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 JT |
132 | /* |
133 | * The pool runs in 3 modes. Ordered in degraded order for comparisons. | |
134 | */ | |
135 | enum pool_mode { | |
136 | PM_WRITE, /* metadata may be changed */ | |
137 | PM_READ_ONLY, /* metadata may not be changed */ | |
138 | PM_FAIL, /* all I/O fails */ | |
139 | }; | |
140 | ||
67e2e2b2 | 141 | struct pool_features { |
e49e5829 JT |
142 | enum pool_mode mode; |
143 | ||
9bc142dd MS |
144 | bool zero_new_blocks:1; |
145 | bool discard_enabled:1; | |
146 | bool discard_passdown:1; | |
67e2e2b2 JT |
147 | }; |
148 | ||
e49e5829 JT |
149 | struct thin_c; |
150 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
151 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
152 | ||
991d9fa0 JT |
153 | struct pool { |
154 | struct list_head list; | |
155 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
156 | ||
157 | struct mapped_device *pool_md; | |
158 | struct block_device *md_dev; | |
159 | struct dm_pool_metadata *pmd; | |
160 | ||
991d9fa0 | 161 | dm_block_t low_water_blocks; |
55f2b8bd | 162 | uint32_t sectors_per_block; |
f9a8e0cd | 163 | int sectors_per_block_shift; |
991d9fa0 | 164 | |
67e2e2b2 | 165 | struct pool_features pf; |
991d9fa0 JT |
166 | unsigned low_water_triggered:1; /* A dm event has been sent */ |
167 | unsigned no_free_space:1; /* A -ENOSPC warning has been issued */ | |
168 | ||
44feb387 | 169 | struct dm_bio_prison *prison; |
991d9fa0 JT |
170 | struct dm_kcopyd_client *copier; |
171 | ||
172 | struct workqueue_struct *wq; | |
173 | struct work_struct worker; | |
905e51b3 | 174 | struct delayed_work waker; |
991d9fa0 | 175 | |
905e51b3 | 176 | unsigned long last_commit_jiffies; |
55f2b8bd | 177 | unsigned ref_count; |
991d9fa0 JT |
178 | |
179 | spinlock_t lock; | |
180 | struct bio_list deferred_bios; | |
181 | struct bio_list deferred_flush_bios; | |
182 | struct list_head prepared_mappings; | |
104655fd | 183 | struct list_head prepared_discards; |
991d9fa0 JT |
184 | |
185 | struct bio_list retry_on_resume_list; | |
186 | ||
44feb387 MS |
187 | struct dm_deferred_set *shared_read_ds; |
188 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 189 | |
a24c2569 | 190 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 191 | mempool_t *mapping_pool; |
e49e5829 JT |
192 | |
193 | process_bio_fn process_bio; | |
194 | process_bio_fn process_discard; | |
195 | ||
196 | process_mapping_fn process_prepared_mapping; | |
197 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
198 | }; |
199 | ||
e49e5829 JT |
200 | static enum pool_mode get_pool_mode(struct pool *pool); |
201 | static void set_pool_mode(struct pool *pool, enum pool_mode mode); | |
202 | ||
991d9fa0 JT |
203 | /* |
204 | * Target context for a pool. | |
205 | */ | |
206 | struct pool_c { | |
207 | struct dm_target *ti; | |
208 | struct pool *pool; | |
209 | struct dm_dev *data_dev; | |
210 | struct dm_dev *metadata_dev; | |
211 | struct dm_target_callbacks callbacks; | |
212 | ||
213 | dm_block_t low_water_blocks; | |
0424caa1 MS |
214 | struct pool_features requested_pf; /* Features requested during table load */ |
215 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
216 | }; |
217 | ||
218 | /* | |
219 | * Target context for a thin. | |
220 | */ | |
221 | struct thin_c { | |
222 | struct dm_dev *pool_dev; | |
2dd9c257 | 223 | struct dm_dev *origin_dev; |
991d9fa0 JT |
224 | dm_thin_id dev_id; |
225 | ||
226 | struct pool *pool; | |
227 | struct dm_thin_device *td; | |
228 | }; | |
229 | ||
230 | /*----------------------------------------------------------------*/ | |
231 | ||
025b9685 JT |
232 | /* |
233 | * wake_worker() is used when new work is queued and when pool_resume is | |
234 | * ready to continue deferred IO processing. | |
235 | */ | |
236 | static void wake_worker(struct pool *pool) | |
237 | { | |
238 | queue_work(pool->wq, &pool->worker); | |
239 | } | |
240 | ||
241 | /*----------------------------------------------------------------*/ | |
242 | ||
6beca5eb JT |
243 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
244 | struct dm_bio_prison_cell **cell_result) | |
245 | { | |
246 | int r; | |
247 | struct dm_bio_prison_cell *cell_prealloc; | |
248 | ||
249 | /* | |
250 | * Allocate a cell from the prison's mempool. | |
251 | * This might block but it can't fail. | |
252 | */ | |
253 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
254 | ||
255 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
256 | if (r) | |
257 | /* | |
258 | * We reused an old cell; we can get rid of | |
259 | * the new one. | |
260 | */ | |
261 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
262 | ||
263 | return r; | |
264 | } | |
265 | ||
266 | static void cell_release(struct pool *pool, | |
267 | struct dm_bio_prison_cell *cell, | |
268 | struct bio_list *bios) | |
269 | { | |
270 | dm_cell_release(pool->prison, cell, bios); | |
271 | dm_bio_prison_free_cell(pool->prison, cell); | |
272 | } | |
273 | ||
274 | static void cell_release_no_holder(struct pool *pool, | |
275 | struct dm_bio_prison_cell *cell, | |
276 | struct bio_list *bios) | |
277 | { | |
278 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
279 | dm_bio_prison_free_cell(pool->prison, cell); | |
280 | } | |
281 | ||
025b9685 JT |
282 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
283 | struct dm_bio_prison_cell *cell) | |
284 | { | |
285 | struct pool *pool = tc->pool; | |
286 | unsigned long flags; | |
287 | ||
288 | spin_lock_irqsave(&pool->lock, flags); | |
289 | dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios); | |
290 | spin_unlock_irqrestore(&pool->lock, flags); | |
291 | ||
292 | wake_worker(pool); | |
293 | } | |
294 | ||
6beca5eb JT |
295 | static void cell_error(struct pool *pool, |
296 | struct dm_bio_prison_cell *cell) | |
297 | { | |
298 | dm_cell_error(pool->prison, cell); | |
299 | dm_bio_prison_free_cell(pool->prison, cell); | |
300 | } | |
301 | ||
302 | /*----------------------------------------------------------------*/ | |
303 | ||
991d9fa0 JT |
304 | /* |
305 | * A global list of pools that uses a struct mapped_device as a key. | |
306 | */ | |
307 | static struct dm_thin_pool_table { | |
308 | struct mutex mutex; | |
309 | struct list_head pools; | |
310 | } dm_thin_pool_table; | |
311 | ||
312 | static void pool_table_init(void) | |
313 | { | |
314 | mutex_init(&dm_thin_pool_table.mutex); | |
315 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
316 | } | |
317 | ||
318 | static void __pool_table_insert(struct pool *pool) | |
319 | { | |
320 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
321 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
322 | } | |
323 | ||
324 | static void __pool_table_remove(struct pool *pool) | |
325 | { | |
326 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
327 | list_del(&pool->list); | |
328 | } | |
329 | ||
330 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
331 | { | |
332 | struct pool *pool = NULL, *tmp; | |
333 | ||
334 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
335 | ||
336 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
337 | if (tmp->pool_md == md) { | |
338 | pool = tmp; | |
339 | break; | |
340 | } | |
341 | } | |
342 | ||
343 | return pool; | |
344 | } | |
345 | ||
346 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
347 | { | |
348 | struct pool *pool = NULL, *tmp; | |
349 | ||
350 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
351 | ||
352 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
353 | if (tmp->md_dev == md_dev) { | |
354 | pool = tmp; | |
355 | break; | |
356 | } | |
357 | } | |
358 | ||
359 | return pool; | |
360 | } | |
361 | ||
362 | /*----------------------------------------------------------------*/ | |
363 | ||
a24c2569 | 364 | struct dm_thin_endio_hook { |
eb2aa48d | 365 | struct thin_c *tc; |
44feb387 MS |
366 | struct dm_deferred_entry *shared_read_entry; |
367 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 368 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
369 | }; |
370 | ||
991d9fa0 JT |
371 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
372 | { | |
373 | struct bio *bio; | |
374 | struct bio_list bios; | |
375 | ||
376 | bio_list_init(&bios); | |
377 | bio_list_merge(&bios, master); | |
378 | bio_list_init(master); | |
379 | ||
380 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 381 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 382 | |
eb2aa48d | 383 | if (h->tc == tc) |
991d9fa0 JT |
384 | bio_endio(bio, DM_ENDIO_REQUEUE); |
385 | else | |
386 | bio_list_add(master, bio); | |
387 | } | |
388 | } | |
389 | ||
390 | static void requeue_io(struct thin_c *tc) | |
391 | { | |
392 | struct pool *pool = tc->pool; | |
393 | unsigned long flags; | |
394 | ||
395 | spin_lock_irqsave(&pool->lock, flags); | |
396 | __requeue_bio_list(tc, &pool->deferred_bios); | |
397 | __requeue_bio_list(tc, &pool->retry_on_resume_list); | |
398 | spin_unlock_irqrestore(&pool->lock, flags); | |
399 | } | |
400 | ||
401 | /* | |
402 | * This section of code contains the logic for processing a thin device's IO. | |
403 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
404 | * but most is exclusively called from the thin target rather than the thin-pool | |
405 | * target. | |
406 | */ | |
407 | ||
58f77a21 MS |
408 | static bool block_size_is_power_of_two(struct pool *pool) |
409 | { | |
410 | return pool->sectors_per_block_shift >= 0; | |
411 | } | |
412 | ||
991d9fa0 JT |
413 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
414 | { | |
58f77a21 | 415 | struct pool *pool = tc->pool; |
55f2b8bd MS |
416 | sector_t block_nr = bio->bi_sector; |
417 | ||
58f77a21 MS |
418 | if (block_size_is_power_of_two(pool)) |
419 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 420 | else |
58f77a21 | 421 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
422 | |
423 | return block_nr; | |
991d9fa0 JT |
424 | } |
425 | ||
426 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
427 | { | |
428 | struct pool *pool = tc->pool; | |
55f2b8bd | 429 | sector_t bi_sector = bio->bi_sector; |
991d9fa0 JT |
430 | |
431 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 432 | if (block_size_is_power_of_two(pool)) |
f9a8e0cd MP |
433 | bio->bi_sector = (block << pool->sectors_per_block_shift) | |
434 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 MS |
435 | else |
436 | bio->bi_sector = (block * pool->sectors_per_block) + | |
437 | sector_div(bi_sector, pool->sectors_per_block); | |
991d9fa0 JT |
438 | } |
439 | ||
2dd9c257 JT |
440 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
441 | { | |
442 | bio->bi_bdev = tc->origin_dev->bdev; | |
443 | } | |
444 | ||
4afdd680 JT |
445 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
446 | { | |
447 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
448 | dm_thin_changed_this_transaction(tc->td); | |
449 | } | |
450 | ||
e8088073 JT |
451 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
452 | { | |
453 | struct dm_thin_endio_hook *h; | |
454 | ||
455 | if (bio->bi_rw & REQ_DISCARD) | |
456 | return; | |
457 | ||
59c3d2c6 | 458 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
459 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
460 | } | |
461 | ||
2dd9c257 | 462 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
463 | { |
464 | struct pool *pool = tc->pool; | |
465 | unsigned long flags; | |
466 | ||
e49e5829 JT |
467 | if (!bio_triggers_commit(tc, bio)) { |
468 | generic_make_request(bio); | |
469 | return; | |
470 | } | |
471 | ||
991d9fa0 | 472 | /* |
e49e5829 JT |
473 | * Complete bio with an error if earlier I/O caused changes to |
474 | * the metadata that can't be committed e.g, due to I/O errors | |
475 | * on the metadata device. | |
991d9fa0 | 476 | */ |
e49e5829 JT |
477 | if (dm_thin_aborted_changes(tc->td)) { |
478 | bio_io_error(bio); | |
479 | return; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Batch together any bios that trigger commits and then issue a | |
484 | * single commit for them in process_deferred_bios(). | |
485 | */ | |
486 | spin_lock_irqsave(&pool->lock, flags); | |
487 | bio_list_add(&pool->deferred_flush_bios, bio); | |
488 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
489 | } |
490 | ||
2dd9c257 JT |
491 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
492 | { | |
493 | remap_to_origin(tc, bio); | |
494 | issue(tc, bio); | |
495 | } | |
496 | ||
497 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
498 | dm_block_t block) | |
499 | { | |
500 | remap(tc, bio, block); | |
501 | issue(tc, bio); | |
502 | } | |
503 | ||
991d9fa0 JT |
504 | /*----------------------------------------------------------------*/ |
505 | ||
506 | /* | |
507 | * Bio endio functions. | |
508 | */ | |
a24c2569 | 509 | struct dm_thin_new_mapping { |
991d9fa0 JT |
510 | struct list_head list; |
511 | ||
7f214665 MS |
512 | bool quiesced:1; |
513 | bool prepared:1; | |
514 | bool pass_discard:1; | |
515 | bool definitely_not_shared:1; | |
991d9fa0 | 516 | |
7f214665 | 517 | int err; |
991d9fa0 JT |
518 | struct thin_c *tc; |
519 | dm_block_t virt_block; | |
520 | dm_block_t data_block; | |
a24c2569 | 521 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
522 | |
523 | /* | |
524 | * If the bio covers the whole area of a block then we can avoid | |
525 | * zeroing or copying. Instead this bio is hooked. The bio will | |
526 | * still be in the cell, so care has to be taken to avoid issuing | |
527 | * the bio twice. | |
528 | */ | |
529 | struct bio *bio; | |
530 | bio_end_io_t *saved_bi_end_io; | |
531 | }; | |
532 | ||
a24c2569 | 533 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
534 | { |
535 | struct pool *pool = m->tc->pool; | |
536 | ||
eb2aa48d | 537 | if (m->quiesced && m->prepared) { |
daec338b | 538 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
539 | wake_worker(pool); |
540 | } | |
541 | } | |
542 | ||
543 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
544 | { | |
545 | unsigned long flags; | |
a24c2569 | 546 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
547 | struct pool *pool = m->tc->pool; |
548 | ||
549 | m->err = read_err || write_err ? -EIO : 0; | |
550 | ||
551 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 552 | m->prepared = true; |
991d9fa0 JT |
553 | __maybe_add_mapping(m); |
554 | spin_unlock_irqrestore(&pool->lock, flags); | |
555 | } | |
556 | ||
557 | static void overwrite_endio(struct bio *bio, int err) | |
558 | { | |
559 | unsigned long flags; | |
59c3d2c6 | 560 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 561 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
562 | struct pool *pool = m->tc->pool; |
563 | ||
564 | m->err = err; | |
565 | ||
566 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 567 | m->prepared = true; |
991d9fa0 JT |
568 | __maybe_add_mapping(m); |
569 | spin_unlock_irqrestore(&pool->lock, flags); | |
570 | } | |
571 | ||
991d9fa0 JT |
572 | /*----------------------------------------------------------------*/ |
573 | ||
574 | /* | |
575 | * Workqueue. | |
576 | */ | |
577 | ||
578 | /* | |
579 | * Prepared mapping jobs. | |
580 | */ | |
581 | ||
582 | /* | |
583 | * This sends the bios in the cell back to the deferred_bios list. | |
584 | */ | |
2aab3850 | 585 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
586 | { |
587 | struct pool *pool = tc->pool; | |
588 | unsigned long flags; | |
589 | ||
590 | spin_lock_irqsave(&pool->lock, flags); | |
6beca5eb | 591 | cell_release(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
592 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
593 | ||
594 | wake_worker(pool); | |
595 | } | |
596 | ||
597 | /* | |
6beca5eb | 598 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 599 | */ |
f286ba0e | 600 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 601 | { |
991d9fa0 JT |
602 | struct pool *pool = tc->pool; |
603 | unsigned long flags; | |
604 | ||
991d9fa0 | 605 | spin_lock_irqsave(&pool->lock, flags); |
6beca5eb | 606 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
607 | spin_unlock_irqrestore(&pool->lock, flags); |
608 | ||
609 | wake_worker(pool); | |
610 | } | |
611 | ||
e49e5829 JT |
612 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
613 | { | |
614 | if (m->bio) | |
615 | m->bio->bi_end_io = m->saved_bi_end_io; | |
6beca5eb | 616 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
617 | list_del(&m->list); |
618 | mempool_free(m, m->tc->pool->mapping_pool); | |
619 | } | |
025b9685 | 620 | |
a24c2569 | 621 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
622 | { |
623 | struct thin_c *tc = m->tc; | |
6beca5eb | 624 | struct pool *pool = tc->pool; |
991d9fa0 JT |
625 | struct bio *bio; |
626 | int r; | |
627 | ||
628 | bio = m->bio; | |
629 | if (bio) | |
630 | bio->bi_end_io = m->saved_bi_end_io; | |
631 | ||
632 | if (m->err) { | |
6beca5eb | 633 | cell_error(pool, m->cell); |
905386f8 | 634 | goto out; |
991d9fa0 JT |
635 | } |
636 | ||
637 | /* | |
638 | * Commit the prepared block into the mapping btree. | |
639 | * Any I/O for this block arriving after this point will get | |
640 | * remapped to it directly. | |
641 | */ | |
642 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
643 | if (r) { | |
fafc7a81 JT |
644 | DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d", |
645 | dm_device_name(pool->pool_md), r); | |
646 | set_pool_mode(pool, PM_READ_ONLY); | |
6beca5eb | 647 | cell_error(pool, m->cell); |
905386f8 | 648 | goto out; |
991d9fa0 JT |
649 | } |
650 | ||
651 | /* | |
652 | * Release any bios held while the block was being provisioned. | |
653 | * If we are processing a write bio that completely covers the block, | |
654 | * we already processed it so can ignore it now when processing | |
655 | * the bios in the cell. | |
656 | */ | |
657 | if (bio) { | |
f286ba0e | 658 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
659 | bio_endio(bio, 0); |
660 | } else | |
2aab3850 | 661 | cell_defer(tc, m->cell); |
991d9fa0 | 662 | |
905386f8 | 663 | out: |
991d9fa0 | 664 | list_del(&m->list); |
6beca5eb | 665 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
666 | } |
667 | ||
e49e5829 | 668 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 669 | { |
104655fd JT |
670 | struct thin_c *tc = m->tc; |
671 | ||
e49e5829 | 672 | bio_io_error(m->bio); |
f286ba0e JT |
673 | cell_defer_no_holder(tc, m->cell); |
674 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
675 | mempool_free(m, tc->pool->mapping_pool); |
676 | } | |
677 | ||
678 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
679 | { | |
680 | struct thin_c *tc = m->tc; | |
104655fd | 681 | |
e8088073 | 682 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
683 | cell_defer_no_holder(tc, m->cell); |
684 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 685 | |
104655fd | 686 | if (m->pass_discard) |
19fa1a67 JT |
687 | if (m->definitely_not_shared) |
688 | remap_and_issue(tc, m->bio, m->data_block); | |
689 | else { | |
690 | bool used = false; | |
691 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
692 | bio_endio(m->bio, 0); | |
693 | else | |
694 | remap_and_issue(tc, m->bio, m->data_block); | |
695 | } | |
104655fd JT |
696 | else |
697 | bio_endio(m->bio, 0); | |
698 | ||
104655fd JT |
699 | mempool_free(m, tc->pool->mapping_pool); |
700 | } | |
701 | ||
e49e5829 JT |
702 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
703 | { | |
704 | int r; | |
705 | struct thin_c *tc = m->tc; | |
706 | ||
707 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
708 | if (r) | |
c397741c | 709 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
710 | |
711 | process_prepared_discard_passdown(m); | |
712 | } | |
713 | ||
104655fd | 714 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 715 | process_mapping_fn *fn) |
991d9fa0 JT |
716 | { |
717 | unsigned long flags; | |
718 | struct list_head maps; | |
a24c2569 | 719 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
720 | |
721 | INIT_LIST_HEAD(&maps); | |
722 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 723 | list_splice_init(head, &maps); |
991d9fa0 JT |
724 | spin_unlock_irqrestore(&pool->lock, flags); |
725 | ||
726 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 727 | (*fn)(m); |
991d9fa0 JT |
728 | } |
729 | ||
730 | /* | |
731 | * Deferred bio jobs. | |
732 | */ | |
104655fd | 733 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 734 | { |
f9a8e0cd | 735 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
736 | } |
737 | ||
738 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
739 | { | |
740 | return (bio_data_dir(bio) == WRITE) && | |
741 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
742 | } |
743 | ||
744 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
745 | bio_end_io_t *fn) | |
746 | { | |
747 | *save = bio->bi_end_io; | |
748 | bio->bi_end_io = fn; | |
749 | } | |
750 | ||
751 | static int ensure_next_mapping(struct pool *pool) | |
752 | { | |
753 | if (pool->next_mapping) | |
754 | return 0; | |
755 | ||
756 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
757 | ||
758 | return pool->next_mapping ? 0 : -ENOMEM; | |
759 | } | |
760 | ||
a24c2569 | 761 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 762 | { |
16961b04 | 763 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
764 | |
765 | BUG_ON(!pool->next_mapping); | |
766 | ||
16961b04 MS |
767 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
768 | INIT_LIST_HEAD(&m->list); | |
769 | m->bio = NULL; | |
770 | ||
991d9fa0 JT |
771 | pool->next_mapping = NULL; |
772 | ||
16961b04 | 773 | return m; |
991d9fa0 JT |
774 | } |
775 | ||
776 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
777 | struct dm_dev *origin, dm_block_t data_origin, |
778 | dm_block_t data_dest, | |
a24c2569 | 779 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
780 | { |
781 | int r; | |
782 | struct pool *pool = tc->pool; | |
a24c2569 | 783 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 784 | |
991d9fa0 JT |
785 | m->tc = tc; |
786 | m->virt_block = virt_block; | |
787 | m->data_block = data_dest; | |
788 | m->cell = cell; | |
991d9fa0 | 789 | |
44feb387 | 790 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 791 | m->quiesced = true; |
991d9fa0 JT |
792 | |
793 | /* | |
794 | * IO to pool_dev remaps to the pool target's data_dev. | |
795 | * | |
796 | * If the whole block of data is being overwritten, we can issue the | |
797 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
798 | */ | |
799 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 800 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 801 | |
eb2aa48d | 802 | h->overwrite_mapping = m; |
991d9fa0 JT |
803 | m->bio = bio; |
804 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 805 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
806 | remap_and_issue(tc, bio, data_dest); |
807 | } else { | |
808 | struct dm_io_region from, to; | |
809 | ||
2dd9c257 | 810 | from.bdev = origin->bdev; |
991d9fa0 JT |
811 | from.sector = data_origin * pool->sectors_per_block; |
812 | from.count = pool->sectors_per_block; | |
813 | ||
814 | to.bdev = tc->pool_dev->bdev; | |
815 | to.sector = data_dest * pool->sectors_per_block; | |
816 | to.count = pool->sectors_per_block; | |
817 | ||
818 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
819 | 0, copy_complete, m); | |
820 | if (r < 0) { | |
821 | mempool_free(m, pool->mapping_pool); | |
c397741c | 822 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 823 | cell_error(pool, cell); |
991d9fa0 JT |
824 | } |
825 | } | |
826 | } | |
827 | ||
2dd9c257 JT |
828 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
829 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 830 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
831 | { |
832 | schedule_copy(tc, virt_block, tc->pool_dev, | |
833 | data_origin, data_dest, cell, bio); | |
834 | } | |
835 | ||
836 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
837 | dm_block_t data_dest, | |
a24c2569 | 838 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
839 | { |
840 | schedule_copy(tc, virt_block, tc->origin_dev, | |
841 | virt_block, data_dest, cell, bio); | |
842 | } | |
843 | ||
991d9fa0 | 844 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 845 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
846 | struct bio *bio) |
847 | { | |
848 | struct pool *pool = tc->pool; | |
a24c2569 | 849 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 850 | |
7f214665 MS |
851 | m->quiesced = true; |
852 | m->prepared = false; | |
991d9fa0 JT |
853 | m->tc = tc; |
854 | m->virt_block = virt_block; | |
855 | m->data_block = data_block; | |
856 | m->cell = cell; | |
991d9fa0 JT |
857 | |
858 | /* | |
859 | * If the whole block of data is being overwritten or we are not | |
860 | * zeroing pre-existing data, we can issue the bio immediately. | |
861 | * Otherwise we use kcopyd to zero the data first. | |
862 | */ | |
67e2e2b2 | 863 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
864 | process_prepared_mapping(m); |
865 | ||
866 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 867 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 868 | |
eb2aa48d | 869 | h->overwrite_mapping = m; |
991d9fa0 JT |
870 | m->bio = bio; |
871 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 872 | inc_all_io_entry(pool, bio); |
991d9fa0 | 873 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
874 | } else { |
875 | int r; | |
876 | struct dm_io_region to; | |
877 | ||
878 | to.bdev = tc->pool_dev->bdev; | |
879 | to.sector = data_block * pool->sectors_per_block; | |
880 | to.count = pool->sectors_per_block; | |
881 | ||
882 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
883 | if (r < 0) { | |
884 | mempool_free(m, pool->mapping_pool); | |
c397741c | 885 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 886 | cell_error(pool, cell); |
991d9fa0 JT |
887 | } |
888 | } | |
889 | } | |
890 | ||
e49e5829 JT |
891 | /* |
892 | * A non-zero return indicates read_only or fail_io mode. | |
893 | * Many callers don't care about the return value. | |
894 | */ | |
020cc3b5 | 895 | static int commit(struct pool *pool) |
e49e5829 JT |
896 | { |
897 | int r; | |
898 | ||
899 | if (get_pool_mode(pool) != PM_WRITE) | |
900 | return -EINVAL; | |
901 | ||
020cc3b5 JT |
902 | r = dm_pool_commit_metadata(pool->pmd); |
903 | if (r) { | |
904 | DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d", | |
905 | dm_device_name(pool->pool_md), r); | |
e49e5829 | 906 | set_pool_mode(pool, PM_READ_ONLY); |
020cc3b5 | 907 | } |
e49e5829 JT |
908 | |
909 | return r; | |
910 | } | |
911 | ||
991d9fa0 JT |
912 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
913 | { | |
914 | int r; | |
915 | dm_block_t free_blocks; | |
916 | unsigned long flags; | |
917 | struct pool *pool = tc->pool; | |
918 | ||
94563bad MS |
919 | /* |
920 | * Once no_free_space is set we must not allow allocation to succeed. | |
921 | * Otherwise it is difficult to explain, debug, test and support. | |
922 | */ | |
923 | if (pool->no_free_space) | |
924 | return -ENOSPC; | |
925 | ||
8d30abff JT |
926 | if (get_pool_mode(pool) != PM_WRITE) |
927 | return -EINVAL; | |
928 | ||
991d9fa0 JT |
929 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
930 | if (r) | |
931 | return r; | |
932 | ||
933 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
b17446df | 934 | DMWARN("%s: reached low water mark for data device: sending event.", |
991d9fa0 JT |
935 | dm_device_name(pool->pool_md)); |
936 | spin_lock_irqsave(&pool->lock, flags); | |
937 | pool->low_water_triggered = 1; | |
938 | spin_unlock_irqrestore(&pool->lock, flags); | |
939 | dm_table_event(pool->ti->table); | |
940 | } | |
941 | ||
942 | if (!free_blocks) { | |
94563bad MS |
943 | /* |
944 | * Try to commit to see if that will free up some | |
945 | * more space. | |
946 | */ | |
020cc3b5 JT |
947 | r = commit(pool); |
948 | if (r) | |
949 | return r; | |
991d9fa0 | 950 | |
94563bad MS |
951 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
952 | if (r) | |
953 | return r; | |
991d9fa0 | 954 | |
94563bad MS |
955 | /* |
956 | * If we still have no space we set a flag to avoid | |
957 | * doing all this checking and return -ENOSPC. This | |
958 | * flag serves as a latch that disallows allocations from | |
959 | * this pool until the admin takes action (e.g. resize or | |
960 | * table reload). | |
961 | */ | |
962 | if (!free_blocks) { | |
4a02b34e | 963 | DMWARN("%s: no free data space available.", |
94563bad MS |
964 | dm_device_name(pool->pool_md)); |
965 | spin_lock_irqsave(&pool->lock, flags); | |
966 | pool->no_free_space = 1; | |
967 | spin_unlock_irqrestore(&pool->lock, flags); | |
968 | return -ENOSPC; | |
991d9fa0 JT |
969 | } |
970 | } | |
971 | ||
972 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e MS |
973 | if (r) { |
974 | if (r == -ENOSPC && | |
975 | !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) && | |
976 | !free_blocks) { | |
977 | DMWARN("%s: no free metadata space available.", | |
978 | dm_device_name(pool->pool_md)); | |
979 | set_pool_mode(pool, PM_READ_ONLY); | |
980 | } | |
991d9fa0 | 981 | return r; |
4a02b34e | 982 | } |
991d9fa0 JT |
983 | |
984 | return 0; | |
985 | } | |
986 | ||
987 | /* | |
988 | * If we have run out of space, queue bios until the device is | |
989 | * resumed, presumably after having been reloaded with more space. | |
990 | */ | |
991 | static void retry_on_resume(struct bio *bio) | |
992 | { | |
59c3d2c6 | 993 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 994 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
995 | struct pool *pool = tc->pool; |
996 | unsigned long flags; | |
997 | ||
998 | spin_lock_irqsave(&pool->lock, flags); | |
999 | bio_list_add(&pool->retry_on_resume_list, bio); | |
1000 | spin_unlock_irqrestore(&pool->lock, flags); | |
1001 | } | |
1002 | ||
6beca5eb | 1003 | static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1004 | { |
1005 | struct bio *bio; | |
1006 | struct bio_list bios; | |
1007 | ||
1008 | bio_list_init(&bios); | |
6beca5eb | 1009 | cell_release(pool, cell, &bios); |
991d9fa0 JT |
1010 | |
1011 | while ((bio = bio_list_pop(&bios))) | |
1012 | retry_on_resume(bio); | |
1013 | } | |
1014 | ||
104655fd JT |
1015 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1016 | { | |
1017 | int r; | |
c3a0ce2e | 1018 | unsigned long flags; |
104655fd | 1019 | struct pool *pool = tc->pool; |
a24c2569 | 1020 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1021 | struct dm_cell_key key, key2; |
104655fd JT |
1022 | dm_block_t block = get_bio_block(tc, bio); |
1023 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1024 | struct dm_thin_new_mapping *m; |
104655fd JT |
1025 | |
1026 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1027 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1028 | return; |
1029 | ||
1030 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1031 | switch (r) { | |
1032 | case 0: | |
1033 | /* | |
1034 | * Check nobody is fiddling with this pool block. This can | |
1035 | * happen if someone's in the process of breaking sharing | |
1036 | * on this block. | |
1037 | */ | |
1038 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1039 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1040 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1041 | break; |
1042 | } | |
1043 | ||
1044 | if (io_overlaps_block(pool, bio)) { | |
1045 | /* | |
1046 | * IO may still be going to the destination block. We must | |
1047 | * quiesce before we can do the removal. | |
1048 | */ | |
1049 | m = get_next_mapping(pool); | |
1050 | m->tc = tc; | |
19fa1a67 JT |
1051 | m->pass_discard = pool->pf.discard_passdown; |
1052 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1053 | m->virt_block = block; |
1054 | m->data_block = lookup_result.block; | |
1055 | m->cell = cell; | |
1056 | m->cell2 = cell2; | |
104655fd JT |
1057 | m->bio = bio; |
1058 | ||
44feb387 | 1059 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1060 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1061 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1062 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1063 | wake_worker(pool); |
1064 | } | |
1065 | } else { | |
e8088073 | 1066 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1067 | cell_defer_no_holder(tc, cell); |
1068 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1069 | |
104655fd | 1070 | /* |
49296309 MP |
1071 | * The DM core makes sure that the discard doesn't span |
1072 | * a block boundary. So we submit the discard of a | |
1073 | * partial block appropriately. | |
104655fd | 1074 | */ |
650d2a06 MP |
1075 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1076 | remap_and_issue(tc, bio, lookup_result.block); | |
1077 | else | |
1078 | bio_endio(bio, 0); | |
104655fd JT |
1079 | } |
1080 | break; | |
1081 | ||
1082 | case -ENODATA: | |
1083 | /* | |
1084 | * It isn't provisioned, just forget it. | |
1085 | */ | |
f286ba0e | 1086 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1087 | bio_endio(bio, 0); |
1088 | break; | |
1089 | ||
1090 | default: | |
c397741c MS |
1091 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1092 | __func__, r); | |
f286ba0e | 1093 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1094 | bio_io_error(bio); |
1095 | break; | |
1096 | } | |
1097 | } | |
1098 | ||
991d9fa0 | 1099 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1100 | struct dm_cell_key *key, |
991d9fa0 | 1101 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1102 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1103 | { |
1104 | int r; | |
1105 | dm_block_t data_block; | |
d6fc2042 | 1106 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1107 | |
1108 | r = alloc_data_block(tc, &data_block); | |
1109 | switch (r) { | |
1110 | case 0: | |
2dd9c257 JT |
1111 | schedule_internal_copy(tc, block, lookup_result->block, |
1112 | data_block, cell, bio); | |
991d9fa0 JT |
1113 | break; |
1114 | ||
1115 | case -ENOSPC: | |
d6fc2042 | 1116 | no_space(pool, cell); |
991d9fa0 JT |
1117 | break; |
1118 | ||
1119 | default: | |
c397741c MS |
1120 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1121 | __func__, r); | |
d6fc2042 MS |
1122 | set_pool_mode(pool, PM_READ_ONLY); |
1123 | cell_error(pool, cell); | |
991d9fa0 JT |
1124 | break; |
1125 | } | |
1126 | } | |
1127 | ||
1128 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1129 | dm_block_t block, | |
1130 | struct dm_thin_lookup_result *lookup_result) | |
1131 | { | |
a24c2569 | 1132 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1133 | struct pool *pool = tc->pool; |
44feb387 | 1134 | struct dm_cell_key key; |
991d9fa0 JT |
1135 | |
1136 | /* | |
1137 | * If cell is already occupied, then sharing is already in the process | |
1138 | * of being broken so we have nothing further to do here. | |
1139 | */ | |
1140 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1141 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1142 | return; |
1143 | ||
60049701 | 1144 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1145 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1146 | else { | |
59c3d2c6 | 1147 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1148 | |
44feb387 | 1149 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1150 | inc_all_io_entry(pool, bio); |
f286ba0e | 1151 | cell_defer_no_holder(tc, cell); |
e8088073 | 1152 | |
991d9fa0 JT |
1153 | remap_and_issue(tc, bio, lookup_result->block); |
1154 | } | |
1155 | } | |
1156 | ||
1157 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1158 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1159 | { |
1160 | int r; | |
1161 | dm_block_t data_block; | |
6beca5eb | 1162 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1163 | |
1164 | /* | |
1165 | * Remap empty bios (flushes) immediately, without provisioning. | |
1166 | */ | |
1167 | if (!bio->bi_size) { | |
6beca5eb | 1168 | inc_all_io_entry(pool, bio); |
f286ba0e | 1169 | cell_defer_no_holder(tc, cell); |
e8088073 | 1170 | |
991d9fa0 JT |
1171 | remap_and_issue(tc, bio, 0); |
1172 | return; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Fill read bios with zeroes and complete them immediately. | |
1177 | */ | |
1178 | if (bio_data_dir(bio) == READ) { | |
1179 | zero_fill_bio(bio); | |
f286ba0e | 1180 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1181 | bio_endio(bio, 0); |
1182 | return; | |
1183 | } | |
1184 | ||
1185 | r = alloc_data_block(tc, &data_block); | |
1186 | switch (r) { | |
1187 | case 0: | |
2dd9c257 JT |
1188 | if (tc->origin_dev) |
1189 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1190 | else | |
1191 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1192 | break; |
1193 | ||
1194 | case -ENOSPC: | |
6beca5eb | 1195 | no_space(pool, cell); |
991d9fa0 JT |
1196 | break; |
1197 | ||
1198 | default: | |
c397741c MS |
1199 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1200 | __func__, r); | |
6beca5eb JT |
1201 | set_pool_mode(pool, PM_READ_ONLY); |
1202 | cell_error(pool, cell); | |
991d9fa0 JT |
1203 | break; |
1204 | } | |
1205 | } | |
1206 | ||
1207 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1208 | { | |
1209 | int r; | |
6beca5eb | 1210 | struct pool *pool = tc->pool; |
991d9fa0 | 1211 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1212 | struct dm_bio_prison_cell *cell; |
44feb387 | 1213 | struct dm_cell_key key; |
991d9fa0 JT |
1214 | struct dm_thin_lookup_result lookup_result; |
1215 | ||
1216 | /* | |
1217 | * If cell is already occupied, then the block is already | |
1218 | * being provisioned so we have nothing further to do here. | |
1219 | */ | |
1220 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1221 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1222 | return; |
1223 | ||
1224 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1225 | switch (r) { | |
1226 | case 0: | |
e8088073 | 1227 | if (lookup_result.shared) { |
991d9fa0 | 1228 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1229 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1230 | } else { |
6beca5eb | 1231 | inc_all_io_entry(pool, bio); |
f286ba0e | 1232 | cell_defer_no_holder(tc, cell); |
e8088073 | 1233 | |
991d9fa0 | 1234 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1235 | } |
991d9fa0 JT |
1236 | break; |
1237 | ||
1238 | case -ENODATA: | |
2dd9c257 | 1239 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1240 | inc_all_io_entry(pool, bio); |
f286ba0e | 1241 | cell_defer_no_holder(tc, cell); |
e8088073 | 1242 | |
2dd9c257 JT |
1243 | remap_to_origin_and_issue(tc, bio); |
1244 | } else | |
1245 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1246 | break; |
1247 | ||
1248 | default: | |
c397741c MS |
1249 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1250 | __func__, r); | |
f286ba0e | 1251 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1252 | bio_io_error(bio); |
1253 | break; | |
1254 | } | |
1255 | } | |
1256 | ||
e49e5829 JT |
1257 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1258 | { | |
1259 | int r; | |
1260 | int rw = bio_data_dir(bio); | |
1261 | dm_block_t block = get_bio_block(tc, bio); | |
1262 | struct dm_thin_lookup_result lookup_result; | |
1263 | ||
1264 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1265 | switch (r) { | |
1266 | case 0: | |
1267 | if (lookup_result.shared && (rw == WRITE) && bio->bi_size) | |
1268 | bio_io_error(bio); | |
e8088073 JT |
1269 | else { |
1270 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1271 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1272 | } |
e49e5829 JT |
1273 | break; |
1274 | ||
1275 | case -ENODATA: | |
1276 | if (rw != READ) { | |
1277 | bio_io_error(bio); | |
1278 | break; | |
1279 | } | |
1280 | ||
1281 | if (tc->origin_dev) { | |
e8088073 | 1282 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1283 | remap_to_origin_and_issue(tc, bio); |
1284 | break; | |
1285 | } | |
1286 | ||
1287 | zero_fill_bio(bio); | |
1288 | bio_endio(bio, 0); | |
1289 | break; | |
1290 | ||
1291 | default: | |
c397741c MS |
1292 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1293 | __func__, r); | |
e49e5829 JT |
1294 | bio_io_error(bio); |
1295 | break; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) | |
1300 | { | |
1301 | bio_io_error(bio); | |
1302 | } | |
1303 | ||
ac8c3f3d JT |
1304 | /* |
1305 | * FIXME: should we also commit due to size of transaction, measured in | |
1306 | * metadata blocks? | |
1307 | */ | |
905e51b3 JT |
1308 | static int need_commit_due_to_time(struct pool *pool) |
1309 | { | |
1310 | return jiffies < pool->last_commit_jiffies || | |
1311 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1312 | } | |
1313 | ||
991d9fa0 JT |
1314 | static void process_deferred_bios(struct pool *pool) |
1315 | { | |
1316 | unsigned long flags; | |
1317 | struct bio *bio; | |
1318 | struct bio_list bios; | |
991d9fa0 JT |
1319 | |
1320 | bio_list_init(&bios); | |
1321 | ||
1322 | spin_lock_irqsave(&pool->lock, flags); | |
1323 | bio_list_merge(&bios, &pool->deferred_bios); | |
1324 | bio_list_init(&pool->deferred_bios); | |
1325 | spin_unlock_irqrestore(&pool->lock, flags); | |
1326 | ||
1327 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 1328 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1329 | struct thin_c *tc = h->tc; |
1330 | ||
991d9fa0 JT |
1331 | /* |
1332 | * If we've got no free new_mapping structs, and processing | |
1333 | * this bio might require one, we pause until there are some | |
1334 | * prepared mappings to process. | |
1335 | */ | |
1336 | if (ensure_next_mapping(pool)) { | |
1337 | spin_lock_irqsave(&pool->lock, flags); | |
1338 | bio_list_merge(&pool->deferred_bios, &bios); | |
1339 | spin_unlock_irqrestore(&pool->lock, flags); | |
1340 | ||
1341 | break; | |
1342 | } | |
104655fd JT |
1343 | |
1344 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1345 | pool->process_discard(tc, bio); |
104655fd | 1346 | else |
e49e5829 | 1347 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1348 | } |
1349 | ||
1350 | /* | |
1351 | * If there are any deferred flush bios, we must commit | |
1352 | * the metadata before issuing them. | |
1353 | */ | |
1354 | bio_list_init(&bios); | |
1355 | spin_lock_irqsave(&pool->lock, flags); | |
1356 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1357 | bio_list_init(&pool->deferred_flush_bios); | |
1358 | spin_unlock_irqrestore(&pool->lock, flags); | |
1359 | ||
905e51b3 | 1360 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1361 | return; |
1362 | ||
020cc3b5 | 1363 | if (commit(pool)) { |
991d9fa0 JT |
1364 | while ((bio = bio_list_pop(&bios))) |
1365 | bio_io_error(bio); | |
1366 | return; | |
1367 | } | |
905e51b3 | 1368 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1369 | |
1370 | while ((bio = bio_list_pop(&bios))) | |
1371 | generic_make_request(bio); | |
1372 | } | |
1373 | ||
1374 | static void do_worker(struct work_struct *ws) | |
1375 | { | |
1376 | struct pool *pool = container_of(ws, struct pool, worker); | |
1377 | ||
e49e5829 JT |
1378 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1379 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1380 | process_deferred_bios(pool); |
1381 | } | |
1382 | ||
905e51b3 JT |
1383 | /* |
1384 | * We want to commit periodically so that not too much | |
1385 | * unwritten data builds up. | |
1386 | */ | |
1387 | static void do_waker(struct work_struct *ws) | |
1388 | { | |
1389 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1390 | wake_worker(pool); | |
1391 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1392 | } | |
1393 | ||
991d9fa0 JT |
1394 | /*----------------------------------------------------------------*/ |
1395 | ||
e49e5829 JT |
1396 | static enum pool_mode get_pool_mode(struct pool *pool) |
1397 | { | |
1398 | return pool->pf.mode; | |
1399 | } | |
1400 | ||
1401 | static void set_pool_mode(struct pool *pool, enum pool_mode mode) | |
1402 | { | |
1403 | int r; | |
1404 | ||
1405 | pool->pf.mode = mode; | |
1406 | ||
1407 | switch (mode) { | |
1408 | case PM_FAIL: | |
4fa5971a MS |
1409 | DMERR("%s: switching pool to failure mode", |
1410 | dm_device_name(pool->pool_md)); | |
5383ef3a | 1411 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1412 | pool->process_bio = process_bio_fail; |
1413 | pool->process_discard = process_bio_fail; | |
1414 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1415 | pool->process_prepared_discard = process_prepared_discard_fail; | |
1416 | break; | |
1417 | ||
1418 | case PM_READ_ONLY: | |
4fa5971a MS |
1419 | DMERR("%s: switching pool to read-only mode", |
1420 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1421 | r = dm_pool_abort_metadata(pool->pmd); |
1422 | if (r) { | |
4fa5971a MS |
1423 | DMERR("%s: aborting transaction failed", |
1424 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1425 | set_pool_mode(pool, PM_FAIL); |
1426 | } else { | |
1427 | dm_pool_metadata_read_only(pool->pmd); | |
1428 | pool->process_bio = process_bio_read_only; | |
1429 | pool->process_discard = process_discard; | |
1430 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1431 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1432 | } | |
1433 | break; | |
1434 | ||
1435 | case PM_WRITE: | |
9b7aaa64 | 1436 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1437 | pool->process_bio = process_bio; |
1438 | pool->process_discard = process_discard; | |
1439 | pool->process_prepared_mapping = process_prepared_mapping; | |
1440 | pool->process_prepared_discard = process_prepared_discard; | |
1441 | break; | |
1442 | } | |
1443 | } | |
1444 | ||
1445 | /*----------------------------------------------------------------*/ | |
1446 | ||
991d9fa0 JT |
1447 | /* |
1448 | * Mapping functions. | |
1449 | */ | |
1450 | ||
1451 | /* | |
1452 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1453 | */ | |
1454 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1455 | { | |
1456 | unsigned long flags; | |
1457 | struct pool *pool = tc->pool; | |
1458 | ||
1459 | spin_lock_irqsave(&pool->lock, flags); | |
1460 | bio_list_add(&pool->deferred_bios, bio); | |
1461 | spin_unlock_irqrestore(&pool->lock, flags); | |
1462 | ||
1463 | wake_worker(pool); | |
1464 | } | |
1465 | ||
59c3d2c6 | 1466 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1467 | { |
59c3d2c6 | 1468 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1469 | |
1470 | h->tc = tc; | |
1471 | h->shared_read_entry = NULL; | |
e8088073 | 1472 | h->all_io_entry = NULL; |
eb2aa48d | 1473 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1474 | } |
1475 | ||
991d9fa0 JT |
1476 | /* |
1477 | * Non-blocking function called from the thin target's map function. | |
1478 | */ | |
7de3ee57 | 1479 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1480 | { |
1481 | int r; | |
1482 | struct thin_c *tc = ti->private; | |
1483 | dm_block_t block = get_bio_block(tc, bio); | |
1484 | struct dm_thin_device *td = tc->td; | |
1485 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1486 | struct dm_bio_prison_cell cell1, cell2; |
1487 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1488 | struct dm_cell_key key; |
991d9fa0 | 1489 | |
59c3d2c6 | 1490 | thin_hook_bio(tc, bio); |
e49e5829 JT |
1491 | |
1492 | if (get_pool_mode(tc->pool) == PM_FAIL) { | |
1493 | bio_io_error(bio); | |
1494 | return DM_MAPIO_SUBMITTED; | |
1495 | } | |
1496 | ||
104655fd | 1497 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1498 | thin_defer_bio(tc, bio); |
1499 | return DM_MAPIO_SUBMITTED; | |
1500 | } | |
1501 | ||
1502 | r = dm_thin_find_block(td, block, 0, &result); | |
1503 | ||
1504 | /* | |
1505 | * Note that we defer readahead too. | |
1506 | */ | |
1507 | switch (r) { | |
1508 | case 0: | |
1509 | if (unlikely(result.shared)) { | |
1510 | /* | |
1511 | * We have a race condition here between the | |
1512 | * result.shared value returned by the lookup and | |
1513 | * snapshot creation, which may cause new | |
1514 | * sharing. | |
1515 | * | |
1516 | * To avoid this always quiesce the origin before | |
1517 | * taking the snap. You want to do this anyway to | |
1518 | * ensure a consistent application view | |
1519 | * (i.e. lockfs). | |
1520 | * | |
1521 | * More distant ancestors are irrelevant. The | |
1522 | * shared flag will be set in their case. | |
1523 | */ | |
1524 | thin_defer_bio(tc, bio); | |
e8088073 | 1525 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1526 | } |
e8088073 JT |
1527 | |
1528 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1529 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1530 | return DM_MAPIO_SUBMITTED; |
1531 | ||
1532 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1533 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1534 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1535 | return DM_MAPIO_SUBMITTED; |
1536 | } | |
1537 | ||
1538 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1539 | cell_defer_no_holder_no_free(tc, &cell2); |
1540 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1541 | |
1542 | remap(tc, bio, result.block); | |
1543 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1544 | |
1545 | case -ENODATA: | |
e49e5829 JT |
1546 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1547 | /* | |
1548 | * This block isn't provisioned, and we have no way | |
1549 | * of doing so. Just error it. | |
1550 | */ | |
1551 | bio_io_error(bio); | |
2aab3850 | 1552 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1553 | } |
1554 | /* fall through */ | |
1555 | ||
1556 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1557 | /* |
1558 | * In future, the failed dm_thin_find_block above could | |
1559 | * provide the hint to load the metadata into cache. | |
1560 | */ | |
991d9fa0 | 1561 | thin_defer_bio(tc, bio); |
2aab3850 | 1562 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1563 | |
1564 | default: | |
1565 | /* | |
1566 | * Must always call bio_io_error on failure. | |
1567 | * dm_thin_find_block can fail with -EINVAL if the | |
1568 | * pool is switched to fail-io mode. | |
1569 | */ | |
1570 | bio_io_error(bio); | |
2aab3850 | 1571 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1572 | } |
991d9fa0 JT |
1573 | } |
1574 | ||
1575 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1576 | { | |
1577 | int r; | |
1578 | unsigned long flags; | |
1579 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1580 | ||
1581 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1582 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1583 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1584 | ||
1585 | if (!r) { | |
1586 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1587 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1588 | } | |
1589 | ||
1590 | return r; | |
1591 | } | |
1592 | ||
1593 | static void __requeue_bios(struct pool *pool) | |
1594 | { | |
1595 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1596 | bio_list_init(&pool->retry_on_resume_list); | |
1597 | } | |
1598 | ||
1599 | /*---------------------------------------------------------------- | |
1600 | * Binding of control targets to a pool object | |
1601 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1602 | static bool data_dev_supports_discard(struct pool_c *pt) |
1603 | { | |
1604 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1605 | ||
1606 | return q && blk_queue_discard(q); | |
1607 | } | |
1608 | ||
58051b94 JT |
1609 | static bool is_factor(sector_t block_size, uint32_t n) |
1610 | { | |
1611 | return !sector_div(block_size, n); | |
1612 | } | |
1613 | ||
9bc142dd MS |
1614 | /* |
1615 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1616 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1617 | */ |
0424caa1 | 1618 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1619 | { |
0424caa1 MS |
1620 | struct pool *pool = pt->pool; |
1621 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1622 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1623 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1624 | const char *reason = NULL; | |
9bc142dd MS |
1625 | char buf[BDEVNAME_SIZE]; |
1626 | ||
0424caa1 | 1627 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1628 | return; |
1629 | ||
0424caa1 MS |
1630 | if (!data_dev_supports_discard(pt)) |
1631 | reason = "discard unsupported"; | |
1632 | ||
1633 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1634 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1635 | |
0424caa1 MS |
1636 | else if (data_limits->discard_granularity > block_size) |
1637 | reason = "discard granularity larger than a block"; | |
1638 | ||
58051b94 | 1639 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1640 | reason = "discard granularity not a factor of block size"; |
1641 | ||
1642 | if (reason) { | |
1643 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1644 | pt->adjusted_pf.discard_passdown = false; | |
1645 | } | |
9bc142dd MS |
1646 | } |
1647 | ||
991d9fa0 JT |
1648 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1649 | { | |
1650 | struct pool_c *pt = ti->private; | |
1651 | ||
e49e5829 | 1652 | /* |
9b7aaa64 | 1653 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 JT |
1654 | */ |
1655 | enum pool_mode old_mode = pool->pf.mode; | |
0424caa1 | 1656 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1657 | |
9b7aaa64 JT |
1658 | /* |
1659 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1660 | * not going to recover without a thin_repair. So we never let the | |
1661 | * pool move out of the old mode. On the other hand a PM_READ_ONLY | |
1662 | * may have been due to a lack of metadata or data space, and may | |
1663 | * now work (ie. if the underlying devices have been resized). | |
1664 | */ | |
1665 | if (old_mode == PM_FAIL) | |
e49e5829 JT |
1666 | new_mode = old_mode; |
1667 | ||
991d9fa0 JT |
1668 | pool->ti = ti; |
1669 | pool->low_water_blocks = pt->low_water_blocks; | |
0424caa1 | 1670 | pool->pf = pt->adjusted_pf; |
991d9fa0 | 1671 | |
9bc142dd | 1672 | set_pool_mode(pool, new_mode); |
f402693d | 1673 | |
991d9fa0 JT |
1674 | return 0; |
1675 | } | |
1676 | ||
1677 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1678 | { | |
1679 | if (pool->ti == ti) | |
1680 | pool->ti = NULL; | |
1681 | } | |
1682 | ||
1683 | /*---------------------------------------------------------------- | |
1684 | * Pool creation | |
1685 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1686 | /* Initialize pool features. */ |
1687 | static void pool_features_init(struct pool_features *pf) | |
1688 | { | |
e49e5829 | 1689 | pf->mode = PM_WRITE; |
9bc142dd MS |
1690 | pf->zero_new_blocks = true; |
1691 | pf->discard_enabled = true; | |
1692 | pf->discard_passdown = true; | |
67e2e2b2 JT |
1693 | } |
1694 | ||
991d9fa0 JT |
1695 | static void __pool_destroy(struct pool *pool) |
1696 | { | |
1697 | __pool_table_remove(pool); | |
1698 | ||
1699 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1700 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1701 | ||
44feb387 | 1702 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1703 | dm_kcopyd_client_destroy(pool->copier); |
1704 | ||
1705 | if (pool->wq) | |
1706 | destroy_workqueue(pool->wq); | |
1707 | ||
1708 | if (pool->next_mapping) | |
1709 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1710 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1711 | dm_deferred_set_destroy(pool->shared_read_ds); |
1712 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1713 | kfree(pool); |
1714 | } | |
1715 | ||
a24c2569 | 1716 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1717 | |
991d9fa0 JT |
1718 | static struct pool *pool_create(struct mapped_device *pool_md, |
1719 | struct block_device *metadata_dev, | |
e49e5829 JT |
1720 | unsigned long block_size, |
1721 | int read_only, char **error) | |
991d9fa0 JT |
1722 | { |
1723 | int r; | |
1724 | void *err_p; | |
1725 | struct pool *pool; | |
1726 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1727 | bool format_device = read_only ? false : true; |
991d9fa0 | 1728 | |
e49e5829 | 1729 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1730 | if (IS_ERR(pmd)) { |
1731 | *error = "Error creating metadata object"; | |
1732 | return (struct pool *)pmd; | |
1733 | } | |
1734 | ||
1735 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1736 | if (!pool) { | |
1737 | *error = "Error allocating memory for pool"; | |
1738 | err_p = ERR_PTR(-ENOMEM); | |
1739 | goto bad_pool; | |
1740 | } | |
1741 | ||
1742 | pool->pmd = pmd; | |
1743 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1744 | if (block_size & (block_size - 1)) |
1745 | pool->sectors_per_block_shift = -1; | |
1746 | else | |
1747 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1748 | pool->low_water_blocks = 0; |
67e2e2b2 | 1749 | pool_features_init(&pool->pf); |
44feb387 | 1750 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
1751 | if (!pool->prison) { |
1752 | *error = "Error creating pool's bio prison"; | |
1753 | err_p = ERR_PTR(-ENOMEM); | |
1754 | goto bad_prison; | |
1755 | } | |
1756 | ||
df5d2e90 | 1757 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
1758 | if (IS_ERR(pool->copier)) { |
1759 | r = PTR_ERR(pool->copier); | |
1760 | *error = "Error creating pool's kcopyd client"; | |
1761 | err_p = ERR_PTR(r); | |
1762 | goto bad_kcopyd_client; | |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * Create singlethreaded workqueue that will service all devices | |
1767 | * that use this metadata. | |
1768 | */ | |
1769 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1770 | if (!pool->wq) { | |
1771 | *error = "Error creating pool's workqueue"; | |
1772 | err_p = ERR_PTR(-ENOMEM); | |
1773 | goto bad_wq; | |
1774 | } | |
1775 | ||
1776 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1777 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1778 | spin_lock_init(&pool->lock); |
1779 | bio_list_init(&pool->deferred_bios); | |
1780 | bio_list_init(&pool->deferred_flush_bios); | |
1781 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1782 | INIT_LIST_HEAD(&pool->prepared_discards); |
991d9fa0 JT |
1783 | pool->low_water_triggered = 0; |
1784 | pool->no_free_space = 0; | |
1785 | bio_list_init(&pool->retry_on_resume_list); | |
44feb387 MS |
1786 | |
1787 | pool->shared_read_ds = dm_deferred_set_create(); | |
1788 | if (!pool->shared_read_ds) { | |
1789 | *error = "Error creating pool's shared read deferred set"; | |
1790 | err_p = ERR_PTR(-ENOMEM); | |
1791 | goto bad_shared_read_ds; | |
1792 | } | |
1793 | ||
1794 | pool->all_io_ds = dm_deferred_set_create(); | |
1795 | if (!pool->all_io_ds) { | |
1796 | *error = "Error creating pool's all io deferred set"; | |
1797 | err_p = ERR_PTR(-ENOMEM); | |
1798 | goto bad_all_io_ds; | |
1799 | } | |
991d9fa0 JT |
1800 | |
1801 | pool->next_mapping = NULL; | |
a24c2569 MS |
1802 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1803 | _new_mapping_cache); | |
991d9fa0 JT |
1804 | if (!pool->mapping_pool) { |
1805 | *error = "Error creating pool's mapping mempool"; | |
1806 | err_p = ERR_PTR(-ENOMEM); | |
1807 | goto bad_mapping_pool; | |
1808 | } | |
1809 | ||
991d9fa0 | 1810 | pool->ref_count = 1; |
905e51b3 | 1811 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1812 | pool->pool_md = pool_md; |
1813 | pool->md_dev = metadata_dev; | |
1814 | __pool_table_insert(pool); | |
1815 | ||
1816 | return pool; | |
1817 | ||
991d9fa0 | 1818 | bad_mapping_pool: |
44feb387 MS |
1819 | dm_deferred_set_destroy(pool->all_io_ds); |
1820 | bad_all_io_ds: | |
1821 | dm_deferred_set_destroy(pool->shared_read_ds); | |
1822 | bad_shared_read_ds: | |
991d9fa0 JT |
1823 | destroy_workqueue(pool->wq); |
1824 | bad_wq: | |
1825 | dm_kcopyd_client_destroy(pool->copier); | |
1826 | bad_kcopyd_client: | |
44feb387 | 1827 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1828 | bad_prison: |
1829 | kfree(pool); | |
1830 | bad_pool: | |
1831 | if (dm_pool_metadata_close(pmd)) | |
1832 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1833 | ||
1834 | return err_p; | |
1835 | } | |
1836 | ||
1837 | static void __pool_inc(struct pool *pool) | |
1838 | { | |
1839 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1840 | pool->ref_count++; | |
1841 | } | |
1842 | ||
1843 | static void __pool_dec(struct pool *pool) | |
1844 | { | |
1845 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1846 | BUG_ON(!pool->ref_count); | |
1847 | if (!--pool->ref_count) | |
1848 | __pool_destroy(pool); | |
1849 | } | |
1850 | ||
1851 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
1852 | struct block_device *metadata_dev, | |
e49e5829 JT |
1853 | unsigned long block_size, int read_only, |
1854 | char **error, int *created) | |
991d9fa0 JT |
1855 | { |
1856 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
1857 | ||
1858 | if (pool) { | |
f09996c9 MS |
1859 | if (pool->pool_md != pool_md) { |
1860 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 1861 | return ERR_PTR(-EBUSY); |
f09996c9 | 1862 | } |
991d9fa0 JT |
1863 | __pool_inc(pool); |
1864 | ||
1865 | } else { | |
1866 | pool = __pool_table_lookup(pool_md); | |
1867 | if (pool) { | |
f09996c9 MS |
1868 | if (pool->md_dev != metadata_dev) { |
1869 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 1870 | return ERR_PTR(-EINVAL); |
f09996c9 | 1871 | } |
991d9fa0 JT |
1872 | __pool_inc(pool); |
1873 | ||
67e2e2b2 | 1874 | } else { |
e49e5829 | 1875 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
1876 | *created = 1; |
1877 | } | |
991d9fa0 JT |
1878 | } |
1879 | ||
1880 | return pool; | |
1881 | } | |
1882 | ||
1883 | /*---------------------------------------------------------------- | |
1884 | * Pool target methods | |
1885 | *--------------------------------------------------------------*/ | |
1886 | static void pool_dtr(struct dm_target *ti) | |
1887 | { | |
1888 | struct pool_c *pt = ti->private; | |
1889 | ||
1890 | mutex_lock(&dm_thin_pool_table.mutex); | |
1891 | ||
1892 | unbind_control_target(pt->pool, ti); | |
1893 | __pool_dec(pt->pool); | |
1894 | dm_put_device(ti, pt->metadata_dev); | |
1895 | dm_put_device(ti, pt->data_dev); | |
1896 | kfree(pt); | |
1897 | ||
1898 | mutex_unlock(&dm_thin_pool_table.mutex); | |
1899 | } | |
1900 | ||
991d9fa0 JT |
1901 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
1902 | struct dm_target *ti) | |
1903 | { | |
1904 | int r; | |
1905 | unsigned argc; | |
1906 | const char *arg_name; | |
1907 | ||
1908 | static struct dm_arg _args[] = { | |
67e2e2b2 | 1909 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
1910 | }; |
1911 | ||
1912 | /* | |
1913 | * No feature arguments supplied. | |
1914 | */ | |
1915 | if (!as->argc) | |
1916 | return 0; | |
1917 | ||
1918 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1919 | if (r) | |
1920 | return -EINVAL; | |
1921 | ||
1922 | while (argc && !r) { | |
1923 | arg_name = dm_shift_arg(as); | |
1924 | argc--; | |
1925 | ||
e49e5829 | 1926 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 1927 | pf->zero_new_blocks = false; |
e49e5829 JT |
1928 | |
1929 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 1930 | pf->discard_enabled = false; |
e49e5829 JT |
1931 | |
1932 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 1933 | pf->discard_passdown = false; |
991d9fa0 | 1934 | |
e49e5829 JT |
1935 | else if (!strcasecmp(arg_name, "read_only")) |
1936 | pf->mode = PM_READ_ONLY; | |
1937 | ||
1938 | else { | |
1939 | ti->error = "Unrecognised pool feature requested"; | |
1940 | r = -EINVAL; | |
1941 | break; | |
1942 | } | |
991d9fa0 JT |
1943 | } |
1944 | ||
1945 | return r; | |
1946 | } | |
1947 | ||
ac8c3f3d JT |
1948 | static void metadata_low_callback(void *context) |
1949 | { | |
1950 | struct pool *pool = context; | |
1951 | ||
1952 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
1953 | dm_device_name(pool->pool_md)); | |
1954 | ||
1955 | dm_table_event(pool->ti->table); | |
1956 | } | |
1957 | ||
b17446df JT |
1958 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
1959 | { | |
1960 | sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
1961 | char buffer[BDEVNAME_SIZE]; | |
1962 | ||
1963 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) { | |
1964 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
1965 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
1966 | metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING; | |
1967 | } | |
1968 | ||
1969 | return metadata_dev_size; | |
1970 | } | |
1971 | ||
24347e95 JT |
1972 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
1973 | { | |
1974 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
1975 | ||
1976 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); | |
1977 | ||
1978 | return metadata_dev_size; | |
1979 | } | |
1980 | ||
ac8c3f3d JT |
1981 | /* |
1982 | * When a metadata threshold is crossed a dm event is triggered, and | |
1983 | * userland should respond by growing the metadata device. We could let | |
1984 | * userland set the threshold, like we do with the data threshold, but I'm | |
1985 | * not sure they know enough to do this well. | |
1986 | */ | |
1987 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
1988 | { | |
1989 | /* | |
1990 | * 4M is ample for all ops with the possible exception of thin | |
1991 | * device deletion which is harmless if it fails (just retry the | |
1992 | * delete after you've grown the device). | |
1993 | */ | |
1994 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
1995 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
1996 | } | |
1997 | ||
991d9fa0 JT |
1998 | /* |
1999 | * thin-pool <metadata dev> <data dev> | |
2000 | * <data block size (sectors)> | |
2001 | * <low water mark (blocks)> | |
2002 | * [<#feature args> [<arg>]*] | |
2003 | * | |
2004 | * Optional feature arguments are: | |
2005 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2006 | * ignore_discard: disable discard |
2007 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
2008 | */ |
2009 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2010 | { | |
67e2e2b2 | 2011 | int r, pool_created = 0; |
991d9fa0 JT |
2012 | struct pool_c *pt; |
2013 | struct pool *pool; | |
2014 | struct pool_features pf; | |
2015 | struct dm_arg_set as; | |
2016 | struct dm_dev *data_dev; | |
2017 | unsigned long block_size; | |
2018 | dm_block_t low_water_blocks; | |
2019 | struct dm_dev *metadata_dev; | |
5d0db96d | 2020 | fmode_t metadata_mode; |
991d9fa0 JT |
2021 | |
2022 | /* | |
2023 | * FIXME Remove validation from scope of lock. | |
2024 | */ | |
2025 | mutex_lock(&dm_thin_pool_table.mutex); | |
2026 | ||
2027 | if (argc < 4) { | |
2028 | ti->error = "Invalid argument count"; | |
2029 | r = -EINVAL; | |
2030 | goto out_unlock; | |
2031 | } | |
5d0db96d | 2032 | |
991d9fa0 JT |
2033 | as.argc = argc; |
2034 | as.argv = argv; | |
2035 | ||
5d0db96d JT |
2036 | /* |
2037 | * Set default pool features. | |
2038 | */ | |
2039 | pool_features_init(&pf); | |
2040 | ||
2041 | dm_consume_args(&as, 4); | |
2042 | r = parse_pool_features(&as, &pf, ti); | |
2043 | if (r) | |
2044 | goto out_unlock; | |
2045 | ||
2046 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2047 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2048 | if (r) { |
2049 | ti->error = "Error opening metadata block device"; | |
2050 | goto out_unlock; | |
2051 | } | |
2052 | ||
b17446df JT |
2053 | /* |
2054 | * Run for the side-effect of possibly issuing a warning if the | |
2055 | * device is too big. | |
2056 | */ | |
2057 | (void) get_metadata_dev_size(metadata_dev->bdev); | |
991d9fa0 JT |
2058 | |
2059 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2060 | if (r) { | |
2061 | ti->error = "Error getting data device"; | |
2062 | goto out_metadata; | |
2063 | } | |
2064 | ||
2065 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2066 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2067 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2068 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2069 | ti->error = "Invalid block size"; |
2070 | r = -EINVAL; | |
2071 | goto out; | |
2072 | } | |
2073 | ||
2074 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2075 | ti->error = "Invalid low water mark"; | |
2076 | r = -EINVAL; | |
2077 | goto out; | |
2078 | } | |
2079 | ||
991d9fa0 JT |
2080 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2081 | if (!pt) { | |
2082 | r = -ENOMEM; | |
2083 | goto out; | |
2084 | } | |
2085 | ||
2086 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2087 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2088 | if (IS_ERR(pool)) { |
2089 | r = PTR_ERR(pool); | |
2090 | goto out_free_pt; | |
2091 | } | |
2092 | ||
67e2e2b2 JT |
2093 | /* |
2094 | * 'pool_created' reflects whether this is the first table load. | |
2095 | * Top level discard support is not allowed to be changed after | |
2096 | * initial load. This would require a pool reload to trigger thin | |
2097 | * device changes. | |
2098 | */ | |
2099 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2100 | ti->error = "Discard support cannot be disabled once enabled"; | |
2101 | r = -EINVAL; | |
2102 | goto out_flags_changed; | |
2103 | } | |
2104 | ||
991d9fa0 JT |
2105 | pt->pool = pool; |
2106 | pt->ti = ti; | |
2107 | pt->metadata_dev = metadata_dev; | |
2108 | pt->data_dev = data_dev; | |
2109 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2110 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2111 | ti->num_flush_bios = 1; |
9bc142dd | 2112 | |
67e2e2b2 JT |
2113 | /* |
2114 | * Only need to enable discards if the pool should pass | |
2115 | * them down to the data device. The thin device's discard | |
2116 | * processing will cause mappings to be removed from the btree. | |
2117 | */ | |
b60ab990 | 2118 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2119 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2120 | ti->num_discard_bios = 1; |
9bc142dd | 2121 | |
67e2e2b2 JT |
2122 | /* |
2123 | * Setting 'discards_supported' circumvents the normal | |
2124 | * stacking of discard limits (this keeps the pool and | |
2125 | * thin devices' discard limits consistent). | |
2126 | */ | |
0ac55489 | 2127 | ti->discards_supported = true; |
67e2e2b2 | 2128 | } |
991d9fa0 JT |
2129 | ti->private = pt; |
2130 | ||
ac8c3f3d JT |
2131 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2132 | calc_metadata_threshold(pt), | |
2133 | metadata_low_callback, | |
2134 | pool); | |
2135 | if (r) | |
2136 | goto out_free_pt; | |
2137 | ||
991d9fa0 JT |
2138 | pt->callbacks.congested_fn = pool_is_congested; |
2139 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2140 | ||
2141 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2142 | ||
2143 | return 0; | |
2144 | ||
67e2e2b2 JT |
2145 | out_flags_changed: |
2146 | __pool_dec(pool); | |
991d9fa0 JT |
2147 | out_free_pt: |
2148 | kfree(pt); | |
2149 | out: | |
2150 | dm_put_device(ti, data_dev); | |
2151 | out_metadata: | |
2152 | dm_put_device(ti, metadata_dev); | |
2153 | out_unlock: | |
2154 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2155 | ||
2156 | return r; | |
2157 | } | |
2158 | ||
7de3ee57 | 2159 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2160 | { |
2161 | int r; | |
2162 | struct pool_c *pt = ti->private; | |
2163 | struct pool *pool = pt->pool; | |
2164 | unsigned long flags; | |
2165 | ||
2166 | /* | |
2167 | * As this is a singleton target, ti->begin is always zero. | |
2168 | */ | |
2169 | spin_lock_irqsave(&pool->lock, flags); | |
2170 | bio->bi_bdev = pt->data_dev->bdev; | |
2171 | r = DM_MAPIO_REMAPPED; | |
2172 | spin_unlock_irqrestore(&pool->lock, flags); | |
2173 | ||
2174 | return r; | |
2175 | } | |
2176 | ||
b17446df | 2177 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2178 | { |
2179 | int r; | |
2180 | struct pool_c *pt = ti->private; | |
2181 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2182 | sector_t data_size = ti->len; |
2183 | dm_block_t sb_data_size; | |
991d9fa0 | 2184 | |
b17446df | 2185 | *need_commit = false; |
991d9fa0 | 2186 | |
55f2b8bd MS |
2187 | (void) sector_div(data_size, pool->sectors_per_block); |
2188 | ||
991d9fa0 JT |
2189 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2190 | if (r) { | |
4fa5971a MS |
2191 | DMERR("%s: failed to retrieve data device size", |
2192 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2193 | return r; |
2194 | } | |
2195 | ||
2196 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2197 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2198 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2199 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2200 | return -EINVAL; |
2201 | ||
2202 | } else if (data_size > sb_data_size) { | |
2203 | r = dm_pool_resize_data_dev(pool->pmd, data_size); | |
2204 | if (r) { | |
4fa5971a MS |
2205 | DMERR("%s: failed to resize data device", |
2206 | dm_device_name(pool->pool_md)); | |
e49e5829 | 2207 | set_pool_mode(pool, PM_READ_ONLY); |
991d9fa0 JT |
2208 | return r; |
2209 | } | |
2210 | ||
b17446df | 2211 | *need_commit = true; |
991d9fa0 JT |
2212 | } |
2213 | ||
2214 | return 0; | |
2215 | } | |
2216 | ||
24347e95 JT |
2217 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2218 | { | |
2219 | int r; | |
2220 | struct pool_c *pt = ti->private; | |
2221 | struct pool *pool = pt->pool; | |
2222 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2223 | ||
2224 | *need_commit = false; | |
2225 | ||
610bba8b | 2226 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2227 | |
2228 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2229 | if (r) { | |
4fa5971a MS |
2230 | DMERR("%s: failed to retrieve metadata device size", |
2231 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2232 | return r; |
2233 | } | |
2234 | ||
2235 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2236 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2237 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2238 | metadata_dev_size, sb_metadata_dev_size); |
2239 | return -EINVAL; | |
2240 | ||
2241 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
2242 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); | |
2243 | if (r) { | |
4fa5971a MS |
2244 | DMERR("%s: failed to resize metadata device", |
2245 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2246 | return r; |
2247 | } | |
2248 | ||
2249 | *need_commit = true; | |
2250 | } | |
2251 | ||
2252 | return 0; | |
2253 | } | |
2254 | ||
b17446df JT |
2255 | /* |
2256 | * Retrieves the number of blocks of the data device from | |
2257 | * the superblock and compares it to the actual device size, | |
2258 | * thus resizing the data device in case it has grown. | |
2259 | * | |
2260 | * This both copes with opening preallocated data devices in the ctr | |
2261 | * being followed by a resume | |
2262 | * -and- | |
2263 | * calling the resume method individually after userspace has | |
2264 | * grown the data device in reaction to a table event. | |
2265 | */ | |
2266 | static int pool_preresume(struct dm_target *ti) | |
2267 | { | |
2268 | int r; | |
24347e95 | 2269 | bool need_commit1, need_commit2; |
b17446df JT |
2270 | struct pool_c *pt = ti->private; |
2271 | struct pool *pool = pt->pool; | |
2272 | ||
2273 | /* | |
2274 | * Take control of the pool object. | |
2275 | */ | |
2276 | r = bind_control_target(pool, ti); | |
2277 | if (r) | |
2278 | return r; | |
2279 | ||
2280 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2281 | if (r) | |
2282 | return r; | |
2283 | ||
24347e95 JT |
2284 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2285 | if (r) | |
2286 | return r; | |
2287 | ||
2288 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2289 | (void) commit(pool); |
b17446df JT |
2290 | |
2291 | return 0; | |
2292 | } | |
2293 | ||
991d9fa0 JT |
2294 | static void pool_resume(struct dm_target *ti) |
2295 | { | |
2296 | struct pool_c *pt = ti->private; | |
2297 | struct pool *pool = pt->pool; | |
2298 | unsigned long flags; | |
2299 | ||
2300 | spin_lock_irqsave(&pool->lock, flags); | |
2301 | pool->low_water_triggered = 0; | |
2302 | pool->no_free_space = 0; | |
2303 | __requeue_bios(pool); | |
2304 | spin_unlock_irqrestore(&pool->lock, flags); | |
2305 | ||
905e51b3 | 2306 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2307 | } |
2308 | ||
2309 | static void pool_postsuspend(struct dm_target *ti) | |
2310 | { | |
991d9fa0 JT |
2311 | struct pool_c *pt = ti->private; |
2312 | struct pool *pool = pt->pool; | |
2313 | ||
905e51b3 | 2314 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2315 | flush_workqueue(pool->wq); |
020cc3b5 | 2316 | (void) commit(pool); |
991d9fa0 JT |
2317 | } |
2318 | ||
2319 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2320 | { | |
2321 | if (argc != args_required) { | |
2322 | DMWARN("Message received with %u arguments instead of %u.", | |
2323 | argc, args_required); | |
2324 | return -EINVAL; | |
2325 | } | |
2326 | ||
2327 | return 0; | |
2328 | } | |
2329 | ||
2330 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2331 | { | |
2332 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2333 | *dev_id <= MAX_DEV_ID) | |
2334 | return 0; | |
2335 | ||
2336 | if (warning) | |
2337 | DMWARN("Message received with invalid device id: %s", arg); | |
2338 | ||
2339 | return -EINVAL; | |
2340 | } | |
2341 | ||
2342 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2343 | { | |
2344 | dm_thin_id dev_id; | |
2345 | int r; | |
2346 | ||
2347 | r = check_arg_count(argc, 2); | |
2348 | if (r) | |
2349 | return r; | |
2350 | ||
2351 | r = read_dev_id(argv[1], &dev_id, 1); | |
2352 | if (r) | |
2353 | return r; | |
2354 | ||
2355 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2356 | if (r) { | |
2357 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2358 | argv[1]); | |
2359 | return r; | |
2360 | } | |
2361 | ||
2362 | return 0; | |
2363 | } | |
2364 | ||
2365 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2366 | { | |
2367 | dm_thin_id dev_id; | |
2368 | dm_thin_id origin_dev_id; | |
2369 | int r; | |
2370 | ||
2371 | r = check_arg_count(argc, 3); | |
2372 | if (r) | |
2373 | return r; | |
2374 | ||
2375 | r = read_dev_id(argv[1], &dev_id, 1); | |
2376 | if (r) | |
2377 | return r; | |
2378 | ||
2379 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2380 | if (r) | |
2381 | return r; | |
2382 | ||
2383 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2384 | if (r) { | |
2385 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2386 | argv[1], argv[2]); | |
2387 | return r; | |
2388 | } | |
2389 | ||
2390 | return 0; | |
2391 | } | |
2392 | ||
2393 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2394 | { | |
2395 | dm_thin_id dev_id; | |
2396 | int r; | |
2397 | ||
2398 | r = check_arg_count(argc, 2); | |
2399 | if (r) | |
2400 | return r; | |
2401 | ||
2402 | r = read_dev_id(argv[1], &dev_id, 1); | |
2403 | if (r) | |
2404 | return r; | |
2405 | ||
2406 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2407 | if (r) | |
2408 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2409 | ||
2410 | return r; | |
2411 | } | |
2412 | ||
2413 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2414 | { | |
2415 | dm_thin_id old_id, new_id; | |
2416 | int r; | |
2417 | ||
2418 | r = check_arg_count(argc, 3); | |
2419 | if (r) | |
2420 | return r; | |
2421 | ||
2422 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2423 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2424 | return -EINVAL; | |
2425 | } | |
2426 | ||
2427 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2428 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2429 | return -EINVAL; | |
2430 | } | |
2431 | ||
2432 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2433 | if (r) { | |
2434 | DMWARN("Failed to change transaction id from %s to %s.", | |
2435 | argv[1], argv[2]); | |
2436 | return r; | |
2437 | } | |
2438 | ||
2439 | return 0; | |
2440 | } | |
2441 | ||
cc8394d8 JT |
2442 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2443 | { | |
2444 | int r; | |
2445 | ||
2446 | r = check_arg_count(argc, 1); | |
2447 | if (r) | |
2448 | return r; | |
2449 | ||
020cc3b5 | 2450 | (void) commit(pool); |
0d200aef | 2451 | |
cc8394d8 JT |
2452 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2453 | if (r) | |
2454 | DMWARN("reserve_metadata_snap message failed."); | |
2455 | ||
2456 | return r; | |
2457 | } | |
2458 | ||
2459 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2460 | { | |
2461 | int r; | |
2462 | ||
2463 | r = check_arg_count(argc, 1); | |
2464 | if (r) | |
2465 | return r; | |
2466 | ||
2467 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2468 | if (r) | |
2469 | DMWARN("release_metadata_snap message failed."); | |
2470 | ||
2471 | return r; | |
2472 | } | |
2473 | ||
991d9fa0 JT |
2474 | /* |
2475 | * Messages supported: | |
2476 | * create_thin <dev_id> | |
2477 | * create_snap <dev_id> <origin_id> | |
2478 | * delete <dev_id> | |
2479 | * trim <dev_id> <new_size_in_sectors> | |
2480 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2481 | * reserve_metadata_snap |
2482 | * release_metadata_snap | |
991d9fa0 JT |
2483 | */ |
2484 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2485 | { | |
2486 | int r = -EINVAL; | |
2487 | struct pool_c *pt = ti->private; | |
2488 | struct pool *pool = pt->pool; | |
2489 | ||
2490 | if (!strcasecmp(argv[0], "create_thin")) | |
2491 | r = process_create_thin_mesg(argc, argv, pool); | |
2492 | ||
2493 | else if (!strcasecmp(argv[0], "create_snap")) | |
2494 | r = process_create_snap_mesg(argc, argv, pool); | |
2495 | ||
2496 | else if (!strcasecmp(argv[0], "delete")) | |
2497 | r = process_delete_mesg(argc, argv, pool); | |
2498 | ||
2499 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2500 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2501 | ||
cc8394d8 JT |
2502 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2503 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2504 | ||
2505 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2506 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2507 | ||
991d9fa0 JT |
2508 | else |
2509 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2510 | ||
e49e5829 | 2511 | if (!r) |
020cc3b5 | 2512 | (void) commit(pool); |
991d9fa0 JT |
2513 | |
2514 | return r; | |
2515 | } | |
2516 | ||
e49e5829 JT |
2517 | static void emit_flags(struct pool_features *pf, char *result, |
2518 | unsigned sz, unsigned maxlen) | |
2519 | { | |
2520 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
2521 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY); | |
2522 | DMEMIT("%u ", count); | |
2523 | ||
2524 | if (!pf->zero_new_blocks) | |
2525 | DMEMIT("skip_block_zeroing "); | |
2526 | ||
2527 | if (!pf->discard_enabled) | |
2528 | DMEMIT("ignore_discard "); | |
2529 | ||
2530 | if (!pf->discard_passdown) | |
2531 | DMEMIT("no_discard_passdown "); | |
2532 | ||
2533 | if (pf->mode == PM_READ_ONLY) | |
2534 | DMEMIT("read_only "); | |
2535 | } | |
2536 | ||
991d9fa0 JT |
2537 | /* |
2538 | * Status line is: | |
2539 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2540 | * <used data sectors>/<total data sectors> <held metadata root> | |
2541 | */ | |
fd7c092e MP |
2542 | static void pool_status(struct dm_target *ti, status_type_t type, |
2543 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2544 | { |
e49e5829 | 2545 | int r; |
991d9fa0 JT |
2546 | unsigned sz = 0; |
2547 | uint64_t transaction_id; | |
2548 | dm_block_t nr_free_blocks_data; | |
2549 | dm_block_t nr_free_blocks_metadata; | |
2550 | dm_block_t nr_blocks_data; | |
2551 | dm_block_t nr_blocks_metadata; | |
2552 | dm_block_t held_root; | |
2553 | char buf[BDEVNAME_SIZE]; | |
2554 | char buf2[BDEVNAME_SIZE]; | |
2555 | struct pool_c *pt = ti->private; | |
2556 | struct pool *pool = pt->pool; | |
2557 | ||
2558 | switch (type) { | |
2559 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2560 | if (get_pool_mode(pool) == PM_FAIL) { |
2561 | DMEMIT("Fail"); | |
2562 | break; | |
2563 | } | |
2564 | ||
1f4e0ff0 AK |
2565 | /* Commit to ensure statistics aren't out-of-date */ |
2566 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2567 | (void) commit(pool); |
1f4e0ff0 | 2568 | |
fd7c092e MP |
2569 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2570 | if (r) { | |
4fa5971a MS |
2571 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2572 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2573 | goto err; |
2574 | } | |
991d9fa0 | 2575 | |
fd7c092e MP |
2576 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2577 | if (r) { | |
4fa5971a MS |
2578 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2579 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2580 | goto err; |
2581 | } | |
991d9fa0 JT |
2582 | |
2583 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2584 | if (r) { |
4fa5971a MS |
2585 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2586 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2587 | goto err; |
2588 | } | |
991d9fa0 | 2589 | |
fd7c092e MP |
2590 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2591 | if (r) { | |
4fa5971a MS |
2592 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2593 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2594 | goto err; |
2595 | } | |
991d9fa0 JT |
2596 | |
2597 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2598 | if (r) { |
4fa5971a MS |
2599 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2600 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2601 | goto err; |
2602 | } | |
991d9fa0 | 2603 | |
cc8394d8 | 2604 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2605 | if (r) { |
4fa5971a MS |
2606 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2607 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2608 | goto err; |
2609 | } | |
991d9fa0 JT |
2610 | |
2611 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2612 | (unsigned long long)transaction_id, | |
2613 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2614 | (unsigned long long)nr_blocks_metadata, | |
2615 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2616 | (unsigned long long)nr_blocks_data); | |
2617 | ||
2618 | if (held_root) | |
e49e5829 JT |
2619 | DMEMIT("%llu ", held_root); |
2620 | else | |
2621 | DMEMIT("- "); | |
2622 | ||
2623 | if (pool->pf.mode == PM_READ_ONLY) | |
2624 | DMEMIT("ro "); | |
991d9fa0 | 2625 | else |
e49e5829 JT |
2626 | DMEMIT("rw "); |
2627 | ||
018debea MS |
2628 | if (!pool->pf.discard_enabled) |
2629 | DMEMIT("ignore_discard"); | |
2630 | else if (pool->pf.discard_passdown) | |
e49e5829 JT |
2631 | DMEMIT("discard_passdown"); |
2632 | else | |
2633 | DMEMIT("no_discard_passdown"); | |
991d9fa0 JT |
2634 | |
2635 | break; | |
2636 | ||
2637 | case STATUSTYPE_TABLE: | |
2638 | DMEMIT("%s %s %lu %llu ", | |
2639 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2640 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2641 | (unsigned long)pool->sectors_per_block, | |
2642 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2643 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2644 | break; |
2645 | } | |
fd7c092e | 2646 | return; |
991d9fa0 | 2647 | |
fd7c092e MP |
2648 | err: |
2649 | DMEMIT("Error"); | |
991d9fa0 JT |
2650 | } |
2651 | ||
2652 | static int pool_iterate_devices(struct dm_target *ti, | |
2653 | iterate_devices_callout_fn fn, void *data) | |
2654 | { | |
2655 | struct pool_c *pt = ti->private; | |
2656 | ||
2657 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2658 | } | |
2659 | ||
2660 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2661 | struct bio_vec *biovec, int max_size) | |
2662 | { | |
2663 | struct pool_c *pt = ti->private; | |
2664 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2665 | ||
2666 | if (!q->merge_bvec_fn) | |
2667 | return max_size; | |
2668 | ||
2669 | bvm->bi_bdev = pt->data_dev->bdev; | |
2670 | ||
2671 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2672 | } | |
2673 | ||
0424caa1 | 2674 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2675 | { |
0424caa1 MS |
2676 | struct pool *pool = pt->pool; |
2677 | struct queue_limits *data_limits; | |
2678 | ||
104655fd JT |
2679 | limits->max_discard_sectors = pool->sectors_per_block; |
2680 | ||
2681 | /* | |
0424caa1 | 2682 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2683 | */ |
0424caa1 MS |
2684 | if (pt->adjusted_pf.discard_passdown) { |
2685 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
2686 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 2687 | } else |
0424caa1 | 2688 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
2689 | } |
2690 | ||
991d9fa0 JT |
2691 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2692 | { | |
2693 | struct pool_c *pt = ti->private; | |
2694 | struct pool *pool = pt->pool; | |
0cc67cd9 | 2695 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 2696 | |
0cc67cd9 MS |
2697 | /* |
2698 | * If the system-determined stacked limits are compatible with the | |
2699 | * pool's blocksize (io_opt is a factor) do not override them. | |
2700 | */ | |
2701 | if (io_opt_sectors < pool->sectors_per_block || | |
2702 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
2703 | blk_limits_io_min(limits, 0); | |
2704 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
2705 | } | |
0424caa1 MS |
2706 | |
2707 | /* | |
2708 | * pt->adjusted_pf is a staging area for the actual features to use. | |
2709 | * They get transferred to the live pool in bind_control_target() | |
2710 | * called from pool_preresume(). | |
2711 | */ | |
b60ab990 MS |
2712 | if (!pt->adjusted_pf.discard_enabled) { |
2713 | /* | |
2714 | * Must explicitly disallow stacking discard limits otherwise the | |
2715 | * block layer will stack them if pool's data device has support. | |
2716 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
2717 | * user to see that, so make sure to set all discard limits to 0. | |
2718 | */ | |
2719 | limits->discard_granularity = 0; | |
0424caa1 | 2720 | return; |
b60ab990 | 2721 | } |
0424caa1 MS |
2722 | |
2723 | disable_passdown_if_not_supported(pt); | |
2724 | ||
2725 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
2726 | } |
2727 | ||
2728 | static struct target_type pool_target = { | |
2729 | .name = "thin-pool", | |
2730 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2731 | DM_TARGET_IMMUTABLE, | |
94563bad | 2732 | .version = {1, 9, 0}, |
991d9fa0 JT |
2733 | .module = THIS_MODULE, |
2734 | .ctr = pool_ctr, | |
2735 | .dtr = pool_dtr, | |
2736 | .map = pool_map, | |
2737 | .postsuspend = pool_postsuspend, | |
2738 | .preresume = pool_preresume, | |
2739 | .resume = pool_resume, | |
2740 | .message = pool_message, | |
2741 | .status = pool_status, | |
2742 | .merge = pool_merge, | |
2743 | .iterate_devices = pool_iterate_devices, | |
2744 | .io_hints = pool_io_hints, | |
2745 | }; | |
2746 | ||
2747 | /*---------------------------------------------------------------- | |
2748 | * Thin target methods | |
2749 | *--------------------------------------------------------------*/ | |
2750 | static void thin_dtr(struct dm_target *ti) | |
2751 | { | |
2752 | struct thin_c *tc = ti->private; | |
2753 | ||
2754 | mutex_lock(&dm_thin_pool_table.mutex); | |
2755 | ||
2756 | __pool_dec(tc->pool); | |
2757 | dm_pool_close_thin_device(tc->td); | |
2758 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2759 | if (tc->origin_dev) |
2760 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2761 | kfree(tc); |
2762 | ||
2763 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2764 | } | |
2765 | ||
2766 | /* | |
2767 | * Thin target parameters: | |
2768 | * | |
2dd9c257 | 2769 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2770 | * |
2771 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2772 | * dev_id: the internal device identifier | |
2dd9c257 | 2773 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2774 | * |
2775 | * If the pool device has discards disabled, they get disabled for the thin | |
2776 | * device as well. | |
991d9fa0 JT |
2777 | */ |
2778 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2779 | { | |
2780 | int r; | |
2781 | struct thin_c *tc; | |
2dd9c257 | 2782 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2783 | struct mapped_device *pool_md; |
2784 | ||
2785 | mutex_lock(&dm_thin_pool_table.mutex); | |
2786 | ||
2dd9c257 | 2787 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2788 | ti->error = "Invalid argument count"; |
2789 | r = -EINVAL; | |
2790 | goto out_unlock; | |
2791 | } | |
2792 | ||
2793 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2794 | if (!tc) { | |
2795 | ti->error = "Out of memory"; | |
2796 | r = -ENOMEM; | |
2797 | goto out_unlock; | |
2798 | } | |
2799 | ||
2dd9c257 JT |
2800 | if (argc == 3) { |
2801 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2802 | if (r) { | |
2803 | ti->error = "Error opening origin device"; | |
2804 | goto bad_origin_dev; | |
2805 | } | |
2806 | tc->origin_dev = origin_dev; | |
2807 | } | |
2808 | ||
991d9fa0 JT |
2809 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2810 | if (r) { | |
2811 | ti->error = "Error opening pool device"; | |
2812 | goto bad_pool_dev; | |
2813 | } | |
2814 | tc->pool_dev = pool_dev; | |
2815 | ||
2816 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2817 | ti->error = "Invalid device id"; | |
2818 | r = -EINVAL; | |
2819 | goto bad_common; | |
2820 | } | |
2821 | ||
2822 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2823 | if (!pool_md) { | |
2824 | ti->error = "Couldn't get pool mapped device"; | |
2825 | r = -EINVAL; | |
2826 | goto bad_common; | |
2827 | } | |
2828 | ||
2829 | tc->pool = __pool_table_lookup(pool_md); | |
2830 | if (!tc->pool) { | |
2831 | ti->error = "Couldn't find pool object"; | |
2832 | r = -EINVAL; | |
2833 | goto bad_pool_lookup; | |
2834 | } | |
2835 | __pool_inc(tc->pool); | |
2836 | ||
e49e5829 JT |
2837 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2838 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
2839 | goto bad_thin_open; | |
2840 | } | |
2841 | ||
991d9fa0 JT |
2842 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
2843 | if (r) { | |
2844 | ti->error = "Couldn't open thin internal device"; | |
2845 | goto bad_thin_open; | |
2846 | } | |
2847 | ||
542f9038 MS |
2848 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2849 | if (r) | |
2850 | goto bad_thin_open; | |
2851 | ||
55a62eef | 2852 | ti->num_flush_bios = 1; |
16ad3d10 | 2853 | ti->flush_supported = true; |
59c3d2c6 | 2854 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
2855 | |
2856 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 2857 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2858 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 2859 | ti->discards_supported = true; |
55a62eef | 2860 | ti->num_discard_bios = 1; |
55a62eef AK |
2861 | /* Discard bios must be split on a block boundary */ |
2862 | ti->split_discard_bios = true; | |
67e2e2b2 | 2863 | } |
991d9fa0 JT |
2864 | |
2865 | dm_put(pool_md); | |
2866 | ||
2867 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2868 | ||
2869 | return 0; | |
2870 | ||
2871 | bad_thin_open: | |
2872 | __pool_dec(tc->pool); | |
2873 | bad_pool_lookup: | |
2874 | dm_put(pool_md); | |
2875 | bad_common: | |
2876 | dm_put_device(ti, tc->pool_dev); | |
2877 | bad_pool_dev: | |
2dd9c257 JT |
2878 | if (tc->origin_dev) |
2879 | dm_put_device(ti, tc->origin_dev); | |
2880 | bad_origin_dev: | |
991d9fa0 JT |
2881 | kfree(tc); |
2882 | out_unlock: | |
2883 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2884 | ||
2885 | return r; | |
2886 | } | |
2887 | ||
7de3ee57 | 2888 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 2889 | { |
6efd6e83 | 2890 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 | 2891 | |
7de3ee57 | 2892 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
2893 | } |
2894 | ||
7de3ee57 | 2895 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
2896 | { |
2897 | unsigned long flags; | |
59c3d2c6 | 2898 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 2899 | struct list_head work; |
a24c2569 | 2900 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2901 | struct pool *pool = h->tc->pool; |
2902 | ||
2903 | if (h->shared_read_entry) { | |
2904 | INIT_LIST_HEAD(&work); | |
44feb387 | 2905 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
2906 | |
2907 | spin_lock_irqsave(&pool->lock, flags); | |
2908 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2909 | list_del(&m->list); | |
7f214665 | 2910 | m->quiesced = true; |
eb2aa48d JT |
2911 | __maybe_add_mapping(m); |
2912 | } | |
2913 | spin_unlock_irqrestore(&pool->lock, flags); | |
2914 | } | |
2915 | ||
104655fd JT |
2916 | if (h->all_io_entry) { |
2917 | INIT_LIST_HEAD(&work); | |
44feb387 | 2918 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
2919 | if (!list_empty(&work)) { |
2920 | spin_lock_irqsave(&pool->lock, flags); | |
2921 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 2922 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
2923 | spin_unlock_irqrestore(&pool->lock, flags); |
2924 | wake_worker(pool); | |
2925 | } | |
104655fd JT |
2926 | } |
2927 | ||
eb2aa48d JT |
2928 | return 0; |
2929 | } | |
2930 | ||
991d9fa0 JT |
2931 | static void thin_postsuspend(struct dm_target *ti) |
2932 | { | |
2933 | if (dm_noflush_suspending(ti)) | |
2934 | requeue_io((struct thin_c *)ti->private); | |
2935 | } | |
2936 | ||
2937 | /* | |
2938 | * <nr mapped sectors> <highest mapped sector> | |
2939 | */ | |
fd7c092e MP |
2940 | static void thin_status(struct dm_target *ti, status_type_t type, |
2941 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
2942 | { |
2943 | int r; | |
2944 | ssize_t sz = 0; | |
2945 | dm_block_t mapped, highest; | |
2946 | char buf[BDEVNAME_SIZE]; | |
2947 | struct thin_c *tc = ti->private; | |
2948 | ||
e49e5829 JT |
2949 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2950 | DMEMIT("Fail"); | |
fd7c092e | 2951 | return; |
e49e5829 JT |
2952 | } |
2953 | ||
991d9fa0 JT |
2954 | if (!tc->td) |
2955 | DMEMIT("-"); | |
2956 | else { | |
2957 | switch (type) { | |
2958 | case STATUSTYPE_INFO: | |
2959 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
2960 | if (r) { |
2961 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
2962 | goto err; | |
2963 | } | |
991d9fa0 JT |
2964 | |
2965 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
2966 | if (r < 0) { |
2967 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
2968 | goto err; | |
2969 | } | |
991d9fa0 JT |
2970 | |
2971 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
2972 | if (r) | |
2973 | DMEMIT("%llu", ((highest + 1) * | |
2974 | tc->pool->sectors_per_block) - 1); | |
2975 | else | |
2976 | DMEMIT("-"); | |
2977 | break; | |
2978 | ||
2979 | case STATUSTYPE_TABLE: | |
2980 | DMEMIT("%s %lu", | |
2981 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
2982 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
2983 | if (tc->origin_dev) |
2984 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
2985 | break; |
2986 | } | |
2987 | } | |
2988 | ||
fd7c092e MP |
2989 | return; |
2990 | ||
2991 | err: | |
2992 | DMEMIT("Error"); | |
991d9fa0 JT |
2993 | } |
2994 | ||
2995 | static int thin_iterate_devices(struct dm_target *ti, | |
2996 | iterate_devices_callout_fn fn, void *data) | |
2997 | { | |
55f2b8bd | 2998 | sector_t blocks; |
991d9fa0 | 2999 | struct thin_c *tc = ti->private; |
55f2b8bd | 3000 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3001 | |
3002 | /* | |
3003 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3004 | * we follow a more convoluted path through to the pool's target. | |
3005 | */ | |
55f2b8bd | 3006 | if (!pool->ti) |
991d9fa0 JT |
3007 | return 0; /* nothing is bound */ |
3008 | ||
55f2b8bd MS |
3009 | blocks = pool->ti->len; |
3010 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3011 | if (blocks) |
55f2b8bd | 3012 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3013 | |
3014 | return 0; | |
3015 | } | |
3016 | ||
991d9fa0 JT |
3017 | static struct target_type thin_target = { |
3018 | .name = "thin", | |
94563bad | 3019 | .version = {1, 9, 0}, |
991d9fa0 JT |
3020 | .module = THIS_MODULE, |
3021 | .ctr = thin_ctr, | |
3022 | .dtr = thin_dtr, | |
3023 | .map = thin_map, | |
eb2aa48d | 3024 | .end_io = thin_endio, |
991d9fa0 JT |
3025 | .postsuspend = thin_postsuspend, |
3026 | .status = thin_status, | |
3027 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3028 | }; |
3029 | ||
3030 | /*----------------------------------------------------------------*/ | |
3031 | ||
3032 | static int __init dm_thin_init(void) | |
3033 | { | |
3034 | int r; | |
3035 | ||
3036 | pool_table_init(); | |
3037 | ||
3038 | r = dm_register_target(&thin_target); | |
3039 | if (r) | |
3040 | return r; | |
3041 | ||
3042 | r = dm_register_target(&pool_target); | |
3043 | if (r) | |
a24c2569 MS |
3044 | goto bad_pool_target; |
3045 | ||
3046 | r = -ENOMEM; | |
3047 | ||
a24c2569 MS |
3048 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3049 | if (!_new_mapping_cache) | |
3050 | goto bad_new_mapping_cache; | |
3051 | ||
a24c2569 MS |
3052 | return 0; |
3053 | ||
a24c2569 | 3054 | bad_new_mapping_cache: |
a24c2569 MS |
3055 | dm_unregister_target(&pool_target); |
3056 | bad_pool_target: | |
3057 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3058 | |
3059 | return r; | |
3060 | } | |
3061 | ||
3062 | static void dm_thin_exit(void) | |
3063 | { | |
3064 | dm_unregister_target(&thin_target); | |
3065 | dm_unregister_target(&pool_target); | |
a24c2569 | 3066 | |
a24c2569 | 3067 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3068 | } |
3069 | ||
3070 | module_init(dm_thin_init); | |
3071 | module_exit(dm_thin_exit); | |
3072 | ||
7cab8bf1 | 3073 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3074 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3075 | MODULE_LICENSE("GPL"); |