]>
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> | |
15 | #include <linux/init.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/slab.h> | |
18 | ||
19 | #define DM_MSG_PREFIX "thin" | |
20 | ||
21 | /* | |
22 | * Tunable constants | |
23 | */ | |
7768ed33 | 24 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
25 | #define MAPPING_POOL_SIZE 1024 |
26 | #define PRISON_CELLS 1024 | |
905e51b3 | 27 | #define COMMIT_PERIOD HZ |
991d9fa0 | 28 | |
df5d2e90 MP |
29 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
30 | "A percentage of time allocated for copy on write"); | |
31 | ||
991d9fa0 JT |
32 | /* |
33 | * The block size of the device holding pool data must be | |
34 | * between 64KB and 1GB. | |
35 | */ | |
36 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
37 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
38 | ||
991d9fa0 JT |
39 | /* |
40 | * Device id is restricted to 24 bits. | |
41 | */ | |
42 | #define MAX_DEV_ID ((1 << 24) - 1) | |
43 | ||
44 | /* | |
45 | * How do we handle breaking sharing of data blocks? | |
46 | * ================================================= | |
47 | * | |
48 | * We use a standard copy-on-write btree to store the mappings for the | |
49 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
50 | * the data). When you take an internal snapshot you clone the root node | |
51 | * of the origin btree. After this there is no concept of an origin or a | |
52 | * snapshot. They are just two device trees that happen to point to the | |
53 | * same data blocks. | |
54 | * | |
55 | * When we get a write in we decide if it's to a shared data block using | |
56 | * some timestamp magic. If it is, we have to break sharing. | |
57 | * | |
58 | * Let's say we write to a shared block in what was the origin. The | |
59 | * steps are: | |
60 | * | |
61 | * i) plug io further to this physical block. (see bio_prison code). | |
62 | * | |
63 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 64 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
65 | * |
66 | * iii) copy the data block to a newly allocate block. This step can be | |
67 | * missed out if the io covers the block. (schedule_copy). | |
68 | * | |
69 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 70 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
71 | * sharing of btree nodes between the two devices. Breaking sharing only |
72 | * effects the btree of that specific device. Btrees for the other | |
73 | * devices that share the block never change. The btree for the origin | |
74 | * device as it was after the last commit is untouched, ie. we're using | |
75 | * persistent data structures in the functional programming sense. | |
76 | * | |
77 | * v) unplug io to this physical block, including the io that triggered | |
78 | * the breaking of sharing. | |
79 | * | |
80 | * Steps (ii) and (iii) occur in parallel. | |
81 | * | |
82 | * The metadata _doesn't_ need to be committed before the io continues. We | |
83 | * get away with this because the io is always written to a _new_ block. | |
84 | * If there's a crash, then: | |
85 | * | |
86 | * - The origin mapping will point to the old origin block (the shared | |
87 | * one). This will contain the data as it was before the io that triggered | |
88 | * the breaking of sharing came in. | |
89 | * | |
90 | * - The snap mapping still points to the old block. As it would after | |
91 | * the commit. | |
92 | * | |
93 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
94 | * will continue to think that data block in the snapshot device is shared | |
95 | * even after the write to the origin has broken sharing. I suspect data | |
96 | * blocks will typically be shared by many different devices, so we're | |
97 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
98 | * devices that reference this data block. At the moment I think the | |
99 | * benefits far, far outweigh the disadvantages. | |
100 | */ | |
101 | ||
102 | /*----------------------------------------------------------------*/ | |
103 | ||
991d9fa0 JT |
104 | /* |
105 | * Key building. | |
106 | */ | |
107 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 108 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
109 | { |
110 | key->virtual = 0; | |
111 | key->dev = dm_thin_dev_id(td); | |
112 | key->block = b; | |
113 | } | |
114 | ||
115 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 116 | struct dm_cell_key *key) |
991d9fa0 JT |
117 | { |
118 | key->virtual = 1; | |
119 | key->dev = dm_thin_dev_id(td); | |
120 | key->block = b; | |
121 | } | |
122 | ||
123 | /*----------------------------------------------------------------*/ | |
124 | ||
125 | /* | |
126 | * A pool device ties together a metadata device and a data device. It | |
127 | * also provides the interface for creating and destroying internal | |
128 | * devices. | |
129 | */ | |
a24c2569 | 130 | struct dm_thin_new_mapping; |
67e2e2b2 | 131 | |
e49e5829 | 132 | /* |
3e1a0699 | 133 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
134 | */ |
135 | enum pool_mode { | |
136 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 137 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
138 | PM_READ_ONLY, /* metadata may not be changed */ |
139 | PM_FAIL, /* all I/O fails */ | |
140 | }; | |
141 | ||
67e2e2b2 | 142 | struct pool_features { |
e49e5829 JT |
143 | enum pool_mode mode; |
144 | ||
9bc142dd MS |
145 | bool zero_new_blocks:1; |
146 | bool discard_enabled:1; | |
147 | bool discard_passdown:1; | |
787a996c | 148 | bool error_if_no_space:1; |
67e2e2b2 JT |
149 | }; |
150 | ||
e49e5829 JT |
151 | struct thin_c; |
152 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
153 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
154 | ||
991d9fa0 JT |
155 | struct pool { |
156 | struct list_head list; | |
157 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
158 | ||
159 | struct mapped_device *pool_md; | |
160 | struct block_device *md_dev; | |
161 | struct dm_pool_metadata *pmd; | |
162 | ||
991d9fa0 | 163 | dm_block_t low_water_blocks; |
55f2b8bd | 164 | uint32_t sectors_per_block; |
f9a8e0cd | 165 | int sectors_per_block_shift; |
991d9fa0 | 166 | |
67e2e2b2 | 167 | struct pool_features pf; |
88a6621b | 168 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 169 | |
44feb387 | 170 | struct dm_bio_prison *prison; |
991d9fa0 JT |
171 | struct dm_kcopyd_client *copier; |
172 | ||
173 | struct workqueue_struct *wq; | |
174 | struct work_struct worker; | |
905e51b3 | 175 | struct delayed_work waker; |
991d9fa0 | 176 | |
905e51b3 | 177 | unsigned long last_commit_jiffies; |
55f2b8bd | 178 | unsigned ref_count; |
991d9fa0 JT |
179 | |
180 | spinlock_t lock; | |
181 | struct bio_list deferred_bios; | |
182 | struct bio_list deferred_flush_bios; | |
183 | struct list_head prepared_mappings; | |
104655fd | 184 | struct list_head prepared_discards; |
991d9fa0 JT |
185 | |
186 | struct bio_list retry_on_resume_list; | |
187 | ||
44feb387 MS |
188 | struct dm_deferred_set *shared_read_ds; |
189 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 190 | |
a24c2569 | 191 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 192 | mempool_t *mapping_pool; |
e49e5829 JT |
193 | |
194 | process_bio_fn process_bio; | |
195 | process_bio_fn process_discard; | |
196 | ||
197 | process_mapping_fn process_prepared_mapping; | |
198 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
199 | }; |
200 | ||
e49e5829 | 201 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 202 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 203 | |
991d9fa0 JT |
204 | /* |
205 | * Target context for a pool. | |
206 | */ | |
207 | struct pool_c { | |
208 | struct dm_target *ti; | |
209 | struct pool *pool; | |
210 | struct dm_dev *data_dev; | |
211 | struct dm_dev *metadata_dev; | |
212 | struct dm_target_callbacks callbacks; | |
213 | ||
214 | dm_block_t low_water_blocks; | |
0424caa1 MS |
215 | struct pool_features requested_pf; /* Features requested during table load */ |
216 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
217 | }; |
218 | ||
219 | /* | |
220 | * Target context for a thin. | |
221 | */ | |
222 | struct thin_c { | |
223 | struct dm_dev *pool_dev; | |
2dd9c257 | 224 | struct dm_dev *origin_dev; |
991d9fa0 JT |
225 | dm_thin_id dev_id; |
226 | ||
227 | struct pool *pool; | |
228 | struct dm_thin_device *td; | |
738211f7 | 229 | bool requeue_mode:1; |
991d9fa0 JT |
230 | }; |
231 | ||
232 | /*----------------------------------------------------------------*/ | |
233 | ||
025b9685 JT |
234 | /* |
235 | * wake_worker() is used when new work is queued and when pool_resume is | |
236 | * ready to continue deferred IO processing. | |
237 | */ | |
238 | static void wake_worker(struct pool *pool) | |
239 | { | |
240 | queue_work(pool->wq, &pool->worker); | |
241 | } | |
242 | ||
243 | /*----------------------------------------------------------------*/ | |
244 | ||
6beca5eb JT |
245 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
246 | struct dm_bio_prison_cell **cell_result) | |
247 | { | |
248 | int r; | |
249 | struct dm_bio_prison_cell *cell_prealloc; | |
250 | ||
251 | /* | |
252 | * Allocate a cell from the prison's mempool. | |
253 | * This might block but it can't fail. | |
254 | */ | |
255 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
256 | ||
257 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
258 | if (r) | |
259 | /* | |
260 | * We reused an old cell; we can get rid of | |
261 | * the new one. | |
262 | */ | |
263 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
264 | ||
265 | return r; | |
266 | } | |
267 | ||
268 | static void cell_release(struct pool *pool, | |
269 | struct dm_bio_prison_cell *cell, | |
270 | struct bio_list *bios) | |
271 | { | |
272 | dm_cell_release(pool->prison, cell, bios); | |
273 | dm_bio_prison_free_cell(pool->prison, cell); | |
274 | } | |
275 | ||
276 | static void cell_release_no_holder(struct pool *pool, | |
277 | struct dm_bio_prison_cell *cell, | |
278 | struct bio_list *bios) | |
279 | { | |
280 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
281 | dm_bio_prison_free_cell(pool->prison, cell); | |
282 | } | |
283 | ||
025b9685 JT |
284 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
285 | struct dm_bio_prison_cell *cell) | |
286 | { | |
287 | struct pool *pool = tc->pool; | |
288 | unsigned long flags; | |
289 | ||
290 | spin_lock_irqsave(&pool->lock, flags); | |
291 | dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios); | |
292 | spin_unlock_irqrestore(&pool->lock, flags); | |
293 | ||
294 | wake_worker(pool); | |
295 | } | |
296 | ||
6beca5eb JT |
297 | static void cell_error(struct pool *pool, |
298 | struct dm_bio_prison_cell *cell) | |
299 | { | |
300 | dm_cell_error(pool->prison, cell); | |
301 | dm_bio_prison_free_cell(pool->prison, cell); | |
302 | } | |
303 | ||
304 | /*----------------------------------------------------------------*/ | |
305 | ||
991d9fa0 JT |
306 | /* |
307 | * A global list of pools that uses a struct mapped_device as a key. | |
308 | */ | |
309 | static struct dm_thin_pool_table { | |
310 | struct mutex mutex; | |
311 | struct list_head pools; | |
312 | } dm_thin_pool_table; | |
313 | ||
314 | static void pool_table_init(void) | |
315 | { | |
316 | mutex_init(&dm_thin_pool_table.mutex); | |
317 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
318 | } | |
319 | ||
320 | static void __pool_table_insert(struct pool *pool) | |
321 | { | |
322 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
323 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
324 | } | |
325 | ||
326 | static void __pool_table_remove(struct pool *pool) | |
327 | { | |
328 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
329 | list_del(&pool->list); | |
330 | } | |
331 | ||
332 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
333 | { | |
334 | struct pool *pool = NULL, *tmp; | |
335 | ||
336 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
337 | ||
338 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
339 | if (tmp->pool_md == md) { | |
340 | pool = tmp; | |
341 | break; | |
342 | } | |
343 | } | |
344 | ||
345 | return pool; | |
346 | } | |
347 | ||
348 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
349 | { | |
350 | struct pool *pool = NULL, *tmp; | |
351 | ||
352 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
353 | ||
354 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
355 | if (tmp->md_dev == md_dev) { | |
356 | pool = tmp; | |
357 | break; | |
358 | } | |
359 | } | |
360 | ||
361 | return pool; | |
362 | } | |
363 | ||
364 | /*----------------------------------------------------------------*/ | |
365 | ||
a24c2569 | 366 | struct dm_thin_endio_hook { |
eb2aa48d | 367 | struct thin_c *tc; |
44feb387 MS |
368 | struct dm_deferred_entry *shared_read_entry; |
369 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 370 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
371 | }; |
372 | ||
18adc577 | 373 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
374 | { |
375 | struct bio *bio; | |
376 | struct bio_list bios; | |
18adc577 | 377 | unsigned long flags; |
991d9fa0 JT |
378 | |
379 | bio_list_init(&bios); | |
18adc577 JT |
380 | |
381 | spin_lock_irqsave(&tc->pool->lock, flags); | |
991d9fa0 JT |
382 | bio_list_merge(&bios, master); |
383 | bio_list_init(master); | |
18adc577 | 384 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
991d9fa0 JT |
385 | |
386 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 387 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 388 | |
eb2aa48d | 389 | if (h->tc == tc) |
991d9fa0 JT |
390 | bio_endio(bio, DM_ENDIO_REQUEUE); |
391 | else | |
392 | bio_list_add(master, bio); | |
393 | } | |
394 | } | |
395 | ||
396 | static void requeue_io(struct thin_c *tc) | |
397 | { | |
398 | struct pool *pool = tc->pool; | |
991d9fa0 | 399 | |
18adc577 JT |
400 | requeue_bio_list(tc, &pool->deferred_bios); |
401 | requeue_bio_list(tc, &pool->retry_on_resume_list); | |
991d9fa0 JT |
402 | } |
403 | ||
3e1a0699 JT |
404 | static void error_retry_list(struct pool *pool) |
405 | { | |
406 | struct bio *bio; | |
407 | unsigned long flags; | |
408 | struct bio_list bios; | |
409 | ||
410 | bio_list_init(&bios); | |
411 | ||
412 | spin_lock_irqsave(&pool->lock, flags); | |
413 | bio_list_merge(&bios, &pool->retry_on_resume_list); | |
414 | bio_list_init(&pool->retry_on_resume_list); | |
415 | spin_unlock_irqrestore(&pool->lock, flags); | |
416 | ||
417 | while ((bio = bio_list_pop(&bios))) | |
418 | bio_io_error(bio); | |
419 | } | |
420 | ||
991d9fa0 JT |
421 | /* |
422 | * This section of code contains the logic for processing a thin device's IO. | |
423 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
424 | * but most is exclusively called from the thin target rather than the thin-pool | |
425 | * target. | |
426 | */ | |
427 | ||
58f77a21 MS |
428 | static bool block_size_is_power_of_two(struct pool *pool) |
429 | { | |
430 | return pool->sectors_per_block_shift >= 0; | |
431 | } | |
432 | ||
991d9fa0 JT |
433 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
434 | { | |
58f77a21 | 435 | struct pool *pool = tc->pool; |
4f024f37 | 436 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 437 | |
58f77a21 MS |
438 | if (block_size_is_power_of_two(pool)) |
439 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 440 | else |
58f77a21 | 441 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
442 | |
443 | return block_nr; | |
991d9fa0 JT |
444 | } |
445 | ||
446 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
447 | { | |
448 | struct pool *pool = tc->pool; | |
4f024f37 | 449 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
450 | |
451 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 452 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
453 | bio->bi_iter.bi_sector = |
454 | (block << pool->sectors_per_block_shift) | | |
455 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 456 | else |
4f024f37 | 457 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 458 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
459 | } |
460 | ||
2dd9c257 JT |
461 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
462 | { | |
463 | bio->bi_bdev = tc->origin_dev->bdev; | |
464 | } | |
465 | ||
4afdd680 JT |
466 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
467 | { | |
468 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
469 | dm_thin_changed_this_transaction(tc->td); | |
470 | } | |
471 | ||
e8088073 JT |
472 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
473 | { | |
474 | struct dm_thin_endio_hook *h; | |
475 | ||
476 | if (bio->bi_rw & REQ_DISCARD) | |
477 | return; | |
478 | ||
59c3d2c6 | 479 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
480 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
481 | } | |
482 | ||
2dd9c257 | 483 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
484 | { |
485 | struct pool *pool = tc->pool; | |
486 | unsigned long flags; | |
487 | ||
e49e5829 JT |
488 | if (!bio_triggers_commit(tc, bio)) { |
489 | generic_make_request(bio); | |
490 | return; | |
491 | } | |
492 | ||
991d9fa0 | 493 | /* |
e49e5829 JT |
494 | * Complete bio with an error if earlier I/O caused changes to |
495 | * the metadata that can't be committed e.g, due to I/O errors | |
496 | * on the metadata device. | |
991d9fa0 | 497 | */ |
e49e5829 JT |
498 | if (dm_thin_aborted_changes(tc->td)) { |
499 | bio_io_error(bio); | |
500 | return; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Batch together any bios that trigger commits and then issue a | |
505 | * single commit for them in process_deferred_bios(). | |
506 | */ | |
507 | spin_lock_irqsave(&pool->lock, flags); | |
508 | bio_list_add(&pool->deferred_flush_bios, bio); | |
509 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
510 | } |
511 | ||
2dd9c257 JT |
512 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
513 | { | |
514 | remap_to_origin(tc, bio); | |
515 | issue(tc, bio); | |
516 | } | |
517 | ||
518 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
519 | dm_block_t block) | |
520 | { | |
521 | remap(tc, bio, block); | |
522 | issue(tc, bio); | |
523 | } | |
524 | ||
991d9fa0 JT |
525 | /*----------------------------------------------------------------*/ |
526 | ||
527 | /* | |
528 | * Bio endio functions. | |
529 | */ | |
a24c2569 | 530 | struct dm_thin_new_mapping { |
991d9fa0 JT |
531 | struct list_head list; |
532 | ||
7f214665 MS |
533 | bool quiesced:1; |
534 | bool prepared:1; | |
535 | bool pass_discard:1; | |
536 | bool definitely_not_shared:1; | |
991d9fa0 | 537 | |
7f214665 | 538 | int err; |
991d9fa0 JT |
539 | struct thin_c *tc; |
540 | dm_block_t virt_block; | |
541 | dm_block_t data_block; | |
a24c2569 | 542 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
543 | |
544 | /* | |
545 | * If the bio covers the whole area of a block then we can avoid | |
546 | * zeroing or copying. Instead this bio is hooked. The bio will | |
547 | * still be in the cell, so care has to be taken to avoid issuing | |
548 | * the bio twice. | |
549 | */ | |
550 | struct bio *bio; | |
551 | bio_end_io_t *saved_bi_end_io; | |
552 | }; | |
553 | ||
a24c2569 | 554 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
555 | { |
556 | struct pool *pool = m->tc->pool; | |
557 | ||
eb2aa48d | 558 | if (m->quiesced && m->prepared) { |
daec338b | 559 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
560 | wake_worker(pool); |
561 | } | |
562 | } | |
563 | ||
564 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
565 | { | |
566 | unsigned long flags; | |
a24c2569 | 567 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
568 | struct pool *pool = m->tc->pool; |
569 | ||
570 | m->err = read_err || write_err ? -EIO : 0; | |
571 | ||
572 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 573 | m->prepared = true; |
991d9fa0 JT |
574 | __maybe_add_mapping(m); |
575 | spin_unlock_irqrestore(&pool->lock, flags); | |
576 | } | |
577 | ||
578 | static void overwrite_endio(struct bio *bio, int err) | |
579 | { | |
580 | unsigned long flags; | |
59c3d2c6 | 581 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 582 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
583 | struct pool *pool = m->tc->pool; |
584 | ||
585 | m->err = err; | |
586 | ||
587 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 588 | m->prepared = true; |
991d9fa0 JT |
589 | __maybe_add_mapping(m); |
590 | spin_unlock_irqrestore(&pool->lock, flags); | |
591 | } | |
592 | ||
991d9fa0 JT |
593 | /*----------------------------------------------------------------*/ |
594 | ||
595 | /* | |
596 | * Workqueue. | |
597 | */ | |
598 | ||
599 | /* | |
600 | * Prepared mapping jobs. | |
601 | */ | |
602 | ||
603 | /* | |
604 | * This sends the bios in the cell back to the deferred_bios list. | |
605 | */ | |
2aab3850 | 606 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
607 | { |
608 | struct pool *pool = tc->pool; | |
609 | unsigned long flags; | |
610 | ||
611 | spin_lock_irqsave(&pool->lock, flags); | |
6beca5eb | 612 | cell_release(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
613 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
614 | ||
615 | wake_worker(pool); | |
616 | } | |
617 | ||
618 | /* | |
6beca5eb | 619 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 620 | */ |
f286ba0e | 621 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 622 | { |
991d9fa0 JT |
623 | struct pool *pool = tc->pool; |
624 | unsigned long flags; | |
625 | ||
991d9fa0 | 626 | spin_lock_irqsave(&pool->lock, flags); |
6beca5eb | 627 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
628 | spin_unlock_irqrestore(&pool->lock, flags); |
629 | ||
630 | wake_worker(pool); | |
631 | } | |
632 | ||
e49e5829 JT |
633 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
634 | { | |
196d38bc | 635 | if (m->bio) { |
e49e5829 | 636 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
637 | atomic_inc(&m->bio->bi_remaining); |
638 | } | |
6beca5eb | 639 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
640 | list_del(&m->list); |
641 | mempool_free(m, m->tc->pool->mapping_pool); | |
642 | } | |
025b9685 | 643 | |
a24c2569 | 644 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
645 | { |
646 | struct thin_c *tc = m->tc; | |
6beca5eb | 647 | struct pool *pool = tc->pool; |
991d9fa0 JT |
648 | struct bio *bio; |
649 | int r; | |
650 | ||
651 | bio = m->bio; | |
196d38bc | 652 | if (bio) { |
991d9fa0 | 653 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
654 | atomic_inc(&bio->bi_remaining); |
655 | } | |
991d9fa0 JT |
656 | |
657 | if (m->err) { | |
6beca5eb | 658 | cell_error(pool, m->cell); |
905386f8 | 659 | goto out; |
991d9fa0 JT |
660 | } |
661 | ||
662 | /* | |
663 | * Commit the prepared block into the mapping btree. | |
664 | * Any I/O for this block arriving after this point will get | |
665 | * remapped to it directly. | |
666 | */ | |
667 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
668 | if (r) { | |
b5330655 | 669 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 670 | cell_error(pool, m->cell); |
905386f8 | 671 | goto out; |
991d9fa0 JT |
672 | } |
673 | ||
674 | /* | |
675 | * Release any bios held while the block was being provisioned. | |
676 | * If we are processing a write bio that completely covers the block, | |
677 | * we already processed it so can ignore it now when processing | |
678 | * the bios in the cell. | |
679 | */ | |
680 | if (bio) { | |
f286ba0e | 681 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
682 | bio_endio(bio, 0); |
683 | } else | |
2aab3850 | 684 | cell_defer(tc, m->cell); |
991d9fa0 | 685 | |
905386f8 | 686 | out: |
991d9fa0 | 687 | list_del(&m->list); |
6beca5eb | 688 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
689 | } |
690 | ||
e49e5829 | 691 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 692 | { |
104655fd JT |
693 | struct thin_c *tc = m->tc; |
694 | ||
e49e5829 | 695 | bio_io_error(m->bio); |
f286ba0e JT |
696 | cell_defer_no_holder(tc, m->cell); |
697 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
698 | mempool_free(m, tc->pool->mapping_pool); |
699 | } | |
700 | ||
701 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
702 | { | |
703 | struct thin_c *tc = m->tc; | |
104655fd | 704 | |
e8088073 | 705 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
706 | cell_defer_no_holder(tc, m->cell); |
707 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 708 | |
104655fd | 709 | if (m->pass_discard) |
19fa1a67 JT |
710 | if (m->definitely_not_shared) |
711 | remap_and_issue(tc, m->bio, m->data_block); | |
712 | else { | |
713 | bool used = false; | |
714 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
715 | bio_endio(m->bio, 0); | |
716 | else | |
717 | remap_and_issue(tc, m->bio, m->data_block); | |
718 | } | |
104655fd JT |
719 | else |
720 | bio_endio(m->bio, 0); | |
721 | ||
104655fd JT |
722 | mempool_free(m, tc->pool->mapping_pool); |
723 | } | |
724 | ||
e49e5829 JT |
725 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
726 | { | |
727 | int r; | |
728 | struct thin_c *tc = m->tc; | |
729 | ||
730 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
731 | if (r) | |
c397741c | 732 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
733 | |
734 | process_prepared_discard_passdown(m); | |
735 | } | |
736 | ||
104655fd | 737 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 738 | process_mapping_fn *fn) |
991d9fa0 JT |
739 | { |
740 | unsigned long flags; | |
741 | struct list_head maps; | |
a24c2569 | 742 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
743 | |
744 | INIT_LIST_HEAD(&maps); | |
745 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 746 | list_splice_init(head, &maps); |
991d9fa0 JT |
747 | spin_unlock_irqrestore(&pool->lock, flags); |
748 | ||
749 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 750 | (*fn)(m); |
991d9fa0 JT |
751 | } |
752 | ||
753 | /* | |
754 | * Deferred bio jobs. | |
755 | */ | |
104655fd | 756 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 757 | { |
4f024f37 KO |
758 | return bio->bi_iter.bi_size == |
759 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
760 | } |
761 | ||
762 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
763 | { | |
764 | return (bio_data_dir(bio) == WRITE) && | |
765 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
766 | } |
767 | ||
768 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
769 | bio_end_io_t *fn) | |
770 | { | |
771 | *save = bio->bi_end_io; | |
772 | bio->bi_end_io = fn; | |
773 | } | |
774 | ||
775 | static int ensure_next_mapping(struct pool *pool) | |
776 | { | |
777 | if (pool->next_mapping) | |
778 | return 0; | |
779 | ||
780 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
781 | ||
782 | return pool->next_mapping ? 0 : -ENOMEM; | |
783 | } | |
784 | ||
a24c2569 | 785 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 786 | { |
16961b04 | 787 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
788 | |
789 | BUG_ON(!pool->next_mapping); | |
790 | ||
16961b04 MS |
791 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
792 | INIT_LIST_HEAD(&m->list); | |
793 | m->bio = NULL; | |
794 | ||
991d9fa0 JT |
795 | pool->next_mapping = NULL; |
796 | ||
16961b04 | 797 | return m; |
991d9fa0 JT |
798 | } |
799 | ||
800 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
801 | struct dm_dev *origin, dm_block_t data_origin, |
802 | dm_block_t data_dest, | |
a24c2569 | 803 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
804 | { |
805 | int r; | |
806 | struct pool *pool = tc->pool; | |
a24c2569 | 807 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 808 | |
991d9fa0 JT |
809 | m->tc = tc; |
810 | m->virt_block = virt_block; | |
811 | m->data_block = data_dest; | |
812 | m->cell = cell; | |
991d9fa0 | 813 | |
44feb387 | 814 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 815 | m->quiesced = true; |
991d9fa0 JT |
816 | |
817 | /* | |
818 | * IO to pool_dev remaps to the pool target's data_dev. | |
819 | * | |
820 | * If the whole block of data is being overwritten, we can issue the | |
821 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
822 | */ | |
823 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 824 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 825 | |
eb2aa48d | 826 | h->overwrite_mapping = m; |
991d9fa0 JT |
827 | m->bio = bio; |
828 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 829 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
830 | remap_and_issue(tc, bio, data_dest); |
831 | } else { | |
832 | struct dm_io_region from, to; | |
833 | ||
2dd9c257 | 834 | from.bdev = origin->bdev; |
991d9fa0 JT |
835 | from.sector = data_origin * pool->sectors_per_block; |
836 | from.count = pool->sectors_per_block; | |
837 | ||
838 | to.bdev = tc->pool_dev->bdev; | |
839 | to.sector = data_dest * pool->sectors_per_block; | |
840 | to.count = pool->sectors_per_block; | |
841 | ||
842 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
843 | 0, copy_complete, m); | |
844 | if (r < 0) { | |
845 | mempool_free(m, pool->mapping_pool); | |
c397741c | 846 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 847 | cell_error(pool, cell); |
991d9fa0 JT |
848 | } |
849 | } | |
850 | } | |
851 | ||
2dd9c257 JT |
852 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
853 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 854 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
855 | { |
856 | schedule_copy(tc, virt_block, tc->pool_dev, | |
857 | data_origin, data_dest, cell, bio); | |
858 | } | |
859 | ||
860 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
861 | dm_block_t data_dest, | |
a24c2569 | 862 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
863 | { |
864 | schedule_copy(tc, virt_block, tc->origin_dev, | |
865 | virt_block, data_dest, cell, bio); | |
866 | } | |
867 | ||
991d9fa0 | 868 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 869 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
870 | struct bio *bio) |
871 | { | |
872 | struct pool *pool = tc->pool; | |
a24c2569 | 873 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 874 | |
7f214665 MS |
875 | m->quiesced = true; |
876 | m->prepared = false; | |
991d9fa0 JT |
877 | m->tc = tc; |
878 | m->virt_block = virt_block; | |
879 | m->data_block = data_block; | |
880 | m->cell = cell; | |
991d9fa0 JT |
881 | |
882 | /* | |
883 | * If the whole block of data is being overwritten or we are not | |
884 | * zeroing pre-existing data, we can issue the bio immediately. | |
885 | * Otherwise we use kcopyd to zero the data first. | |
886 | */ | |
67e2e2b2 | 887 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
888 | process_prepared_mapping(m); |
889 | ||
890 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 891 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 892 | |
eb2aa48d | 893 | h->overwrite_mapping = m; |
991d9fa0 JT |
894 | m->bio = bio; |
895 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 896 | inc_all_io_entry(pool, bio); |
991d9fa0 | 897 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
898 | } else { |
899 | int r; | |
900 | struct dm_io_region to; | |
901 | ||
902 | to.bdev = tc->pool_dev->bdev; | |
903 | to.sector = data_block * pool->sectors_per_block; | |
904 | to.count = pool->sectors_per_block; | |
905 | ||
906 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
907 | if (r < 0) { | |
908 | mempool_free(m, pool->mapping_pool); | |
c397741c | 909 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 910 | cell_error(pool, cell); |
991d9fa0 JT |
911 | } |
912 | } | |
913 | } | |
914 | ||
e49e5829 JT |
915 | /* |
916 | * A non-zero return indicates read_only or fail_io mode. | |
917 | * Many callers don't care about the return value. | |
918 | */ | |
020cc3b5 | 919 | static int commit(struct pool *pool) |
e49e5829 JT |
920 | { |
921 | int r; | |
922 | ||
923 | if (get_pool_mode(pool) != PM_WRITE) | |
924 | return -EINVAL; | |
925 | ||
020cc3b5 | 926 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
927 | if (r) |
928 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
929 | |
930 | return r; | |
931 | } | |
932 | ||
88a6621b JT |
933 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
934 | { | |
935 | unsigned long flags; | |
936 | ||
937 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
938 | DMWARN("%s: reached low water mark for data device: sending event.", | |
939 | dm_device_name(pool->pool_md)); | |
940 | spin_lock_irqsave(&pool->lock, flags); | |
941 | pool->low_water_triggered = true; | |
942 | spin_unlock_irqrestore(&pool->lock, flags); | |
943 | dm_table_event(pool->ti->table); | |
944 | } | |
945 | } | |
946 | ||
3e1a0699 JT |
947 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
948 | ||
991d9fa0 JT |
949 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
950 | { | |
951 | int r; | |
952 | dm_block_t free_blocks; | |
991d9fa0 JT |
953 | struct pool *pool = tc->pool; |
954 | ||
3e1a0699 | 955 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
956 | return -EINVAL; |
957 | ||
991d9fa0 | 958 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
959 | if (r) { |
960 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 961 | return r; |
b5330655 | 962 | } |
991d9fa0 | 963 | |
88a6621b | 964 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
965 | |
966 | if (!free_blocks) { | |
94563bad MS |
967 | /* |
968 | * Try to commit to see if that will free up some | |
969 | * more space. | |
970 | */ | |
020cc3b5 JT |
971 | r = commit(pool); |
972 | if (r) | |
973 | return r; | |
991d9fa0 | 974 | |
94563bad | 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); | |
94563bad | 978 | return r; |
b5330655 | 979 | } |
991d9fa0 | 980 | |
94563bad | 981 | if (!free_blocks) { |
3e1a0699 | 982 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 983 | return -ENOSPC; |
991d9fa0 JT |
984 | } |
985 | } | |
986 | ||
987 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 988 | if (r) { |
b5330655 | 989 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 990 | return r; |
4a02b34e | 991 | } |
991d9fa0 JT |
992 | |
993 | return 0; | |
994 | } | |
995 | ||
996 | /* | |
997 | * If we have run out of space, queue bios until the device is | |
998 | * resumed, presumably after having been reloaded with more space. | |
999 | */ | |
1000 | static void retry_on_resume(struct bio *bio) | |
1001 | { | |
59c3d2c6 | 1002 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1003 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1004 | struct pool *pool = tc->pool; |
1005 | unsigned long flags; | |
1006 | ||
1007 | spin_lock_irqsave(&pool->lock, flags); | |
1008 | bio_list_add(&pool->retry_on_resume_list, bio); | |
1009 | spin_unlock_irqrestore(&pool->lock, flags); | |
1010 | } | |
1011 | ||
3e1a0699 | 1012 | static bool should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1013 | { |
3e1a0699 JT |
1014 | enum pool_mode m = get_pool_mode(pool); |
1015 | ||
1016 | switch (m) { | |
1017 | case PM_WRITE: | |
1018 | /* Shouldn't get here */ | |
1019 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
1020 | return true; | |
1021 | ||
1022 | case PM_OUT_OF_DATA_SPACE: | |
1023 | return pool->pf.error_if_no_space; | |
1024 | ||
1025 | case PM_READ_ONLY: | |
1026 | case PM_FAIL: | |
1027 | return true; | |
1028 | default: | |
1029 | /* Shouldn't get here */ | |
1030 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
1031 | return true; | |
1032 | } | |
1033 | } | |
8c0f0e8c | 1034 | |
3e1a0699 JT |
1035 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1036 | { | |
1037 | if (should_error_unserviceable_bio(pool)) | |
8c0f0e8c | 1038 | bio_io_error(bio); |
6d16202b MS |
1039 | else |
1040 | retry_on_resume(bio); | |
8c0f0e8c MS |
1041 | } |
1042 | ||
399caddf | 1043 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1044 | { |
1045 | struct bio *bio; | |
1046 | struct bio_list bios; | |
1047 | ||
3e1a0699 JT |
1048 | if (should_error_unserviceable_bio(pool)) { |
1049 | cell_error(pool, cell); | |
1050 | return; | |
1051 | } | |
1052 | ||
991d9fa0 | 1053 | bio_list_init(&bios); |
6beca5eb | 1054 | cell_release(pool, cell, &bios); |
991d9fa0 | 1055 | |
3e1a0699 JT |
1056 | if (should_error_unserviceable_bio(pool)) |
1057 | while ((bio = bio_list_pop(&bios))) | |
1058 | bio_io_error(bio); | |
1059 | else | |
1060 | while ((bio = bio_list_pop(&bios))) | |
1061 | retry_on_resume(bio); | |
991d9fa0 JT |
1062 | } |
1063 | ||
104655fd JT |
1064 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1065 | { | |
1066 | int r; | |
c3a0ce2e | 1067 | unsigned long flags; |
104655fd | 1068 | struct pool *pool = tc->pool; |
a24c2569 | 1069 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1070 | struct dm_cell_key key, key2; |
104655fd JT |
1071 | dm_block_t block = get_bio_block(tc, bio); |
1072 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1073 | struct dm_thin_new_mapping *m; |
104655fd JT |
1074 | |
1075 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1076 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1077 | return; |
1078 | ||
1079 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1080 | switch (r) { | |
1081 | case 0: | |
1082 | /* | |
1083 | * Check nobody is fiddling with this pool block. This can | |
1084 | * happen if someone's in the process of breaking sharing | |
1085 | * on this block. | |
1086 | */ | |
1087 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1088 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1089 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1090 | break; |
1091 | } | |
1092 | ||
1093 | if (io_overlaps_block(pool, bio)) { | |
1094 | /* | |
1095 | * IO may still be going to the destination block. We must | |
1096 | * quiesce before we can do the removal. | |
1097 | */ | |
1098 | m = get_next_mapping(pool); | |
1099 | m->tc = tc; | |
19fa1a67 JT |
1100 | m->pass_discard = pool->pf.discard_passdown; |
1101 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1102 | m->virt_block = block; |
1103 | m->data_block = lookup_result.block; | |
1104 | m->cell = cell; | |
1105 | m->cell2 = cell2; | |
104655fd JT |
1106 | m->bio = bio; |
1107 | ||
44feb387 | 1108 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1109 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1110 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1111 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1112 | wake_worker(pool); |
1113 | } | |
1114 | } else { | |
e8088073 | 1115 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1116 | cell_defer_no_holder(tc, cell); |
1117 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1118 | |
104655fd | 1119 | /* |
49296309 MP |
1120 | * The DM core makes sure that the discard doesn't span |
1121 | * a block boundary. So we submit the discard of a | |
1122 | * partial block appropriately. | |
104655fd | 1123 | */ |
650d2a06 MP |
1124 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1125 | remap_and_issue(tc, bio, lookup_result.block); | |
1126 | else | |
1127 | bio_endio(bio, 0); | |
104655fd JT |
1128 | } |
1129 | break; | |
1130 | ||
1131 | case -ENODATA: | |
1132 | /* | |
1133 | * It isn't provisioned, just forget it. | |
1134 | */ | |
f286ba0e | 1135 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1136 | bio_endio(bio, 0); |
1137 | break; | |
1138 | ||
1139 | default: | |
c397741c MS |
1140 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1141 | __func__, r); | |
f286ba0e | 1142 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1143 | bio_io_error(bio); |
1144 | break; | |
1145 | } | |
1146 | } | |
1147 | ||
991d9fa0 | 1148 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1149 | struct dm_cell_key *key, |
991d9fa0 | 1150 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1151 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1152 | { |
1153 | int r; | |
1154 | dm_block_t data_block; | |
d6fc2042 | 1155 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1156 | |
1157 | r = alloc_data_block(tc, &data_block); | |
1158 | switch (r) { | |
1159 | case 0: | |
2dd9c257 JT |
1160 | schedule_internal_copy(tc, block, lookup_result->block, |
1161 | data_block, cell, bio); | |
991d9fa0 JT |
1162 | break; |
1163 | ||
1164 | case -ENOSPC: | |
399caddf | 1165 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1166 | break; |
1167 | ||
1168 | default: | |
c397741c MS |
1169 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1170 | __func__, r); | |
d6fc2042 | 1171 | cell_error(pool, cell); |
991d9fa0 JT |
1172 | break; |
1173 | } | |
1174 | } | |
1175 | ||
1176 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1177 | dm_block_t block, | |
1178 | struct dm_thin_lookup_result *lookup_result) | |
1179 | { | |
a24c2569 | 1180 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1181 | struct pool *pool = tc->pool; |
44feb387 | 1182 | struct dm_cell_key key; |
991d9fa0 JT |
1183 | |
1184 | /* | |
1185 | * If cell is already occupied, then sharing is already in the process | |
1186 | * of being broken so we have nothing further to do here. | |
1187 | */ | |
1188 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1189 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1190 | return; |
1191 | ||
4f024f37 | 1192 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
991d9fa0 JT |
1193 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1194 | else { | |
59c3d2c6 | 1195 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1196 | |
44feb387 | 1197 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1198 | inc_all_io_entry(pool, bio); |
f286ba0e | 1199 | cell_defer_no_holder(tc, cell); |
e8088073 | 1200 | |
991d9fa0 JT |
1201 | remap_and_issue(tc, bio, lookup_result->block); |
1202 | } | |
1203 | } | |
1204 | ||
1205 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1206 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1207 | { |
1208 | int r; | |
1209 | dm_block_t data_block; | |
6beca5eb | 1210 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1211 | |
1212 | /* | |
1213 | * Remap empty bios (flushes) immediately, without provisioning. | |
1214 | */ | |
4f024f37 | 1215 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1216 | inc_all_io_entry(pool, bio); |
f286ba0e | 1217 | cell_defer_no_holder(tc, cell); |
e8088073 | 1218 | |
991d9fa0 JT |
1219 | remap_and_issue(tc, bio, 0); |
1220 | return; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Fill read bios with zeroes and complete them immediately. | |
1225 | */ | |
1226 | if (bio_data_dir(bio) == READ) { | |
1227 | zero_fill_bio(bio); | |
f286ba0e | 1228 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1229 | bio_endio(bio, 0); |
1230 | return; | |
1231 | } | |
1232 | ||
1233 | r = alloc_data_block(tc, &data_block); | |
1234 | switch (r) { | |
1235 | case 0: | |
2dd9c257 JT |
1236 | if (tc->origin_dev) |
1237 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1238 | else | |
1239 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1240 | break; |
1241 | ||
1242 | case -ENOSPC: | |
399caddf | 1243 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1244 | break; |
1245 | ||
1246 | default: | |
c397741c MS |
1247 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1248 | __func__, r); | |
6beca5eb | 1249 | cell_error(pool, cell); |
991d9fa0 JT |
1250 | break; |
1251 | } | |
1252 | } | |
1253 | ||
1254 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1255 | { | |
1256 | int r; | |
6beca5eb | 1257 | struct pool *pool = tc->pool; |
991d9fa0 | 1258 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1259 | struct dm_bio_prison_cell *cell; |
44feb387 | 1260 | struct dm_cell_key key; |
991d9fa0 JT |
1261 | struct dm_thin_lookup_result lookup_result; |
1262 | ||
1263 | /* | |
1264 | * If cell is already occupied, then the block is already | |
1265 | * being provisioned so we have nothing further to do here. | |
1266 | */ | |
1267 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1268 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1269 | return; |
1270 | ||
1271 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1272 | switch (r) { | |
1273 | case 0: | |
e8088073 | 1274 | if (lookup_result.shared) { |
991d9fa0 | 1275 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1276 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1277 | } else { |
6beca5eb | 1278 | inc_all_io_entry(pool, bio); |
f286ba0e | 1279 | cell_defer_no_holder(tc, cell); |
e8088073 | 1280 | |
991d9fa0 | 1281 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1282 | } |
991d9fa0 JT |
1283 | break; |
1284 | ||
1285 | case -ENODATA: | |
2dd9c257 | 1286 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1287 | inc_all_io_entry(pool, bio); |
f286ba0e | 1288 | cell_defer_no_holder(tc, cell); |
e8088073 | 1289 | |
2dd9c257 JT |
1290 | remap_to_origin_and_issue(tc, bio); |
1291 | } else | |
1292 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1293 | break; |
1294 | ||
1295 | default: | |
c397741c MS |
1296 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1297 | __func__, r); | |
f286ba0e | 1298 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1299 | bio_io_error(bio); |
1300 | break; | |
1301 | } | |
1302 | } | |
1303 | ||
e49e5829 JT |
1304 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1305 | { | |
1306 | int r; | |
1307 | int rw = bio_data_dir(bio); | |
1308 | dm_block_t block = get_bio_block(tc, bio); | |
1309 | struct dm_thin_lookup_result lookup_result; | |
1310 | ||
1311 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1312 | switch (r) { | |
1313 | case 0: | |
4f024f37 | 1314 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
8c0f0e8c | 1315 | handle_unserviceable_bio(tc->pool, bio); |
e8088073 JT |
1316 | else { |
1317 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1318 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1319 | } |
e49e5829 JT |
1320 | break; |
1321 | ||
1322 | case -ENODATA: | |
1323 | if (rw != READ) { | |
8c0f0e8c | 1324 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1325 | break; |
1326 | } | |
1327 | ||
1328 | if (tc->origin_dev) { | |
e8088073 | 1329 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1330 | remap_to_origin_and_issue(tc, bio); |
1331 | break; | |
1332 | } | |
1333 | ||
1334 | zero_fill_bio(bio); | |
1335 | bio_endio(bio, 0); | |
1336 | break; | |
1337 | ||
1338 | default: | |
c397741c MS |
1339 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1340 | __func__, r); | |
e49e5829 JT |
1341 | bio_io_error(bio); |
1342 | break; | |
1343 | } | |
1344 | } | |
1345 | ||
3e1a0699 JT |
1346 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1347 | { | |
1348 | bio_endio(bio, 0); | |
1349 | } | |
1350 | ||
e49e5829 JT |
1351 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1352 | { | |
1353 | bio_io_error(bio); | |
1354 | } | |
1355 | ||
ac8c3f3d JT |
1356 | /* |
1357 | * FIXME: should we also commit due to size of transaction, measured in | |
1358 | * metadata blocks? | |
1359 | */ | |
905e51b3 JT |
1360 | static int need_commit_due_to_time(struct pool *pool) |
1361 | { | |
1362 | return jiffies < pool->last_commit_jiffies || | |
1363 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1364 | } | |
1365 | ||
991d9fa0 JT |
1366 | static void process_deferred_bios(struct pool *pool) |
1367 | { | |
1368 | unsigned long flags; | |
1369 | struct bio *bio; | |
1370 | struct bio_list bios; | |
991d9fa0 JT |
1371 | |
1372 | bio_list_init(&bios); | |
1373 | ||
1374 | spin_lock_irqsave(&pool->lock, flags); | |
1375 | bio_list_merge(&bios, &pool->deferred_bios); | |
1376 | bio_list_init(&pool->deferred_bios); | |
1377 | spin_unlock_irqrestore(&pool->lock, flags); | |
1378 | ||
1379 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 1380 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1381 | struct thin_c *tc = h->tc; |
1382 | ||
738211f7 JT |
1383 | if (tc->requeue_mode) { |
1384 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1385 | continue; | |
1386 | } | |
1387 | ||
991d9fa0 JT |
1388 | /* |
1389 | * If we've got no free new_mapping structs, and processing | |
1390 | * this bio might require one, we pause until there are some | |
1391 | * prepared mappings to process. | |
1392 | */ | |
1393 | if (ensure_next_mapping(pool)) { | |
1394 | spin_lock_irqsave(&pool->lock, flags); | |
1395 | bio_list_merge(&pool->deferred_bios, &bios); | |
1396 | spin_unlock_irqrestore(&pool->lock, flags); | |
1397 | ||
1398 | break; | |
1399 | } | |
104655fd JT |
1400 | |
1401 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1402 | pool->process_discard(tc, bio); |
104655fd | 1403 | else |
e49e5829 | 1404 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1405 | } |
1406 | ||
1407 | /* | |
1408 | * If there are any deferred flush bios, we must commit | |
1409 | * the metadata before issuing them. | |
1410 | */ | |
1411 | bio_list_init(&bios); | |
1412 | spin_lock_irqsave(&pool->lock, flags); | |
1413 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1414 | bio_list_init(&pool->deferred_flush_bios); | |
1415 | spin_unlock_irqrestore(&pool->lock, flags); | |
1416 | ||
4d1662a3 MS |
1417 | if (bio_list_empty(&bios) && |
1418 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1419 | return; |
1420 | ||
020cc3b5 | 1421 | if (commit(pool)) { |
991d9fa0 JT |
1422 | while ((bio = bio_list_pop(&bios))) |
1423 | bio_io_error(bio); | |
1424 | return; | |
1425 | } | |
905e51b3 | 1426 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1427 | |
1428 | while ((bio = bio_list_pop(&bios))) | |
1429 | generic_make_request(bio); | |
1430 | } | |
1431 | ||
1432 | static void do_worker(struct work_struct *ws) | |
1433 | { | |
1434 | struct pool *pool = container_of(ws, struct pool, worker); | |
1435 | ||
e49e5829 JT |
1436 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1437 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1438 | process_deferred_bios(pool); |
1439 | } | |
1440 | ||
905e51b3 JT |
1441 | /* |
1442 | * We want to commit periodically so that not too much | |
1443 | * unwritten data builds up. | |
1444 | */ | |
1445 | static void do_waker(struct work_struct *ws) | |
1446 | { | |
1447 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1448 | wake_worker(pool); | |
1449 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1450 | } | |
1451 | ||
991d9fa0 JT |
1452 | /*----------------------------------------------------------------*/ |
1453 | ||
738211f7 JT |
1454 | struct noflush_work { |
1455 | struct work_struct worker; | |
1456 | struct thin_c *tc; | |
1457 | ||
1458 | atomic_t complete; | |
1459 | wait_queue_head_t wait; | |
1460 | }; | |
1461 | ||
1462 | static void complete_noflush_work(struct noflush_work *w) | |
1463 | { | |
1464 | atomic_set(&w->complete, 1); | |
1465 | wake_up(&w->wait); | |
1466 | } | |
1467 | ||
1468 | static void do_noflush_start(struct work_struct *ws) | |
1469 | { | |
1470 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1471 | w->tc->requeue_mode = true; | |
1472 | requeue_io(w->tc); | |
1473 | complete_noflush_work(w); | |
1474 | } | |
1475 | ||
1476 | static void do_noflush_stop(struct work_struct *ws) | |
1477 | { | |
1478 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1479 | w->tc->requeue_mode = false; | |
1480 | complete_noflush_work(w); | |
1481 | } | |
1482 | ||
1483 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
1484 | { | |
1485 | struct noflush_work w; | |
1486 | ||
1487 | INIT_WORK(&w.worker, fn); | |
1488 | w.tc = tc; | |
1489 | atomic_set(&w.complete, 0); | |
1490 | init_waitqueue_head(&w.wait); | |
1491 | ||
1492 | queue_work(tc->pool->wq, &w.worker); | |
1493 | ||
1494 | wait_event(w.wait, atomic_read(&w.complete)); | |
1495 | } | |
1496 | ||
1497 | /*----------------------------------------------------------------*/ | |
1498 | ||
e49e5829 JT |
1499 | static enum pool_mode get_pool_mode(struct pool *pool) |
1500 | { | |
1501 | return pool->pf.mode; | |
1502 | } | |
1503 | ||
3e1a0699 JT |
1504 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
1505 | { | |
1506 | dm_table_event(pool->ti->table); | |
1507 | DMINFO("%s: switching pool to %s mode", | |
1508 | dm_device_name(pool->pool_md), new_mode); | |
1509 | } | |
1510 | ||
8b64e881 | 1511 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 1512 | { |
cdc2b415 | 1513 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
1514 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
1515 | enum pool_mode old_mode = get_pool_mode(pool); | |
1516 | ||
1517 | /* | |
1518 | * Never allow the pool to transition to PM_WRITE mode if user | |
1519 | * intervention is required to verify metadata and data consistency. | |
1520 | */ | |
1521 | if (new_mode == PM_WRITE && needs_check) { | |
1522 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
1523 | dm_device_name(pool->pool_md)); | |
1524 | if (old_mode != new_mode) | |
1525 | new_mode = old_mode; | |
1526 | else | |
1527 | new_mode = PM_READ_ONLY; | |
1528 | } | |
1529 | /* | |
1530 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1531 | * not going to recover without a thin_repair. So we never let the | |
1532 | * pool move out of the old mode. | |
1533 | */ | |
1534 | if (old_mode == PM_FAIL) | |
1535 | new_mode = old_mode; | |
e49e5829 | 1536 | |
8b64e881 | 1537 | switch (new_mode) { |
e49e5829 | 1538 | case PM_FAIL: |
8b64e881 | 1539 | if (old_mode != new_mode) |
3e1a0699 | 1540 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 1541 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1542 | pool->process_bio = process_bio_fail; |
1543 | pool->process_discard = process_bio_fail; | |
1544 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1545 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
1546 | |
1547 | error_retry_list(pool); | |
e49e5829 JT |
1548 | break; |
1549 | ||
1550 | case PM_READ_ONLY: | |
8b64e881 | 1551 | if (old_mode != new_mode) |
3e1a0699 JT |
1552 | notify_of_pool_mode_change(pool, "read-only"); |
1553 | dm_pool_metadata_read_only(pool->pmd); | |
1554 | pool->process_bio = process_bio_read_only; | |
1555 | pool->process_discard = process_bio_success; | |
1556 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1557 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1558 | ||
1559 | error_retry_list(pool); | |
1560 | break; | |
1561 | ||
1562 | case PM_OUT_OF_DATA_SPACE: | |
1563 | /* | |
1564 | * Ideally we'd never hit this state; the low water mark | |
1565 | * would trigger userland to extend the pool before we | |
1566 | * completely run out of data space. However, many small | |
1567 | * IOs to unprovisioned space can consume data space at an | |
1568 | * alarming rate. Adjust your low water mark if you're | |
1569 | * frequently seeing this mode. | |
1570 | */ | |
1571 | if (old_mode != new_mode) | |
1572 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
1573 | pool->process_bio = process_bio_read_only; | |
1574 | pool->process_discard = process_discard; | |
1575 | pool->process_prepared_mapping = process_prepared_mapping; | |
1576 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
e49e5829 JT |
1577 | break; |
1578 | ||
1579 | case PM_WRITE: | |
8b64e881 | 1580 | if (old_mode != new_mode) |
3e1a0699 | 1581 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 1582 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1583 | pool->process_bio = process_bio; |
1584 | pool->process_discard = process_discard; | |
1585 | pool->process_prepared_mapping = process_prepared_mapping; | |
1586 | pool->process_prepared_discard = process_prepared_discard; | |
1587 | break; | |
1588 | } | |
8b64e881 MS |
1589 | |
1590 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
1591 | /* |
1592 | * The pool mode may have changed, sync it so bind_control_target() | |
1593 | * doesn't cause an unexpected mode transition on resume. | |
1594 | */ | |
1595 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
1596 | } |
1597 | ||
07f2b6e0 | 1598 | static void abort_transaction(struct pool *pool) |
b5330655 | 1599 | { |
07f2b6e0 MS |
1600 | const char *dev_name = dm_device_name(pool->pool_md); |
1601 | ||
1602 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
1603 | if (dm_pool_abort_metadata(pool->pmd)) { | |
1604 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
1605 | set_pool_mode(pool, PM_FAIL); | |
1606 | } | |
1607 | ||
1608 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
1609 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
1610 | set_pool_mode(pool, PM_FAIL); | |
1611 | } | |
1612 | } | |
399caddf | 1613 | |
07f2b6e0 MS |
1614 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1615 | { | |
b5330655 JT |
1616 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1617 | dm_device_name(pool->pool_md), op, r); | |
1618 | ||
07f2b6e0 | 1619 | abort_transaction(pool); |
b5330655 JT |
1620 | set_pool_mode(pool, PM_READ_ONLY); |
1621 | } | |
1622 | ||
e49e5829 JT |
1623 | /*----------------------------------------------------------------*/ |
1624 | ||
991d9fa0 JT |
1625 | /* |
1626 | * Mapping functions. | |
1627 | */ | |
1628 | ||
1629 | /* | |
1630 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1631 | */ | |
1632 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1633 | { | |
1634 | unsigned long flags; | |
1635 | struct pool *pool = tc->pool; | |
1636 | ||
1637 | spin_lock_irqsave(&pool->lock, flags); | |
1638 | bio_list_add(&pool->deferred_bios, bio); | |
1639 | spin_unlock_irqrestore(&pool->lock, flags); | |
1640 | ||
1641 | wake_worker(pool); | |
1642 | } | |
1643 | ||
59c3d2c6 | 1644 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1645 | { |
59c3d2c6 | 1646 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1647 | |
1648 | h->tc = tc; | |
1649 | h->shared_read_entry = NULL; | |
e8088073 | 1650 | h->all_io_entry = NULL; |
eb2aa48d | 1651 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1652 | } |
1653 | ||
991d9fa0 JT |
1654 | /* |
1655 | * Non-blocking function called from the thin target's map function. | |
1656 | */ | |
7de3ee57 | 1657 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1658 | { |
1659 | int r; | |
1660 | struct thin_c *tc = ti->private; | |
1661 | dm_block_t block = get_bio_block(tc, bio); | |
1662 | struct dm_thin_device *td = tc->td; | |
1663 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1664 | struct dm_bio_prison_cell cell1, cell2; |
1665 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1666 | struct dm_cell_key key; |
991d9fa0 | 1667 | |
59c3d2c6 | 1668 | thin_hook_bio(tc, bio); |
e49e5829 | 1669 | |
738211f7 JT |
1670 | if (tc->requeue_mode) { |
1671 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1672 | return DM_MAPIO_SUBMITTED; | |
1673 | } | |
1674 | ||
e49e5829 JT |
1675 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
1676 | bio_io_error(bio); | |
1677 | return DM_MAPIO_SUBMITTED; | |
1678 | } | |
1679 | ||
104655fd | 1680 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1681 | thin_defer_bio(tc, bio); |
1682 | return DM_MAPIO_SUBMITTED; | |
1683 | } | |
1684 | ||
1685 | r = dm_thin_find_block(td, block, 0, &result); | |
1686 | ||
1687 | /* | |
1688 | * Note that we defer readahead too. | |
1689 | */ | |
1690 | switch (r) { | |
1691 | case 0: | |
1692 | if (unlikely(result.shared)) { | |
1693 | /* | |
1694 | * We have a race condition here between the | |
1695 | * result.shared value returned by the lookup and | |
1696 | * snapshot creation, which may cause new | |
1697 | * sharing. | |
1698 | * | |
1699 | * To avoid this always quiesce the origin before | |
1700 | * taking the snap. You want to do this anyway to | |
1701 | * ensure a consistent application view | |
1702 | * (i.e. lockfs). | |
1703 | * | |
1704 | * More distant ancestors are irrelevant. The | |
1705 | * shared flag will be set in their case. | |
1706 | */ | |
1707 | thin_defer_bio(tc, bio); | |
e8088073 | 1708 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1709 | } |
e8088073 JT |
1710 | |
1711 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1712 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1713 | return DM_MAPIO_SUBMITTED; |
1714 | ||
1715 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1716 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1717 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1718 | return DM_MAPIO_SUBMITTED; |
1719 | } | |
1720 | ||
1721 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1722 | cell_defer_no_holder_no_free(tc, &cell2); |
1723 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1724 | |
1725 | remap(tc, bio, result.block); | |
1726 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1727 | |
1728 | case -ENODATA: | |
e49e5829 JT |
1729 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1730 | /* | |
1731 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 1732 | * of doing so. |
e49e5829 | 1733 | */ |
8c0f0e8c | 1734 | handle_unserviceable_bio(tc->pool, bio); |
2aab3850 | 1735 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1736 | } |
1737 | /* fall through */ | |
1738 | ||
1739 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1740 | /* |
1741 | * In future, the failed dm_thin_find_block above could | |
1742 | * provide the hint to load the metadata into cache. | |
1743 | */ | |
991d9fa0 | 1744 | thin_defer_bio(tc, bio); |
2aab3850 | 1745 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1746 | |
1747 | default: | |
1748 | /* | |
1749 | * Must always call bio_io_error on failure. | |
1750 | * dm_thin_find_block can fail with -EINVAL if the | |
1751 | * pool is switched to fail-io mode. | |
1752 | */ | |
1753 | bio_io_error(bio); | |
2aab3850 | 1754 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1755 | } |
991d9fa0 JT |
1756 | } |
1757 | ||
1758 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1759 | { | |
1760 | int r; | |
1761 | unsigned long flags; | |
1762 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1763 | ||
1764 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1765 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1766 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1767 | ||
1768 | if (!r) { | |
1769 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1770 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1771 | } | |
1772 | ||
1773 | return r; | |
1774 | } | |
1775 | ||
1776 | static void __requeue_bios(struct pool *pool) | |
1777 | { | |
1778 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1779 | bio_list_init(&pool->retry_on_resume_list); | |
1780 | } | |
1781 | ||
1782 | /*---------------------------------------------------------------- | |
1783 | * Binding of control targets to a pool object | |
1784 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1785 | static bool data_dev_supports_discard(struct pool_c *pt) |
1786 | { | |
1787 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1788 | ||
1789 | return q && blk_queue_discard(q); | |
1790 | } | |
1791 | ||
58051b94 JT |
1792 | static bool is_factor(sector_t block_size, uint32_t n) |
1793 | { | |
1794 | return !sector_div(block_size, n); | |
1795 | } | |
1796 | ||
9bc142dd MS |
1797 | /* |
1798 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1799 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1800 | */ |
0424caa1 | 1801 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1802 | { |
0424caa1 MS |
1803 | struct pool *pool = pt->pool; |
1804 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1805 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1806 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1807 | const char *reason = NULL; | |
9bc142dd MS |
1808 | char buf[BDEVNAME_SIZE]; |
1809 | ||
0424caa1 | 1810 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1811 | return; |
1812 | ||
0424caa1 MS |
1813 | if (!data_dev_supports_discard(pt)) |
1814 | reason = "discard unsupported"; | |
1815 | ||
1816 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1817 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1818 | |
0424caa1 MS |
1819 | else if (data_limits->discard_granularity > block_size) |
1820 | reason = "discard granularity larger than a block"; | |
1821 | ||
58051b94 | 1822 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1823 | reason = "discard granularity not a factor of block size"; |
1824 | ||
1825 | if (reason) { | |
1826 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1827 | pt->adjusted_pf.discard_passdown = false; | |
1828 | } | |
9bc142dd MS |
1829 | } |
1830 | ||
991d9fa0 JT |
1831 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1832 | { | |
1833 | struct pool_c *pt = ti->private; | |
1834 | ||
e49e5829 | 1835 | /* |
9b7aaa64 | 1836 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 1837 | */ |
07f2b6e0 | 1838 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 1839 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1840 | |
8b64e881 MS |
1841 | /* |
1842 | * Don't change the pool's mode until set_pool_mode() below. | |
1843 | * Otherwise the pool's process_* function pointers may | |
1844 | * not match the desired pool mode. | |
1845 | */ | |
1846 | pt->adjusted_pf.mode = old_mode; | |
1847 | ||
1848 | pool->ti = ti; | |
1849 | pool->pf = pt->adjusted_pf; | |
1850 | pool->low_water_blocks = pt->low_water_blocks; | |
1851 | ||
9bc142dd | 1852 | set_pool_mode(pool, new_mode); |
f402693d | 1853 | |
991d9fa0 JT |
1854 | return 0; |
1855 | } | |
1856 | ||
1857 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1858 | { | |
1859 | if (pool->ti == ti) | |
1860 | pool->ti = NULL; | |
1861 | } | |
1862 | ||
1863 | /*---------------------------------------------------------------- | |
1864 | * Pool creation | |
1865 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1866 | /* Initialize pool features. */ |
1867 | static void pool_features_init(struct pool_features *pf) | |
1868 | { | |
e49e5829 | 1869 | pf->mode = PM_WRITE; |
9bc142dd MS |
1870 | pf->zero_new_blocks = true; |
1871 | pf->discard_enabled = true; | |
1872 | pf->discard_passdown = true; | |
787a996c | 1873 | pf->error_if_no_space = false; |
67e2e2b2 JT |
1874 | } |
1875 | ||
991d9fa0 JT |
1876 | static void __pool_destroy(struct pool *pool) |
1877 | { | |
1878 | __pool_table_remove(pool); | |
1879 | ||
1880 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1881 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1882 | ||
44feb387 | 1883 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1884 | dm_kcopyd_client_destroy(pool->copier); |
1885 | ||
1886 | if (pool->wq) | |
1887 | destroy_workqueue(pool->wq); | |
1888 | ||
1889 | if (pool->next_mapping) | |
1890 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1891 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1892 | dm_deferred_set_destroy(pool->shared_read_ds); |
1893 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1894 | kfree(pool); |
1895 | } | |
1896 | ||
a24c2569 | 1897 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1898 | |
991d9fa0 JT |
1899 | static struct pool *pool_create(struct mapped_device *pool_md, |
1900 | struct block_device *metadata_dev, | |
e49e5829 JT |
1901 | unsigned long block_size, |
1902 | int read_only, char **error) | |
991d9fa0 JT |
1903 | { |
1904 | int r; | |
1905 | void *err_p; | |
1906 | struct pool *pool; | |
1907 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1908 | bool format_device = read_only ? false : true; |
991d9fa0 | 1909 | |
e49e5829 | 1910 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1911 | if (IS_ERR(pmd)) { |
1912 | *error = "Error creating metadata object"; | |
1913 | return (struct pool *)pmd; | |
1914 | } | |
1915 | ||
1916 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1917 | if (!pool) { | |
1918 | *error = "Error allocating memory for pool"; | |
1919 | err_p = ERR_PTR(-ENOMEM); | |
1920 | goto bad_pool; | |
1921 | } | |
1922 | ||
1923 | pool->pmd = pmd; | |
1924 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1925 | if (block_size & (block_size - 1)) |
1926 | pool->sectors_per_block_shift = -1; | |
1927 | else | |
1928 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1929 | pool->low_water_blocks = 0; |
67e2e2b2 | 1930 | pool_features_init(&pool->pf); |
44feb387 | 1931 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
1932 | if (!pool->prison) { |
1933 | *error = "Error creating pool's bio prison"; | |
1934 | err_p = ERR_PTR(-ENOMEM); | |
1935 | goto bad_prison; | |
1936 | } | |
1937 | ||
df5d2e90 | 1938 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
1939 | if (IS_ERR(pool->copier)) { |
1940 | r = PTR_ERR(pool->copier); | |
1941 | *error = "Error creating pool's kcopyd client"; | |
1942 | err_p = ERR_PTR(r); | |
1943 | goto bad_kcopyd_client; | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Create singlethreaded workqueue that will service all devices | |
1948 | * that use this metadata. | |
1949 | */ | |
1950 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1951 | if (!pool->wq) { | |
1952 | *error = "Error creating pool's workqueue"; | |
1953 | err_p = ERR_PTR(-ENOMEM); | |
1954 | goto bad_wq; | |
1955 | } | |
1956 | ||
1957 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1958 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1959 | spin_lock_init(&pool->lock); |
1960 | bio_list_init(&pool->deferred_bios); | |
1961 | bio_list_init(&pool->deferred_flush_bios); | |
1962 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1963 | INIT_LIST_HEAD(&pool->prepared_discards); |
88a6621b | 1964 | pool->low_water_triggered = false; |
991d9fa0 | 1965 | bio_list_init(&pool->retry_on_resume_list); |
44feb387 MS |
1966 | |
1967 | pool->shared_read_ds = dm_deferred_set_create(); | |
1968 | if (!pool->shared_read_ds) { | |
1969 | *error = "Error creating pool's shared read deferred set"; | |
1970 | err_p = ERR_PTR(-ENOMEM); | |
1971 | goto bad_shared_read_ds; | |
1972 | } | |
1973 | ||
1974 | pool->all_io_ds = dm_deferred_set_create(); | |
1975 | if (!pool->all_io_ds) { | |
1976 | *error = "Error creating pool's all io deferred set"; | |
1977 | err_p = ERR_PTR(-ENOMEM); | |
1978 | goto bad_all_io_ds; | |
1979 | } | |
991d9fa0 JT |
1980 | |
1981 | pool->next_mapping = NULL; | |
a24c2569 MS |
1982 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1983 | _new_mapping_cache); | |
991d9fa0 JT |
1984 | if (!pool->mapping_pool) { |
1985 | *error = "Error creating pool's mapping mempool"; | |
1986 | err_p = ERR_PTR(-ENOMEM); | |
1987 | goto bad_mapping_pool; | |
1988 | } | |
1989 | ||
991d9fa0 | 1990 | pool->ref_count = 1; |
905e51b3 | 1991 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1992 | pool->pool_md = pool_md; |
1993 | pool->md_dev = metadata_dev; | |
1994 | __pool_table_insert(pool); | |
1995 | ||
1996 | return pool; | |
1997 | ||
991d9fa0 | 1998 | bad_mapping_pool: |
44feb387 MS |
1999 | dm_deferred_set_destroy(pool->all_io_ds); |
2000 | bad_all_io_ds: | |
2001 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2002 | bad_shared_read_ds: | |
991d9fa0 JT |
2003 | destroy_workqueue(pool->wq); |
2004 | bad_wq: | |
2005 | dm_kcopyd_client_destroy(pool->copier); | |
2006 | bad_kcopyd_client: | |
44feb387 | 2007 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2008 | bad_prison: |
2009 | kfree(pool); | |
2010 | bad_pool: | |
2011 | if (dm_pool_metadata_close(pmd)) | |
2012 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2013 | ||
2014 | return err_p; | |
2015 | } | |
2016 | ||
2017 | static void __pool_inc(struct pool *pool) | |
2018 | { | |
2019 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2020 | pool->ref_count++; | |
2021 | } | |
2022 | ||
2023 | static void __pool_dec(struct pool *pool) | |
2024 | { | |
2025 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2026 | BUG_ON(!pool->ref_count); | |
2027 | if (!--pool->ref_count) | |
2028 | __pool_destroy(pool); | |
2029 | } | |
2030 | ||
2031 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2032 | struct block_device *metadata_dev, | |
e49e5829 JT |
2033 | unsigned long block_size, int read_only, |
2034 | char **error, int *created) | |
991d9fa0 JT |
2035 | { |
2036 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2037 | ||
2038 | if (pool) { | |
f09996c9 MS |
2039 | if (pool->pool_md != pool_md) { |
2040 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2041 | return ERR_PTR(-EBUSY); |
f09996c9 | 2042 | } |
991d9fa0 JT |
2043 | __pool_inc(pool); |
2044 | ||
2045 | } else { | |
2046 | pool = __pool_table_lookup(pool_md); | |
2047 | if (pool) { | |
f09996c9 MS |
2048 | if (pool->md_dev != metadata_dev) { |
2049 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2050 | return ERR_PTR(-EINVAL); |
f09996c9 | 2051 | } |
991d9fa0 JT |
2052 | __pool_inc(pool); |
2053 | ||
67e2e2b2 | 2054 | } else { |
e49e5829 | 2055 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2056 | *created = 1; |
2057 | } | |
991d9fa0 JT |
2058 | } |
2059 | ||
2060 | return pool; | |
2061 | } | |
2062 | ||
2063 | /*---------------------------------------------------------------- | |
2064 | * Pool target methods | |
2065 | *--------------------------------------------------------------*/ | |
2066 | static void pool_dtr(struct dm_target *ti) | |
2067 | { | |
2068 | struct pool_c *pt = ti->private; | |
2069 | ||
2070 | mutex_lock(&dm_thin_pool_table.mutex); | |
2071 | ||
2072 | unbind_control_target(pt->pool, ti); | |
2073 | __pool_dec(pt->pool); | |
2074 | dm_put_device(ti, pt->metadata_dev); | |
2075 | dm_put_device(ti, pt->data_dev); | |
2076 | kfree(pt); | |
2077 | ||
2078 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2079 | } | |
2080 | ||
991d9fa0 JT |
2081 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2082 | struct dm_target *ti) | |
2083 | { | |
2084 | int r; | |
2085 | unsigned argc; | |
2086 | const char *arg_name; | |
2087 | ||
2088 | static struct dm_arg _args[] = { | |
74aa45c3 | 2089 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2090 | }; |
2091 | ||
2092 | /* | |
2093 | * No feature arguments supplied. | |
2094 | */ | |
2095 | if (!as->argc) | |
2096 | return 0; | |
2097 | ||
2098 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2099 | if (r) | |
2100 | return -EINVAL; | |
2101 | ||
2102 | while (argc && !r) { | |
2103 | arg_name = dm_shift_arg(as); | |
2104 | argc--; | |
2105 | ||
e49e5829 | 2106 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2107 | pf->zero_new_blocks = false; |
e49e5829 JT |
2108 | |
2109 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2110 | pf->discard_enabled = false; |
e49e5829 JT |
2111 | |
2112 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2113 | pf->discard_passdown = false; |
991d9fa0 | 2114 | |
e49e5829 JT |
2115 | else if (!strcasecmp(arg_name, "read_only")) |
2116 | pf->mode = PM_READ_ONLY; | |
2117 | ||
787a996c MS |
2118 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2119 | pf->error_if_no_space = true; | |
2120 | ||
e49e5829 JT |
2121 | else { |
2122 | ti->error = "Unrecognised pool feature requested"; | |
2123 | r = -EINVAL; | |
2124 | break; | |
2125 | } | |
991d9fa0 JT |
2126 | } |
2127 | ||
2128 | return r; | |
2129 | } | |
2130 | ||
ac8c3f3d JT |
2131 | static void metadata_low_callback(void *context) |
2132 | { | |
2133 | struct pool *pool = context; | |
2134 | ||
2135 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2136 | dm_device_name(pool->pool_md)); | |
2137 | ||
2138 | dm_table_event(pool->ti->table); | |
2139 | } | |
2140 | ||
7d48935e MS |
2141 | static sector_t get_dev_size(struct block_device *bdev) |
2142 | { | |
2143 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2144 | } | |
2145 | ||
2146 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2147 | { |
7d48935e | 2148 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2149 | char buffer[BDEVNAME_SIZE]; |
2150 | ||
7d48935e | 2151 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2152 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2153 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2154 | } |
2155 | ||
2156 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2157 | { | |
2158 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2159 | ||
2160 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2161 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2162 | |
2163 | return metadata_dev_size; | |
2164 | } | |
2165 | ||
24347e95 JT |
2166 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2167 | { | |
2168 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2169 | ||
7d48935e | 2170 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2171 | |
2172 | return metadata_dev_size; | |
2173 | } | |
2174 | ||
ac8c3f3d JT |
2175 | /* |
2176 | * When a metadata threshold is crossed a dm event is triggered, and | |
2177 | * userland should respond by growing the metadata device. We could let | |
2178 | * userland set the threshold, like we do with the data threshold, but I'm | |
2179 | * not sure they know enough to do this well. | |
2180 | */ | |
2181 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2182 | { | |
2183 | /* | |
2184 | * 4M is ample for all ops with the possible exception of thin | |
2185 | * device deletion which is harmless if it fails (just retry the | |
2186 | * delete after you've grown the device). | |
2187 | */ | |
2188 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2189 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2190 | } | |
2191 | ||
991d9fa0 JT |
2192 | /* |
2193 | * thin-pool <metadata dev> <data dev> | |
2194 | * <data block size (sectors)> | |
2195 | * <low water mark (blocks)> | |
2196 | * [<#feature args> [<arg>]*] | |
2197 | * | |
2198 | * Optional feature arguments are: | |
2199 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2200 | * ignore_discard: disable discard |
2201 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2202 | * read_only: Don't allow any changes to be made to the pool metadata. |
2203 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2204 | */ |
2205 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2206 | { | |
67e2e2b2 | 2207 | int r, pool_created = 0; |
991d9fa0 JT |
2208 | struct pool_c *pt; |
2209 | struct pool *pool; | |
2210 | struct pool_features pf; | |
2211 | struct dm_arg_set as; | |
2212 | struct dm_dev *data_dev; | |
2213 | unsigned long block_size; | |
2214 | dm_block_t low_water_blocks; | |
2215 | struct dm_dev *metadata_dev; | |
5d0db96d | 2216 | fmode_t metadata_mode; |
991d9fa0 JT |
2217 | |
2218 | /* | |
2219 | * FIXME Remove validation from scope of lock. | |
2220 | */ | |
2221 | mutex_lock(&dm_thin_pool_table.mutex); | |
2222 | ||
2223 | if (argc < 4) { | |
2224 | ti->error = "Invalid argument count"; | |
2225 | r = -EINVAL; | |
2226 | goto out_unlock; | |
2227 | } | |
5d0db96d | 2228 | |
991d9fa0 JT |
2229 | as.argc = argc; |
2230 | as.argv = argv; | |
2231 | ||
5d0db96d JT |
2232 | /* |
2233 | * Set default pool features. | |
2234 | */ | |
2235 | pool_features_init(&pf); | |
2236 | ||
2237 | dm_consume_args(&as, 4); | |
2238 | r = parse_pool_features(&as, &pf, ti); | |
2239 | if (r) | |
2240 | goto out_unlock; | |
2241 | ||
2242 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2243 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2244 | if (r) { |
2245 | ti->error = "Error opening metadata block device"; | |
2246 | goto out_unlock; | |
2247 | } | |
7d48935e | 2248 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2249 | |
2250 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2251 | if (r) { | |
2252 | ti->error = "Error getting data device"; | |
2253 | goto out_metadata; | |
2254 | } | |
2255 | ||
2256 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2257 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2258 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2259 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2260 | ti->error = "Invalid block size"; |
2261 | r = -EINVAL; | |
2262 | goto out; | |
2263 | } | |
2264 | ||
2265 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2266 | ti->error = "Invalid low water mark"; | |
2267 | r = -EINVAL; | |
2268 | goto out; | |
2269 | } | |
2270 | ||
991d9fa0 JT |
2271 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2272 | if (!pt) { | |
2273 | r = -ENOMEM; | |
2274 | goto out; | |
2275 | } | |
2276 | ||
2277 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2278 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2279 | if (IS_ERR(pool)) { |
2280 | r = PTR_ERR(pool); | |
2281 | goto out_free_pt; | |
2282 | } | |
2283 | ||
67e2e2b2 JT |
2284 | /* |
2285 | * 'pool_created' reflects whether this is the first table load. | |
2286 | * Top level discard support is not allowed to be changed after | |
2287 | * initial load. This would require a pool reload to trigger thin | |
2288 | * device changes. | |
2289 | */ | |
2290 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2291 | ti->error = "Discard support cannot be disabled once enabled"; | |
2292 | r = -EINVAL; | |
2293 | goto out_flags_changed; | |
2294 | } | |
2295 | ||
991d9fa0 JT |
2296 | pt->pool = pool; |
2297 | pt->ti = ti; | |
2298 | pt->metadata_dev = metadata_dev; | |
2299 | pt->data_dev = data_dev; | |
2300 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2301 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2302 | ti->num_flush_bios = 1; |
9bc142dd | 2303 | |
67e2e2b2 JT |
2304 | /* |
2305 | * Only need to enable discards if the pool should pass | |
2306 | * them down to the data device. The thin device's discard | |
2307 | * processing will cause mappings to be removed from the btree. | |
2308 | */ | |
b60ab990 | 2309 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2310 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2311 | ti->num_discard_bios = 1; |
9bc142dd | 2312 | |
67e2e2b2 JT |
2313 | /* |
2314 | * Setting 'discards_supported' circumvents the normal | |
2315 | * stacking of discard limits (this keeps the pool and | |
2316 | * thin devices' discard limits consistent). | |
2317 | */ | |
0ac55489 | 2318 | ti->discards_supported = true; |
67e2e2b2 | 2319 | } |
991d9fa0 JT |
2320 | ti->private = pt; |
2321 | ||
ac8c3f3d JT |
2322 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2323 | calc_metadata_threshold(pt), | |
2324 | metadata_low_callback, | |
2325 | pool); | |
2326 | if (r) | |
2327 | goto out_free_pt; | |
2328 | ||
991d9fa0 JT |
2329 | pt->callbacks.congested_fn = pool_is_congested; |
2330 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2331 | ||
2332 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2333 | ||
2334 | return 0; | |
2335 | ||
67e2e2b2 JT |
2336 | out_flags_changed: |
2337 | __pool_dec(pool); | |
991d9fa0 JT |
2338 | out_free_pt: |
2339 | kfree(pt); | |
2340 | out: | |
2341 | dm_put_device(ti, data_dev); | |
2342 | out_metadata: | |
2343 | dm_put_device(ti, metadata_dev); | |
2344 | out_unlock: | |
2345 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2346 | ||
2347 | return r; | |
2348 | } | |
2349 | ||
7de3ee57 | 2350 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2351 | { |
2352 | int r; | |
2353 | struct pool_c *pt = ti->private; | |
2354 | struct pool *pool = pt->pool; | |
2355 | unsigned long flags; | |
2356 | ||
2357 | /* | |
2358 | * As this is a singleton target, ti->begin is always zero. | |
2359 | */ | |
2360 | spin_lock_irqsave(&pool->lock, flags); | |
2361 | bio->bi_bdev = pt->data_dev->bdev; | |
2362 | r = DM_MAPIO_REMAPPED; | |
2363 | spin_unlock_irqrestore(&pool->lock, flags); | |
2364 | ||
2365 | return r; | |
2366 | } | |
2367 | ||
b17446df | 2368 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2369 | { |
2370 | int r; | |
2371 | struct pool_c *pt = ti->private; | |
2372 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2373 | sector_t data_size = ti->len; |
2374 | dm_block_t sb_data_size; | |
991d9fa0 | 2375 | |
b17446df | 2376 | *need_commit = false; |
991d9fa0 | 2377 | |
55f2b8bd MS |
2378 | (void) sector_div(data_size, pool->sectors_per_block); |
2379 | ||
991d9fa0 JT |
2380 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2381 | if (r) { | |
4fa5971a MS |
2382 | DMERR("%s: failed to retrieve data device size", |
2383 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2384 | return r; |
2385 | } | |
2386 | ||
2387 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2388 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2389 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2390 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2391 | return -EINVAL; |
2392 | ||
2393 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2394 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2395 | DMERR("%s: unable to grow the data device until repaired.", | |
2396 | dm_device_name(pool->pool_md)); | |
2397 | return 0; | |
2398 | } | |
2399 | ||
6f7f51d4 MS |
2400 | if (sb_data_size) |
2401 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2402 | dm_device_name(pool->pool_md), | |
2403 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2404 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2405 | if (r) { | |
b5330655 | 2406 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2407 | return r; |
2408 | } | |
2409 | ||
b17446df | 2410 | *need_commit = true; |
991d9fa0 JT |
2411 | } |
2412 | ||
2413 | return 0; | |
2414 | } | |
2415 | ||
24347e95 JT |
2416 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2417 | { | |
2418 | int r; | |
2419 | struct pool_c *pt = ti->private; | |
2420 | struct pool *pool = pt->pool; | |
2421 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2422 | ||
2423 | *need_commit = false; | |
2424 | ||
610bba8b | 2425 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2426 | |
2427 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2428 | if (r) { | |
4fa5971a MS |
2429 | DMERR("%s: failed to retrieve metadata device size", |
2430 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2431 | return r; |
2432 | } | |
2433 | ||
2434 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2435 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2436 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2437 | metadata_dev_size, sb_metadata_dev_size); |
2438 | return -EINVAL; | |
2439 | ||
2440 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
2441 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2442 | DMERR("%s: unable to grow the metadata device until repaired.", | |
2443 | dm_device_name(pool->pool_md)); | |
2444 | return 0; | |
2445 | } | |
2446 | ||
7d48935e | 2447 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
2448 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2449 | dm_device_name(pool->pool_md), | |
2450 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2451 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2452 | if (r) { | |
b5330655 | 2453 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2454 | return r; |
2455 | } | |
2456 | ||
2457 | *need_commit = true; | |
2458 | } | |
2459 | ||
2460 | return 0; | |
2461 | } | |
2462 | ||
b17446df JT |
2463 | /* |
2464 | * Retrieves the number of blocks of the data device from | |
2465 | * the superblock and compares it to the actual device size, | |
2466 | * thus resizing the data device in case it has grown. | |
2467 | * | |
2468 | * This both copes with opening preallocated data devices in the ctr | |
2469 | * being followed by a resume | |
2470 | * -and- | |
2471 | * calling the resume method individually after userspace has | |
2472 | * grown the data device in reaction to a table event. | |
2473 | */ | |
2474 | static int pool_preresume(struct dm_target *ti) | |
2475 | { | |
2476 | int r; | |
24347e95 | 2477 | bool need_commit1, need_commit2; |
b17446df JT |
2478 | struct pool_c *pt = ti->private; |
2479 | struct pool *pool = pt->pool; | |
2480 | ||
2481 | /* | |
2482 | * Take control of the pool object. | |
2483 | */ | |
2484 | r = bind_control_target(pool, ti); | |
2485 | if (r) | |
2486 | return r; | |
2487 | ||
2488 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2489 | if (r) | |
2490 | return r; | |
2491 | ||
24347e95 JT |
2492 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2493 | if (r) | |
2494 | return r; | |
2495 | ||
2496 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2497 | (void) commit(pool); |
b17446df JT |
2498 | |
2499 | return 0; | |
2500 | } | |
2501 | ||
991d9fa0 JT |
2502 | static void pool_resume(struct dm_target *ti) |
2503 | { | |
2504 | struct pool_c *pt = ti->private; | |
2505 | struct pool *pool = pt->pool; | |
2506 | unsigned long flags; | |
2507 | ||
2508 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 2509 | pool->low_water_triggered = false; |
991d9fa0 JT |
2510 | __requeue_bios(pool); |
2511 | spin_unlock_irqrestore(&pool->lock, flags); | |
2512 | ||
905e51b3 | 2513 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2514 | } |
2515 | ||
2516 | static void pool_postsuspend(struct dm_target *ti) | |
2517 | { | |
991d9fa0 JT |
2518 | struct pool_c *pt = ti->private; |
2519 | struct pool *pool = pt->pool; | |
2520 | ||
905e51b3 | 2521 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2522 | flush_workqueue(pool->wq); |
020cc3b5 | 2523 | (void) commit(pool); |
991d9fa0 JT |
2524 | } |
2525 | ||
2526 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2527 | { | |
2528 | if (argc != args_required) { | |
2529 | DMWARN("Message received with %u arguments instead of %u.", | |
2530 | argc, args_required); | |
2531 | return -EINVAL; | |
2532 | } | |
2533 | ||
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2538 | { | |
2539 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2540 | *dev_id <= MAX_DEV_ID) | |
2541 | return 0; | |
2542 | ||
2543 | if (warning) | |
2544 | DMWARN("Message received with invalid device id: %s", arg); | |
2545 | ||
2546 | return -EINVAL; | |
2547 | } | |
2548 | ||
2549 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2550 | { | |
2551 | dm_thin_id dev_id; | |
2552 | int r; | |
2553 | ||
2554 | r = check_arg_count(argc, 2); | |
2555 | if (r) | |
2556 | return r; | |
2557 | ||
2558 | r = read_dev_id(argv[1], &dev_id, 1); | |
2559 | if (r) | |
2560 | return r; | |
2561 | ||
2562 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2563 | if (r) { | |
2564 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2565 | argv[1]); | |
2566 | return r; | |
2567 | } | |
2568 | ||
2569 | return 0; | |
2570 | } | |
2571 | ||
2572 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2573 | { | |
2574 | dm_thin_id dev_id; | |
2575 | dm_thin_id origin_dev_id; | |
2576 | int r; | |
2577 | ||
2578 | r = check_arg_count(argc, 3); | |
2579 | if (r) | |
2580 | return r; | |
2581 | ||
2582 | r = read_dev_id(argv[1], &dev_id, 1); | |
2583 | if (r) | |
2584 | return r; | |
2585 | ||
2586 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2587 | if (r) | |
2588 | return r; | |
2589 | ||
2590 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2591 | if (r) { | |
2592 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2593 | argv[1], argv[2]); | |
2594 | return r; | |
2595 | } | |
2596 | ||
2597 | return 0; | |
2598 | } | |
2599 | ||
2600 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2601 | { | |
2602 | dm_thin_id dev_id; | |
2603 | int r; | |
2604 | ||
2605 | r = check_arg_count(argc, 2); | |
2606 | if (r) | |
2607 | return r; | |
2608 | ||
2609 | r = read_dev_id(argv[1], &dev_id, 1); | |
2610 | if (r) | |
2611 | return r; | |
2612 | ||
2613 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2614 | if (r) | |
2615 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2616 | ||
2617 | return r; | |
2618 | } | |
2619 | ||
2620 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2621 | { | |
2622 | dm_thin_id old_id, new_id; | |
2623 | int r; | |
2624 | ||
2625 | r = check_arg_count(argc, 3); | |
2626 | if (r) | |
2627 | return r; | |
2628 | ||
2629 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2630 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2631 | return -EINVAL; | |
2632 | } | |
2633 | ||
2634 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2635 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2636 | return -EINVAL; | |
2637 | } | |
2638 | ||
2639 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2640 | if (r) { | |
2641 | DMWARN("Failed to change transaction id from %s to %s.", | |
2642 | argv[1], argv[2]); | |
2643 | return r; | |
2644 | } | |
2645 | ||
2646 | return 0; | |
2647 | } | |
2648 | ||
cc8394d8 JT |
2649 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2650 | { | |
2651 | int r; | |
2652 | ||
2653 | r = check_arg_count(argc, 1); | |
2654 | if (r) | |
2655 | return r; | |
2656 | ||
020cc3b5 | 2657 | (void) commit(pool); |
0d200aef | 2658 | |
cc8394d8 JT |
2659 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2660 | if (r) | |
2661 | DMWARN("reserve_metadata_snap message failed."); | |
2662 | ||
2663 | return r; | |
2664 | } | |
2665 | ||
2666 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2667 | { | |
2668 | int r; | |
2669 | ||
2670 | r = check_arg_count(argc, 1); | |
2671 | if (r) | |
2672 | return r; | |
2673 | ||
2674 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2675 | if (r) | |
2676 | DMWARN("release_metadata_snap message failed."); | |
2677 | ||
2678 | return r; | |
2679 | } | |
2680 | ||
991d9fa0 JT |
2681 | /* |
2682 | * Messages supported: | |
2683 | * create_thin <dev_id> | |
2684 | * create_snap <dev_id> <origin_id> | |
2685 | * delete <dev_id> | |
2686 | * trim <dev_id> <new_size_in_sectors> | |
2687 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2688 | * reserve_metadata_snap |
2689 | * release_metadata_snap | |
991d9fa0 JT |
2690 | */ |
2691 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2692 | { | |
2693 | int r = -EINVAL; | |
2694 | struct pool_c *pt = ti->private; | |
2695 | struct pool *pool = pt->pool; | |
2696 | ||
2697 | if (!strcasecmp(argv[0], "create_thin")) | |
2698 | r = process_create_thin_mesg(argc, argv, pool); | |
2699 | ||
2700 | else if (!strcasecmp(argv[0], "create_snap")) | |
2701 | r = process_create_snap_mesg(argc, argv, pool); | |
2702 | ||
2703 | else if (!strcasecmp(argv[0], "delete")) | |
2704 | r = process_delete_mesg(argc, argv, pool); | |
2705 | ||
2706 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2707 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2708 | ||
cc8394d8 JT |
2709 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2710 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2711 | ||
2712 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2713 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2714 | ||
991d9fa0 JT |
2715 | else |
2716 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2717 | ||
e49e5829 | 2718 | if (!r) |
020cc3b5 | 2719 | (void) commit(pool); |
991d9fa0 JT |
2720 | |
2721 | return r; | |
2722 | } | |
2723 | ||
e49e5829 JT |
2724 | static void emit_flags(struct pool_features *pf, char *result, |
2725 | unsigned sz, unsigned maxlen) | |
2726 | { | |
2727 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
2728 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
2729 | pf->error_if_no_space; | |
e49e5829 JT |
2730 | DMEMIT("%u ", count); |
2731 | ||
2732 | if (!pf->zero_new_blocks) | |
2733 | DMEMIT("skip_block_zeroing "); | |
2734 | ||
2735 | if (!pf->discard_enabled) | |
2736 | DMEMIT("ignore_discard "); | |
2737 | ||
2738 | if (!pf->discard_passdown) | |
2739 | DMEMIT("no_discard_passdown "); | |
2740 | ||
2741 | if (pf->mode == PM_READ_ONLY) | |
2742 | DMEMIT("read_only "); | |
787a996c MS |
2743 | |
2744 | if (pf->error_if_no_space) | |
2745 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
2746 | } |
2747 | ||
991d9fa0 JT |
2748 | /* |
2749 | * Status line is: | |
2750 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2751 | * <used data sectors>/<total data sectors> <held metadata root> | |
2752 | */ | |
fd7c092e MP |
2753 | static void pool_status(struct dm_target *ti, status_type_t type, |
2754 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2755 | { |
e49e5829 | 2756 | int r; |
991d9fa0 JT |
2757 | unsigned sz = 0; |
2758 | uint64_t transaction_id; | |
2759 | dm_block_t nr_free_blocks_data; | |
2760 | dm_block_t nr_free_blocks_metadata; | |
2761 | dm_block_t nr_blocks_data; | |
2762 | dm_block_t nr_blocks_metadata; | |
2763 | dm_block_t held_root; | |
2764 | char buf[BDEVNAME_SIZE]; | |
2765 | char buf2[BDEVNAME_SIZE]; | |
2766 | struct pool_c *pt = ti->private; | |
2767 | struct pool *pool = pt->pool; | |
2768 | ||
2769 | switch (type) { | |
2770 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2771 | if (get_pool_mode(pool) == PM_FAIL) { |
2772 | DMEMIT("Fail"); | |
2773 | break; | |
2774 | } | |
2775 | ||
1f4e0ff0 AK |
2776 | /* Commit to ensure statistics aren't out-of-date */ |
2777 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2778 | (void) commit(pool); |
1f4e0ff0 | 2779 | |
fd7c092e MP |
2780 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2781 | if (r) { | |
4fa5971a MS |
2782 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2783 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2784 | goto err; |
2785 | } | |
991d9fa0 | 2786 | |
fd7c092e MP |
2787 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2788 | if (r) { | |
4fa5971a MS |
2789 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2790 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2791 | goto err; |
2792 | } | |
991d9fa0 JT |
2793 | |
2794 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2795 | if (r) { |
4fa5971a MS |
2796 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2797 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2798 | goto err; |
2799 | } | |
991d9fa0 | 2800 | |
fd7c092e MP |
2801 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2802 | if (r) { | |
4fa5971a MS |
2803 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2804 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2805 | goto err; |
2806 | } | |
991d9fa0 JT |
2807 | |
2808 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2809 | if (r) { |
4fa5971a MS |
2810 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2811 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2812 | goto err; |
2813 | } | |
991d9fa0 | 2814 | |
cc8394d8 | 2815 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2816 | if (r) { |
4fa5971a MS |
2817 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2818 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2819 | goto err; |
2820 | } | |
991d9fa0 JT |
2821 | |
2822 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2823 | (unsigned long long)transaction_id, | |
2824 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2825 | (unsigned long long)nr_blocks_metadata, | |
2826 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2827 | (unsigned long long)nr_blocks_data); | |
2828 | ||
2829 | if (held_root) | |
e49e5829 JT |
2830 | DMEMIT("%llu ", held_root); |
2831 | else | |
2832 | DMEMIT("- "); | |
2833 | ||
3e1a0699 JT |
2834 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
2835 | DMEMIT("out_of_data_space "); | |
2836 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 2837 | DMEMIT("ro "); |
991d9fa0 | 2838 | else |
e49e5829 JT |
2839 | DMEMIT("rw "); |
2840 | ||
018debea | 2841 | if (!pool->pf.discard_enabled) |
787a996c | 2842 | DMEMIT("ignore_discard "); |
018debea | 2843 | else if (pool->pf.discard_passdown) |
787a996c MS |
2844 | DMEMIT("discard_passdown "); |
2845 | else | |
2846 | DMEMIT("no_discard_passdown "); | |
2847 | ||
2848 | if (pool->pf.error_if_no_space) | |
2849 | DMEMIT("error_if_no_space "); | |
e49e5829 | 2850 | else |
787a996c | 2851 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
2852 | |
2853 | break; | |
2854 | ||
2855 | case STATUSTYPE_TABLE: | |
2856 | DMEMIT("%s %s %lu %llu ", | |
2857 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2858 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2859 | (unsigned long)pool->sectors_per_block, | |
2860 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2861 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2862 | break; |
2863 | } | |
fd7c092e | 2864 | return; |
991d9fa0 | 2865 | |
fd7c092e MP |
2866 | err: |
2867 | DMEMIT("Error"); | |
991d9fa0 JT |
2868 | } |
2869 | ||
2870 | static int pool_iterate_devices(struct dm_target *ti, | |
2871 | iterate_devices_callout_fn fn, void *data) | |
2872 | { | |
2873 | struct pool_c *pt = ti->private; | |
2874 | ||
2875 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2876 | } | |
2877 | ||
2878 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2879 | struct bio_vec *biovec, int max_size) | |
2880 | { | |
2881 | struct pool_c *pt = ti->private; | |
2882 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2883 | ||
2884 | if (!q->merge_bvec_fn) | |
2885 | return max_size; | |
2886 | ||
2887 | bvm->bi_bdev = pt->data_dev->bdev; | |
2888 | ||
2889 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2890 | } | |
2891 | ||
0424caa1 | 2892 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2893 | { |
0424caa1 MS |
2894 | struct pool *pool = pt->pool; |
2895 | struct queue_limits *data_limits; | |
2896 | ||
104655fd JT |
2897 | limits->max_discard_sectors = pool->sectors_per_block; |
2898 | ||
2899 | /* | |
0424caa1 | 2900 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2901 | */ |
0424caa1 MS |
2902 | if (pt->adjusted_pf.discard_passdown) { |
2903 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
2904 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 2905 | } else |
0424caa1 | 2906 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
2907 | } |
2908 | ||
991d9fa0 JT |
2909 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2910 | { | |
2911 | struct pool_c *pt = ti->private; | |
2912 | struct pool *pool = pt->pool; | |
0cc67cd9 | 2913 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 2914 | |
0cc67cd9 MS |
2915 | /* |
2916 | * If the system-determined stacked limits are compatible with the | |
2917 | * pool's blocksize (io_opt is a factor) do not override them. | |
2918 | */ | |
2919 | if (io_opt_sectors < pool->sectors_per_block || | |
2920 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
2921 | blk_limits_io_min(limits, 0); | |
2922 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
2923 | } | |
0424caa1 MS |
2924 | |
2925 | /* | |
2926 | * pt->adjusted_pf is a staging area for the actual features to use. | |
2927 | * They get transferred to the live pool in bind_control_target() | |
2928 | * called from pool_preresume(). | |
2929 | */ | |
b60ab990 MS |
2930 | if (!pt->adjusted_pf.discard_enabled) { |
2931 | /* | |
2932 | * Must explicitly disallow stacking discard limits otherwise the | |
2933 | * block layer will stack them if pool's data device has support. | |
2934 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
2935 | * user to see that, so make sure to set all discard limits to 0. | |
2936 | */ | |
2937 | limits->discard_granularity = 0; | |
0424caa1 | 2938 | return; |
b60ab990 | 2939 | } |
0424caa1 MS |
2940 | |
2941 | disable_passdown_if_not_supported(pt); | |
2942 | ||
2943 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
2944 | } |
2945 | ||
2946 | static struct target_type pool_target = { | |
2947 | .name = "thin-pool", | |
2948 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2949 | DM_TARGET_IMMUTABLE, | |
07f2b6e0 | 2950 | .version = {1, 11, 0}, |
991d9fa0 JT |
2951 | .module = THIS_MODULE, |
2952 | .ctr = pool_ctr, | |
2953 | .dtr = pool_dtr, | |
2954 | .map = pool_map, | |
2955 | .postsuspend = pool_postsuspend, | |
2956 | .preresume = pool_preresume, | |
2957 | .resume = pool_resume, | |
2958 | .message = pool_message, | |
2959 | .status = pool_status, | |
2960 | .merge = pool_merge, | |
2961 | .iterate_devices = pool_iterate_devices, | |
2962 | .io_hints = pool_io_hints, | |
2963 | }; | |
2964 | ||
2965 | /*---------------------------------------------------------------- | |
2966 | * Thin target methods | |
2967 | *--------------------------------------------------------------*/ | |
2968 | static void thin_dtr(struct dm_target *ti) | |
2969 | { | |
2970 | struct thin_c *tc = ti->private; | |
2971 | ||
2972 | mutex_lock(&dm_thin_pool_table.mutex); | |
2973 | ||
2974 | __pool_dec(tc->pool); | |
2975 | dm_pool_close_thin_device(tc->td); | |
2976 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2977 | if (tc->origin_dev) |
2978 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2979 | kfree(tc); |
2980 | ||
2981 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2982 | } | |
2983 | ||
2984 | /* | |
2985 | * Thin target parameters: | |
2986 | * | |
2dd9c257 | 2987 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2988 | * |
2989 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2990 | * dev_id: the internal device identifier | |
2dd9c257 | 2991 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2992 | * |
2993 | * If the pool device has discards disabled, they get disabled for the thin | |
2994 | * device as well. | |
991d9fa0 JT |
2995 | */ |
2996 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2997 | { | |
2998 | int r; | |
2999 | struct thin_c *tc; | |
2dd9c257 | 3000 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
3001 | struct mapped_device *pool_md; |
3002 | ||
3003 | mutex_lock(&dm_thin_pool_table.mutex); | |
3004 | ||
2dd9c257 | 3005 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3006 | ti->error = "Invalid argument count"; |
3007 | r = -EINVAL; | |
3008 | goto out_unlock; | |
3009 | } | |
3010 | ||
3011 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3012 | if (!tc) { | |
3013 | ti->error = "Out of memory"; | |
3014 | r = -ENOMEM; | |
3015 | goto out_unlock; | |
3016 | } | |
3017 | ||
2dd9c257 JT |
3018 | if (argc == 3) { |
3019 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3020 | if (r) { | |
3021 | ti->error = "Error opening origin device"; | |
3022 | goto bad_origin_dev; | |
3023 | } | |
3024 | tc->origin_dev = origin_dev; | |
3025 | } | |
3026 | ||
991d9fa0 JT |
3027 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3028 | if (r) { | |
3029 | ti->error = "Error opening pool device"; | |
3030 | goto bad_pool_dev; | |
3031 | } | |
3032 | tc->pool_dev = pool_dev; | |
3033 | ||
3034 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3035 | ti->error = "Invalid device id"; | |
3036 | r = -EINVAL; | |
3037 | goto bad_common; | |
3038 | } | |
3039 | ||
3040 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3041 | if (!pool_md) { | |
3042 | ti->error = "Couldn't get pool mapped device"; | |
3043 | r = -EINVAL; | |
3044 | goto bad_common; | |
3045 | } | |
3046 | ||
3047 | tc->pool = __pool_table_lookup(pool_md); | |
3048 | if (!tc->pool) { | |
3049 | ti->error = "Couldn't find pool object"; | |
3050 | r = -EINVAL; | |
3051 | goto bad_pool_lookup; | |
3052 | } | |
3053 | __pool_inc(tc->pool); | |
3054 | ||
e49e5829 JT |
3055 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3056 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3057 | r = -EINVAL; |
e49e5829 JT |
3058 | goto bad_thin_open; |
3059 | } | |
3060 | ||
991d9fa0 JT |
3061 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3062 | if (r) { | |
3063 | ti->error = "Couldn't open thin internal device"; | |
3064 | goto bad_thin_open; | |
3065 | } | |
3066 | ||
542f9038 MS |
3067 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3068 | if (r) | |
1acacc07 | 3069 | goto bad_target_max_io_len; |
542f9038 | 3070 | |
55a62eef | 3071 | ti->num_flush_bios = 1; |
16ad3d10 | 3072 | ti->flush_supported = true; |
59c3d2c6 | 3073 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3074 | |
3075 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3076 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3077 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3078 | ti->discards_supported = true; |
55a62eef | 3079 | ti->num_discard_bios = 1; |
55a62eef AK |
3080 | /* Discard bios must be split on a block boundary */ |
3081 | ti->split_discard_bios = true; | |
67e2e2b2 | 3082 | } |
991d9fa0 JT |
3083 | |
3084 | dm_put(pool_md); | |
3085 | ||
3086 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3087 | ||
3088 | return 0; | |
3089 | ||
1acacc07 MS |
3090 | bad_target_max_io_len: |
3091 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3092 | bad_thin_open: |
3093 | __pool_dec(tc->pool); | |
3094 | bad_pool_lookup: | |
3095 | dm_put(pool_md); | |
3096 | bad_common: | |
3097 | dm_put_device(ti, tc->pool_dev); | |
3098 | bad_pool_dev: | |
2dd9c257 JT |
3099 | if (tc->origin_dev) |
3100 | dm_put_device(ti, tc->origin_dev); | |
3101 | bad_origin_dev: | |
991d9fa0 JT |
3102 | kfree(tc); |
3103 | out_unlock: | |
3104 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3105 | ||
3106 | return r; | |
3107 | } | |
3108 | ||
7de3ee57 | 3109 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3110 | { |
4f024f37 | 3111 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3112 | |
7de3ee57 | 3113 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3114 | } |
3115 | ||
7de3ee57 | 3116 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3117 | { |
3118 | unsigned long flags; | |
59c3d2c6 | 3119 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3120 | struct list_head work; |
a24c2569 | 3121 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3122 | struct pool *pool = h->tc->pool; |
3123 | ||
3124 | if (h->shared_read_entry) { | |
3125 | INIT_LIST_HEAD(&work); | |
44feb387 | 3126 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3127 | |
3128 | spin_lock_irqsave(&pool->lock, flags); | |
3129 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3130 | list_del(&m->list); | |
7f214665 | 3131 | m->quiesced = true; |
eb2aa48d JT |
3132 | __maybe_add_mapping(m); |
3133 | } | |
3134 | spin_unlock_irqrestore(&pool->lock, flags); | |
3135 | } | |
3136 | ||
104655fd JT |
3137 | if (h->all_io_entry) { |
3138 | INIT_LIST_HEAD(&work); | |
44feb387 | 3139 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3140 | if (!list_empty(&work)) { |
3141 | spin_lock_irqsave(&pool->lock, flags); | |
3142 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3143 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3144 | spin_unlock_irqrestore(&pool->lock, flags); |
3145 | wake_worker(pool); | |
3146 | } | |
104655fd JT |
3147 | } |
3148 | ||
eb2aa48d JT |
3149 | return 0; |
3150 | } | |
3151 | ||
738211f7 | 3152 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3153 | { |
738211f7 JT |
3154 | struct thin_c *tc = ti->private; |
3155 | ||
991d9fa0 | 3156 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3157 | noflush_work(tc, do_noflush_start); |
3158 | } | |
3159 | ||
3160 | static void thin_postsuspend(struct dm_target *ti) | |
3161 | { | |
3162 | struct thin_c *tc = ti->private; | |
3163 | ||
3164 | /* | |
3165 | * The dm_noflush_suspending flag has been cleared by now, so | |
3166 | * unfortunately we must always run this. | |
3167 | */ | |
3168 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3169 | } |
3170 | ||
3171 | /* | |
3172 | * <nr mapped sectors> <highest mapped sector> | |
3173 | */ | |
fd7c092e MP |
3174 | static void thin_status(struct dm_target *ti, status_type_t type, |
3175 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3176 | { |
3177 | int r; | |
3178 | ssize_t sz = 0; | |
3179 | dm_block_t mapped, highest; | |
3180 | char buf[BDEVNAME_SIZE]; | |
3181 | struct thin_c *tc = ti->private; | |
3182 | ||
e49e5829 JT |
3183 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3184 | DMEMIT("Fail"); | |
fd7c092e | 3185 | return; |
e49e5829 JT |
3186 | } |
3187 | ||
991d9fa0 JT |
3188 | if (!tc->td) |
3189 | DMEMIT("-"); | |
3190 | else { | |
3191 | switch (type) { | |
3192 | case STATUSTYPE_INFO: | |
3193 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3194 | if (r) { |
3195 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3196 | goto err; | |
3197 | } | |
991d9fa0 JT |
3198 | |
3199 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3200 | if (r < 0) { |
3201 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3202 | goto err; | |
3203 | } | |
991d9fa0 JT |
3204 | |
3205 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3206 | if (r) | |
3207 | DMEMIT("%llu", ((highest + 1) * | |
3208 | tc->pool->sectors_per_block) - 1); | |
3209 | else | |
3210 | DMEMIT("-"); | |
3211 | break; | |
3212 | ||
3213 | case STATUSTYPE_TABLE: | |
3214 | DMEMIT("%s %lu", | |
3215 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3216 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3217 | if (tc->origin_dev) |
3218 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3219 | break; |
3220 | } | |
3221 | } | |
3222 | ||
fd7c092e MP |
3223 | return; |
3224 | ||
3225 | err: | |
3226 | DMEMIT("Error"); | |
991d9fa0 JT |
3227 | } |
3228 | ||
3229 | static int thin_iterate_devices(struct dm_target *ti, | |
3230 | iterate_devices_callout_fn fn, void *data) | |
3231 | { | |
55f2b8bd | 3232 | sector_t blocks; |
991d9fa0 | 3233 | struct thin_c *tc = ti->private; |
55f2b8bd | 3234 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3235 | |
3236 | /* | |
3237 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3238 | * we follow a more convoluted path through to the pool's target. | |
3239 | */ | |
55f2b8bd | 3240 | if (!pool->ti) |
991d9fa0 JT |
3241 | return 0; /* nothing is bound */ |
3242 | ||
55f2b8bd MS |
3243 | blocks = pool->ti->len; |
3244 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3245 | if (blocks) |
55f2b8bd | 3246 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3247 | |
3248 | return 0; | |
3249 | } | |
3250 | ||
991d9fa0 JT |
3251 | static struct target_type thin_target = { |
3252 | .name = "thin", | |
07f2b6e0 | 3253 | .version = {1, 11, 0}, |
991d9fa0 JT |
3254 | .module = THIS_MODULE, |
3255 | .ctr = thin_ctr, | |
3256 | .dtr = thin_dtr, | |
3257 | .map = thin_map, | |
eb2aa48d | 3258 | .end_io = thin_endio, |
738211f7 | 3259 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3260 | .postsuspend = thin_postsuspend, |
3261 | .status = thin_status, | |
3262 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3263 | }; |
3264 | ||
3265 | /*----------------------------------------------------------------*/ | |
3266 | ||
3267 | static int __init dm_thin_init(void) | |
3268 | { | |
3269 | int r; | |
3270 | ||
3271 | pool_table_init(); | |
3272 | ||
3273 | r = dm_register_target(&thin_target); | |
3274 | if (r) | |
3275 | return r; | |
3276 | ||
3277 | r = dm_register_target(&pool_target); | |
3278 | if (r) | |
a24c2569 MS |
3279 | goto bad_pool_target; |
3280 | ||
3281 | r = -ENOMEM; | |
3282 | ||
a24c2569 MS |
3283 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3284 | if (!_new_mapping_cache) | |
3285 | goto bad_new_mapping_cache; | |
3286 | ||
a24c2569 MS |
3287 | return 0; |
3288 | ||
a24c2569 | 3289 | bad_new_mapping_cache: |
a24c2569 MS |
3290 | dm_unregister_target(&pool_target); |
3291 | bad_pool_target: | |
3292 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3293 | |
3294 | return r; | |
3295 | } | |
3296 | ||
3297 | static void dm_thin_exit(void) | |
3298 | { | |
3299 | dm_unregister_target(&thin_target); | |
3300 | dm_unregister_target(&pool_target); | |
a24c2569 | 3301 | |
a24c2569 | 3302 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3303 | } |
3304 | ||
3305 | module_init(dm_thin_init); | |
3306 | module_exit(dm_thin_exit); | |
3307 | ||
7cab8bf1 | 3308 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3309 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3310 | MODULE_LICENSE("GPL"); |