
2034 | pf->discard_enabled = true; | |
2035 | pf->discard_passdown = true; | |
787a996c | 2036 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2037 | } |
2038 | ||
991d9fa0 JT |
2039 | static void __pool_destroy(struct pool *pool) |
2040 | { | |
2041 | __pool_table_remove(pool); | |
2042 | ||
2043 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2044 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2045 | ||
44feb387 | 2046 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2047 | dm_kcopyd_client_destroy(pool->copier); |
2048 | ||
2049 | if (pool->wq) | |
2050 | destroy_workqueue(pool->wq); | |
2051 | ||
2052 | if (pool->next_mapping) | |
2053 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2054 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2055 | dm_deferred_set_destroy(pool->shared_read_ds); |
2056 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2057 | kfree(pool); |
2058 | } | |
2059 | ||
a24c2569 | 2060 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2061 | |
991d9fa0 JT |
2062 | static struct pool *pool_create(struct mapped_device *pool_md, |
2063 | struct block_device *metadata_dev, | |
e49e5829 JT |
2064 | unsigned long block_size, |
2065 | int read_only, char **error) | |
991d9fa0 JT |
2066 | { |
2067 | int r; | |
2068 | void *err_p; | |
2069 | struct pool *pool; | |
2070 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2071 | bool format_device = read_only ? false : true; |
991d9fa0 | 2072 | |
e49e5829 | 2073 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2074 | if (IS_ERR(pmd)) { |
2075 | *error = "Error creating metadata object"; | |
2076 | return (struct pool *)pmd; | |
2077 | } | |
2078 | ||
2079 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2080 | if (!pool) { | |
2081 | *error = "Error allocating memory for pool"; | |
2082 | err_p = ERR_PTR(-ENOMEM); | |
2083 | goto bad_pool; | |
2084 | } | |
2085 | ||
2086 | pool->pmd = pmd; | |
2087 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2088 | if (block_size & (block_size - 1)) |
2089 | pool->sectors_per_block_shift = -1; | |
2090 | else | |
2091 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2092 | pool->low_water_blocks = 0; |
67e2e2b2 | 2093 | pool_features_init(&pool->pf); |
44feb387 | 2094 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
2095 | if (!pool->prison) { |
2096 | *error = "Error creating pool's bio prison"; | |
2097 | err_p = ERR_PTR(-ENOMEM); | |
2098 | goto bad_prison; | |
2099 | } | |
2100 | ||
df5d2e90 | 2101 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2102 | if (IS_ERR(pool->copier)) { |
2103 | r = PTR_ERR(pool->copier); | |
2104 | *error = "Error creating pool's kcopyd client"; | |
2105 | err_p = ERR_PTR(r); | |
2106 | goto bad_kcopyd_client; | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * Create singlethreaded workqueue that will service all devices | |
2111 | * that use this metadata. | |
2112 | */ | |
2113 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2114 | if (!pool->wq) { | |
2115 | *error = "Error creating pool's workqueue"; | |
2116 | err_p = ERR_PTR(-ENOMEM); | |
2117 | goto bad_wq; | |
2118 | } | |
2119 | ||
2120 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 2121 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2122 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2123 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2124 | bio_list_init(&pool->deferred_flush_bios); |
2125 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2126 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2127 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2128 | pool->low_water_triggered = false; |
44feb387 MS |
2129 | |
2130 | pool->shared_read_ds = dm_deferred_set_create(); | |
2131 | if (!pool->shared_read_ds) { | |
2132 | *error = "Error creating pool's shared read deferred set"; | |
2133 | err_p = ERR_PTR(-ENOMEM); | |
2134 | goto bad_shared_read_ds; | |
2135 | } | |
2136 | ||
2137 | pool->all_io_ds = dm_deferred_set_create(); | |
2138 | if (!pool->all_io_ds) { | |
2139 | *error = "Error creating pool's all io deferred set"; | |
2140 | err_p = ERR_PTR(-ENOMEM); | |
2141 | goto bad_all_io_ds; | |
2142 | } | |
991d9fa0 JT |
2143 | |
2144 | pool->next_mapping = NULL; | |
a24c2569 MS |
2145 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2146 | _new_mapping_cache); | |
991d9fa0 JT |
2147 | if (!pool->mapping_pool) { |
2148 | *error = "Error creating pool's mapping mempool"; | |
2149 | err_p = ERR_PTR(-ENOMEM); | |
2150 | goto bad_mapping_pool; | |
2151 | } | |
2152 | ||
991d9fa0 | 2153 | pool->ref_count = 1; |
905e51b3 | 2154 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2155 | pool->pool_md = pool_md; |
2156 | pool->md_dev = metadata_dev; | |
2157 | __pool_table_insert(pool); | |
2158 | ||
2159 | return pool; | |
2160 | ||
991d9fa0 | 2161 | bad_mapping_pool: |
44feb387 MS |
2162 | dm_deferred_set_destroy(pool->all_io_ds); |
2163 | bad_all_io_ds: | |
2164 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2165 | bad_shared_read_ds: | |
991d9fa0 JT |
2166 | destroy_workqueue(pool->wq); |
2167 | bad_wq: | |
2168 | dm_kcopyd_client_destroy(pool->copier); | |
2169 | bad_kcopyd_client: | |
44feb387 | 2170 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2171 | bad_prison: |
2172 | kfree(pool); | |
2173 | bad_pool: | |
2174 | if (dm_pool_metadata_close(pmd)) | |
2175 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2176 | ||
2177 | return err_p; | |
2178 | } | |
2179 | ||
2180 | static void __pool_inc(struct pool *pool) | |
2181 | { | |
2182 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2183 | pool->ref_count++; | |
2184 | } | |
2185 | ||
2186 | static void __pool_dec(struct pool *pool) | |
2187 | { | |
2188 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2189 | BUG_ON(!pool->ref_count); | |
2190 | if (!--pool->ref_count) | |
2191 | __pool_destroy(pool); | |
2192 | } | |
2193 | ||
2194 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2195 | struct block_device *metadata_dev, | |
e49e5829 JT |
2196 | unsigned long block_size, int read_only, |
2197 | char **error, int *created) | |
991d9fa0 JT |
2198 | { |
2199 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2200 | ||
2201 | if (pool) { | |
f09996c9 MS |
2202 | if (pool->pool_md != pool_md) { |
2203 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2204 | return ERR_PTR(-EBUSY); |
f09996c9 | 2205 | } |
991d9fa0 JT |
2206 | __pool_inc(pool); |
2207 | ||
2208 | } else { | |
2209 | pool = __pool_table_lookup(pool_md); | |
2210 | if (pool) { | |
f09996c9 MS |
2211 | if (pool->md_dev != metadata_dev) { |
2212 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2213 | return ERR_PTR(-EINVAL); |
f09996c9 | 2214 | } |
991d9fa0 JT |
2215 | __pool_inc(pool); |
2216 | ||
67e2e2b2 | 2217 | } else { |
e49e5829 | 2218 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2219 | *created = 1; |
2220 | } | |
991d9fa0 JT |
2221 | } |
2222 | ||
2223 | return pool; | |
2224 | } | |
2225 | ||
2226 | /*---------------------------------------------------------------- | |
2227 | * Pool target methods | |
2228 | *--------------------------------------------------------------*/ | |
2229 | static void pool_dtr(struct dm_target *ti) | |
2230 | { | |
2231 | struct pool_c *pt = ti->private; | |
2232 | ||
2233 | mutex_lock(&dm_thin_pool_table.mutex); | |
2234 | ||
2235 | unbind_control_target(pt->pool, ti); | |
2236 | __pool_dec(pt->pool); | |
2237 | dm_put_device(ti, pt->metadata_dev); | |
2238 | dm_put_device(ti, pt->data_dev); | |
2239 | kfree(pt); | |
2240 | ||
2241 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2242 | } | |
2243 | ||
991d9fa0 JT |
2244 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2245 | struct dm_target *ti) | |
2246 | { | |
2247 | int r; | |
2248 | unsigned argc; | |
2249 | const char *arg_name; | |
2250 | ||
2251 | static struct dm_arg _args[] = { | |
74aa45c3 | 2252 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2253 | }; |
2254 | ||
2255 | /* | |
2256 | * No feature arguments supplied. | |
2257 | */ | |
2258 | if (!as->argc) | |
2259 | return 0; | |
2260 | ||
2261 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2262 | if (r) | |
2263 | return -EINVAL; | |
2264 | ||
2265 | while (argc && !r) { | |
2266 | arg_name = dm_shift_arg(as); | |
2267 | argc--; | |
2268 | ||
e49e5829 | 2269 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2270 | pf->zero_new_blocks = false; |
e49e5829 JT |
2271 | |
2272 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2273 | pf->discard_enabled = false; |
e49e5829 JT |
2274 | |
2275 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2276 | pf->discard_passdown = false; |
991d9fa0 | 2277 | |
e49e5829 JT |
2278 | else if (!strcasecmp(arg_name, "read_only")) |
2279 | pf->mode = PM_READ_ONLY; | |
2280 | ||
787a996c MS |
2281 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2282 | pf->error_if_no_space = true; | |
2283 | ||
e49e5829 JT |
2284 | else { |
2285 | ti->error = "Unrecognised pool feature requested"; | |
2286 | r = -EINVAL; | |
2287 | break; | |
2288 | } | |
991d9fa0 JT |
2289 | } |
2290 | ||
2291 | return r; | |
2292 | } | |
2293 | ||
ac8c3f3d JT |
2294 | static void metadata_low_callback(void *context) |
2295 | { | |
2296 | struct pool *pool = context; | |
2297 | ||
2298 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2299 | dm_device_name(pool->pool_md)); | |
2300 | ||
2301 | dm_table_event(pool->ti->table); | |
2302 | } | |
2303 | ||
7d48935e MS |
2304 | static sector_t get_dev_size(struct block_device *bdev) |
2305 | { | |
2306 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2307 | } | |
2308 | ||
2309 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2310 | { |
7d48935e | 2311 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2312 | char buffer[BDEVNAME_SIZE]; |
2313 | ||
7d48935e | 2314 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2315 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2316 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2317 | } |
2318 | ||
2319 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2320 | { | |
2321 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2322 | ||
2323 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2324 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2325 | |
2326 | return metadata_dev_size; | |
2327 | } | |
2328 | ||
24347e95 JT |
2329 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2330 | { | |
2331 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2332 | ||
7d48935e | 2333 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2334 | |
2335 | return metadata_dev_size; | |
2336 | } | |
2337 | ||
ac8c3f3d JT |
2338 | /* |
2339 | * When a metadata threshold is crossed a dm event is triggered, and | |
2340 | * userland should respond by growing the metadata device. We could let | |
2341 | * userland set the threshold, like we do with the data threshold, but I'm | |
2342 | * not sure they know enough to do this well. | |
2343 | */ | |
2344 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2345 | { | |
2346 | /* | |
2347 | * 4M is ample for all ops with the possible exception of thin | |
2348 | * device deletion which is harmless if it fails (just retry the | |
2349 | * delete after you've grown the device). | |
2350 | */ | |
2351 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2352 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2353 | } | |
2354 | ||
991d9fa0 JT |
2355 | /* |
2356 | * thin-pool <metadata dev> <data dev> | |
2357 | * <data block size (sectors)> | |
2358 | * <low water mark (blocks)> | |
2359 | * [<#feature args> [<arg>]*] | |
2360 | * | |
2361 | * Optional feature arguments are: | |
2362 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2363 | * ignore_discard: disable discard |
2364 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2365 | * read_only: Don't allow any changes to be made to the pool metadata. |
2366 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2367 | */ |
2368 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2369 | { | |
67e2e2b2 | 2370 | int r, pool_created = 0; |
991d9fa0 JT |
2371 | struct pool_c *pt; |
2372 | struct pool *pool; | |
2373 | struct pool_features pf; | |
2374 | struct dm_arg_set as; | |
2375 | struct dm_dev *data_dev; | |
2376 | unsigned long block_size; | |
2377 | dm_block_t low_water_blocks; | |
2378 | struct dm_dev *metadata_dev; | |
5d0db96d | 2379 | fmode_t metadata_mode; |
991d9fa0 JT |
2380 | |
2381 | /* | |
2382 | * FIXME Remove validation from scope of lock. | |
2383 | */ | |
2384 | mutex_lock(&dm_thin_pool_table.mutex); | |
2385 | ||
2386 | if (argc < 4) { | |
2387 | ti->error = "Invalid argument count"; | |
2388 | r = -EINVAL; | |
2389 | goto out_unlock; | |
2390 | } | |
5d0db96d | 2391 | |
991d9fa0 JT |
2392 | as.argc = argc; |
2393 | as.argv = argv; | |
2394 | ||
5d0db96d JT |
2395 | /* |
2396 | * Set default pool features. | |
2397 | */ | |
2398 | pool_features_init(&pf); | |
2399 | ||
2400 | dm_consume_args(&as, 4); | |
2401 | r = parse_pool_features(&as, &pf, ti); | |
2402 | if (r) | |
2403 | goto out_unlock; | |
2404 | ||
2405 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2406 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2407 | if (r) { |
2408 | ti->error = "Error opening metadata block device"; | |
2409 | goto out_unlock; | |
2410 | } | |
7d48935e | 2411 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2412 | |
2413 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2414 | if (r) { | |
2415 | ti->error = "Error getting data device"; | |
2416 | goto out_metadata; | |
2417 | } | |
2418 | ||
2419 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2420 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2421 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2422 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2423 | ti->error = "Invalid block size"; |
2424 | r = -EINVAL; | |
2425 | goto out; | |
2426 | } | |
2427 | ||
2428 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2429 | ti->error = "Invalid low water mark"; | |
2430 | r = -EINVAL; | |
2431 | goto out; | |
2432 | } | |
2433 | ||
991d9fa0 JT |
2434 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2435 | if (!pt) { | |
2436 | r = -ENOMEM; | |
2437 | goto out; | |
2438 | } | |
2439 | ||
2440 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2441 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2442 | if (IS_ERR(pool)) { |
2443 | r = PTR_ERR(pool); | |
2444 | goto out_free_pt; | |
2445 | } | |
2446 | ||
67e2e2b2 JT |
2447 | /* |
2448 | * 'pool_created' reflects whether this is the first table load. | |
2449 | * Top level discard support is not allowed to be changed after | |
2450 | * initial load. This would require a pool reload to trigger thin | |
2451 | * device changes. | |
2452 | */ | |
2453 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2454 | ti->error = "Discard support cannot be disabled once enabled"; | |
2455 | r = -EINVAL; | |
2456 | goto out_flags_changed; | |
2457 | } | |
2458 | ||
991d9fa0 JT |
2459 | pt->pool = pool; |
2460 | pt->ti = ti; | |
2461 | pt->metadata_dev = metadata_dev; | |
2462 | pt->data_dev = data_dev; | |
2463 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2464 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2465 | ti->num_flush_bios = 1; |
9bc142dd | 2466 | |
67e2e2b2 JT |
2467 | /* |
2468 | * Only need to enable discards if the pool should pass | |
2469 | * them down to the data device. The thin device's discard | |
2470 | * processing will cause mappings to be removed from the btree. | |
2471 | */ | |
b60ab990 | 2472 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2473 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2474 | ti->num_discard_bios = 1; |
9bc142dd | 2475 | |
67e2e2b2 JT |
2476 | /* |
2477 | * Setting 'discards_supported' circumvents the normal | |
2478 | * stacking of discard limits (this keeps the pool and | |
2479 | * thin devices' discard limits consistent). | |
2480 | */ | |
0ac55489 | 2481 | ti->discards_supported = true; |
67e2e2b2 | 2482 | } |
991d9fa0 JT |
2483 | ti->private = pt; |
2484 | ||
ac8c3f3d JT |
2485 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2486 | calc_metadata_threshold(pt), | |
2487 | metadata_low_callback, | |
2488 | pool); | |
2489 | if (r) | |
2490 | goto out_free_pt; | |
2491 | ||
991d9fa0 JT |
2492 | pt->callbacks.congested_fn = pool_is_congested; |
2493 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2494 | ||
2495 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2496 | ||
2497 | return 0; | |
2498 | ||
67e2e2b2 JT |
2499 | out_flags_changed: |
2500 | __pool_dec(pool); | |
991d9fa0 JT |
2501 | out_free_pt: |
2502 | kfree(pt); | |
2503 | out: | |
2504 | dm_put_device(ti, data_dev); | |
2505 | out_metadata: | |
2506 | dm_put_device(ti, metadata_dev); | |
2507 | out_unlock: | |
2508 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2509 | ||
2510 | return r; | |
2511 | } | |
2512 | ||
7de3ee57 | 2513 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2514 | { |
2515 | int r; | |
2516 | struct pool_c *pt = ti->private; | |
2517 | struct pool *pool = pt->pool; | |
2518 | unsigned long flags; | |
2519 | ||
2520 | /* | |
2521 | * As this is a singleton target, ti->begin is always zero. | |
2522 | */ | |
2523 | spin_lock_irqsave(&pool->lock, flags); | |
2524 | bio->bi_bdev = pt->data_dev->bdev; | |
2525 | r = DM_MAPIO_REMAPPED; | |
2526 | spin_unlock_irqrestore(&pool->lock, flags); | |
2527 | ||
2528 | return r; | |
2529 | } | |
2530 | ||
b17446df | 2531 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2532 | { |
2533 | int r; | |
2534 | struct pool_c *pt = ti->private; | |
2535 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2536 | sector_t data_size = ti->len; |
2537 | dm_block_t sb_data_size; | |
991d9fa0 | 2538 | |
b17446df | 2539 | *need_commit = false; |
991d9fa0 | 2540 | |
55f2b8bd MS |
2541 | (void) sector_div(data_size, pool->sectors_per_block); |
2542 | ||
991d9fa0 JT |
2543 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2544 | if (r) { | |
4fa5971a MS |
2545 | DMERR("%s: failed to retrieve data device size", |
2546 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2547 | return r; |
2548 | } | |
2549 | ||
2550 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2551 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2552 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2553 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2554 | return -EINVAL; |
2555 | ||
2556 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2557 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2558 | DMERR("%s: unable to grow the data device until repaired.", | |
2559 | dm_device_name(pool->pool_md)); | |
2560 | return 0; | |
2561 | } | |
2562 | ||
6f7f51d4 MS |
2563 | if (sb_data_size) |
2564 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2565 | dm_device_name(pool->pool_md), | |
2566 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2567 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2568 | if (r) { | |
b5330655 | 2569 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2570 | return r; |
2571 | } | |
2572 | ||
b17446df | 2573 | *need_commit = true; |
991d9fa0 JT |
2574 | } |
2575 | ||
2576 | return 0; | |
2577 | } | |
2578 | ||
24347e95 JT |
2579 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2580 | { | |
2581 | int r; | |
2582 | struct pool_c *pt = ti->private; | |
2583 | struct pool *pool = pt->pool; | |
2584 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2585 | ||
2586 | *need_commit = false; | |
2587 | ||
610bba8b | 2588 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2589 | |
2590 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2591 | if (r) { | |
4fa5971a MS |
2592 | DMERR("%s: failed to retrieve metadata device size", |
2593 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2594 | return r; |
2595 | } | |
2596 | ||
2597 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2598 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2599 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2600 | metadata_dev_size, sb_metadata_dev_size); |
2601 | return -EINVAL; | |
2602 | ||
2603 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
2604 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2605 | DMERR("%s: unable to grow the metadata device until repaired.", | |
2606 | dm_device_name(pool->pool_md)); | |
2607 | return 0; | |
2608 | } | |
2609 | ||
7d48935e | 2610 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
2611 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2612 | dm_device_name(pool->pool_md), | |
2613 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2614 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2615 | if (r) { | |
b5330655 | 2616 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2617 | return r; |
2618 | } | |
2619 | ||
2620 | *need_commit = true; | |
2621 | } | |
2622 | ||
2623 | return 0; | |
2624 | } | |
2625 | ||
b17446df JT |
2626 | /* |
2627 | * Retrieves the number of blocks of the data device from | |
2628 | * the superblock and compares it to the actual device size, | |
2629 | * thus resizing the data device in case it has grown. | |
2630 | * | |
2631 | * This both copes with opening preallocated data devices in the ctr | |
2632 | * being followed by a resume | |
2633 | * -and- | |
2634 | * calling the resume method individually after userspace has | |
2635 | * grown the data device in reaction to a table event. | |
2636 | */ | |
2637 | static int pool_preresume(struct dm_target *ti) | |
2638 | { | |
2639 | int r; | |
24347e95 | 2640 | bool need_commit1, need_commit2; |
b17446df JT |
2641 | struct pool_c *pt = ti->private; |
2642 | struct pool *pool = pt->pool; | |
2643 | ||
2644 | /* | |
2645 | * Take control of the pool object. | |
2646 | */ | |
2647 | r = bind_control_target(pool, ti); | |
2648 | if (r) | |
2649 | return r; | |
2650 | ||
2651 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2652 | if (r) | |
2653 | return r; | |
2654 | ||
24347e95 JT |
2655 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2656 | if (r) | |
2657 | return r; | |
2658 | ||
2659 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2660 | (void) commit(pool); |
b17446df JT |
2661 | |
2662 | return 0; | |
2663 | } | |
2664 | ||
991d9fa0 JT |
2665 | static void pool_resume(struct dm_target *ti) |
2666 | { | |
2667 | struct pool_c *pt = ti->private; | |
2668 | struct pool *pool = pt->pool; | |
2669 | unsigned long flags; | |
2670 | ||
2671 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 2672 | pool->low_water_triggered = false; |
991d9fa0 | 2673 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 2674 | requeue_bios(pool); |
991d9fa0 | 2675 | |
905e51b3 | 2676 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2677 | } |
2678 | ||
2679 | static void pool_postsuspend(struct dm_target *ti) | |
2680 | { | |
991d9fa0 JT |
2681 | struct pool_c *pt = ti->private; |
2682 | struct pool *pool = pt->pool; | |
2683 | ||
905e51b3 | 2684 | cancel_delayed_work(&pool->waker); |
85ad643b | 2685 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 2686 | flush_workqueue(pool->wq); |
020cc3b5 | 2687 | (void) commit(pool); |
991d9fa0 JT |
2688 | } |
2689 | ||
2690 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2691 | { | |
2692 | if (argc != args_required) { | |
2693 | DMWARN("Message received with %u arguments instead of %u.", | |
2694 | argc, args_required); | |
2695 | return -EINVAL; | |
2696 | } | |
2697 | ||
2698 | return 0; | |
2699 | } | |
2700 | ||
2701 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2702 | { | |
2703 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2704 | *dev_id <= MAX_DEV_ID) | |
2705 | return 0; | |
2706 | ||
2707 | if (warning) | |
2708 | DMWARN("Message received with invalid device id: %s", arg); | |
2709 | ||
2710 | return -EINVAL; | |
2711 | } | |
2712 | ||
2713 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2714 | { | |
2715 | dm_thin_id dev_id; | |
2716 | int r; | |
2717 | ||
2718 | r = check_arg_count(argc, 2); | |
2719 | if (r) | |
2720 | return r; | |
2721 | ||
2722 | r = read_dev_id(argv[1], &dev_id, 1); | |
2723 | if (r) | |
2724 | return r; | |
2725 | ||
2726 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2727 | if (r) { | |
2728 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2729 | argv[1]); | |
2730 | return r; | |
2731 | } | |
2732 | ||
2733 | return 0; | |
2734 | } | |
2735 | ||
2736 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2737 | { | |
2738 | dm_thin_id dev_id; | |
2739 | dm_thin_id origin_dev_id; | |
2740 | int r; | |
2741 | ||
2742 | r = check_arg_count(argc, 3); | |
2743 | if (r) | |
2744 | return r; | |
2745 | ||
2746 | r = read_dev_id(argv[1], &dev_id, 1); | |
2747 | if (r) | |
2748 | return r; | |
2749 | ||
2750 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2751 | if (r) | |
2752 | return r; | |
2753 | ||
2754 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2755 | if (r) { | |
2756 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2757 | argv[1], argv[2]); | |
2758 | return r; | |
2759 | } | |
2760 | ||
2761 | return 0; | |
2762 | } | |
2763 | ||
2764 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2765 | { | |
2766 | dm_thin_id dev_id; | |
2767 | int r; | |
2768 | ||
2769 | r = check_arg_count(argc, 2); | |
2770 | if (r) | |
2771 | return r; | |
2772 | ||
2773 | r = read_dev_id(argv[1], &dev_id, 1); | |
2774 | if (r) | |
2775 | return r; | |
2776 | ||
2777 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2778 | if (r) | |
2779 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2780 | ||
2781 | return r; | |
2782 | } | |
2783 | ||
2784 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2785 | { | |
2786 | dm_thin_id old_id, new_id; | |
2787 | int r; | |
2788 | ||
2789 | r = check_arg_count(argc, 3); | |
2790 | if (r) | |
2791 | return r; | |
2792 | ||
2793 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2794 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2795 | return -EINVAL; | |
2796 | } | |
2797 | ||
2798 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2799 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2800 | return -EINVAL; | |
2801 | } | |
2802 | ||
2803 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2804 | if (r) { | |
2805 | DMWARN("Failed to change transaction id from %s to %s.", | |
2806 | argv[1], argv[2]); | |
2807 | return r; | |
2808 | } | |
2809 | ||
2810 | return 0; | |
2811 | } | |
2812 | ||
cc8394d8 JT |
2813 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2814 | { | |
2815 | int r; | |
2816 | ||
2817 | r = check_arg_count(argc, 1); | |
2818 | if (r) | |
2819 | return r; | |
2820 | ||
020cc3b5 | 2821 | (void) commit(pool); |
0d200aef | 2822 | |
cc8394d8 JT |
2823 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2824 | if (r) | |
2825 | DMWARN("reserve_metadata_snap message failed."); | |
2826 | ||
2827 | return r; | |
2828 | } | |
2829 | ||
2830 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2831 | { | |
2832 | int r; | |
2833 | ||
2834 | r = check_arg_count(argc, 1); | |
2835 | if (r) | |
2836 | return r; | |
2837 | ||
2838 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2839 | if (r) | |
2840 | DMWARN("release_metadata_snap message failed."); | |
2841 | ||
2842 | return r; | |
2843 | } | |
2844 | ||
991d9fa0 JT |
2845 | /* |
2846 | * Messages supported: | |
2847 | * create_thin <dev_id> | |
2848 | * create_snap <dev_id> <origin_id> | |
2849 | * delete <dev_id> | |
2850 | * trim <dev_id> <new_size_in_sectors> | |
2851 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2852 | * reserve_metadata_snap |
2853 | * release_metadata_snap | |
991d9fa0 JT |
2854 | */ |
2855 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2856 | { | |
2857 | int r = -EINVAL; | |
2858 | struct pool_c *pt = ti->private; | |
2859 | struct pool *pool = pt->pool; | |
2860 | ||
2861 | if (!strcasecmp(argv[0], "create_thin")) | |
2862 | r = process_create_thin_mesg(argc, argv, pool); | |
2863 | ||
2864 | else if (!strcasecmp(argv[0], "create_snap")) | |
2865 | r = process_create_snap_mesg(argc, argv, pool); | |
2866 | ||
2867 | else if (!strcasecmp(argv[0], "delete")) | |
2868 | r = process_delete_mesg(argc, argv, pool); | |
2869 | ||
2870 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2871 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2872 | ||
cc8394d8 JT |
2873 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2874 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2875 | ||
2876 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2877 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2878 | ||
991d9fa0 JT |
2879 | else |
2880 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2881 | ||
e49e5829 | 2882 | if (!r) |
020cc3b5 | 2883 | (void) commit(pool); |
991d9fa0 JT |
2884 | |
2885 | return r; | |
2886 | } | |
2887 | ||
e49e5829 JT |
2888 | static void emit_flags(struct pool_features *pf, char *result, |
2889 | unsigned sz, unsigned maxlen) | |
2890 | { | |
2891 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
2892 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
2893 | pf->error_if_no_space; | |
e49e5829 JT |
2894 | DMEMIT("%u ", count); |
2895 | ||
2896 | if (!pf->zero_new_blocks) | |
2897 | DMEMIT("skip_block_zeroing "); | |
2898 | ||
2899 | if (!pf->discard_enabled) | |
2900 | DMEMIT("ignore_discard "); | |
2901 | ||
2902 | if (!pf->discard_passdown) | |
2903 | DMEMIT("no_discard_passdown "); | |
2904 | ||
2905 | if (pf->mode == PM_READ_ONLY) | |
2906 | DMEMIT("read_only "); | |
787a996c MS |
2907 | |
2908 | if (pf->error_if_no_space) | |
2909 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
2910 | } |
2911 | ||
991d9fa0 JT |
2912 | /* |
2913 | * Status line is: | |
2914 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2915 | * <used data sectors>/<total data sectors> <held metadata root> | |
2916 | */ | |
fd7c092e MP |
2917 | static void pool_status(struct dm_target *ti, status_type_t type, |
2918 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2919 | { |
e49e5829 | 2920 | int r; |
991d9fa0 JT |
2921 | unsigned sz = 0; |
2922 | uint64_t transaction_id; | |
2923 | dm_block_t nr_free_blocks_data; | |
2924 | dm_block_t nr_free_blocks_metadata; | |
2925 | dm_block_t nr_blocks_data; | |
2926 | dm_block_t nr_blocks_metadata; | |
2927 | dm_block_t held_root; | |
2928 | char buf[BDEVNAME_SIZE]; | |
2929 | char buf2[BDEVNAME_SIZE]; | |
2930 | struct pool_c *pt = ti->private; | |
2931 | struct pool *pool = pt->pool; | |
2932 | ||
2933 | switch (type) { | |
2934 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2935 | if (get_pool_mode(pool) == PM_FAIL) { |
2936 | DMEMIT("Fail"); | |
2937 | break; | |
2938 | } | |
2939 | ||
1f4e0ff0 AK |
2940 | /* Commit to ensure statistics aren't out-of-date */ |
2941 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2942 | (void) commit(pool); |
1f4e0ff0 | 2943 | |
fd7c092e MP |
2944 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2945 | if (r) { | |
4fa5971a MS |
2946 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2947 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2948 | goto err; |
2949 | } | |
991d9fa0 | 2950 | |
fd7c092e MP |
2951 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2952 | if (r) { | |
4fa5971a MS |
2953 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2954 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2955 | goto err; |
2956 | } | |
991d9fa0 JT |
2957 | |
2958 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2959 | if (r) { |
4fa5971a MS |
2960 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2961 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2962 | goto err; |
2963 | } | |
991d9fa0 | 2964 | |
fd7c092e MP |
2965 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2966 | if (r) { | |
4fa5971a MS |
2967 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2968 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2969 | goto err; |
2970 | } | |
991d9fa0 JT |
2971 | |
2972 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2973 | if (r) { |
4fa5971a MS |
2974 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2975 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2976 | goto err; |
2977 | } | |
991d9fa0 | 2978 | |
cc8394d8 | 2979 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2980 | if (r) { |
4fa5971a MS |
2981 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2982 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2983 | goto err; |
2984 | } | |
991d9fa0 JT |
2985 | |
2986 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2987 | (unsigned long long)transaction_id, | |
2988 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2989 | (unsigned long long)nr_blocks_metadata, | |
2990 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2991 | (unsigned long long)nr_blocks_data); | |
2992 | ||
2993 | if (held_root) | |
e49e5829 JT |
2994 | DMEMIT("%llu ", held_root); |
2995 | else | |
2996 | DMEMIT("- "); | |
2997 | ||
3e1a0699 JT |
2998 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
2999 | DMEMIT("out_of_data_space "); | |
3000 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3001 | DMEMIT("ro "); |
991d9fa0 | 3002 | else |
e49e5829 JT |
3003 | DMEMIT("rw "); |
3004 | ||
018debea | 3005 | if (!pool->pf.discard_enabled) |
787a996c | 3006 | DMEMIT("ignore_discard "); |
018debea | 3007 | else if (pool->pf.discard_passdown) |
787a996c MS |
3008 | DMEMIT("discard_passdown "); |
3009 | else | |
3010 | DMEMIT("no_discard_passdown "); | |
3011 | ||
3012 | if (pool->pf.error_if_no_space) | |
3013 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3014 | else |
787a996c | 3015 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3016 | |
3017 | break; | |
3018 | ||
3019 | case STATUSTYPE_TABLE: | |
3020 | DMEMIT("%s %s %lu %llu ", | |
3021 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3022 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3023 | (unsigned long)pool->sectors_per_block, | |
3024 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3025 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3026 | break; |
3027 | } | |
fd7c092e | 3028 | return; |
991d9fa0 | 3029 | |
fd7c092e MP |
3030 | err: |
3031 | DMEMIT("Error"); | |
991d9fa0 JT |
3032 | } |
3033 | ||
3034 | static int pool_iterate_devices(struct dm_target *ti, | |
3035 | iterate_devices_callout_fn fn, void *data) | |
3036 | { | |
3037 | struct pool_c *pt = ti->private; | |
3038 | ||
3039 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3040 | } | |
3041 | ||
3042 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3043 | struct bio_vec *biovec, int max_size) | |
3044 | { | |
3045 | struct pool_c *pt = ti->private; | |
3046 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3047 | ||
3048 | if (!q->merge_bvec_fn) | |
3049 | return max_size; | |
3050 | ||
3051 | bvm->bi_bdev = pt->data_dev->bdev; | |
3052 | ||
3053 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3054 | } | |
3055 | ||
0424caa1 | 3056 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 3057 | { |
0424caa1 MS |
3058 | struct pool *pool = pt->pool; |
3059 | struct queue_limits *data_limits; | |
3060 | ||
104655fd JT |
3061 | limits->max_discard_sectors = pool->sectors_per_block; |
3062 | ||
3063 | /* | |
0424caa1 | 3064 | * discard_granularity is just a hint, and not enforced. |
104655fd | 3065 | */ |
0424caa1 MS |
3066 | if (pt->adjusted_pf.discard_passdown) { |
3067 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
3068 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 3069 | } else |
0424caa1 | 3070 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3071 | } |
3072 | ||
991d9fa0 JT |
3073 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3074 | { | |
3075 | struct pool_c *pt = ti->private; | |
3076 | struct pool *pool = pt->pool; | |
0cc67cd9 | 3077 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 3078 | |
0cc67cd9 MS |
3079 | /* |
3080 | * If the system-determined stacked limits are compatible with the | |
3081 | * pool's blocksize (io_opt is a factor) do not override them. | |
3082 | */ | |
3083 | if (io_opt_sectors < pool->sectors_per_block || | |
3084 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
3085 | blk_limits_io_min(limits, 0); | |
3086 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
3087 | } | |
0424caa1 MS |
3088 | |
3089 | /* | |
3090 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3091 | * They get transferred to the live pool in bind_control_target() | |
3092 | * called from pool_preresume(). | |
3093 | */ | |
b60ab990 MS |
3094 | if (!pt->adjusted_pf.discard_enabled) { |
3095 | /* | |
3096 | * Must explicitly disallow stacking discard limits otherwise the | |
3097 | * block layer will stack them if pool's data device has support. | |
3098 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3099 | * user to see that, so make sure to set all discard limits to 0. | |
3100 | */ | |
3101 | limits->discard_granularity = 0; | |
0424caa1 | 3102 | return; |
b60ab990 | 3103 | } |
0424caa1 MS |
3104 | |
3105 | disable_passdown_if_not_supported(pt); | |
3106 | ||
3107 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3108 | } |
3109 | ||
3110 | static struct target_type pool_target = { | |
3111 | .name = "thin-pool", | |
3112 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3113 | DM_TARGET_IMMUTABLE, | |
67324ea1 | 3114 | .version = {1, 12, 0}, |
991d9fa0 JT |
3115 | .module = THIS_MODULE, |
3116 | .ctr = pool_ctr, | |
3117 | .dtr = pool_dtr, | |
3118 | .map = pool_map, | |
3119 | .postsuspend = pool_postsuspend, | |
3120 | .preresume = pool_preresume, | |
3121 | .resume = pool_resume, | |
3122 | .message = pool_message, | |
3123 | .status = pool_status, | |
3124 | .merge = pool_merge, | |
3125 | .iterate_devices = pool_iterate_devices, | |
3126 | .io_hints = pool_io_hints, | |
3127 | }; | |
3128 | ||
3129 | /*---------------------------------------------------------------- | |
3130 | * Thin target methods | |
3131 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3132 | static void thin_get(struct thin_c *tc) |
3133 | { | |
3134 | atomic_inc(&tc->refcount); | |
3135 | } | |
3136 | ||
3137 | static void thin_put(struct thin_c *tc) | |
3138 | { | |
3139 | if (atomic_dec_and_test(&tc->refcount)) | |
3140 | complete(&tc->can_destroy); | |
3141 | } | |
3142 | ||
991d9fa0 JT |
3143 | static void thin_dtr(struct dm_target *ti) |
3144 | { | |
3145 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3146 | unsigned long flags; |
3147 | ||
b10ebd34 JT |
3148 | thin_put(tc); |
3149 | wait_for_completion(&tc->can_destroy); | |
3150 | ||
c140e1c4 MS |
3151 | spin_lock_irqsave(&tc->pool->lock, flags); |
3152 | list_del_rcu(&tc->list); | |
3153 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3154 | synchronize_rcu(); | |
991d9fa0 JT |
3155 | |
3156 | mutex_lock(&dm_thin_pool_table.mutex); | |
3157 | ||
3158 | __pool_dec(tc->pool); | |
3159 | dm_pool_close_thin_device(tc->td); | |
3160 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3161 | if (tc->origin_dev) |
3162 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3163 | kfree(tc); |
3164 | ||
3165 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * Thin target parameters: | |
3170 | * | |
2dd9c257 | 3171 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3172 | * |
3173 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3174 | * dev_id: the internal device identifier | |
2dd9c257 | 3175 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3176 | * |
3177 | * If the pool device has discards disabled, they get disabled for the thin | |
3178 | * device as well. | |
991d9fa0 JT |
3179 | */ |
3180 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3181 | { | |
3182 | int r; | |
3183 | struct thin_c *tc; | |
2dd9c257 | 3184 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3185 | struct mapped_device *pool_md; |
5e3283e2 | 3186 | unsigned long flags; |
991d9fa0 JT |
3187 | |
3188 | mutex_lock(&dm_thin_pool_table.mutex); | |
3189 | ||
2dd9c257 | 3190 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3191 | ti->error = "Invalid argument count"; |
3192 | r = -EINVAL; | |
3193 | goto out_unlock; | |
3194 | } | |
3195 | ||
3196 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3197 | if (!tc) { | |
3198 | ti->error = "Out of memory"; | |
3199 | r = -ENOMEM; | |
3200 | goto out_unlock; | |
3201 | } | |
c140e1c4 MS |
3202 | spin_lock_init(&tc->lock); |
3203 | bio_list_init(&tc->deferred_bio_list); | |
3204 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3205 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3206 | |
2dd9c257 JT |
3207 | if (argc == 3) { |
3208 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3209 | if (r) { | |
3210 | ti->error = "Error opening origin device"; | |
3211 | goto bad_origin_dev; | |
3212 | } | |
3213 | tc->origin_dev = origin_dev; | |
3214 | } | |
3215 | ||
991d9fa0 JT |
3216 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3217 | if (r) { | |
3218 | ti->error = "Error opening pool device"; | |
3219 | goto bad_pool_dev; | |
3220 | } | |
3221 | tc->pool_dev = pool_dev; | |
3222 | ||
3223 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3224 | ti->error = "Invalid device id"; | |
3225 | r = -EINVAL; | |
3226 | goto bad_common; | |
3227 | } | |
3228 | ||
3229 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3230 | if (!pool_md) { | |
3231 | ti->error = "Couldn't get pool mapped device"; | |
3232 | r = -EINVAL; | |
3233 | goto bad_common; | |
3234 | } | |
3235 | ||
3236 | tc->pool = __pool_table_lookup(pool_md); | |
3237 | if (!tc->pool) { | |
3238 | ti->error = "Couldn't find pool object"; | |
3239 | r = -EINVAL; | |
3240 | goto bad_pool_lookup; | |
3241 | } | |
3242 | __pool_inc(tc->pool); | |
3243 | ||
e49e5829 JT |
3244 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3245 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3246 | r = -EINVAL; |
e49e5829 JT |
3247 | goto bad_thin_open; |
3248 | } | |
3249 | ||
991d9fa0 JT |
3250 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3251 | if (r) { | |
3252 | ti->error = "Couldn't open thin internal device"; | |
3253 | goto bad_thin_open; | |
3254 | } | |
3255 | ||
542f9038 MS |
3256 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3257 | if (r) | |
1acacc07 | 3258 | goto bad_target_max_io_len; |
542f9038 | 3259 | |
55a62eef | 3260 | ti->num_flush_bios = 1; |
16ad3d10 | 3261 | ti->flush_supported = true; |
59c3d2c6 | 3262 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3263 | |
3264 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3265 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3266 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3267 | ti->discards_supported = true; |
55a62eef | 3268 | ti->num_discard_bios = 1; |
55a62eef AK |
3269 | /* Discard bios must be split on a block boundary */ |
3270 | ti->split_discard_bios = true; | |
67e2e2b2 | 3271 | } |
991d9fa0 JT |
3272 | |
3273 | dm_put(pool_md); | |
3274 | ||
3275 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3276 | ||
b10ebd34 JT |
3277 | atomic_set(&tc->refcount, 1); |
3278 | init_completion(&tc->can_destroy); | |
3279 | ||
5e3283e2 | 3280 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3281 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3282 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3283 | /* |
3284 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3285 | * wake_worker() call finding no bios to process (because the newly | |
3286 | * added tc isn't yet visible). So this reduces latency since we | |
3287 | * aren't then dependent on the periodic commit to wake_worker(). | |
3288 | */ | |
3289 | synchronize_rcu(); | |
3290 | ||
991d9fa0 JT |
3291 | return 0; |
3292 | ||
1acacc07 MS |
3293 | bad_target_max_io_len: |
3294 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3295 | bad_thin_open: |
3296 | __pool_dec(tc->pool); | |
3297 | bad_pool_lookup: | |
3298 | dm_put(pool_md); | |
3299 | bad_common: | |
3300 | dm_put_device(ti, tc->pool_dev); | |
3301 | bad_pool_dev: | |
2dd9c257 JT |
3302 | if (tc->origin_dev) |
3303 | dm_put_device(ti, tc->origin_dev); | |
3304 | bad_origin_dev: | |
991d9fa0 JT |
3305 | kfree(tc); |
3306 | out_unlock: | |
3307 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3308 | ||
3309 | return r; | |
3310 | } | |
3311 | ||
7de3ee57 | 3312 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3313 | { |
4f024f37 | 3314 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3315 | |
7de3ee57 | 3316 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3317 | } |
3318 | ||
7de3ee57 | 3319 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3320 | { |
3321 | unsigned long flags; | |
59c3d2c6 | 3322 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3323 | struct list_head work; |
a24c2569 | 3324 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3325 | struct pool *pool = h->tc->pool; |
3326 | ||
3327 | if (h->shared_read_entry) { | |
3328 | INIT_LIST_HEAD(&work); | |
44feb387 | 3329 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3330 | |
3331 | spin_lock_irqsave(&pool->lock, flags); | |
3332 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3333 | list_del(&m->list); | |
7f214665 | 3334 | m->quiesced = true; |
eb2aa48d JT |
3335 | __maybe_add_mapping(m); |
3336 | } | |
3337 | spin_unlock_irqrestore(&pool->lock, flags); | |
3338 | } | |
3339 | ||
104655fd JT |
3340 | if (h->all_io_entry) { |
3341 | INIT_LIST_HEAD(&work); | |
44feb387 | 3342 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3343 | if (!list_empty(&work)) { |
3344 | spin_lock_irqsave(&pool->lock, flags); | |
3345 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3346 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3347 | spin_unlock_irqrestore(&pool->lock, flags); |
3348 | wake_worker(pool); | |
3349 | } | |
104655fd JT |
3350 | } |
3351 | ||
eb2aa48d JT |
3352 | return 0; |
3353 | } | |
3354 | ||
738211f7 | 3355 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3356 | { |
738211f7 JT |
3357 | struct thin_c *tc = ti->private; |
3358 | ||
991d9fa0 | 3359 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3360 | noflush_work(tc, do_noflush_start); |
3361 | } | |
3362 | ||
3363 | static void thin_postsuspend(struct dm_target *ti) | |
3364 | { | |
3365 | struct thin_c *tc = ti->private; | |
3366 | ||
3367 | /* | |
3368 | * The dm_noflush_suspending flag has been cleared by now, so | |
3369 | * unfortunately we must always run this. | |
3370 | */ | |
3371 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3372 | } |
3373 | ||
3374 | /* | |
3375 | * <nr mapped sectors> <highest mapped sector> | |
3376 | */ | |
fd7c092e MP |
3377 | static void thin_status(struct dm_target *ti, status_type_t type, |
3378 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3379 | { |
3380 | int r; | |
3381 | ssize_t sz = 0; | |
3382 | dm_block_t mapped, highest; | |
3383 | char buf[BDEVNAME_SIZE]; | |
3384 | struct thin_c *tc = ti->private; | |
3385 | ||
e49e5829 JT |
3386 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3387 | DMEMIT("Fail"); | |
fd7c092e | 3388 | return; |
e49e5829 JT |
3389 | } |
3390 | ||
991d9fa0 JT |
3391 | if (!tc->td) |
3392 | DMEMIT("-"); | |
3393 | else { | |
3394 | switch (type) { | |
3395 | case STATUSTYPE_INFO: | |
3396 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3397 | if (r) { |
3398 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3399 | goto err; | |
3400 | } | |
991d9fa0 JT |
3401 | |
3402 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3403 | if (r < 0) { |
3404 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3405 | goto err; | |
3406 | } | |
991d9fa0 JT |
3407 | |
3408 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3409 | if (r) | |
3410 | DMEMIT("%llu", ((highest + 1) * | |
3411 | tc->pool->sectors_per_block) - 1); | |
3412 | else | |
3413 | DMEMIT("-"); | |
3414 | break; | |
3415 | ||
3416 | case STATUSTYPE_TABLE: | |
3417 | DMEMIT("%s %lu", | |
3418 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3419 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3420 | if (tc->origin_dev) |
3421 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3422 | break; |
3423 | } | |
3424 | } | |
3425 | ||
fd7c092e MP |
3426 | return; |
3427 | ||
3428 | err: | |
3429 | DMEMIT("Error"); | |
991d9fa0 JT |
3430 | } |
3431 | ||
3432 | static int thin_iterate_devices(struct dm_target *ti, | |
3433 | iterate_devices_callout_fn fn, void *data) | |
3434 | { | |
55f2b8bd | 3435 | sector_t blocks; |
991d9fa0 | 3436 | struct thin_c *tc = ti->private; |
55f2b8bd | 3437 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3438 | |
3439 | /* | |
3440 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3441 | * we follow a more convoluted path through to the pool's target. | |
3442 | */ | |
55f2b8bd | 3443 | if (!pool->ti) |
991d9fa0 JT |
3444 | return 0; /* nothing is bound */ |
3445 | ||
55f2b8bd MS |
3446 | blocks = pool->ti->len; |
3447 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3448 | if (blocks) |
55f2b8bd | 3449 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3450 | |
3451 | return 0; | |
3452 | } | |
3453 | ||
991d9fa0 JT |
3454 | static struct target_type thin_target = { |
3455 | .name = "thin", | |
67324ea1 | 3456 | .version = {1, 12, 0}, |
991d9fa0 JT |
3457 | .module = THIS_MODULE, |
3458 | .ctr = thin_ctr, | |
3459 | .dtr = thin_dtr, | |
3460 | .map = thin_map, | |
eb2aa48d | 3461 | .end_io = thin_endio, |
738211f7 | 3462 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3463 | .postsuspend = thin_postsuspend, |
3464 | .status = thin_status, | |
3465 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3466 | }; |
3467 | ||
3468 | /*----------------------------------------------------------------*/ | |
3469 | ||
3470 | static int __init dm_thin_init(void) | |
3471 | { | |
3472 | int r; | |
3473 | ||
3474 | pool_table_init(); | |
3475 | ||
3476 | r = dm_register_target(&thin_target); | |
3477 | if (r) | |
3478 | return r; | |
3479 | ||
3480 | r = dm_register_target(&pool_target); | |
3481 | if (r) | |
a24c2569 MS |
3482 | goto bad_pool_target; |
3483 | ||
3484 | r = -ENOMEM; | |
3485 | ||
a24c2569 MS |
3486 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3487 | if (!_new_mapping_cache) | |
3488 | goto bad_new_mapping_cache; | |
3489 | ||
a24c2569 MS |
3490 | return 0; |
3491 | ||
a24c2569 | 3492 | bad_new_mapping_cache: |
a24c2569 MS |
3493 | dm_unregister_target(&pool_target); |
3494 | bad_pool_target: | |
3495 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3496 | |
3497 | return r; | |
3498 | } | |
3499 | ||
3500 | static void dm_thin_exit(void) | |
3501 | { | |
3502 | dm_unregister_target(&thin_target); | |
3503 | dm_unregister_target(&pool_target); | |
a24c2569 | 3504 | |
a24c2569 | 3505 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3506 | } |
3507 | ||
3508 | module_init(dm_thin_init); | |
3509 | module_exit(dm_thin_exit); | |
3510 | ||
7cab8bf1 | 3511 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3512 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3513 | MODULE_LICENSE("GPL"); |