]>
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 | ||
eb2aa48d JT |
512 | unsigned quiesced:1; |
513 | unsigned prepared:1; | |
104655fd | 514 | unsigned pass_discard:1; |
991d9fa0 JT |
515 | |
516 | struct thin_c *tc; | |
517 | dm_block_t virt_block; | |
518 | dm_block_t data_block; | |
a24c2569 | 519 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
520 | int err; |
521 | ||
522 | /* | |
523 | * If the bio covers the whole area of a block then we can avoid | |
524 | * zeroing or copying. Instead this bio is hooked. The bio will | |
525 | * still be in the cell, so care has to be taken to avoid issuing | |
526 | * the bio twice. | |
527 | */ | |
528 | struct bio *bio; | |
529 | bio_end_io_t *saved_bi_end_io; | |
530 | }; | |
531 | ||
a24c2569 | 532 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
533 | { |
534 | struct pool *pool = m->tc->pool; | |
535 | ||
eb2aa48d | 536 | if (m->quiesced && m->prepared) { |
991d9fa0 JT |
537 | list_add(&m->list, &pool->prepared_mappings); |
538 | wake_worker(pool); | |
539 | } | |
540 | } | |
541 | ||
542 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
543 | { | |
544 | unsigned long flags; | |
a24c2569 | 545 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
546 | struct pool *pool = m->tc->pool; |
547 | ||
548 | m->err = read_err || write_err ? -EIO : 0; | |
549 | ||
550 | spin_lock_irqsave(&pool->lock, flags); | |
551 | m->prepared = 1; | |
552 | __maybe_add_mapping(m); | |
553 | spin_unlock_irqrestore(&pool->lock, flags); | |
554 | } | |
555 | ||
556 | static void overwrite_endio(struct bio *bio, int err) | |
557 | { | |
558 | unsigned long flags; | |
59c3d2c6 | 559 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 560 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
561 | struct pool *pool = m->tc->pool; |
562 | ||
563 | m->err = err; | |
564 | ||
565 | spin_lock_irqsave(&pool->lock, flags); | |
566 | m->prepared = 1; | |
567 | __maybe_add_mapping(m); | |
568 | spin_unlock_irqrestore(&pool->lock, flags); | |
569 | } | |
570 | ||
991d9fa0 JT |
571 | /*----------------------------------------------------------------*/ |
572 | ||
573 | /* | |
574 | * Workqueue. | |
575 | */ | |
576 | ||
577 | /* | |
578 | * Prepared mapping jobs. | |
579 | */ | |
580 | ||
581 | /* | |
582 | * This sends the bios in the cell back to the deferred_bios list. | |
583 | */ | |
2aab3850 | 584 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
585 | { |
586 | struct pool *pool = tc->pool; | |
587 | unsigned long flags; | |
588 | ||
589 | spin_lock_irqsave(&pool->lock, flags); | |
6beca5eb | 590 | cell_release(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
591 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
592 | ||
593 | wake_worker(pool); | |
594 | } | |
595 | ||
596 | /* | |
6beca5eb | 597 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 598 | */ |
f286ba0e | 599 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 600 | { |
991d9fa0 JT |
601 | struct pool *pool = tc->pool; |
602 | unsigned long flags; | |
603 | ||
991d9fa0 | 604 | spin_lock_irqsave(&pool->lock, flags); |
6beca5eb | 605 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
606 | spin_unlock_irqrestore(&pool->lock, flags); |
607 | ||
608 | wake_worker(pool); | |
609 | } | |
610 | ||
e49e5829 JT |
611 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
612 | { | |
613 | if (m->bio) | |
614 | m->bio->bi_end_io = m->saved_bi_end_io; | |
6beca5eb | 615 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
616 | list_del(&m->list); |
617 | mempool_free(m, m->tc->pool->mapping_pool); | |
618 | } | |
025b9685 | 619 | |
a24c2569 | 620 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
621 | { |
622 | struct thin_c *tc = m->tc; | |
6beca5eb | 623 | struct pool *pool = tc->pool; |
991d9fa0 JT |
624 | struct bio *bio; |
625 | int r; | |
626 | ||
627 | bio = m->bio; | |
628 | if (bio) | |
629 | bio->bi_end_io = m->saved_bi_end_io; | |
630 | ||
631 | if (m->err) { | |
6beca5eb | 632 | cell_error(pool, m->cell); |
905386f8 | 633 | goto out; |
991d9fa0 JT |
634 | } |
635 | ||
636 | /* | |
637 | * Commit the prepared block into the mapping btree. | |
638 | * Any I/O for this block arriving after this point will get | |
639 | * remapped to it directly. | |
640 | */ | |
641 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
642 | if (r) { | |
fafc7a81 JT |
643 | DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d", |
644 | dm_device_name(pool->pool_md), r); | |
645 | set_pool_mode(pool, PM_READ_ONLY); | |
6beca5eb | 646 | cell_error(pool, m->cell); |
905386f8 | 647 | goto out; |
991d9fa0 JT |
648 | } |
649 | ||
650 | /* | |
651 | * Release any bios held while the block was being provisioned. | |
652 | * If we are processing a write bio that completely covers the block, | |
653 | * we already processed it so can ignore it now when processing | |
654 | * the bios in the cell. | |
655 | */ | |
656 | if (bio) { | |
f286ba0e | 657 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
658 | bio_endio(bio, 0); |
659 | } else | |
2aab3850 | 660 | cell_defer(tc, m->cell); |
991d9fa0 | 661 | |
905386f8 | 662 | out: |
991d9fa0 | 663 | list_del(&m->list); |
6beca5eb | 664 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
665 | } |
666 | ||
e49e5829 | 667 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 668 | { |
104655fd JT |
669 | struct thin_c *tc = m->tc; |
670 | ||
e49e5829 | 671 | bio_io_error(m->bio); |
f286ba0e JT |
672 | cell_defer_no_holder(tc, m->cell); |
673 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
674 | mempool_free(m, tc->pool->mapping_pool); |
675 | } | |
676 | ||
677 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
678 | { | |
679 | struct thin_c *tc = m->tc; | |
104655fd | 680 | |
e8088073 | 681 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
682 | cell_defer_no_holder(tc, m->cell); |
683 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 684 | |
104655fd JT |
685 | if (m->pass_discard) |
686 | remap_and_issue(tc, m->bio, m->data_block); | |
687 | else | |
688 | bio_endio(m->bio, 0); | |
689 | ||
104655fd JT |
690 | mempool_free(m, tc->pool->mapping_pool); |
691 | } | |
692 | ||
e49e5829 JT |
693 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
694 | { | |
695 | int r; | |
696 | struct thin_c *tc = m->tc; | |
697 | ||
698 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
699 | if (r) | |
c397741c | 700 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
701 | |
702 | process_prepared_discard_passdown(m); | |
703 | } | |
704 | ||
104655fd | 705 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 706 | process_mapping_fn *fn) |
991d9fa0 JT |
707 | { |
708 | unsigned long flags; | |
709 | struct list_head maps; | |
a24c2569 | 710 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
711 | |
712 | INIT_LIST_HEAD(&maps); | |
713 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 714 | list_splice_init(head, &maps); |
991d9fa0 JT |
715 | spin_unlock_irqrestore(&pool->lock, flags); |
716 | ||
717 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 718 | (*fn)(m); |
991d9fa0 JT |
719 | } |
720 | ||
721 | /* | |
722 | * Deferred bio jobs. | |
723 | */ | |
104655fd | 724 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 725 | { |
f9a8e0cd | 726 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
727 | } |
728 | ||
729 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
730 | { | |
731 | return (bio_data_dir(bio) == WRITE) && | |
732 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
733 | } |
734 | ||
735 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
736 | bio_end_io_t *fn) | |
737 | { | |
738 | *save = bio->bi_end_io; | |
739 | bio->bi_end_io = fn; | |
740 | } | |
741 | ||
742 | static int ensure_next_mapping(struct pool *pool) | |
743 | { | |
744 | if (pool->next_mapping) | |
745 | return 0; | |
746 | ||
747 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
748 | ||
749 | return pool->next_mapping ? 0 : -ENOMEM; | |
750 | } | |
751 | ||
a24c2569 | 752 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 753 | { |
16961b04 | 754 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
755 | |
756 | BUG_ON(!pool->next_mapping); | |
757 | ||
16961b04 MS |
758 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
759 | INIT_LIST_HEAD(&m->list); | |
760 | m->bio = NULL; | |
761 | ||
991d9fa0 JT |
762 | pool->next_mapping = NULL; |
763 | ||
16961b04 | 764 | return m; |
991d9fa0 JT |
765 | } |
766 | ||
767 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
768 | struct dm_dev *origin, dm_block_t data_origin, |
769 | dm_block_t data_dest, | |
a24c2569 | 770 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
771 | { |
772 | int r; | |
773 | struct pool *pool = tc->pool; | |
a24c2569 | 774 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 775 | |
991d9fa0 JT |
776 | m->tc = tc; |
777 | m->virt_block = virt_block; | |
778 | m->data_block = data_dest; | |
779 | m->cell = cell; | |
991d9fa0 | 780 | |
44feb387 | 781 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
eb2aa48d | 782 | m->quiesced = 1; |
991d9fa0 JT |
783 | |
784 | /* | |
785 | * IO to pool_dev remaps to the pool target's data_dev. | |
786 | * | |
787 | * If the whole block of data is being overwritten, we can issue the | |
788 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
789 | */ | |
790 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 791 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 792 | |
eb2aa48d | 793 | h->overwrite_mapping = m; |
991d9fa0 JT |
794 | m->bio = bio; |
795 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 796 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
797 | remap_and_issue(tc, bio, data_dest); |
798 | } else { | |
799 | struct dm_io_region from, to; | |
800 | ||
2dd9c257 | 801 | from.bdev = origin->bdev; |
991d9fa0 JT |
802 | from.sector = data_origin * pool->sectors_per_block; |
803 | from.count = pool->sectors_per_block; | |
804 | ||
805 | to.bdev = tc->pool_dev->bdev; | |
806 | to.sector = data_dest * pool->sectors_per_block; | |
807 | to.count = pool->sectors_per_block; | |
808 | ||
809 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
810 | 0, copy_complete, m); | |
811 | if (r < 0) { | |
812 | mempool_free(m, pool->mapping_pool); | |
c397741c | 813 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 814 | cell_error(pool, cell); |
991d9fa0 JT |
815 | } |
816 | } | |
817 | } | |
818 | ||
2dd9c257 JT |
819 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
820 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 821 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
822 | { |
823 | schedule_copy(tc, virt_block, tc->pool_dev, | |
824 | data_origin, data_dest, cell, bio); | |
825 | } | |
826 | ||
827 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
828 | dm_block_t data_dest, | |
a24c2569 | 829 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
830 | { |
831 | schedule_copy(tc, virt_block, tc->origin_dev, | |
832 | virt_block, data_dest, cell, bio); | |
833 | } | |
834 | ||
991d9fa0 | 835 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 836 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
837 | struct bio *bio) |
838 | { | |
839 | struct pool *pool = tc->pool; | |
a24c2569 | 840 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 841 | |
eb2aa48d | 842 | m->quiesced = 1; |
991d9fa0 JT |
843 | m->prepared = 0; |
844 | m->tc = tc; | |
845 | m->virt_block = virt_block; | |
846 | m->data_block = data_block; | |
847 | m->cell = cell; | |
991d9fa0 JT |
848 | |
849 | /* | |
850 | * If the whole block of data is being overwritten or we are not | |
851 | * zeroing pre-existing data, we can issue the bio immediately. | |
852 | * Otherwise we use kcopyd to zero the data first. | |
853 | */ | |
67e2e2b2 | 854 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
855 | process_prepared_mapping(m); |
856 | ||
857 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 858 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 859 | |
eb2aa48d | 860 | h->overwrite_mapping = m; |
991d9fa0 JT |
861 | m->bio = bio; |
862 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 863 | inc_all_io_entry(pool, bio); |
991d9fa0 | 864 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
865 | } else { |
866 | int r; | |
867 | struct dm_io_region to; | |
868 | ||
869 | to.bdev = tc->pool_dev->bdev; | |
870 | to.sector = data_block * pool->sectors_per_block; | |
871 | to.count = pool->sectors_per_block; | |
872 | ||
873 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
874 | if (r < 0) { | |
875 | mempool_free(m, pool->mapping_pool); | |
c397741c | 876 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 877 | cell_error(pool, cell); |
991d9fa0 JT |
878 | } |
879 | } | |
880 | } | |
881 | ||
e49e5829 JT |
882 | /* |
883 | * A non-zero return indicates read_only or fail_io mode. | |
884 | * Many callers don't care about the return value. | |
885 | */ | |
020cc3b5 | 886 | static int commit(struct pool *pool) |
e49e5829 JT |
887 | { |
888 | int r; | |
889 | ||
890 | if (get_pool_mode(pool) != PM_WRITE) | |
891 | return -EINVAL; | |
892 | ||
020cc3b5 JT |
893 | r = dm_pool_commit_metadata(pool->pmd); |
894 | if (r) { | |
895 | DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d", | |
896 | dm_device_name(pool->pool_md), r); | |
e49e5829 | 897 | set_pool_mode(pool, PM_READ_ONLY); |
020cc3b5 | 898 | } |
e49e5829 JT |
899 | |
900 | return r; | |
901 | } | |
902 | ||
991d9fa0 JT |
903 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
904 | { | |
905 | int r; | |
906 | dm_block_t free_blocks; | |
907 | unsigned long flags; | |
908 | struct pool *pool = tc->pool; | |
909 | ||
94563bad MS |
910 | /* |
911 | * Once no_free_space is set we must not allow allocation to succeed. | |
912 | * Otherwise it is difficult to explain, debug, test and support. | |
913 | */ | |
914 | if (pool->no_free_space) | |
915 | return -ENOSPC; | |
916 | ||
991d9fa0 JT |
917 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
918 | if (r) | |
919 | return r; | |
920 | ||
921 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
b17446df | 922 | DMWARN("%s: reached low water mark for data device: sending event.", |
991d9fa0 JT |
923 | dm_device_name(pool->pool_md)); |
924 | spin_lock_irqsave(&pool->lock, flags); | |
925 | pool->low_water_triggered = 1; | |
926 | spin_unlock_irqrestore(&pool->lock, flags); | |
927 | dm_table_event(pool->ti->table); | |
928 | } | |
929 | ||
930 | if (!free_blocks) { | |
94563bad MS |
931 | /* |
932 | * Try to commit to see if that will free up some | |
933 | * more space. | |
934 | */ | |
020cc3b5 JT |
935 | r = commit(pool); |
936 | if (r) | |
937 | return r; | |
991d9fa0 | 938 | |
94563bad MS |
939 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
940 | if (r) | |
941 | return r; | |
991d9fa0 | 942 | |
94563bad MS |
943 | /* |
944 | * If we still have no space we set a flag to avoid | |
945 | * doing all this checking and return -ENOSPC. This | |
946 | * flag serves as a latch that disallows allocations from | |
947 | * this pool until the admin takes action (e.g. resize or | |
948 | * table reload). | |
949 | */ | |
950 | if (!free_blocks) { | |
4a02b34e | 951 | DMWARN("%s: no free data space available.", |
94563bad MS |
952 | dm_device_name(pool->pool_md)); |
953 | spin_lock_irqsave(&pool->lock, flags); | |
954 | pool->no_free_space = 1; | |
955 | spin_unlock_irqrestore(&pool->lock, flags); | |
956 | return -ENOSPC; | |
991d9fa0 JT |
957 | } |
958 | } | |
959 | ||
960 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e MS |
961 | if (r) { |
962 | if (r == -ENOSPC && | |
963 | !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) && | |
964 | !free_blocks) { | |
965 | DMWARN("%s: no free metadata space available.", | |
966 | dm_device_name(pool->pool_md)); | |
967 | set_pool_mode(pool, PM_READ_ONLY); | |
968 | } | |
991d9fa0 | 969 | return r; |
4a02b34e | 970 | } |
991d9fa0 JT |
971 | |
972 | return 0; | |
973 | } | |
974 | ||
975 | /* | |
976 | * If we have run out of space, queue bios until the device is | |
977 | * resumed, presumably after having been reloaded with more space. | |
978 | */ | |
979 | static void retry_on_resume(struct bio *bio) | |
980 | { | |
59c3d2c6 | 981 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 982 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
983 | struct pool *pool = tc->pool; |
984 | unsigned long flags; | |
985 | ||
986 | spin_lock_irqsave(&pool->lock, flags); | |
987 | bio_list_add(&pool->retry_on_resume_list, bio); | |
988 | spin_unlock_irqrestore(&pool->lock, flags); | |
989 | } | |
990 | ||
6beca5eb | 991 | static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
992 | { |
993 | struct bio *bio; | |
994 | struct bio_list bios; | |
995 | ||
996 | bio_list_init(&bios); | |
6beca5eb | 997 | cell_release(pool, cell, &bios); |
991d9fa0 JT |
998 | |
999 | while ((bio = bio_list_pop(&bios))) | |
1000 | retry_on_resume(bio); | |
1001 | } | |
1002 | ||
104655fd JT |
1003 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1004 | { | |
1005 | int r; | |
c3a0ce2e | 1006 | unsigned long flags; |
104655fd | 1007 | struct pool *pool = tc->pool; |
a24c2569 | 1008 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1009 | struct dm_cell_key key, key2; |
104655fd JT |
1010 | dm_block_t block = get_bio_block(tc, bio); |
1011 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1012 | struct dm_thin_new_mapping *m; |
104655fd JT |
1013 | |
1014 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1015 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1016 | return; |
1017 | ||
1018 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1019 | switch (r) { | |
1020 | case 0: | |
1021 | /* | |
1022 | * Check nobody is fiddling with this pool block. This can | |
1023 | * happen if someone's in the process of breaking sharing | |
1024 | * on this block. | |
1025 | */ | |
1026 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1027 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1028 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1029 | break; |
1030 | } | |
1031 | ||
1032 | if (io_overlaps_block(pool, bio)) { | |
1033 | /* | |
1034 | * IO may still be going to the destination block. We must | |
1035 | * quiesce before we can do the removal. | |
1036 | */ | |
1037 | m = get_next_mapping(pool); | |
1038 | m->tc = tc; | |
17b7d63f | 1039 | m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown; |
104655fd JT |
1040 | m->virt_block = block; |
1041 | m->data_block = lookup_result.block; | |
1042 | m->cell = cell; | |
1043 | m->cell2 = cell2; | |
104655fd JT |
1044 | m->bio = bio; |
1045 | ||
44feb387 | 1046 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1047 | spin_lock_irqsave(&pool->lock, flags); |
104655fd | 1048 | list_add(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1049 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1050 | wake_worker(pool); |
1051 | } | |
1052 | } else { | |
e8088073 | 1053 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1054 | cell_defer_no_holder(tc, cell); |
1055 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1056 | |
104655fd | 1057 | /* |
49296309 MP |
1058 | * The DM core makes sure that the discard doesn't span |
1059 | * a block boundary. So we submit the discard of a | |
1060 | * partial block appropriately. | |
104655fd | 1061 | */ |
650d2a06 MP |
1062 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1063 | remap_and_issue(tc, bio, lookup_result.block); | |
1064 | else | |
1065 | bio_endio(bio, 0); | |
104655fd JT |
1066 | } |
1067 | break; | |
1068 | ||
1069 | case -ENODATA: | |
1070 | /* | |
1071 | * It isn't provisioned, just forget it. | |
1072 | */ | |
f286ba0e | 1073 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1074 | bio_endio(bio, 0); |
1075 | break; | |
1076 | ||
1077 | default: | |
c397741c MS |
1078 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1079 | __func__, r); | |
f286ba0e | 1080 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1081 | bio_io_error(bio); |
1082 | break; | |
1083 | } | |
1084 | } | |
1085 | ||
991d9fa0 | 1086 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1087 | struct dm_cell_key *key, |
991d9fa0 | 1088 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1089 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1090 | { |
1091 | int r; | |
1092 | dm_block_t data_block; | |
d6fc2042 | 1093 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1094 | |
1095 | r = alloc_data_block(tc, &data_block); | |
1096 | switch (r) { | |
1097 | case 0: | |
2dd9c257 JT |
1098 | schedule_internal_copy(tc, block, lookup_result->block, |
1099 | data_block, cell, bio); | |
991d9fa0 JT |
1100 | break; |
1101 | ||
1102 | case -ENOSPC: | |
d6fc2042 | 1103 | no_space(pool, cell); |
991d9fa0 JT |
1104 | break; |
1105 | ||
1106 | default: | |
c397741c MS |
1107 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1108 | __func__, r); | |
d6fc2042 MS |
1109 | set_pool_mode(pool, PM_READ_ONLY); |
1110 | cell_error(pool, cell); | |
991d9fa0 JT |
1111 | break; |
1112 | } | |
1113 | } | |
1114 | ||
1115 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1116 | dm_block_t block, | |
1117 | struct dm_thin_lookup_result *lookup_result) | |
1118 | { | |
a24c2569 | 1119 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1120 | struct pool *pool = tc->pool; |
44feb387 | 1121 | struct dm_cell_key key; |
991d9fa0 JT |
1122 | |
1123 | /* | |
1124 | * If cell is already occupied, then sharing is already in the process | |
1125 | * of being broken so we have nothing further to do here. | |
1126 | */ | |
1127 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1128 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1129 | return; |
1130 | ||
60049701 | 1131 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1132 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1133 | else { | |
59c3d2c6 | 1134 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1135 | |
44feb387 | 1136 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1137 | inc_all_io_entry(pool, bio); |
f286ba0e | 1138 | cell_defer_no_holder(tc, cell); |
e8088073 | 1139 | |
991d9fa0 JT |
1140 | remap_and_issue(tc, bio, lookup_result->block); |
1141 | } | |
1142 | } | |
1143 | ||
1144 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1145 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1146 | { |
1147 | int r; | |
1148 | dm_block_t data_block; | |
6beca5eb | 1149 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1150 | |
1151 | /* | |
1152 | * Remap empty bios (flushes) immediately, without provisioning. | |
1153 | */ | |
1154 | if (!bio->bi_size) { | |
6beca5eb | 1155 | inc_all_io_entry(pool, bio); |
f286ba0e | 1156 | cell_defer_no_holder(tc, cell); |
e8088073 | 1157 | |
991d9fa0 JT |
1158 | remap_and_issue(tc, bio, 0); |
1159 | return; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Fill read bios with zeroes and complete them immediately. | |
1164 | */ | |
1165 | if (bio_data_dir(bio) == READ) { | |
1166 | zero_fill_bio(bio); | |
f286ba0e | 1167 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1168 | bio_endio(bio, 0); |
1169 | return; | |
1170 | } | |
1171 | ||
1172 | r = alloc_data_block(tc, &data_block); | |
1173 | switch (r) { | |
1174 | case 0: | |
2dd9c257 JT |
1175 | if (tc->origin_dev) |
1176 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1177 | else | |
1178 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1179 | break; |
1180 | ||
1181 | case -ENOSPC: | |
6beca5eb | 1182 | no_space(pool, cell); |
991d9fa0 JT |
1183 | break; |
1184 | ||
1185 | default: | |
c397741c MS |
1186 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1187 | __func__, r); | |
6beca5eb JT |
1188 | set_pool_mode(pool, PM_READ_ONLY); |
1189 | cell_error(pool, cell); | |
991d9fa0 JT |
1190 | break; |
1191 | } | |
1192 | } | |
1193 | ||
1194 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1195 | { | |
1196 | int r; | |
6beca5eb | 1197 | struct pool *pool = tc->pool; |
991d9fa0 | 1198 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1199 | struct dm_bio_prison_cell *cell; |
44feb387 | 1200 | struct dm_cell_key key; |
991d9fa0 JT |
1201 | struct dm_thin_lookup_result lookup_result; |
1202 | ||
1203 | /* | |
1204 | * If cell is already occupied, then the block is already | |
1205 | * being provisioned so we have nothing further to do here. | |
1206 | */ | |
1207 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1208 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1209 | return; |
1210 | ||
1211 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1212 | switch (r) { | |
1213 | case 0: | |
e8088073 | 1214 | if (lookup_result.shared) { |
991d9fa0 | 1215 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1216 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1217 | } else { |
6beca5eb | 1218 | inc_all_io_entry(pool, bio); |
f286ba0e | 1219 | cell_defer_no_holder(tc, cell); |
e8088073 | 1220 | |
991d9fa0 | 1221 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1222 | } |
991d9fa0 JT |
1223 | break; |
1224 | ||
1225 | case -ENODATA: | |
2dd9c257 | 1226 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1227 | inc_all_io_entry(pool, bio); |
f286ba0e | 1228 | cell_defer_no_holder(tc, cell); |
e8088073 | 1229 | |
2dd9c257 JT |
1230 | remap_to_origin_and_issue(tc, bio); |
1231 | } else | |
1232 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1233 | break; |
1234 | ||
1235 | default: | |
c397741c MS |
1236 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1237 | __func__, r); | |
f286ba0e | 1238 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1239 | bio_io_error(bio); |
1240 | break; | |
1241 | } | |
1242 | } | |
1243 | ||
e49e5829 JT |
1244 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1245 | { | |
1246 | int r; | |
1247 | int rw = bio_data_dir(bio); | |
1248 | dm_block_t block = get_bio_block(tc, bio); | |
1249 | struct dm_thin_lookup_result lookup_result; | |
1250 | ||
1251 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1252 | switch (r) { | |
1253 | case 0: | |
1254 | if (lookup_result.shared && (rw == WRITE) && bio->bi_size) | |
1255 | bio_io_error(bio); | |
e8088073 JT |
1256 | else { |
1257 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1258 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1259 | } |
e49e5829 JT |
1260 | break; |
1261 | ||
1262 | case -ENODATA: | |
1263 | if (rw != READ) { | |
1264 | bio_io_error(bio); | |
1265 | break; | |
1266 | } | |
1267 | ||
1268 | if (tc->origin_dev) { | |
e8088073 | 1269 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1270 | remap_to_origin_and_issue(tc, bio); |
1271 | break; | |
1272 | } | |
1273 | ||
1274 | zero_fill_bio(bio); | |
1275 | bio_endio(bio, 0); | |
1276 | break; | |
1277 | ||
1278 | default: | |
c397741c MS |
1279 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1280 | __func__, r); | |
e49e5829 JT |
1281 | bio_io_error(bio); |
1282 | break; | |
1283 | } | |
1284 | } | |
1285 | ||
1286 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) | |
1287 | { | |
1288 | bio_io_error(bio); | |
1289 | } | |
1290 | ||
ac8c3f3d JT |
1291 | /* |
1292 | * FIXME: should we also commit due to size of transaction, measured in | |
1293 | * metadata blocks? | |
1294 | */ | |
905e51b3 JT |
1295 | static int need_commit_due_to_time(struct pool *pool) |
1296 | { | |
1297 | return jiffies < pool->last_commit_jiffies || | |
1298 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1299 | } | |
1300 | ||
991d9fa0 JT |
1301 | static void process_deferred_bios(struct pool *pool) |
1302 | { | |
1303 | unsigned long flags; | |
1304 | struct bio *bio; | |
1305 | struct bio_list bios; | |
991d9fa0 JT |
1306 | |
1307 | bio_list_init(&bios); | |
1308 | ||
1309 | spin_lock_irqsave(&pool->lock, flags); | |
1310 | bio_list_merge(&bios, &pool->deferred_bios); | |
1311 | bio_list_init(&pool->deferred_bios); | |
1312 | spin_unlock_irqrestore(&pool->lock, flags); | |
1313 | ||
1314 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 1315 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1316 | struct thin_c *tc = h->tc; |
1317 | ||
991d9fa0 JT |
1318 | /* |
1319 | * If we've got no free new_mapping structs, and processing | |
1320 | * this bio might require one, we pause until there are some | |
1321 | * prepared mappings to process. | |
1322 | */ | |
1323 | if (ensure_next_mapping(pool)) { | |
1324 | spin_lock_irqsave(&pool->lock, flags); | |
1325 | bio_list_merge(&pool->deferred_bios, &bios); | |
1326 | spin_unlock_irqrestore(&pool->lock, flags); | |
1327 | ||
1328 | break; | |
1329 | } | |
104655fd JT |
1330 | |
1331 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1332 | pool->process_discard(tc, bio); |
104655fd | 1333 | else |
e49e5829 | 1334 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * If there are any deferred flush bios, we must commit | |
1339 | * the metadata before issuing them. | |
1340 | */ | |
1341 | bio_list_init(&bios); | |
1342 | spin_lock_irqsave(&pool->lock, flags); | |
1343 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1344 | bio_list_init(&pool->deferred_flush_bios); | |
1345 | spin_unlock_irqrestore(&pool->lock, flags); | |
1346 | ||
905e51b3 | 1347 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1348 | return; |
1349 | ||
020cc3b5 | 1350 | if (commit(pool)) { |
991d9fa0 JT |
1351 | while ((bio = bio_list_pop(&bios))) |
1352 | bio_io_error(bio); | |
1353 | return; | |
1354 | } | |
905e51b3 | 1355 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1356 | |
1357 | while ((bio = bio_list_pop(&bios))) | |
1358 | generic_make_request(bio); | |
1359 | } | |
1360 | ||
1361 | static void do_worker(struct work_struct *ws) | |
1362 | { | |
1363 | struct pool *pool = container_of(ws, struct pool, worker); | |
1364 | ||
e49e5829 JT |
1365 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1366 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1367 | process_deferred_bios(pool); |
1368 | } | |
1369 | ||
905e51b3 JT |
1370 | /* |
1371 | * We want to commit periodically so that not too much | |
1372 | * unwritten data builds up. | |
1373 | */ | |
1374 | static void do_waker(struct work_struct *ws) | |
1375 | { | |
1376 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1377 | wake_worker(pool); | |
1378 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1379 | } | |
1380 | ||
991d9fa0 JT |
1381 | /*----------------------------------------------------------------*/ |
1382 | ||
e49e5829 JT |
1383 | static enum pool_mode get_pool_mode(struct pool *pool) |
1384 | { | |
1385 | return pool->pf.mode; | |
1386 | } | |
1387 | ||
1388 | static void set_pool_mode(struct pool *pool, enum pool_mode mode) | |
1389 | { | |
1390 | int r; | |
1391 | ||
1392 | pool->pf.mode = mode; | |
1393 | ||
1394 | switch (mode) { | |
1395 | case PM_FAIL: | |
4fa5971a MS |
1396 | DMERR("%s: switching pool to failure mode", |
1397 | dm_device_name(pool->pool_md)); | |
5383ef3a | 1398 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1399 | pool->process_bio = process_bio_fail; |
1400 | pool->process_discard = process_bio_fail; | |
1401 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1402 | pool->process_prepared_discard = process_prepared_discard_fail; | |
1403 | break; | |
1404 | ||
1405 | case PM_READ_ONLY: | |
4fa5971a MS |
1406 | DMERR("%s: switching pool to read-only mode", |
1407 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1408 | r = dm_pool_abort_metadata(pool->pmd); |
1409 | if (r) { | |
4fa5971a MS |
1410 | DMERR("%s: aborting transaction failed", |
1411 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1412 | set_pool_mode(pool, PM_FAIL); |
1413 | } else { | |
1414 | dm_pool_metadata_read_only(pool->pmd); | |
1415 | pool->process_bio = process_bio_read_only; | |
1416 | pool->process_discard = process_discard; | |
1417 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1418 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1419 | } | |
1420 | break; | |
1421 | ||
1422 | case PM_WRITE: | |
9b7aaa64 | 1423 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1424 | pool->process_bio = process_bio; |
1425 | pool->process_discard = process_discard; | |
1426 | pool->process_prepared_mapping = process_prepared_mapping; | |
1427 | pool->process_prepared_discard = process_prepared_discard; | |
1428 | break; | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | /*----------------------------------------------------------------*/ | |
1433 | ||
991d9fa0 JT |
1434 | /* |
1435 | * Mapping functions. | |
1436 | */ | |
1437 | ||
1438 | /* | |
1439 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1440 | */ | |
1441 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1442 | { | |
1443 | unsigned long flags; | |
1444 | struct pool *pool = tc->pool; | |
1445 | ||
1446 | spin_lock_irqsave(&pool->lock, flags); | |
1447 | bio_list_add(&pool->deferred_bios, bio); | |
1448 | spin_unlock_irqrestore(&pool->lock, flags); | |
1449 | ||
1450 | wake_worker(pool); | |
1451 | } | |
1452 | ||
59c3d2c6 | 1453 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1454 | { |
59c3d2c6 | 1455 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1456 | |
1457 | h->tc = tc; | |
1458 | h->shared_read_entry = NULL; | |
e8088073 | 1459 | h->all_io_entry = NULL; |
eb2aa48d | 1460 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1461 | } |
1462 | ||
991d9fa0 JT |
1463 | /* |
1464 | * Non-blocking function called from the thin target's map function. | |
1465 | */ | |
7de3ee57 | 1466 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1467 | { |
1468 | int r; | |
1469 | struct thin_c *tc = ti->private; | |
1470 | dm_block_t block = get_bio_block(tc, bio); | |
1471 | struct dm_thin_device *td = tc->td; | |
1472 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1473 | struct dm_bio_prison_cell cell1, cell2; |
1474 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1475 | struct dm_cell_key key; |
991d9fa0 | 1476 | |
59c3d2c6 | 1477 | thin_hook_bio(tc, bio); |
e49e5829 JT |
1478 | |
1479 | if (get_pool_mode(tc->pool) == PM_FAIL) { | |
1480 | bio_io_error(bio); | |
1481 | return DM_MAPIO_SUBMITTED; | |
1482 | } | |
1483 | ||
104655fd | 1484 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1485 | thin_defer_bio(tc, bio); |
1486 | return DM_MAPIO_SUBMITTED; | |
1487 | } | |
1488 | ||
1489 | r = dm_thin_find_block(td, block, 0, &result); | |
1490 | ||
1491 | /* | |
1492 | * Note that we defer readahead too. | |
1493 | */ | |
1494 | switch (r) { | |
1495 | case 0: | |
1496 | if (unlikely(result.shared)) { | |
1497 | /* | |
1498 | * We have a race condition here between the | |
1499 | * result.shared value returned by the lookup and | |
1500 | * snapshot creation, which may cause new | |
1501 | * sharing. | |
1502 | * | |
1503 | * To avoid this always quiesce the origin before | |
1504 | * taking the snap. You want to do this anyway to | |
1505 | * ensure a consistent application view | |
1506 | * (i.e. lockfs). | |
1507 | * | |
1508 | * More distant ancestors are irrelevant. The | |
1509 | * shared flag will be set in their case. | |
1510 | */ | |
1511 | thin_defer_bio(tc, bio); | |
e8088073 | 1512 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1513 | } |
e8088073 JT |
1514 | |
1515 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1516 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1517 | return DM_MAPIO_SUBMITTED; |
1518 | ||
1519 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1520 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1521 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1522 | return DM_MAPIO_SUBMITTED; |
1523 | } | |
1524 | ||
1525 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1526 | cell_defer_no_holder_no_free(tc, &cell2); |
1527 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1528 | |
1529 | remap(tc, bio, result.block); | |
1530 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1531 | |
1532 | case -ENODATA: | |
e49e5829 JT |
1533 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1534 | /* | |
1535 | * This block isn't provisioned, and we have no way | |
1536 | * of doing so. Just error it. | |
1537 | */ | |
1538 | bio_io_error(bio); | |
2aab3850 | 1539 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1540 | } |
1541 | /* fall through */ | |
1542 | ||
1543 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1544 | /* |
1545 | * In future, the failed dm_thin_find_block above could | |
1546 | * provide the hint to load the metadata into cache. | |
1547 | */ | |
991d9fa0 | 1548 | thin_defer_bio(tc, bio); |
2aab3850 | 1549 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1550 | |
1551 | default: | |
1552 | /* | |
1553 | * Must always call bio_io_error on failure. | |
1554 | * dm_thin_find_block can fail with -EINVAL if the | |
1555 | * pool is switched to fail-io mode. | |
1556 | */ | |
1557 | bio_io_error(bio); | |
2aab3850 | 1558 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1559 | } |
991d9fa0 JT |
1560 | } |
1561 | ||
1562 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1563 | { | |
1564 | int r; | |
1565 | unsigned long flags; | |
1566 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1567 | ||
1568 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1569 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1570 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1571 | ||
1572 | if (!r) { | |
1573 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1574 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1575 | } | |
1576 | ||
1577 | return r; | |
1578 | } | |
1579 | ||
1580 | static void __requeue_bios(struct pool *pool) | |
1581 | { | |
1582 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1583 | bio_list_init(&pool->retry_on_resume_list); | |
1584 | } | |
1585 | ||
1586 | /*---------------------------------------------------------------- | |
1587 | * Binding of control targets to a pool object | |
1588 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1589 | static bool data_dev_supports_discard(struct pool_c *pt) |
1590 | { | |
1591 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1592 | ||
1593 | return q && blk_queue_discard(q); | |
1594 | } | |
1595 | ||
58051b94 JT |
1596 | static bool is_factor(sector_t block_size, uint32_t n) |
1597 | { | |
1598 | return !sector_div(block_size, n); | |
1599 | } | |
1600 | ||
9bc142dd MS |
1601 | /* |
1602 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1603 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1604 | */ |
0424caa1 | 1605 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1606 | { |
0424caa1 MS |
1607 | struct pool *pool = pt->pool; |
1608 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1609 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1610 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1611 | const char *reason = NULL; | |
9bc142dd MS |
1612 | char buf[BDEVNAME_SIZE]; |
1613 | ||
0424caa1 | 1614 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1615 | return; |
1616 | ||
0424caa1 MS |
1617 | if (!data_dev_supports_discard(pt)) |
1618 | reason = "discard unsupported"; | |
1619 | ||
1620 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1621 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1622 | |
0424caa1 MS |
1623 | else if (data_limits->discard_granularity > block_size) |
1624 | reason = "discard granularity larger than a block"; | |
1625 | ||
58051b94 | 1626 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1627 | reason = "discard granularity not a factor of block size"; |
1628 | ||
1629 | if (reason) { | |
1630 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1631 | pt->adjusted_pf.discard_passdown = false; | |
1632 | } | |
9bc142dd MS |
1633 | } |
1634 | ||
991d9fa0 JT |
1635 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1636 | { | |
1637 | struct pool_c *pt = ti->private; | |
1638 | ||
e49e5829 | 1639 | /* |
9b7aaa64 | 1640 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 JT |
1641 | */ |
1642 | enum pool_mode old_mode = pool->pf.mode; | |
0424caa1 | 1643 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1644 | |
9b7aaa64 JT |
1645 | /* |
1646 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1647 | * not going to recover without a thin_repair. So we never let the | |
1648 | * pool move out of the old mode. On the other hand a PM_READ_ONLY | |
1649 | * may have been due to a lack of metadata or data space, and may | |
1650 | * now work (ie. if the underlying devices have been resized). | |
1651 | */ | |
1652 | if (old_mode == PM_FAIL) | |
e49e5829 JT |
1653 | new_mode = old_mode; |
1654 | ||
991d9fa0 JT |
1655 | pool->ti = ti; |
1656 | pool->low_water_blocks = pt->low_water_blocks; | |
0424caa1 | 1657 | pool->pf = pt->adjusted_pf; |
991d9fa0 | 1658 | |
9bc142dd | 1659 | set_pool_mode(pool, new_mode); |
f402693d | 1660 | |
991d9fa0 JT |
1661 | return 0; |
1662 | } | |
1663 | ||
1664 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1665 | { | |
1666 | if (pool->ti == ti) | |
1667 | pool->ti = NULL; | |
1668 | } | |
1669 | ||
1670 | /*---------------------------------------------------------------- | |
1671 | * Pool creation | |
1672 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1673 | /* Initialize pool features. */ |
1674 | static void pool_features_init(struct pool_features *pf) | |
1675 | { | |
e49e5829 | 1676 | pf->mode = PM_WRITE; |
9bc142dd MS |
1677 | pf->zero_new_blocks = true; |
1678 | pf->discard_enabled = true; | |
1679 | pf->discard_passdown = true; | |
67e2e2b2 JT |
1680 | } |
1681 | ||
991d9fa0 JT |
1682 | static void __pool_destroy(struct pool *pool) |
1683 | { | |
1684 | __pool_table_remove(pool); | |
1685 | ||
1686 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1687 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1688 | ||
44feb387 | 1689 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1690 | dm_kcopyd_client_destroy(pool->copier); |
1691 | ||
1692 | if (pool->wq) | |
1693 | destroy_workqueue(pool->wq); | |
1694 | ||
1695 | if (pool->next_mapping) | |
1696 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1697 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1698 | dm_deferred_set_destroy(pool->shared_read_ds); |
1699 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1700 | kfree(pool); |
1701 | } | |
1702 | ||
a24c2569 | 1703 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1704 | |
991d9fa0 JT |
1705 | static struct pool *pool_create(struct mapped_device *pool_md, |
1706 | struct block_device *metadata_dev, | |
e49e5829 JT |
1707 | unsigned long block_size, |
1708 | int read_only, char **error) | |
991d9fa0 JT |
1709 | { |
1710 | int r; | |
1711 | void *err_p; | |
1712 | struct pool *pool; | |
1713 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1714 | bool format_device = read_only ? false : true; |
991d9fa0 | 1715 | |
e49e5829 | 1716 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1717 | if (IS_ERR(pmd)) { |
1718 | *error = "Error creating metadata object"; | |
1719 | return (struct pool *)pmd; | |
1720 | } | |
1721 | ||
1722 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1723 | if (!pool) { | |
1724 | *error = "Error allocating memory for pool"; | |
1725 | err_p = ERR_PTR(-ENOMEM); | |
1726 | goto bad_pool; | |
1727 | } | |
1728 | ||
1729 | pool->pmd = pmd; | |
1730 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1731 | if (block_size & (block_size - 1)) |
1732 | pool->sectors_per_block_shift = -1; | |
1733 | else | |
1734 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1735 | pool->low_water_blocks = 0; |
67e2e2b2 | 1736 | pool_features_init(&pool->pf); |
44feb387 | 1737 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
1738 | if (!pool->prison) { |
1739 | *error = "Error creating pool's bio prison"; | |
1740 | err_p = ERR_PTR(-ENOMEM); | |
1741 | goto bad_prison; | |
1742 | } | |
1743 | ||
df5d2e90 | 1744 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
1745 | if (IS_ERR(pool->copier)) { |
1746 | r = PTR_ERR(pool->copier); | |
1747 | *error = "Error creating pool's kcopyd client"; | |
1748 | err_p = ERR_PTR(r); | |
1749 | goto bad_kcopyd_client; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * Create singlethreaded workqueue that will service all devices | |
1754 | * that use this metadata. | |
1755 | */ | |
1756 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1757 | if (!pool->wq) { | |
1758 | *error = "Error creating pool's workqueue"; | |
1759 | err_p = ERR_PTR(-ENOMEM); | |
1760 | goto bad_wq; | |
1761 | } | |
1762 | ||
1763 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1764 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1765 | spin_lock_init(&pool->lock); |
1766 | bio_list_init(&pool->deferred_bios); | |
1767 | bio_list_init(&pool->deferred_flush_bios); | |
1768 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1769 | INIT_LIST_HEAD(&pool->prepared_discards); |
991d9fa0 JT |
1770 | pool->low_water_triggered = 0; |
1771 | pool->no_free_space = 0; | |
1772 | bio_list_init(&pool->retry_on_resume_list); | |
44feb387 MS |
1773 | |
1774 | pool->shared_read_ds = dm_deferred_set_create(); | |
1775 | if (!pool->shared_read_ds) { | |
1776 | *error = "Error creating pool's shared read deferred set"; | |
1777 | err_p = ERR_PTR(-ENOMEM); | |
1778 | goto bad_shared_read_ds; | |
1779 | } | |
1780 | ||
1781 | pool->all_io_ds = dm_deferred_set_create(); | |
1782 | if (!pool->all_io_ds) { | |
1783 | *error = "Error creating pool's all io deferred set"; | |
1784 | err_p = ERR_PTR(-ENOMEM); | |
1785 | goto bad_all_io_ds; | |
1786 | } | |
991d9fa0 JT |
1787 | |
1788 | pool->next_mapping = NULL; | |
a24c2569 MS |
1789 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1790 | _new_mapping_cache); | |
991d9fa0 JT |
1791 | if (!pool->mapping_pool) { |
1792 | *error = "Error creating pool's mapping mempool"; | |
1793 | err_p = ERR_PTR(-ENOMEM); | |
1794 | goto bad_mapping_pool; | |
1795 | } | |
1796 | ||
991d9fa0 | 1797 | pool->ref_count = 1; |
905e51b3 | 1798 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1799 | pool->pool_md = pool_md; |
1800 | pool->md_dev = metadata_dev; | |
1801 | __pool_table_insert(pool); | |
1802 | ||
1803 | return pool; | |
1804 | ||
991d9fa0 | 1805 | bad_mapping_pool: |
44feb387 MS |
1806 | dm_deferred_set_destroy(pool->all_io_ds); |
1807 | bad_all_io_ds: | |
1808 | dm_deferred_set_destroy(pool->shared_read_ds); | |
1809 | bad_shared_read_ds: | |
991d9fa0 JT |
1810 | destroy_workqueue(pool->wq); |
1811 | bad_wq: | |
1812 | dm_kcopyd_client_destroy(pool->copier); | |
1813 | bad_kcopyd_client: | |
44feb387 | 1814 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1815 | bad_prison: |
1816 | kfree(pool); | |
1817 | bad_pool: | |
1818 | if (dm_pool_metadata_close(pmd)) | |
1819 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1820 | ||
1821 | return err_p; | |
1822 | } | |
1823 | ||
1824 | static void __pool_inc(struct pool *pool) | |
1825 | { | |
1826 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1827 | pool->ref_count++; | |
1828 | } | |
1829 | ||
1830 | static void __pool_dec(struct pool *pool) | |
1831 | { | |
1832 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1833 | BUG_ON(!pool->ref_count); | |
1834 | if (!--pool->ref_count) | |
1835 | __pool_destroy(pool); | |
1836 | } | |
1837 | ||
1838 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
1839 | struct block_device *metadata_dev, | |
e49e5829 JT |
1840 | unsigned long block_size, int read_only, |
1841 | char **error, int *created) | |
991d9fa0 JT |
1842 | { |
1843 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
1844 | ||
1845 | if (pool) { | |
f09996c9 MS |
1846 | if (pool->pool_md != pool_md) { |
1847 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 1848 | return ERR_PTR(-EBUSY); |
f09996c9 | 1849 | } |
991d9fa0 JT |
1850 | __pool_inc(pool); |
1851 | ||
1852 | } else { | |
1853 | pool = __pool_table_lookup(pool_md); | |
1854 | if (pool) { | |
f09996c9 MS |
1855 | if (pool->md_dev != metadata_dev) { |
1856 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 1857 | return ERR_PTR(-EINVAL); |
f09996c9 | 1858 | } |
991d9fa0 JT |
1859 | __pool_inc(pool); |
1860 | ||
67e2e2b2 | 1861 | } else { |
e49e5829 | 1862 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
1863 | *created = 1; |
1864 | } | |
991d9fa0 JT |
1865 | } |
1866 | ||
1867 | return pool; | |
1868 | } | |
1869 | ||
1870 | /*---------------------------------------------------------------- | |
1871 | * Pool target methods | |
1872 | *--------------------------------------------------------------*/ | |
1873 | static void pool_dtr(struct dm_target *ti) | |
1874 | { | |
1875 | struct pool_c *pt = ti->private; | |
1876 | ||
1877 | mutex_lock(&dm_thin_pool_table.mutex); | |
1878 | ||
1879 | unbind_control_target(pt->pool, ti); | |
1880 | __pool_dec(pt->pool); | |
1881 | dm_put_device(ti, pt->metadata_dev); | |
1882 | dm_put_device(ti, pt->data_dev); | |
1883 | kfree(pt); | |
1884 | ||
1885 | mutex_unlock(&dm_thin_pool_table.mutex); | |
1886 | } | |
1887 | ||
991d9fa0 JT |
1888 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
1889 | struct dm_target *ti) | |
1890 | { | |
1891 | int r; | |
1892 | unsigned argc; | |
1893 | const char *arg_name; | |
1894 | ||
1895 | static struct dm_arg _args[] = { | |
67e2e2b2 | 1896 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
1897 | }; |
1898 | ||
1899 | /* | |
1900 | * No feature arguments supplied. | |
1901 | */ | |
1902 | if (!as->argc) | |
1903 | return 0; | |
1904 | ||
1905 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1906 | if (r) | |
1907 | return -EINVAL; | |
1908 | ||
1909 | while (argc && !r) { | |
1910 | arg_name = dm_shift_arg(as); | |
1911 | argc--; | |
1912 | ||
e49e5829 | 1913 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 1914 | pf->zero_new_blocks = false; |
e49e5829 JT |
1915 | |
1916 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 1917 | pf->discard_enabled = false; |
e49e5829 JT |
1918 | |
1919 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 1920 | pf->discard_passdown = false; |
991d9fa0 | 1921 | |
e49e5829 JT |
1922 | else if (!strcasecmp(arg_name, "read_only")) |
1923 | pf->mode = PM_READ_ONLY; | |
1924 | ||
1925 | else { | |
1926 | ti->error = "Unrecognised pool feature requested"; | |
1927 | r = -EINVAL; | |
1928 | break; | |
1929 | } | |
991d9fa0 JT |
1930 | } |
1931 | ||
1932 | return r; | |
1933 | } | |
1934 | ||
ac8c3f3d JT |
1935 | static void metadata_low_callback(void *context) |
1936 | { | |
1937 | struct pool *pool = context; | |
1938 | ||
1939 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
1940 | dm_device_name(pool->pool_md)); | |
1941 | ||
1942 | dm_table_event(pool->ti->table); | |
1943 | } | |
1944 | ||
b17446df JT |
1945 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
1946 | { | |
1947 | sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
1948 | char buffer[BDEVNAME_SIZE]; | |
1949 | ||
1950 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) { | |
1951 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
1952 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
1953 | metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING; | |
1954 | } | |
1955 | ||
1956 | return metadata_dev_size; | |
1957 | } | |
1958 | ||
24347e95 JT |
1959 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
1960 | { | |
1961 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
1962 | ||
1963 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); | |
1964 | ||
1965 | return metadata_dev_size; | |
1966 | } | |
1967 | ||
ac8c3f3d JT |
1968 | /* |
1969 | * When a metadata threshold is crossed a dm event is triggered, and | |
1970 | * userland should respond by growing the metadata device. We could let | |
1971 | * userland set the threshold, like we do with the data threshold, but I'm | |
1972 | * not sure they know enough to do this well. | |
1973 | */ | |
1974 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
1975 | { | |
1976 | /* | |
1977 | * 4M is ample for all ops with the possible exception of thin | |
1978 | * device deletion which is harmless if it fails (just retry the | |
1979 | * delete after you've grown the device). | |
1980 | */ | |
1981 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
1982 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
1983 | } | |
1984 | ||
991d9fa0 JT |
1985 | /* |
1986 | * thin-pool <metadata dev> <data dev> | |
1987 | * <data block size (sectors)> | |
1988 | * <low water mark (blocks)> | |
1989 | * [<#feature args> [<arg>]*] | |
1990 | * | |
1991 | * Optional feature arguments are: | |
1992 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
1993 | * ignore_discard: disable discard |
1994 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
1995 | */ |
1996 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
1997 | { | |
67e2e2b2 | 1998 | int r, pool_created = 0; |
991d9fa0 JT |
1999 | struct pool_c *pt; |
2000 | struct pool *pool; | |
2001 | struct pool_features pf; | |
2002 | struct dm_arg_set as; | |
2003 | struct dm_dev *data_dev; | |
2004 | unsigned long block_size; | |
2005 | dm_block_t low_water_blocks; | |
2006 | struct dm_dev *metadata_dev; | |
5d0db96d | 2007 | fmode_t metadata_mode; |
991d9fa0 JT |
2008 | |
2009 | /* | |
2010 | * FIXME Remove validation from scope of lock. | |
2011 | */ | |
2012 | mutex_lock(&dm_thin_pool_table.mutex); | |
2013 | ||
2014 | if (argc < 4) { | |
2015 | ti->error = "Invalid argument count"; | |
2016 | r = -EINVAL; | |
2017 | goto out_unlock; | |
2018 | } | |
5d0db96d | 2019 | |
991d9fa0 JT |
2020 | as.argc = argc; |
2021 | as.argv = argv; | |
2022 | ||
5d0db96d JT |
2023 | /* |
2024 | * Set default pool features. | |
2025 | */ | |
2026 | pool_features_init(&pf); | |
2027 | ||
2028 | dm_consume_args(&as, 4); | |
2029 | r = parse_pool_features(&as, &pf, ti); | |
2030 | if (r) | |
2031 | goto out_unlock; | |
2032 | ||
2033 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2034 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2035 | if (r) { |
2036 | ti->error = "Error opening metadata block device"; | |
2037 | goto out_unlock; | |
2038 | } | |
2039 | ||
b17446df JT |
2040 | /* |
2041 | * Run for the side-effect of possibly issuing a warning if the | |
2042 | * device is too big. | |
2043 | */ | |
2044 | (void) get_metadata_dev_size(metadata_dev->bdev); | |
991d9fa0 JT |
2045 | |
2046 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2047 | if (r) { | |
2048 | ti->error = "Error getting data device"; | |
2049 | goto out_metadata; | |
2050 | } | |
2051 | ||
2052 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2053 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2054 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2055 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2056 | ti->error = "Invalid block size"; |
2057 | r = -EINVAL; | |
2058 | goto out; | |
2059 | } | |
2060 | ||
2061 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2062 | ti->error = "Invalid low water mark"; | |
2063 | r = -EINVAL; | |
2064 | goto out; | |
2065 | } | |
2066 | ||
991d9fa0 JT |
2067 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2068 | if (!pt) { | |
2069 | r = -ENOMEM; | |
2070 | goto out; | |
2071 | } | |
2072 | ||
2073 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2074 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2075 | if (IS_ERR(pool)) { |
2076 | r = PTR_ERR(pool); | |
2077 | goto out_free_pt; | |
2078 | } | |
2079 | ||
67e2e2b2 JT |
2080 | /* |
2081 | * 'pool_created' reflects whether this is the first table load. | |
2082 | * Top level discard support is not allowed to be changed after | |
2083 | * initial load. This would require a pool reload to trigger thin | |
2084 | * device changes. | |
2085 | */ | |
2086 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2087 | ti->error = "Discard support cannot be disabled once enabled"; | |
2088 | r = -EINVAL; | |
2089 | goto out_flags_changed; | |
2090 | } | |
2091 | ||
991d9fa0 JT |
2092 | pt->pool = pool; |
2093 | pt->ti = ti; | |
2094 | pt->metadata_dev = metadata_dev; | |
2095 | pt->data_dev = data_dev; | |
2096 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2097 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2098 | ti->num_flush_bios = 1; |
9bc142dd | 2099 | |
67e2e2b2 JT |
2100 | /* |
2101 | * Only need to enable discards if the pool should pass | |
2102 | * them down to the data device. The thin device's discard | |
2103 | * processing will cause mappings to be removed from the btree. | |
2104 | */ | |
b60ab990 | 2105 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2106 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2107 | ti->num_discard_bios = 1; |
9bc142dd | 2108 | |
67e2e2b2 JT |
2109 | /* |
2110 | * Setting 'discards_supported' circumvents the normal | |
2111 | * stacking of discard limits (this keeps the pool and | |
2112 | * thin devices' discard limits consistent). | |
2113 | */ | |
0ac55489 | 2114 | ti->discards_supported = true; |
67e2e2b2 | 2115 | } |
991d9fa0 JT |
2116 | ti->private = pt; |
2117 | ||
ac8c3f3d JT |
2118 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2119 | calc_metadata_threshold(pt), | |
2120 | metadata_low_callback, | |
2121 | pool); | |
2122 | if (r) | |
2123 | goto out_free_pt; | |
2124 | ||
991d9fa0 JT |
2125 | pt->callbacks.congested_fn = pool_is_congested; |
2126 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2127 | ||
2128 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2129 | ||
2130 | return 0; | |
2131 | ||
67e2e2b2 JT |
2132 | out_flags_changed: |
2133 | __pool_dec(pool); | |
991d9fa0 JT |
2134 | out_free_pt: |
2135 | kfree(pt); | |
2136 | out: | |
2137 | dm_put_device(ti, data_dev); | |
2138 | out_metadata: | |
2139 | dm_put_device(ti, metadata_dev); | |
2140 | out_unlock: | |
2141 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2142 | ||
2143 | return r; | |
2144 | } | |
2145 | ||
7de3ee57 | 2146 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2147 | { |
2148 | int r; | |
2149 | struct pool_c *pt = ti->private; | |
2150 | struct pool *pool = pt->pool; | |
2151 | unsigned long flags; | |
2152 | ||
2153 | /* | |
2154 | * As this is a singleton target, ti->begin is always zero. | |
2155 | */ | |
2156 | spin_lock_irqsave(&pool->lock, flags); | |
2157 | bio->bi_bdev = pt->data_dev->bdev; | |
2158 | r = DM_MAPIO_REMAPPED; | |
2159 | spin_unlock_irqrestore(&pool->lock, flags); | |
2160 | ||
2161 | return r; | |
2162 | } | |
2163 | ||
b17446df | 2164 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2165 | { |
2166 | int r; | |
2167 | struct pool_c *pt = ti->private; | |
2168 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2169 | sector_t data_size = ti->len; |
2170 | dm_block_t sb_data_size; | |
991d9fa0 | 2171 | |
b17446df | 2172 | *need_commit = false; |
991d9fa0 | 2173 | |
55f2b8bd MS |
2174 | (void) sector_div(data_size, pool->sectors_per_block); |
2175 | ||
991d9fa0 JT |
2176 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2177 | if (r) { | |
4fa5971a MS |
2178 | DMERR("%s: failed to retrieve data device size", |
2179 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2180 | return r; |
2181 | } | |
2182 | ||
2183 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2184 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2185 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2186 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2187 | return -EINVAL; |
2188 | ||
2189 | } else if (data_size > sb_data_size) { | |
2190 | r = dm_pool_resize_data_dev(pool->pmd, data_size); | |
2191 | if (r) { | |
4fa5971a MS |
2192 | DMERR("%s: failed to resize data device", |
2193 | dm_device_name(pool->pool_md)); | |
e49e5829 | 2194 | set_pool_mode(pool, PM_READ_ONLY); |
991d9fa0 JT |
2195 | return r; |
2196 | } | |
2197 | ||
b17446df | 2198 | *need_commit = true; |
991d9fa0 JT |
2199 | } |
2200 | ||
2201 | return 0; | |
2202 | } | |
2203 | ||
24347e95 JT |
2204 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2205 | { | |
2206 | int r; | |
2207 | struct pool_c *pt = ti->private; | |
2208 | struct pool *pool = pt->pool; | |
2209 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2210 | ||
2211 | *need_commit = false; | |
2212 | ||
610bba8b | 2213 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2214 | |
2215 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2216 | if (r) { | |
4fa5971a MS |
2217 | DMERR("%s: failed to retrieve metadata device size", |
2218 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2219 | return r; |
2220 | } | |
2221 | ||
2222 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2223 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2224 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2225 | metadata_dev_size, sb_metadata_dev_size); |
2226 | return -EINVAL; | |
2227 | ||
2228 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
2229 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); | |
2230 | if (r) { | |
4fa5971a MS |
2231 | DMERR("%s: failed to resize metadata device", |
2232 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2233 | return r; |
2234 | } | |
2235 | ||
2236 | *need_commit = true; | |
2237 | } | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
b17446df JT |
2242 | /* |
2243 | * Retrieves the number of blocks of the data device from | |
2244 | * the superblock and compares it to the actual device size, | |
2245 | * thus resizing the data device in case it has grown. | |
2246 | * | |
2247 | * This both copes with opening preallocated data devices in the ctr | |
2248 | * being followed by a resume | |
2249 | * -and- | |
2250 | * calling the resume method individually after userspace has | |
2251 | * grown the data device in reaction to a table event. | |
2252 | */ | |
2253 | static int pool_preresume(struct dm_target *ti) | |
2254 | { | |
2255 | int r; | |
24347e95 | 2256 | bool need_commit1, need_commit2; |
b17446df JT |
2257 | struct pool_c *pt = ti->private; |
2258 | struct pool *pool = pt->pool; | |
2259 | ||
2260 | /* | |
2261 | * Take control of the pool object. | |
2262 | */ | |
2263 | r = bind_control_target(pool, ti); | |
2264 | if (r) | |
2265 | return r; | |
2266 | ||
2267 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2268 | if (r) | |
2269 | return r; | |
2270 | ||
24347e95 JT |
2271 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2272 | if (r) | |
2273 | return r; | |
2274 | ||
2275 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2276 | (void) commit(pool); |
b17446df JT |
2277 | |
2278 | return 0; | |
2279 | } | |
2280 | ||
991d9fa0 JT |
2281 | static void pool_resume(struct dm_target *ti) |
2282 | { | |
2283 | struct pool_c *pt = ti->private; | |
2284 | struct pool *pool = pt->pool; | |
2285 | unsigned long flags; | |
2286 | ||
2287 | spin_lock_irqsave(&pool->lock, flags); | |
2288 | pool->low_water_triggered = 0; | |
2289 | pool->no_free_space = 0; | |
2290 | __requeue_bios(pool); | |
2291 | spin_unlock_irqrestore(&pool->lock, flags); | |
2292 | ||
905e51b3 | 2293 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2294 | } |
2295 | ||
2296 | static void pool_postsuspend(struct dm_target *ti) | |
2297 | { | |
991d9fa0 JT |
2298 | struct pool_c *pt = ti->private; |
2299 | struct pool *pool = pt->pool; | |
2300 | ||
905e51b3 | 2301 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2302 | flush_workqueue(pool->wq); |
020cc3b5 | 2303 | (void) commit(pool); |
991d9fa0 JT |
2304 | } |
2305 | ||
2306 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2307 | { | |
2308 | if (argc != args_required) { | |
2309 | DMWARN("Message received with %u arguments instead of %u.", | |
2310 | argc, args_required); | |
2311 | return -EINVAL; | |
2312 | } | |
2313 | ||
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2318 | { | |
2319 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2320 | *dev_id <= MAX_DEV_ID) | |
2321 | return 0; | |
2322 | ||
2323 | if (warning) | |
2324 | DMWARN("Message received with invalid device id: %s", arg); | |
2325 | ||
2326 | return -EINVAL; | |
2327 | } | |
2328 | ||
2329 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2330 | { | |
2331 | dm_thin_id dev_id; | |
2332 | int r; | |
2333 | ||
2334 | r = check_arg_count(argc, 2); | |
2335 | if (r) | |
2336 | return r; | |
2337 | ||
2338 | r = read_dev_id(argv[1], &dev_id, 1); | |
2339 | if (r) | |
2340 | return r; | |
2341 | ||
2342 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2343 | if (r) { | |
2344 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2345 | argv[1]); | |
2346 | return r; | |
2347 | } | |
2348 | ||
2349 | return 0; | |
2350 | } | |
2351 | ||
2352 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2353 | { | |
2354 | dm_thin_id dev_id; | |
2355 | dm_thin_id origin_dev_id; | |
2356 | int r; | |
2357 | ||
2358 | r = check_arg_count(argc, 3); | |
2359 | if (r) | |
2360 | return r; | |
2361 | ||
2362 | r = read_dev_id(argv[1], &dev_id, 1); | |
2363 | if (r) | |
2364 | return r; | |
2365 | ||
2366 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2367 | if (r) | |
2368 | return r; | |
2369 | ||
2370 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2371 | if (r) { | |
2372 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2373 | argv[1], argv[2]); | |
2374 | return r; | |
2375 | } | |
2376 | ||
2377 | return 0; | |
2378 | } | |
2379 | ||
2380 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2381 | { | |
2382 | dm_thin_id dev_id; | |
2383 | int r; | |
2384 | ||
2385 | r = check_arg_count(argc, 2); | |
2386 | if (r) | |
2387 | return r; | |
2388 | ||
2389 | r = read_dev_id(argv[1], &dev_id, 1); | |
2390 | if (r) | |
2391 | return r; | |
2392 | ||
2393 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2394 | if (r) | |
2395 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2396 | ||
2397 | return r; | |
2398 | } | |
2399 | ||
2400 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2401 | { | |
2402 | dm_thin_id old_id, new_id; | |
2403 | int r; | |
2404 | ||
2405 | r = check_arg_count(argc, 3); | |
2406 | if (r) | |
2407 | return r; | |
2408 | ||
2409 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2410 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2411 | return -EINVAL; | |
2412 | } | |
2413 | ||
2414 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2415 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2416 | return -EINVAL; | |
2417 | } | |
2418 | ||
2419 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2420 | if (r) { | |
2421 | DMWARN("Failed to change transaction id from %s to %s.", | |
2422 | argv[1], argv[2]); | |
2423 | return r; | |
2424 | } | |
2425 | ||
2426 | return 0; | |
2427 | } | |
2428 | ||
cc8394d8 JT |
2429 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2430 | { | |
2431 | int r; | |
2432 | ||
2433 | r = check_arg_count(argc, 1); | |
2434 | if (r) | |
2435 | return r; | |
2436 | ||
020cc3b5 | 2437 | (void) commit(pool); |
0d200aef | 2438 | |
cc8394d8 JT |
2439 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2440 | if (r) | |
2441 | DMWARN("reserve_metadata_snap message failed."); | |
2442 | ||
2443 | return r; | |
2444 | } | |
2445 | ||
2446 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2447 | { | |
2448 | int r; | |
2449 | ||
2450 | r = check_arg_count(argc, 1); | |
2451 | if (r) | |
2452 | return r; | |
2453 | ||
2454 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2455 | if (r) | |
2456 | DMWARN("release_metadata_snap message failed."); | |
2457 | ||
2458 | return r; | |
2459 | } | |
2460 | ||
991d9fa0 JT |
2461 | /* |
2462 | * Messages supported: | |
2463 | * create_thin <dev_id> | |
2464 | * create_snap <dev_id> <origin_id> | |
2465 | * delete <dev_id> | |
2466 | * trim <dev_id> <new_size_in_sectors> | |
2467 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2468 | * reserve_metadata_snap |
2469 | * release_metadata_snap | |
991d9fa0 JT |
2470 | */ |
2471 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2472 | { | |
2473 | int r = -EINVAL; | |
2474 | struct pool_c *pt = ti->private; | |
2475 | struct pool *pool = pt->pool; | |
2476 | ||
2477 | if (!strcasecmp(argv[0], "create_thin")) | |
2478 | r = process_create_thin_mesg(argc, argv, pool); | |
2479 | ||
2480 | else if (!strcasecmp(argv[0], "create_snap")) | |
2481 | r = process_create_snap_mesg(argc, argv, pool); | |
2482 | ||
2483 | else if (!strcasecmp(argv[0], "delete")) | |
2484 | r = process_delete_mesg(argc, argv, pool); | |
2485 | ||
2486 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2487 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2488 | ||
cc8394d8 JT |
2489 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2490 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2491 | ||
2492 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2493 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2494 | ||
991d9fa0 JT |
2495 | else |
2496 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2497 | ||
e49e5829 | 2498 | if (!r) |
020cc3b5 | 2499 | (void) commit(pool); |
991d9fa0 JT |
2500 | |
2501 | return r; | |
2502 | } | |
2503 | ||
e49e5829 JT |
2504 | static void emit_flags(struct pool_features *pf, char *result, |
2505 | unsigned sz, unsigned maxlen) | |
2506 | { | |
2507 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
2508 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY); | |
2509 | DMEMIT("%u ", count); | |
2510 | ||
2511 | if (!pf->zero_new_blocks) | |
2512 | DMEMIT("skip_block_zeroing "); | |
2513 | ||
2514 | if (!pf->discard_enabled) | |
2515 | DMEMIT("ignore_discard "); | |
2516 | ||
2517 | if (!pf->discard_passdown) | |
2518 | DMEMIT("no_discard_passdown "); | |
2519 | ||
2520 | if (pf->mode == PM_READ_ONLY) | |
2521 | DMEMIT("read_only "); | |
2522 | } | |
2523 | ||
991d9fa0 JT |
2524 | /* |
2525 | * Status line is: | |
2526 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2527 | * <used data sectors>/<total data sectors> <held metadata root> | |
2528 | */ | |
fd7c092e MP |
2529 | static void pool_status(struct dm_target *ti, status_type_t type, |
2530 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2531 | { |
e49e5829 | 2532 | int r; |
991d9fa0 JT |
2533 | unsigned sz = 0; |
2534 | uint64_t transaction_id; | |
2535 | dm_block_t nr_free_blocks_data; | |
2536 | dm_block_t nr_free_blocks_metadata; | |
2537 | dm_block_t nr_blocks_data; | |
2538 | dm_block_t nr_blocks_metadata; | |
2539 | dm_block_t held_root; | |
2540 | char buf[BDEVNAME_SIZE]; | |
2541 | char buf2[BDEVNAME_SIZE]; | |
2542 | struct pool_c *pt = ti->private; | |
2543 | struct pool *pool = pt->pool; | |
2544 | ||
2545 | switch (type) { | |
2546 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2547 | if (get_pool_mode(pool) == PM_FAIL) { |
2548 | DMEMIT("Fail"); | |
2549 | break; | |
2550 | } | |
2551 | ||
1f4e0ff0 AK |
2552 | /* Commit to ensure statistics aren't out-of-date */ |
2553 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2554 | (void) commit(pool); |
1f4e0ff0 | 2555 | |
fd7c092e MP |
2556 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2557 | if (r) { | |
4fa5971a MS |
2558 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2559 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2560 | goto err; |
2561 | } | |
991d9fa0 | 2562 | |
fd7c092e MP |
2563 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2564 | if (r) { | |
4fa5971a MS |
2565 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2566 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2567 | goto err; |
2568 | } | |
991d9fa0 JT |
2569 | |
2570 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2571 | if (r) { |
4fa5971a MS |
2572 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2573 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2574 | goto err; |
2575 | } | |
991d9fa0 | 2576 | |
fd7c092e MP |
2577 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2578 | if (r) { | |
4fa5971a MS |
2579 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2580 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2581 | goto err; |
2582 | } | |
991d9fa0 JT |
2583 | |
2584 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2585 | if (r) { |
4fa5971a MS |
2586 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2587 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2588 | goto err; |
2589 | } | |
991d9fa0 | 2590 | |
cc8394d8 | 2591 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2592 | if (r) { |
4fa5971a MS |
2593 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2594 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2595 | goto err; |
2596 | } | |
991d9fa0 JT |
2597 | |
2598 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2599 | (unsigned long long)transaction_id, | |
2600 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2601 | (unsigned long long)nr_blocks_metadata, | |
2602 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2603 | (unsigned long long)nr_blocks_data); | |
2604 | ||
2605 | if (held_root) | |
e49e5829 JT |
2606 | DMEMIT("%llu ", held_root); |
2607 | else | |
2608 | DMEMIT("- "); | |
2609 | ||
2610 | if (pool->pf.mode == PM_READ_ONLY) | |
2611 | DMEMIT("ro "); | |
991d9fa0 | 2612 | else |
e49e5829 JT |
2613 | DMEMIT("rw "); |
2614 | ||
018debea MS |
2615 | if (!pool->pf.discard_enabled) |
2616 | DMEMIT("ignore_discard"); | |
2617 | else if (pool->pf.discard_passdown) | |
e49e5829 JT |
2618 | DMEMIT("discard_passdown"); |
2619 | else | |
2620 | DMEMIT("no_discard_passdown"); | |
991d9fa0 JT |
2621 | |
2622 | break; | |
2623 | ||
2624 | case STATUSTYPE_TABLE: | |
2625 | DMEMIT("%s %s %lu %llu ", | |
2626 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2627 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2628 | (unsigned long)pool->sectors_per_block, | |
2629 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2630 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2631 | break; |
2632 | } | |
fd7c092e | 2633 | return; |
991d9fa0 | 2634 | |
fd7c092e MP |
2635 | err: |
2636 | DMEMIT("Error"); | |
991d9fa0 JT |
2637 | } |
2638 | ||
2639 | static int pool_iterate_devices(struct dm_target *ti, | |
2640 | iterate_devices_callout_fn fn, void *data) | |
2641 | { | |
2642 | struct pool_c *pt = ti->private; | |
2643 | ||
2644 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2645 | } | |
2646 | ||
2647 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2648 | struct bio_vec *biovec, int max_size) | |
2649 | { | |
2650 | struct pool_c *pt = ti->private; | |
2651 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2652 | ||
2653 | if (!q->merge_bvec_fn) | |
2654 | return max_size; | |
2655 | ||
2656 | bvm->bi_bdev = pt->data_dev->bdev; | |
2657 | ||
2658 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2659 | } | |
2660 | ||
0424caa1 | 2661 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2662 | { |
0424caa1 MS |
2663 | struct pool *pool = pt->pool; |
2664 | struct queue_limits *data_limits; | |
2665 | ||
104655fd JT |
2666 | limits->max_discard_sectors = pool->sectors_per_block; |
2667 | ||
2668 | /* | |
0424caa1 | 2669 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2670 | */ |
0424caa1 MS |
2671 | if (pt->adjusted_pf.discard_passdown) { |
2672 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
2673 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 2674 | } else |
0424caa1 | 2675 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
2676 | } |
2677 | ||
991d9fa0 JT |
2678 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2679 | { | |
2680 | struct pool_c *pt = ti->private; | |
2681 | struct pool *pool = pt->pool; | |
0cc67cd9 | 2682 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 2683 | |
0cc67cd9 MS |
2684 | /* |
2685 | * If the system-determined stacked limits are compatible with the | |
2686 | * pool's blocksize (io_opt is a factor) do not override them. | |
2687 | */ | |
2688 | if (io_opt_sectors < pool->sectors_per_block || | |
2689 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
2690 | blk_limits_io_min(limits, 0); | |
2691 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
2692 | } | |
0424caa1 MS |
2693 | |
2694 | /* | |
2695 | * pt->adjusted_pf is a staging area for the actual features to use. | |
2696 | * They get transferred to the live pool in bind_control_target() | |
2697 | * called from pool_preresume(). | |
2698 | */ | |
b60ab990 MS |
2699 | if (!pt->adjusted_pf.discard_enabled) { |
2700 | /* | |
2701 | * Must explicitly disallow stacking discard limits otherwise the | |
2702 | * block layer will stack them if pool's data device has support. | |
2703 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
2704 | * user to see that, so make sure to set all discard limits to 0. | |
2705 | */ | |
2706 | limits->discard_granularity = 0; | |
0424caa1 | 2707 | return; |
b60ab990 | 2708 | } |
0424caa1 MS |
2709 | |
2710 | disable_passdown_if_not_supported(pt); | |
2711 | ||
2712 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
2713 | } |
2714 | ||
2715 | static struct target_type pool_target = { | |
2716 | .name = "thin-pool", | |
2717 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2718 | DM_TARGET_IMMUTABLE, | |
94563bad | 2719 | .version = {1, 9, 0}, |
991d9fa0 JT |
2720 | .module = THIS_MODULE, |
2721 | .ctr = pool_ctr, | |
2722 | .dtr = pool_dtr, | |
2723 | .map = pool_map, | |
2724 | .postsuspend = pool_postsuspend, | |
2725 | .preresume = pool_preresume, | |
2726 | .resume = pool_resume, | |
2727 | .message = pool_message, | |
2728 | .status = pool_status, | |
2729 | .merge = pool_merge, | |
2730 | .iterate_devices = pool_iterate_devices, | |
2731 | .io_hints = pool_io_hints, | |
2732 | }; | |
2733 | ||
2734 | /*---------------------------------------------------------------- | |
2735 | * Thin target methods | |
2736 | *--------------------------------------------------------------*/ | |
2737 | static void thin_dtr(struct dm_target *ti) | |
2738 | { | |
2739 | struct thin_c *tc = ti->private; | |
2740 | ||
2741 | mutex_lock(&dm_thin_pool_table.mutex); | |
2742 | ||
2743 | __pool_dec(tc->pool); | |
2744 | dm_pool_close_thin_device(tc->td); | |
2745 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2746 | if (tc->origin_dev) |
2747 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2748 | kfree(tc); |
2749 | ||
2750 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2751 | } | |
2752 | ||
2753 | /* | |
2754 | * Thin target parameters: | |
2755 | * | |
2dd9c257 | 2756 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2757 | * |
2758 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2759 | * dev_id: the internal device identifier | |
2dd9c257 | 2760 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2761 | * |
2762 | * If the pool device has discards disabled, they get disabled for the thin | |
2763 | * device as well. | |
991d9fa0 JT |
2764 | */ |
2765 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2766 | { | |
2767 | int r; | |
2768 | struct thin_c *tc; | |
2dd9c257 | 2769 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2770 | struct mapped_device *pool_md; |
2771 | ||
2772 | mutex_lock(&dm_thin_pool_table.mutex); | |
2773 | ||
2dd9c257 | 2774 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2775 | ti->error = "Invalid argument count"; |
2776 | r = -EINVAL; | |
2777 | goto out_unlock; | |
2778 | } | |
2779 | ||
2780 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2781 | if (!tc) { | |
2782 | ti->error = "Out of memory"; | |
2783 | r = -ENOMEM; | |
2784 | goto out_unlock; | |
2785 | } | |
2786 | ||
2dd9c257 JT |
2787 | if (argc == 3) { |
2788 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2789 | if (r) { | |
2790 | ti->error = "Error opening origin device"; | |
2791 | goto bad_origin_dev; | |
2792 | } | |
2793 | tc->origin_dev = origin_dev; | |
2794 | } | |
2795 | ||
991d9fa0 JT |
2796 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2797 | if (r) { | |
2798 | ti->error = "Error opening pool device"; | |
2799 | goto bad_pool_dev; | |
2800 | } | |
2801 | tc->pool_dev = pool_dev; | |
2802 | ||
2803 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2804 | ti->error = "Invalid device id"; | |
2805 | r = -EINVAL; | |
2806 | goto bad_common; | |
2807 | } | |
2808 | ||
2809 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2810 | if (!pool_md) { | |
2811 | ti->error = "Couldn't get pool mapped device"; | |
2812 | r = -EINVAL; | |
2813 | goto bad_common; | |
2814 | } | |
2815 | ||
2816 | tc->pool = __pool_table_lookup(pool_md); | |
2817 | if (!tc->pool) { | |
2818 | ti->error = "Couldn't find pool object"; | |
2819 | r = -EINVAL; | |
2820 | goto bad_pool_lookup; | |
2821 | } | |
2822 | __pool_inc(tc->pool); | |
2823 | ||
e49e5829 JT |
2824 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2825 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
2826 | goto bad_thin_open; | |
2827 | } | |
2828 | ||
991d9fa0 JT |
2829 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
2830 | if (r) { | |
2831 | ti->error = "Couldn't open thin internal device"; | |
2832 | goto bad_thin_open; | |
2833 | } | |
2834 | ||
542f9038 MS |
2835 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2836 | if (r) | |
2837 | goto bad_thin_open; | |
2838 | ||
55a62eef | 2839 | ti->num_flush_bios = 1; |
16ad3d10 | 2840 | ti->flush_supported = true; |
59c3d2c6 | 2841 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
2842 | |
2843 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 2844 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2845 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 2846 | ti->discards_supported = true; |
55a62eef | 2847 | ti->num_discard_bios = 1; |
55a62eef AK |
2848 | /* Discard bios must be split on a block boundary */ |
2849 | ti->split_discard_bios = true; | |
67e2e2b2 | 2850 | } |
991d9fa0 JT |
2851 | |
2852 | dm_put(pool_md); | |
2853 | ||
2854 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2855 | ||
2856 | return 0; | |
2857 | ||
2858 | bad_thin_open: | |
2859 | __pool_dec(tc->pool); | |
2860 | bad_pool_lookup: | |
2861 | dm_put(pool_md); | |
2862 | bad_common: | |
2863 | dm_put_device(ti, tc->pool_dev); | |
2864 | bad_pool_dev: | |
2dd9c257 JT |
2865 | if (tc->origin_dev) |
2866 | dm_put_device(ti, tc->origin_dev); | |
2867 | bad_origin_dev: | |
991d9fa0 JT |
2868 | kfree(tc); |
2869 | out_unlock: | |
2870 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2871 | ||
2872 | return r; | |
2873 | } | |
2874 | ||
7de3ee57 | 2875 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 2876 | { |
6efd6e83 | 2877 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 | 2878 | |
7de3ee57 | 2879 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
2880 | } |
2881 | ||
7de3ee57 | 2882 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
2883 | { |
2884 | unsigned long flags; | |
59c3d2c6 | 2885 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 2886 | struct list_head work; |
a24c2569 | 2887 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2888 | struct pool *pool = h->tc->pool; |
2889 | ||
2890 | if (h->shared_read_entry) { | |
2891 | INIT_LIST_HEAD(&work); | |
44feb387 | 2892 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
2893 | |
2894 | spin_lock_irqsave(&pool->lock, flags); | |
2895 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2896 | list_del(&m->list); | |
2897 | m->quiesced = 1; | |
2898 | __maybe_add_mapping(m); | |
2899 | } | |
2900 | spin_unlock_irqrestore(&pool->lock, flags); | |
2901 | } | |
2902 | ||
104655fd JT |
2903 | if (h->all_io_entry) { |
2904 | INIT_LIST_HEAD(&work); | |
44feb387 | 2905 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
2906 | if (!list_empty(&work)) { |
2907 | spin_lock_irqsave(&pool->lock, flags); | |
2908 | list_for_each_entry_safe(m, tmp, &work, list) | |
2909 | list_add(&m->list, &pool->prepared_discards); | |
2910 | spin_unlock_irqrestore(&pool->lock, flags); | |
2911 | wake_worker(pool); | |
2912 | } | |
104655fd JT |
2913 | } |
2914 | ||
eb2aa48d JT |
2915 | return 0; |
2916 | } | |
2917 | ||
991d9fa0 JT |
2918 | static void thin_postsuspend(struct dm_target *ti) |
2919 | { | |
2920 | if (dm_noflush_suspending(ti)) | |
2921 | requeue_io((struct thin_c *)ti->private); | |
2922 | } | |
2923 | ||
2924 | /* | |
2925 | * <nr mapped sectors> <highest mapped sector> | |
2926 | */ | |
fd7c092e MP |
2927 | static void thin_status(struct dm_target *ti, status_type_t type, |
2928 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
2929 | { |
2930 | int r; | |
2931 | ssize_t sz = 0; | |
2932 | dm_block_t mapped, highest; | |
2933 | char buf[BDEVNAME_SIZE]; | |
2934 | struct thin_c *tc = ti->private; | |
2935 | ||
e49e5829 JT |
2936 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2937 | DMEMIT("Fail"); | |
fd7c092e | 2938 | return; |
e49e5829 JT |
2939 | } |
2940 | ||
991d9fa0 JT |
2941 | if (!tc->td) |
2942 | DMEMIT("-"); | |
2943 | else { | |
2944 | switch (type) { | |
2945 | case STATUSTYPE_INFO: | |
2946 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
2947 | if (r) { |
2948 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
2949 | goto err; | |
2950 | } | |
991d9fa0 JT |
2951 | |
2952 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
2953 | if (r < 0) { |
2954 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
2955 | goto err; | |
2956 | } | |
991d9fa0 JT |
2957 | |
2958 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
2959 | if (r) | |
2960 | DMEMIT("%llu", ((highest + 1) * | |
2961 | tc->pool->sectors_per_block) - 1); | |
2962 | else | |
2963 | DMEMIT("-"); | |
2964 | break; | |
2965 | ||
2966 | case STATUSTYPE_TABLE: | |
2967 | DMEMIT("%s %lu", | |
2968 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
2969 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
2970 | if (tc->origin_dev) |
2971 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
2972 | break; |
2973 | } | |
2974 | } | |
2975 | ||
fd7c092e MP |
2976 | return; |
2977 | ||
2978 | err: | |
2979 | DMEMIT("Error"); | |
991d9fa0 JT |
2980 | } |
2981 | ||
2982 | static int thin_iterate_devices(struct dm_target *ti, | |
2983 | iterate_devices_callout_fn fn, void *data) | |
2984 | { | |
55f2b8bd | 2985 | sector_t blocks; |
991d9fa0 | 2986 | struct thin_c *tc = ti->private; |
55f2b8bd | 2987 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2988 | |
2989 | /* | |
2990 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
2991 | * we follow a more convoluted path through to the pool's target. | |
2992 | */ | |
55f2b8bd | 2993 | if (!pool->ti) |
991d9fa0 JT |
2994 | return 0; /* nothing is bound */ |
2995 | ||
55f2b8bd MS |
2996 | blocks = pool->ti->len; |
2997 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 2998 | if (blocks) |
55f2b8bd | 2999 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3000 | |
3001 | return 0; | |
3002 | } | |
3003 | ||
991d9fa0 JT |
3004 | static struct target_type thin_target = { |
3005 | .name = "thin", | |
94563bad | 3006 | .version = {1, 9, 0}, |
991d9fa0 JT |
3007 | .module = THIS_MODULE, |
3008 | .ctr = thin_ctr, | |
3009 | .dtr = thin_dtr, | |
3010 | .map = thin_map, | |
eb2aa48d | 3011 | .end_io = thin_endio, |
991d9fa0 JT |
3012 | .postsuspend = thin_postsuspend, |
3013 | .status = thin_status, | |
3014 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3015 | }; |
3016 | ||
3017 | /*----------------------------------------------------------------*/ | |
3018 | ||
3019 | static int __init dm_thin_init(void) | |
3020 | { | |
3021 | int r; | |
3022 | ||
3023 | pool_table_init(); | |
3024 | ||
3025 | r = dm_register_target(&thin_target); | |
3026 | if (r) | |
3027 | return r; | |
3028 | ||
3029 | r = dm_register_target(&pool_target); | |
3030 | if (r) | |
a24c2569 MS |
3031 | goto bad_pool_target; |
3032 | ||
3033 | r = -ENOMEM; | |
3034 | ||
a24c2569 MS |
3035 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3036 | if (!_new_mapping_cache) | |
3037 | goto bad_new_mapping_cache; | |
3038 | ||
a24c2569 MS |
3039 | return 0; |
3040 | ||
a24c2569 | 3041 | bad_new_mapping_cache: |
a24c2569 MS |
3042 | dm_unregister_target(&pool_target); |
3043 | bad_pool_target: | |
3044 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3045 | |
3046 | return r; | |
3047 | } | |
3048 | ||
3049 | static void dm_thin_exit(void) | |
3050 | { | |
3051 | dm_unregister_target(&thin_target); | |
3052 | dm_unregister_target(&pool_target); | |
a24c2569 | 3053 | |
a24c2569 | 3054 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3055 | } |
3056 | ||
3057 | module_init(dm_thin_init); | |
3058 | module_exit(dm_thin_exit); | |
3059 | ||
7cab8bf1 | 3060 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3061 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3062 | MODULE_LICENSE("GPL"); |