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