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