]>
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> | |
c140e1c4 | 15 | #include <linux/rculist.h> |
991d9fa0 JT |
16 | #include <linux/init.h> |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
67324ea1 | 19 | #include <linux/rbtree.h> |
991d9fa0 JT |
20 | |
21 | #define DM_MSG_PREFIX "thin" | |
22 | ||
23 | /* | |
24 | * Tunable constants | |
25 | */ | |
7768ed33 | 26 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
27 | #define MAPPING_POOL_SIZE 1024 |
28 | #define PRISON_CELLS 1024 | |
905e51b3 | 29 | #define COMMIT_PERIOD HZ |
991d9fa0 | 30 | |
df5d2e90 MP |
31 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
32 | "A percentage of time allocated for copy on write"); | |
33 | ||
991d9fa0 JT |
34 | /* |
35 | * The block size of the device holding pool data must be | |
36 | * between 64KB and 1GB. | |
37 | */ | |
38 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
39 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
40 | ||
991d9fa0 JT |
41 | /* |
42 | * Device id is restricted to 24 bits. | |
43 | */ | |
44 | #define MAX_DEV_ID ((1 << 24) - 1) | |
45 | ||
46 | /* | |
47 | * How do we handle breaking sharing of data blocks? | |
48 | * ================================================= | |
49 | * | |
50 | * We use a standard copy-on-write btree to store the mappings for the | |
51 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
52 | * the data). When you take an internal snapshot you clone the root node | |
53 | * of the origin btree. After this there is no concept of an origin or a | |
54 | * snapshot. They are just two device trees that happen to point to the | |
55 | * same data blocks. | |
56 | * | |
57 | * When we get a write in we decide if it's to a shared data block using | |
58 | * some timestamp magic. If it is, we have to break sharing. | |
59 | * | |
60 | * Let's say we write to a shared block in what was the origin. The | |
61 | * steps are: | |
62 | * | |
63 | * i) plug io further to this physical block. (see bio_prison code). | |
64 | * | |
65 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 66 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
67 | * |
68 | * iii) copy the data block to a newly allocate block. This step can be | |
69 | * missed out if the io covers the block. (schedule_copy). | |
70 | * | |
71 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 72 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
73 | * sharing of btree nodes between the two devices. Breaking sharing only |
74 | * effects the btree of that specific device. Btrees for the other | |
75 | * devices that share the block never change. The btree for the origin | |
76 | * device as it was after the last commit is untouched, ie. we're using | |
77 | * persistent data structures in the functional programming sense. | |
78 | * | |
79 | * v) unplug io to this physical block, including the io that triggered | |
80 | * the breaking of sharing. | |
81 | * | |
82 | * Steps (ii) and (iii) occur in parallel. | |
83 | * | |
84 | * The metadata _doesn't_ need to be committed before the io continues. We | |
85 | * get away with this because the io is always written to a _new_ block. | |
86 | * If there's a crash, then: | |
87 | * | |
88 | * - The origin mapping will point to the old origin block (the shared | |
89 | * one). This will contain the data as it was before the io that triggered | |
90 | * the breaking of sharing came in. | |
91 | * | |
92 | * - The snap mapping still points to the old block. As it would after | |
93 | * the commit. | |
94 | * | |
95 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
96 | * will continue to think that data block in the snapshot device is shared | |
97 | * even after the write to the origin has broken sharing. I suspect data | |
98 | * blocks will typically be shared by many different devices, so we're | |
99 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
100 | * devices that reference this data block. At the moment I think the | |
101 | * benefits far, far outweigh the disadvantages. | |
102 | */ | |
103 | ||
104 | /*----------------------------------------------------------------*/ | |
105 | ||
991d9fa0 JT |
106 | /* |
107 | * Key building. | |
108 | */ | |
109 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 110 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
111 | { |
112 | key->virtual = 0; | |
113 | key->dev = dm_thin_dev_id(td); | |
114 | key->block = b; | |
115 | } | |
116 | ||
117 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 118 | struct dm_cell_key *key) |
991d9fa0 JT |
119 | { |
120 | key->virtual = 1; | |
121 | key->dev = dm_thin_dev_id(td); | |
122 | key->block = b; | |
123 | } | |
124 | ||
125 | /*----------------------------------------------------------------*/ | |
126 | ||
127 | /* | |
128 | * A pool device ties together a metadata device and a data device. It | |
129 | * also provides the interface for creating and destroying internal | |
130 | * devices. | |
131 | */ | |
a24c2569 | 132 | struct dm_thin_new_mapping; |
67e2e2b2 | 133 | |
e49e5829 | 134 | /* |
3e1a0699 | 135 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
136 | */ |
137 | enum pool_mode { | |
138 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 139 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
140 | PM_READ_ONLY, /* metadata may not be changed */ |
141 | PM_FAIL, /* all I/O fails */ | |
142 | }; | |
143 | ||
67e2e2b2 | 144 | struct pool_features { |
e49e5829 JT |
145 | enum pool_mode mode; |
146 | ||
9bc142dd MS |
147 | bool zero_new_blocks:1; |
148 | bool discard_enabled:1; | |
149 | bool discard_passdown:1; | |
787a996c | 150 | bool error_if_no_space:1; |
67e2e2b2 JT |
151 | }; |
152 | ||
e49e5829 JT |
153 | struct thin_c; |
154 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
155 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
156 | ||
991d9fa0 JT |
157 | struct pool { |
158 | struct list_head list; | |
159 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
160 | ||
161 | struct mapped_device *pool_md; | |
162 | struct block_device *md_dev; | |
163 | struct dm_pool_metadata *pmd; | |
164 | ||
991d9fa0 | 165 | dm_block_t low_water_blocks; |
55f2b8bd | 166 | uint32_t sectors_per_block; |
f9a8e0cd | 167 | int sectors_per_block_shift; |
991d9fa0 | 168 | |
67e2e2b2 | 169 | struct pool_features pf; |
88a6621b | 170 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 171 | |
44feb387 | 172 | struct dm_bio_prison *prison; |
991d9fa0 JT |
173 | struct dm_kcopyd_client *copier; |
174 | ||
175 | struct workqueue_struct *wq; | |
176 | struct work_struct worker; | |
905e51b3 | 177 | struct delayed_work waker; |
991d9fa0 | 178 | |
905e51b3 | 179 | unsigned long last_commit_jiffies; |
55f2b8bd | 180 | unsigned ref_count; |
991d9fa0 JT |
181 | |
182 | spinlock_t lock; | |
991d9fa0 JT |
183 | struct bio_list deferred_flush_bios; |
184 | struct list_head prepared_mappings; | |
104655fd | 185 | struct list_head prepared_discards; |
c140e1c4 | 186 | struct list_head active_thins; |
991d9fa0 | 187 | |
44feb387 MS |
188 | struct dm_deferred_set *shared_read_ds; |
189 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 190 | |
a24c2569 | 191 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 192 | mempool_t *mapping_pool; |
e49e5829 JT |
193 | |
194 | process_bio_fn process_bio; | |
195 | process_bio_fn process_discard; | |
196 | ||
197 | process_mapping_fn process_prepared_mapping; | |
198 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
199 | }; |
200 | ||
e49e5829 | 201 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 202 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 203 | |
991d9fa0 JT |
204 | /* |
205 | * Target context for a pool. | |
206 | */ | |
207 | struct pool_c { | |
208 | struct dm_target *ti; | |
209 | struct pool *pool; | |
210 | struct dm_dev *data_dev; | |
211 | struct dm_dev *metadata_dev; | |
212 | struct dm_target_callbacks callbacks; | |
213 | ||
214 | dm_block_t low_water_blocks; | |
0424caa1 MS |
215 | struct pool_features requested_pf; /* Features requested during table load */ |
216 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
217 | }; |
218 | ||
219 | /* | |
220 | * Target context for a thin. | |
221 | */ | |
222 | struct thin_c { | |
c140e1c4 | 223 | struct list_head list; |
991d9fa0 | 224 | struct dm_dev *pool_dev; |
2dd9c257 | 225 | struct dm_dev *origin_dev; |
991d9fa0 JT |
226 | dm_thin_id dev_id; |
227 | ||
228 | struct pool *pool; | |
229 | struct dm_thin_device *td; | |
738211f7 | 230 | bool requeue_mode:1; |
c140e1c4 MS |
231 | spinlock_t lock; |
232 | struct bio_list deferred_bio_list; | |
233 | struct bio_list retry_on_resume_list; | |
67324ea1 | 234 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
991d9fa0 JT |
235 | }; |
236 | ||
237 | /*----------------------------------------------------------------*/ | |
238 | ||
025b9685 JT |
239 | /* |
240 | * wake_worker() is used when new work is queued and when pool_resume is | |
241 | * ready to continue deferred IO processing. | |
242 | */ | |
243 | static void wake_worker(struct pool *pool) | |
244 | { | |
245 | queue_work(pool->wq, &pool->worker); | |
246 | } | |
247 | ||
248 | /*----------------------------------------------------------------*/ | |
249 | ||
6beca5eb JT |
250 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
251 | struct dm_bio_prison_cell **cell_result) | |
252 | { | |
253 | int r; | |
254 | struct dm_bio_prison_cell *cell_prealloc; | |
255 | ||
256 | /* | |
257 | * Allocate a cell from the prison's mempool. | |
258 | * This might block but it can't fail. | |
259 | */ | |
260 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
261 | ||
262 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
263 | if (r) | |
264 | /* | |
265 | * We reused an old cell; we can get rid of | |
266 | * the new one. | |
267 | */ | |
268 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
269 | ||
270 | return r; | |
271 | } | |
272 | ||
273 | static void cell_release(struct pool *pool, | |
274 | struct dm_bio_prison_cell *cell, | |
275 | struct bio_list *bios) | |
276 | { | |
277 | dm_cell_release(pool->prison, cell, bios); | |
278 | dm_bio_prison_free_cell(pool->prison, cell); | |
279 | } | |
280 | ||
281 | static void cell_release_no_holder(struct pool *pool, | |
282 | struct dm_bio_prison_cell *cell, | |
283 | struct bio_list *bios) | |
284 | { | |
285 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
286 | dm_bio_prison_free_cell(pool->prison, cell); | |
287 | } | |
288 | ||
025b9685 JT |
289 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
290 | struct dm_bio_prison_cell *cell) | |
291 | { | |
292 | struct pool *pool = tc->pool; | |
293 | unsigned long flags; | |
294 | ||
c140e1c4 MS |
295 | spin_lock_irqsave(&tc->lock, flags); |
296 | dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list); | |
297 | spin_unlock_irqrestore(&tc->lock, flags); | |
025b9685 JT |
298 | |
299 | wake_worker(pool); | |
300 | } | |
301 | ||
6beca5eb JT |
302 | static void cell_error(struct pool *pool, |
303 | struct dm_bio_prison_cell *cell) | |
304 | { | |
305 | dm_cell_error(pool->prison, cell); | |
306 | dm_bio_prison_free_cell(pool->prison, cell); | |
307 | } | |
308 | ||
309 | /*----------------------------------------------------------------*/ | |
310 | ||
991d9fa0 JT |
311 | /* |
312 | * A global list of pools that uses a struct mapped_device as a key. | |
313 | */ | |
314 | static struct dm_thin_pool_table { | |
315 | struct mutex mutex; | |
316 | struct list_head pools; | |
317 | } dm_thin_pool_table; | |
318 | ||
319 | static void pool_table_init(void) | |
320 | { | |
321 | mutex_init(&dm_thin_pool_table.mutex); | |
322 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
323 | } | |
324 | ||
325 | static void __pool_table_insert(struct pool *pool) | |
326 | { | |
327 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
328 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
329 | } | |
330 | ||
331 | static void __pool_table_remove(struct pool *pool) | |
332 | { | |
333 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
334 | list_del(&pool->list); | |
335 | } | |
336 | ||
337 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
338 | { | |
339 | struct pool *pool = NULL, *tmp; | |
340 | ||
341 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
342 | ||
343 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
344 | if (tmp->pool_md == md) { | |
345 | pool = tmp; | |
346 | break; | |
347 | } | |
348 | } | |
349 | ||
350 | return pool; | |
351 | } | |
352 | ||
353 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
354 | { | |
355 | struct pool *pool = NULL, *tmp; | |
356 | ||
357 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
358 | ||
359 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
360 | if (tmp->md_dev == md_dev) { | |
361 | pool = tmp; | |
362 | break; | |
363 | } | |
364 | } | |
365 | ||
366 | return pool; | |
367 | } | |
368 | ||
369 | /*----------------------------------------------------------------*/ | |
370 | ||
a24c2569 | 371 | struct dm_thin_endio_hook { |
eb2aa48d | 372 | struct thin_c *tc; |
44feb387 MS |
373 | struct dm_deferred_entry *shared_read_entry; |
374 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 375 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 376 | struct rb_node rb_node; |
eb2aa48d JT |
377 | }; |
378 | ||
18adc577 | 379 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
380 | { |
381 | struct bio *bio; | |
382 | struct bio_list bios; | |
18adc577 | 383 | unsigned long flags; |
991d9fa0 JT |
384 | |
385 | bio_list_init(&bios); | |
18adc577 | 386 | |
c140e1c4 | 387 | spin_lock_irqsave(&tc->lock, flags); |
991d9fa0 JT |
388 | bio_list_merge(&bios, master); |
389 | bio_list_init(master); | |
c140e1c4 | 390 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 391 | |
c140e1c4 MS |
392 | while ((bio = bio_list_pop(&bios))) |
393 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
991d9fa0 JT |
394 | } |
395 | ||
396 | static void requeue_io(struct thin_c *tc) | |
397 | { | |
c140e1c4 MS |
398 | requeue_bio_list(tc, &tc->deferred_bio_list); |
399 | requeue_bio_list(tc, &tc->retry_on_resume_list); | |
991d9fa0 JT |
400 | } |
401 | ||
c140e1c4 | 402 | static void error_thin_retry_list(struct thin_c *tc) |
3e1a0699 JT |
403 | { |
404 | struct bio *bio; | |
405 | unsigned long flags; | |
406 | struct bio_list bios; | |
407 | ||
408 | bio_list_init(&bios); | |
409 | ||
c140e1c4 MS |
410 | spin_lock_irqsave(&tc->lock, flags); |
411 | bio_list_merge(&bios, &tc->retry_on_resume_list); | |
412 | bio_list_init(&tc->retry_on_resume_list); | |
413 | spin_unlock_irqrestore(&tc->lock, flags); | |
3e1a0699 JT |
414 | |
415 | while ((bio = bio_list_pop(&bios))) | |
416 | bio_io_error(bio); | |
417 | } | |
418 | ||
c140e1c4 MS |
419 | static void error_retry_list(struct pool *pool) |
420 | { | |
421 | struct thin_c *tc; | |
422 | ||
423 | rcu_read_lock(); | |
424 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
425 | error_thin_retry_list(tc); | |
426 | rcu_read_unlock(); | |
427 | } | |
428 | ||
991d9fa0 JT |
429 | /* |
430 | * This section of code contains the logic for processing a thin device's IO. | |
431 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
432 | * but most is exclusively called from the thin target rather than the thin-pool | |
433 | * target. | |
434 | */ | |
435 | ||
58f77a21 MS |
436 | static bool block_size_is_power_of_two(struct pool *pool) |
437 | { | |
438 | return pool->sectors_per_block_shift >= 0; | |
439 | } | |
440 | ||
991d9fa0 JT |
441 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
442 | { | |
58f77a21 | 443 | struct pool *pool = tc->pool; |
4f024f37 | 444 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 445 | |
58f77a21 MS |
446 | if (block_size_is_power_of_two(pool)) |
447 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 448 | else |
58f77a21 | 449 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
450 | |
451 | return block_nr; | |
991d9fa0 JT |
452 | } |
453 | ||
454 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
455 | { | |
456 | struct pool *pool = tc->pool; | |
4f024f37 | 457 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
458 | |
459 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 460 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
461 | bio->bi_iter.bi_sector = |
462 | (block << pool->sectors_per_block_shift) | | |
463 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 464 | else |
4f024f37 | 465 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 466 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
467 | } |
468 | ||
2dd9c257 JT |
469 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
470 | { | |
471 | bio->bi_bdev = tc->origin_dev->bdev; | |
472 | } | |
473 | ||
4afdd680 JT |
474 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
475 | { | |
476 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
477 | dm_thin_changed_this_transaction(tc->td); | |
478 | } | |
479 | ||
e8088073 JT |
480 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
481 | { | |
482 | struct dm_thin_endio_hook *h; | |
483 | ||
484 | if (bio->bi_rw & REQ_DISCARD) | |
485 | return; | |
486 | ||
59c3d2c6 | 487 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
488 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
489 | } | |
490 | ||
2dd9c257 | 491 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
492 | { |
493 | struct pool *pool = tc->pool; | |
494 | unsigned long flags; | |
495 | ||
e49e5829 JT |
496 | if (!bio_triggers_commit(tc, bio)) { |
497 | generic_make_request(bio); | |
498 | return; | |
499 | } | |
500 | ||
991d9fa0 | 501 | /* |
e49e5829 JT |
502 | * Complete bio with an error if earlier I/O caused changes to |
503 | * the metadata that can't be committed e.g, due to I/O errors | |
504 | * on the metadata device. | |
991d9fa0 | 505 | */ |
e49e5829 JT |
506 | if (dm_thin_aborted_changes(tc->td)) { |
507 | bio_io_error(bio); | |
508 | return; | |
509 | } | |
510 | ||
511 | /* | |
512 | * Batch together any bios that trigger commits and then issue a | |
513 | * single commit for them in process_deferred_bios(). | |
514 | */ | |
515 | spin_lock_irqsave(&pool->lock, flags); | |
516 | bio_list_add(&pool->deferred_flush_bios, bio); | |
517 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
518 | } |
519 | ||
2dd9c257 JT |
520 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
521 | { | |
522 | remap_to_origin(tc, bio); | |
523 | issue(tc, bio); | |
524 | } | |
525 | ||
526 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
527 | dm_block_t block) | |
528 | { | |
529 | remap(tc, bio, block); | |
530 | issue(tc, bio); | |
531 | } | |
532 | ||
991d9fa0 JT |
533 | /*----------------------------------------------------------------*/ |
534 | ||
535 | /* | |
536 | * Bio endio functions. | |
537 | */ | |
a24c2569 | 538 | struct dm_thin_new_mapping { |
991d9fa0 JT |
539 | struct list_head list; |
540 | ||
7f214665 MS |
541 | bool quiesced:1; |
542 | bool prepared:1; | |
543 | bool pass_discard:1; | |
544 | bool definitely_not_shared:1; | |
991d9fa0 | 545 | |
7f214665 | 546 | int err; |
991d9fa0 JT |
547 | struct thin_c *tc; |
548 | dm_block_t virt_block; | |
549 | dm_block_t data_block; | |
a24c2569 | 550 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
551 | |
552 | /* | |
553 | * If the bio covers the whole area of a block then we can avoid | |
554 | * zeroing or copying. Instead this bio is hooked. The bio will | |
555 | * still be in the cell, so care has to be taken to avoid issuing | |
556 | * the bio twice. | |
557 | */ | |
558 | struct bio *bio; | |
559 | bio_end_io_t *saved_bi_end_io; | |
560 | }; | |
561 | ||
a24c2569 | 562 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
563 | { |
564 | struct pool *pool = m->tc->pool; | |
565 | ||
eb2aa48d | 566 | if (m->quiesced && m->prepared) { |
daec338b | 567 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
568 | wake_worker(pool); |
569 | } | |
570 | } | |
571 | ||
572 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
573 | { | |
574 | unsigned long flags; | |
a24c2569 | 575 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
576 | struct pool *pool = m->tc->pool; |
577 | ||
578 | m->err = read_err || write_err ? -EIO : 0; | |
579 | ||
580 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 581 | m->prepared = true; |
991d9fa0 JT |
582 | __maybe_add_mapping(m); |
583 | spin_unlock_irqrestore(&pool->lock, flags); | |
584 | } | |
585 | ||
586 | static void overwrite_endio(struct bio *bio, int err) | |
587 | { | |
588 | unsigned long flags; | |
59c3d2c6 | 589 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 590 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
591 | struct pool *pool = m->tc->pool; |
592 | ||
593 | m->err = err; | |
594 | ||
595 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 596 | m->prepared = true; |
991d9fa0 JT |
597 | __maybe_add_mapping(m); |
598 | spin_unlock_irqrestore(&pool->lock, flags); | |
599 | } | |
600 | ||
991d9fa0 JT |
601 | /*----------------------------------------------------------------*/ |
602 | ||
603 | /* | |
604 | * Workqueue. | |
605 | */ | |
606 | ||
607 | /* | |
608 | * Prepared mapping jobs. | |
609 | */ | |
610 | ||
611 | /* | |
612 | * This sends the bios in the cell back to the deferred_bios list. | |
613 | */ | |
2aab3850 | 614 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
615 | { |
616 | struct pool *pool = tc->pool; | |
617 | unsigned long flags; | |
618 | ||
c140e1c4 MS |
619 | spin_lock_irqsave(&tc->lock, flags); |
620 | cell_release(pool, cell, &tc->deferred_bio_list); | |
621 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
622 | |
623 | wake_worker(pool); | |
624 | } | |
625 | ||
626 | /* | |
6beca5eb | 627 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 628 | */ |
f286ba0e | 629 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 630 | { |
991d9fa0 JT |
631 | struct pool *pool = tc->pool; |
632 | unsigned long flags; | |
633 | ||
c140e1c4 MS |
634 | spin_lock_irqsave(&tc->lock, flags); |
635 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
636 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
637 | |
638 | wake_worker(pool); | |
639 | } | |
640 | ||
e49e5829 JT |
641 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
642 | { | |
196d38bc | 643 | if (m->bio) { |
e49e5829 | 644 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
645 | atomic_inc(&m->bio->bi_remaining); |
646 | } | |
6beca5eb | 647 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
648 | list_del(&m->list); |
649 | mempool_free(m, m->tc->pool->mapping_pool); | |
650 | } | |
025b9685 | 651 | |
a24c2569 | 652 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
653 | { |
654 | struct thin_c *tc = m->tc; | |
6beca5eb | 655 | struct pool *pool = tc->pool; |
991d9fa0 JT |
656 | struct bio *bio; |
657 | int r; | |
658 | ||
659 | bio = m->bio; | |
196d38bc | 660 | if (bio) { |
991d9fa0 | 661 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
662 | atomic_inc(&bio->bi_remaining); |
663 | } | |
991d9fa0 JT |
664 | |
665 | if (m->err) { | |
6beca5eb | 666 | cell_error(pool, m->cell); |
905386f8 | 667 | goto out; |
991d9fa0 JT |
668 | } |
669 | ||
670 | /* | |
671 | * Commit the prepared block into the mapping btree. | |
672 | * Any I/O for this block arriving after this point will get | |
673 | * remapped to it directly. | |
674 | */ | |
675 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
676 | if (r) { | |
b5330655 | 677 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 678 | cell_error(pool, m->cell); |
905386f8 | 679 | goto out; |
991d9fa0 JT |
680 | } |
681 | ||
682 | /* | |
683 | * Release any bios held while the block was being provisioned. | |
684 | * If we are processing a write bio that completely covers the block, | |
685 | * we already processed it so can ignore it now when processing | |
686 | * the bios in the cell. | |
687 | */ | |
688 | if (bio) { | |
f286ba0e | 689 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
690 | bio_endio(bio, 0); |
691 | } else | |
2aab3850 | 692 | cell_defer(tc, m->cell); |
991d9fa0 | 693 | |
905386f8 | 694 | out: |
991d9fa0 | 695 | list_del(&m->list); |
6beca5eb | 696 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
697 | } |
698 | ||
e49e5829 | 699 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 700 | { |
104655fd JT |
701 | struct thin_c *tc = m->tc; |
702 | ||
e49e5829 | 703 | bio_io_error(m->bio); |
f286ba0e JT |
704 | cell_defer_no_holder(tc, m->cell); |
705 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
706 | mempool_free(m, tc->pool->mapping_pool); |
707 | } | |
708 | ||
709 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
710 | { | |
711 | struct thin_c *tc = m->tc; | |
104655fd | 712 | |
e8088073 | 713 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
714 | cell_defer_no_holder(tc, m->cell); |
715 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 716 | |
104655fd | 717 | if (m->pass_discard) |
19fa1a67 JT |
718 | if (m->definitely_not_shared) |
719 | remap_and_issue(tc, m->bio, m->data_block); | |
720 | else { | |
721 | bool used = false; | |
722 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
723 | bio_endio(m->bio, 0); | |
724 | else | |
725 | remap_and_issue(tc, m->bio, m->data_block); | |
726 | } | |
104655fd JT |
727 | else |
728 | bio_endio(m->bio, 0); | |
729 | ||
104655fd JT |
730 | mempool_free(m, tc->pool->mapping_pool); |
731 | } | |
732 | ||
e49e5829 JT |
733 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
734 | { | |
735 | int r; | |
736 | struct thin_c *tc = m->tc; | |
737 | ||
738 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
739 | if (r) | |
c397741c | 740 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
741 | |
742 | process_prepared_discard_passdown(m); | |
743 | } | |
744 | ||
104655fd | 745 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 746 | process_mapping_fn *fn) |
991d9fa0 JT |
747 | { |
748 | unsigned long flags; | |
749 | struct list_head maps; | |
a24c2569 | 750 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
751 | |
752 | INIT_LIST_HEAD(&maps); | |
753 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 754 | list_splice_init(head, &maps); |
991d9fa0 JT |
755 | spin_unlock_irqrestore(&pool->lock, flags); |
756 | ||
757 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 758 | (*fn)(m); |
991d9fa0 JT |
759 | } |
760 | ||
761 | /* | |
762 | * Deferred bio jobs. | |
763 | */ | |
104655fd | 764 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 765 | { |
4f024f37 KO |
766 | return bio->bi_iter.bi_size == |
767 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
768 | } |
769 | ||
770 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
771 | { | |
772 | return (bio_data_dir(bio) == WRITE) && | |
773 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
774 | } |
775 | ||
776 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
777 | bio_end_io_t *fn) | |
778 | { | |
779 | *save = bio->bi_end_io; | |
780 | bio->bi_end_io = fn; | |
781 | } | |
782 | ||
783 | static int ensure_next_mapping(struct pool *pool) | |
784 | { | |
785 | if (pool->next_mapping) | |
786 | return 0; | |
787 | ||
788 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
789 | ||
790 | return pool->next_mapping ? 0 : -ENOMEM; | |
791 | } | |
792 | ||
a24c2569 | 793 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 794 | { |
16961b04 | 795 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
796 | |
797 | BUG_ON(!pool->next_mapping); | |
798 | ||
16961b04 MS |
799 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
800 | INIT_LIST_HEAD(&m->list); | |
801 | m->bio = NULL; | |
802 | ||
991d9fa0 JT |
803 | pool->next_mapping = NULL; |
804 | ||
16961b04 | 805 | return m; |
991d9fa0 JT |
806 | } |
807 | ||
808 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
809 | struct dm_dev *origin, dm_block_t data_origin, |
810 | dm_block_t data_dest, | |
a24c2569 | 811 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
812 | { |
813 | int r; | |
814 | struct pool *pool = tc->pool; | |
a24c2569 | 815 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 816 | |
991d9fa0 JT |
817 | m->tc = tc; |
818 | m->virt_block = virt_block; | |
819 | m->data_block = data_dest; | |
820 | m->cell = cell; | |
991d9fa0 | 821 | |
44feb387 | 822 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 823 | m->quiesced = true; |
991d9fa0 JT |
824 | |
825 | /* | |
826 | * IO to pool_dev remaps to the pool target's data_dev. | |
827 | * | |
828 | * If the whole block of data is being overwritten, we can issue the | |
829 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
830 | */ | |
831 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 832 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 833 | |
eb2aa48d | 834 | h->overwrite_mapping = m; |
991d9fa0 JT |
835 | m->bio = bio; |
836 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 837 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
838 | remap_and_issue(tc, bio, data_dest); |
839 | } else { | |
840 | struct dm_io_region from, to; | |
841 | ||
2dd9c257 | 842 | from.bdev = origin->bdev; |
991d9fa0 JT |
843 | from.sector = data_origin * pool->sectors_per_block; |
844 | from.count = pool->sectors_per_block; | |
845 | ||
846 | to.bdev = tc->pool_dev->bdev; | |
847 | to.sector = data_dest * pool->sectors_per_block; | |
848 | to.count = pool->sectors_per_block; | |
849 | ||
850 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
851 | 0, copy_complete, m); | |
852 | if (r < 0) { | |
853 | mempool_free(m, pool->mapping_pool); | |
c397741c | 854 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 855 | cell_error(pool, cell); |
991d9fa0 JT |
856 | } |
857 | } | |
858 | } | |
859 | ||
2dd9c257 JT |
860 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
861 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 862 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
863 | { |
864 | schedule_copy(tc, virt_block, tc->pool_dev, | |
865 | data_origin, data_dest, cell, bio); | |
866 | } | |
867 | ||
868 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
869 | dm_block_t data_dest, | |
a24c2569 | 870 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
871 | { |
872 | schedule_copy(tc, virt_block, tc->origin_dev, | |
873 | virt_block, data_dest, cell, bio); | |
874 | } | |
875 | ||
991d9fa0 | 876 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 877 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
878 | struct bio *bio) |
879 | { | |
880 | struct pool *pool = tc->pool; | |
a24c2569 | 881 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 882 | |
7f214665 MS |
883 | m->quiesced = true; |
884 | m->prepared = false; | |
991d9fa0 JT |
885 | m->tc = tc; |
886 | m->virt_block = virt_block; | |
887 | m->data_block = data_block; | |
888 | m->cell = cell; | |
991d9fa0 JT |
889 | |
890 | /* | |
891 | * If the whole block of data is being overwritten or we are not | |
892 | * zeroing pre-existing data, we can issue the bio immediately. | |
893 | * Otherwise we use kcopyd to zero the data first. | |
894 | */ | |
67e2e2b2 | 895 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
896 | process_prepared_mapping(m); |
897 | ||
898 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 899 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 900 | |
eb2aa48d | 901 | h->overwrite_mapping = m; |
991d9fa0 JT |
902 | m->bio = bio; |
903 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 904 | inc_all_io_entry(pool, bio); |
991d9fa0 | 905 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
906 | } else { |
907 | int r; | |
908 | struct dm_io_region to; | |
909 | ||
910 | to.bdev = tc->pool_dev->bdev; | |
911 | to.sector = data_block * pool->sectors_per_block; | |
912 | to.count = pool->sectors_per_block; | |
913 | ||
914 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
915 | if (r < 0) { | |
916 | mempool_free(m, pool->mapping_pool); | |
c397741c | 917 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 918 | cell_error(pool, cell); |
991d9fa0 JT |
919 | } |
920 | } | |
921 | } | |
922 | ||
e49e5829 JT |
923 | /* |
924 | * A non-zero return indicates read_only or fail_io mode. | |
925 | * Many callers don't care about the return value. | |
926 | */ | |
020cc3b5 | 927 | static int commit(struct pool *pool) |
e49e5829 JT |
928 | { |
929 | int r; | |
930 | ||
931 | if (get_pool_mode(pool) != PM_WRITE) | |
932 | return -EINVAL; | |
933 | ||
020cc3b5 | 934 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
935 | if (r) |
936 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
937 | |
938 | return r; | |
939 | } | |
940 | ||
88a6621b JT |
941 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
942 | { | |
943 | unsigned long flags; | |
944 | ||
945 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
946 | DMWARN("%s: reached low water mark for data device: sending event.", | |
947 | dm_device_name(pool->pool_md)); | |
948 | spin_lock_irqsave(&pool->lock, flags); | |
949 | pool->low_water_triggered = true; | |
950 | spin_unlock_irqrestore(&pool->lock, flags); | |
951 | dm_table_event(pool->ti->table); | |
952 | } | |
953 | } | |
954 | ||
3e1a0699 JT |
955 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
956 | ||
991d9fa0 JT |
957 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
958 | { | |
959 | int r; | |
960 | dm_block_t free_blocks; | |
991d9fa0 JT |
961 | struct pool *pool = tc->pool; |
962 | ||
3e1a0699 | 963 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
964 | return -EINVAL; |
965 | ||
991d9fa0 | 966 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
967 | if (r) { |
968 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 969 | return r; |
b5330655 | 970 | } |
991d9fa0 | 971 | |
88a6621b | 972 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
973 | |
974 | if (!free_blocks) { | |
94563bad MS |
975 | /* |
976 | * Try to commit to see if that will free up some | |
977 | * more space. | |
978 | */ | |
020cc3b5 JT |
979 | r = commit(pool); |
980 | if (r) | |
981 | return r; | |
991d9fa0 | 982 | |
94563bad | 983 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
984 | if (r) { |
985 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 986 | return r; |
b5330655 | 987 | } |
991d9fa0 | 988 | |
94563bad | 989 | if (!free_blocks) { |
3e1a0699 | 990 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 991 | return -ENOSPC; |
991d9fa0 JT |
992 | } |
993 | } | |
994 | ||
995 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 996 | if (r) { |
b5330655 | 997 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 998 | return r; |
4a02b34e | 999 | } |
991d9fa0 JT |
1000 | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * If we have run out of space, queue bios until the device is | |
1006 | * resumed, presumably after having been reloaded with more space. | |
1007 | */ | |
1008 | static void retry_on_resume(struct bio *bio) | |
1009 | { | |
59c3d2c6 | 1010 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1011 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1012 | unsigned long flags; |
1013 | ||
c140e1c4 MS |
1014 | spin_lock_irqsave(&tc->lock, flags); |
1015 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1016 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1017 | } |
1018 | ||
3e1a0699 | 1019 | static bool should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1020 | { |
3e1a0699 JT |
1021 | enum pool_mode m = get_pool_mode(pool); |
1022 | ||
1023 | switch (m) { | |
1024 | case PM_WRITE: | |
1025 | /* Shouldn't get here */ | |
1026 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
1027 | return true; | |
1028 | ||
1029 | case PM_OUT_OF_DATA_SPACE: | |
1030 | return pool->pf.error_if_no_space; | |
1031 | ||
1032 | case PM_READ_ONLY: | |
1033 | case PM_FAIL: | |
1034 | return true; | |
1035 | default: | |
1036 | /* Shouldn't get here */ | |
1037 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
1038 | return true; | |
1039 | } | |
1040 | } | |
8c0f0e8c | 1041 | |
3e1a0699 JT |
1042 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1043 | { | |
1044 | if (should_error_unserviceable_bio(pool)) | |
8c0f0e8c | 1045 | bio_io_error(bio); |
6d16202b MS |
1046 | else |
1047 | retry_on_resume(bio); | |
8c0f0e8c MS |
1048 | } |
1049 | ||
399caddf | 1050 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1051 | { |
1052 | struct bio *bio; | |
1053 | struct bio_list bios; | |
1054 | ||
3e1a0699 JT |
1055 | if (should_error_unserviceable_bio(pool)) { |
1056 | cell_error(pool, cell); | |
1057 | return; | |
1058 | } | |
1059 | ||
991d9fa0 | 1060 | bio_list_init(&bios); |
6beca5eb | 1061 | cell_release(pool, cell, &bios); |
991d9fa0 | 1062 | |
3e1a0699 JT |
1063 | if (should_error_unserviceable_bio(pool)) |
1064 | while ((bio = bio_list_pop(&bios))) | |
1065 | bio_io_error(bio); | |
1066 | else | |
1067 | while ((bio = bio_list_pop(&bios))) | |
1068 | retry_on_resume(bio); | |
991d9fa0 JT |
1069 | } |
1070 | ||
104655fd JT |
1071 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1072 | { | |
1073 | int r; | |
c3a0ce2e | 1074 | unsigned long flags; |
104655fd | 1075 | struct pool *pool = tc->pool; |
a24c2569 | 1076 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1077 | struct dm_cell_key key, key2; |
104655fd JT |
1078 | dm_block_t block = get_bio_block(tc, bio); |
1079 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1080 | struct dm_thin_new_mapping *m; |
104655fd JT |
1081 | |
1082 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1083 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1084 | return; |
1085 | ||
1086 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1087 | switch (r) { | |
1088 | case 0: | |
1089 | /* | |
1090 | * Check nobody is fiddling with this pool block. This can | |
1091 | * happen if someone's in the process of breaking sharing | |
1092 | * on this block. | |
1093 | */ | |
1094 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1095 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1096 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1097 | break; |
1098 | } | |
1099 | ||
1100 | if (io_overlaps_block(pool, bio)) { | |
1101 | /* | |
1102 | * IO may still be going to the destination block. We must | |
1103 | * quiesce before we can do the removal. | |
1104 | */ | |
1105 | m = get_next_mapping(pool); | |
1106 | m->tc = tc; | |
19fa1a67 JT |
1107 | m->pass_discard = pool->pf.discard_passdown; |
1108 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1109 | m->virt_block = block; |
1110 | m->data_block = lookup_result.block; | |
1111 | m->cell = cell; | |
1112 | m->cell2 = cell2; | |
104655fd JT |
1113 | m->bio = bio; |
1114 | ||
44feb387 | 1115 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1116 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1117 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1118 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1119 | wake_worker(pool); |
1120 | } | |
1121 | } else { | |
e8088073 | 1122 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1123 | cell_defer_no_holder(tc, cell); |
1124 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1125 | |
104655fd | 1126 | /* |
49296309 MP |
1127 | * The DM core makes sure that the discard doesn't span |
1128 | * a block boundary. So we submit the discard of a | |
1129 | * partial block appropriately. | |
104655fd | 1130 | */ |
650d2a06 MP |
1131 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1132 | remap_and_issue(tc, bio, lookup_result.block); | |
1133 | else | |
1134 | bio_endio(bio, 0); | |
104655fd JT |
1135 | } |
1136 | break; | |
1137 | ||
1138 | case -ENODATA: | |
1139 | /* | |
1140 | * It isn't provisioned, just forget it. | |
1141 | */ | |
f286ba0e | 1142 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1143 | bio_endio(bio, 0); |
1144 | break; | |
1145 | ||
1146 | default: | |
c397741c MS |
1147 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1148 | __func__, r); | |
f286ba0e | 1149 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1150 | bio_io_error(bio); |
1151 | break; | |
1152 | } | |
1153 | } | |
1154 | ||
991d9fa0 | 1155 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1156 | struct dm_cell_key *key, |
991d9fa0 | 1157 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1158 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1159 | { |
1160 | int r; | |
1161 | dm_block_t data_block; | |
d6fc2042 | 1162 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1163 | |
1164 | r = alloc_data_block(tc, &data_block); | |
1165 | switch (r) { | |
1166 | case 0: | |
2dd9c257 JT |
1167 | schedule_internal_copy(tc, block, lookup_result->block, |
1168 | data_block, cell, bio); | |
991d9fa0 JT |
1169 | break; |
1170 | ||
1171 | case -ENOSPC: | |
399caddf | 1172 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1173 | break; |
1174 | ||
1175 | default: | |
c397741c MS |
1176 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1177 | __func__, r); | |
d6fc2042 | 1178 | cell_error(pool, cell); |
991d9fa0 JT |
1179 | break; |
1180 | } | |
1181 | } | |
1182 | ||
1183 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1184 | dm_block_t block, | |
1185 | struct dm_thin_lookup_result *lookup_result) | |
1186 | { | |
a24c2569 | 1187 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1188 | struct pool *pool = tc->pool; |
44feb387 | 1189 | struct dm_cell_key key; |
991d9fa0 JT |
1190 | |
1191 | /* | |
1192 | * If cell is already occupied, then sharing is already in the process | |
1193 | * of being broken so we have nothing further to do here. | |
1194 | */ | |
1195 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1196 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1197 | return; |
1198 | ||
4f024f37 | 1199 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
991d9fa0 JT |
1200 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1201 | else { | |
59c3d2c6 | 1202 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1203 | |
44feb387 | 1204 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1205 | inc_all_io_entry(pool, bio); |
f286ba0e | 1206 | cell_defer_no_holder(tc, cell); |
e8088073 | 1207 | |
991d9fa0 JT |
1208 | remap_and_issue(tc, bio, lookup_result->block); |
1209 | } | |
1210 | } | |
1211 | ||
1212 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1213 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1214 | { |
1215 | int r; | |
1216 | dm_block_t data_block; | |
6beca5eb | 1217 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1218 | |
1219 | /* | |
1220 | * Remap empty bios (flushes) immediately, without provisioning. | |
1221 | */ | |
4f024f37 | 1222 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1223 | inc_all_io_entry(pool, bio); |
f286ba0e | 1224 | cell_defer_no_holder(tc, cell); |
e8088073 | 1225 | |
991d9fa0 JT |
1226 | remap_and_issue(tc, bio, 0); |
1227 | return; | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * Fill read bios with zeroes and complete them immediately. | |
1232 | */ | |
1233 | if (bio_data_dir(bio) == READ) { | |
1234 | zero_fill_bio(bio); | |
f286ba0e | 1235 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1236 | bio_endio(bio, 0); |
1237 | return; | |
1238 | } | |
1239 | ||
1240 | r = alloc_data_block(tc, &data_block); | |
1241 | switch (r) { | |
1242 | case 0: | |
2dd9c257 JT |
1243 | if (tc->origin_dev) |
1244 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1245 | else | |
1246 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1247 | break; |
1248 | ||
1249 | case -ENOSPC: | |
399caddf | 1250 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1251 | break; |
1252 | ||
1253 | default: | |
c397741c MS |
1254 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1255 | __func__, r); | |
6beca5eb | 1256 | cell_error(pool, cell); |
991d9fa0 JT |
1257 | break; |
1258 | } | |
1259 | } | |
1260 | ||
1261 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1262 | { | |
1263 | int r; | |
6beca5eb | 1264 | struct pool *pool = tc->pool; |
991d9fa0 | 1265 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1266 | struct dm_bio_prison_cell *cell; |
44feb387 | 1267 | struct dm_cell_key key; |
991d9fa0 JT |
1268 | struct dm_thin_lookup_result lookup_result; |
1269 | ||
1270 | /* | |
1271 | * If cell is already occupied, then the block is already | |
1272 | * being provisioned so we have nothing further to do here. | |
1273 | */ | |
1274 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1275 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1276 | return; |
1277 | ||
1278 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1279 | switch (r) { | |
1280 | case 0: | |
e8088073 | 1281 | if (lookup_result.shared) { |
991d9fa0 | 1282 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1283 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1284 | } else { |
6beca5eb | 1285 | inc_all_io_entry(pool, bio); |
f286ba0e | 1286 | cell_defer_no_holder(tc, cell); |
e8088073 | 1287 | |
991d9fa0 | 1288 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1289 | } |
991d9fa0 JT |
1290 | break; |
1291 | ||
1292 | case -ENODATA: | |
2dd9c257 | 1293 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1294 | inc_all_io_entry(pool, bio); |
f286ba0e | 1295 | cell_defer_no_holder(tc, cell); |
e8088073 | 1296 | |
2dd9c257 JT |
1297 | remap_to_origin_and_issue(tc, bio); |
1298 | } else | |
1299 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1300 | break; |
1301 | ||
1302 | default: | |
c397741c MS |
1303 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1304 | __func__, r); | |
f286ba0e | 1305 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1306 | bio_io_error(bio); |
1307 | break; | |
1308 | } | |
1309 | } | |
1310 | ||
e49e5829 JT |
1311 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1312 | { | |
1313 | int r; | |
1314 | int rw = bio_data_dir(bio); | |
1315 | dm_block_t block = get_bio_block(tc, bio); | |
1316 | struct dm_thin_lookup_result lookup_result; | |
1317 | ||
1318 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1319 | switch (r) { | |
1320 | case 0: | |
4f024f37 | 1321 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
8c0f0e8c | 1322 | handle_unserviceable_bio(tc->pool, bio); |
e8088073 JT |
1323 | else { |
1324 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1325 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1326 | } |
e49e5829 JT |
1327 | break; |
1328 | ||
1329 | case -ENODATA: | |
1330 | if (rw != READ) { | |
8c0f0e8c | 1331 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1332 | break; |
1333 | } | |
1334 | ||
1335 | if (tc->origin_dev) { | |
e8088073 | 1336 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1337 | remap_to_origin_and_issue(tc, bio); |
1338 | break; | |
1339 | } | |
1340 | ||
1341 | zero_fill_bio(bio); | |
1342 | bio_endio(bio, 0); | |
1343 | break; | |
1344 | ||
1345 | default: | |
c397741c MS |
1346 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1347 | __func__, r); | |
e49e5829 JT |
1348 | bio_io_error(bio); |
1349 | break; | |
1350 | } | |
1351 | } | |
1352 | ||
3e1a0699 JT |
1353 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1354 | { | |
1355 | bio_endio(bio, 0); | |
1356 | } | |
1357 | ||
e49e5829 JT |
1358 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1359 | { | |
1360 | bio_io_error(bio); | |
1361 | } | |
1362 | ||
ac8c3f3d JT |
1363 | /* |
1364 | * FIXME: should we also commit due to size of transaction, measured in | |
1365 | * metadata blocks? | |
1366 | */ | |
905e51b3 JT |
1367 | static int need_commit_due_to_time(struct pool *pool) |
1368 | { | |
1369 | return jiffies < pool->last_commit_jiffies || | |
1370 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1371 | } | |
1372 | ||
67324ea1 MS |
1373 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1374 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1375 | ||
1376 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1377 | { | |
1378 | struct rb_node **rbp, *parent; | |
1379 | struct dm_thin_endio_hook *pbd; | |
1380 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1381 | ||
1382 | rbp = &tc->sort_bio_list.rb_node; | |
1383 | parent = NULL; | |
1384 | while (*rbp) { | |
1385 | parent = *rbp; | |
1386 | pbd = thin_pbd(parent); | |
1387 | ||
1388 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1389 | rbp = &(*rbp)->rb_left; | |
1390 | else | |
1391 | rbp = &(*rbp)->rb_right; | |
1392 | } | |
1393 | ||
1394 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1395 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1396 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1397 | } | |
1398 | ||
1399 | static void __extract_sorted_bios(struct thin_c *tc) | |
1400 | { | |
1401 | struct rb_node *node; | |
1402 | struct dm_thin_endio_hook *pbd; | |
1403 | struct bio *bio; | |
1404 | ||
1405 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1406 | pbd = thin_pbd(node); | |
1407 | bio = thin_bio(pbd); | |
1408 | ||
1409 | bio_list_add(&tc->deferred_bio_list, bio); | |
1410 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1411 | } | |
1412 | ||
1413 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1414 | } | |
1415 | ||
1416 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1417 | { | |
1418 | struct bio *bio; | |
1419 | struct bio_list bios; | |
1420 | ||
1421 | bio_list_init(&bios); | |
1422 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1423 | bio_list_init(&tc->deferred_bio_list); | |
1424 | ||
1425 | /* Sort deferred_bio_list using rb-tree */ | |
1426 | while ((bio = bio_list_pop(&bios))) | |
1427 | __thin_bio_rb_add(tc, bio); | |
1428 | ||
1429 | /* | |
1430 | * Transfer the sorted bios in sort_bio_list back to | |
1431 | * deferred_bio_list to allow lockless submission of | |
1432 | * all bios. | |
1433 | */ | |
1434 | __extract_sorted_bios(tc); | |
1435 | } | |
1436 | ||
c140e1c4 | 1437 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1438 | { |
c140e1c4 | 1439 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1440 | unsigned long flags; |
1441 | struct bio *bio; | |
1442 | struct bio_list bios; | |
67324ea1 | 1443 | struct blk_plug plug; |
991d9fa0 | 1444 | |
c140e1c4 MS |
1445 | if (tc->requeue_mode) { |
1446 | requeue_bio_list(tc, &tc->deferred_bio_list); | |
1447 | return; | |
1448 | } | |
1449 | ||
991d9fa0 JT |
1450 | bio_list_init(&bios); |
1451 | ||
c140e1c4 | 1452 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1453 | |
1454 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1455 | spin_unlock_irqrestore(&tc->lock, flags); | |
1456 | return; | |
1457 | } | |
1458 | ||
1459 | __sort_thin_deferred_bios(tc); | |
1460 | ||
c140e1c4 MS |
1461 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1462 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1463 | |
c140e1c4 | 1464 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1465 | |
67324ea1 | 1466 | blk_start_plug(&plug); |
991d9fa0 | 1467 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1468 | /* |
1469 | * If we've got no free new_mapping structs, and processing | |
1470 | * this bio might require one, we pause until there are some | |
1471 | * prepared mappings to process. | |
1472 | */ | |
1473 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1474 | spin_lock_irqsave(&tc->lock, flags); |
1475 | bio_list_add(&tc->deferred_bio_list, bio); | |
1476 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1477 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1478 | break; |
1479 | } | |
104655fd JT |
1480 | |
1481 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1482 | pool->process_discard(tc, bio); |
104655fd | 1483 | else |
e49e5829 | 1484 | pool->process_bio(tc, bio); |
991d9fa0 | 1485 | } |
67324ea1 | 1486 | blk_finish_plug(&plug); |
c140e1c4 MS |
1487 | } |
1488 | ||
1489 | static void process_deferred_bios(struct pool *pool) | |
1490 | { | |
1491 | unsigned long flags; | |
1492 | struct bio *bio; | |
1493 | struct bio_list bios; | |
1494 | struct thin_c *tc; | |
1495 | ||
1496 | rcu_read_lock(); | |
1497 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
1498 | process_thin_deferred_bios(tc); | |
1499 | rcu_read_unlock(); | |
991d9fa0 JT |
1500 | |
1501 | /* | |
1502 | * If there are any deferred flush bios, we must commit | |
1503 | * the metadata before issuing them. | |
1504 | */ | |
1505 | bio_list_init(&bios); | |
1506 | spin_lock_irqsave(&pool->lock, flags); | |
1507 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1508 | bio_list_init(&pool->deferred_flush_bios); | |
1509 | spin_unlock_irqrestore(&pool->lock, flags); | |
1510 | ||
4d1662a3 MS |
1511 | if (bio_list_empty(&bios) && |
1512 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1513 | return; |
1514 | ||
020cc3b5 | 1515 | if (commit(pool)) { |
991d9fa0 JT |
1516 | while ((bio = bio_list_pop(&bios))) |
1517 | bio_io_error(bio); | |
1518 | return; | |
1519 | } | |
905e51b3 | 1520 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1521 | |
1522 | while ((bio = bio_list_pop(&bios))) | |
1523 | generic_make_request(bio); | |
1524 | } | |
1525 | ||
1526 | static void do_worker(struct work_struct *ws) | |
1527 | { | |
1528 | struct pool *pool = container_of(ws, struct pool, worker); | |
1529 | ||
e49e5829 JT |
1530 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1531 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1532 | process_deferred_bios(pool); |
1533 | } | |
1534 | ||
905e51b3 JT |
1535 | /* |
1536 | * We want to commit periodically so that not too much | |
1537 | * unwritten data builds up. | |
1538 | */ | |
1539 | static void do_waker(struct work_struct *ws) | |
1540 | { | |
1541 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1542 | wake_worker(pool); | |
1543 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1544 | } | |
1545 | ||
991d9fa0 JT |
1546 | /*----------------------------------------------------------------*/ |
1547 | ||
738211f7 JT |
1548 | struct noflush_work { |
1549 | struct work_struct worker; | |
1550 | struct thin_c *tc; | |
1551 | ||
1552 | atomic_t complete; | |
1553 | wait_queue_head_t wait; | |
1554 | }; | |
1555 | ||
1556 | static void complete_noflush_work(struct noflush_work *w) | |
1557 | { | |
1558 | atomic_set(&w->complete, 1); | |
1559 | wake_up(&w->wait); | |
1560 | } | |
1561 | ||
1562 | static void do_noflush_start(struct work_struct *ws) | |
1563 | { | |
1564 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1565 | w->tc->requeue_mode = true; | |
1566 | requeue_io(w->tc); | |
1567 | complete_noflush_work(w); | |
1568 | } | |
1569 | ||
1570 | static void do_noflush_stop(struct work_struct *ws) | |
1571 | { | |
1572 | struct noflush_work *w = container_of(ws, struct noflush_work, worker); | |
1573 | w->tc->requeue_mode = false; | |
1574 | complete_noflush_work(w); | |
1575 | } | |
1576 | ||
1577 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
1578 | { | |
1579 | struct noflush_work w; | |
1580 | ||
1581 | INIT_WORK(&w.worker, fn); | |
1582 | w.tc = tc; | |
1583 | atomic_set(&w.complete, 0); | |
1584 | init_waitqueue_head(&w.wait); | |
1585 | ||
1586 | queue_work(tc->pool->wq, &w.worker); | |
1587 | ||
1588 | wait_event(w.wait, atomic_read(&w.complete)); | |
1589 | } | |
1590 | ||
1591 | /*----------------------------------------------------------------*/ | |
1592 | ||
e49e5829 JT |
1593 | static enum pool_mode get_pool_mode(struct pool *pool) |
1594 | { | |
1595 | return pool->pf.mode; | |
1596 | } | |
1597 | ||
3e1a0699 JT |
1598 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
1599 | { | |
1600 | dm_table_event(pool->ti->table); | |
1601 | DMINFO("%s: switching pool to %s mode", | |
1602 | dm_device_name(pool->pool_md), new_mode); | |
1603 | } | |
1604 | ||
8b64e881 | 1605 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 1606 | { |
cdc2b415 | 1607 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
1608 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
1609 | enum pool_mode old_mode = get_pool_mode(pool); | |
1610 | ||
1611 | /* | |
1612 | * Never allow the pool to transition to PM_WRITE mode if user | |
1613 | * intervention is required to verify metadata and data consistency. | |
1614 | */ | |
1615 | if (new_mode == PM_WRITE && needs_check) { | |
1616 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
1617 | dm_device_name(pool->pool_md)); | |
1618 | if (old_mode != new_mode) | |
1619 | new_mode = old_mode; | |
1620 | else | |
1621 | new_mode = PM_READ_ONLY; | |
1622 | } | |
1623 | /* | |
1624 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1625 | * not going to recover without a thin_repair. So we never let the | |
1626 | * pool move out of the old mode. | |
1627 | */ | |
1628 | if (old_mode == PM_FAIL) | |
1629 | new_mode = old_mode; | |
e49e5829 | 1630 | |
8b64e881 | 1631 | switch (new_mode) { |
e49e5829 | 1632 | case PM_FAIL: |
8b64e881 | 1633 | if (old_mode != new_mode) |
3e1a0699 | 1634 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 1635 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1636 | pool->process_bio = process_bio_fail; |
1637 | pool->process_discard = process_bio_fail; | |
1638 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1639 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
1640 | |
1641 | error_retry_list(pool); | |
e49e5829 JT |
1642 | break; |
1643 | ||
1644 | case PM_READ_ONLY: | |
8b64e881 | 1645 | if (old_mode != new_mode) |
3e1a0699 JT |
1646 | notify_of_pool_mode_change(pool, "read-only"); |
1647 | dm_pool_metadata_read_only(pool->pmd); | |
1648 | pool->process_bio = process_bio_read_only; | |
1649 | pool->process_discard = process_bio_success; | |
1650 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1651 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1652 | ||
1653 | error_retry_list(pool); | |
1654 | break; | |
1655 | ||
1656 | case PM_OUT_OF_DATA_SPACE: | |
1657 | /* | |
1658 | * Ideally we'd never hit this state; the low water mark | |
1659 | * would trigger userland to extend the pool before we | |
1660 | * completely run out of data space. However, many small | |
1661 | * IOs to unprovisioned space can consume data space at an | |
1662 | * alarming rate. Adjust your low water mark if you're | |
1663 | * frequently seeing this mode. | |
1664 | */ | |
1665 | if (old_mode != new_mode) | |
1666 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
1667 | pool->process_bio = process_bio_read_only; | |
1668 | pool->process_discard = process_discard; | |
1669 | pool->process_prepared_mapping = process_prepared_mapping; | |
1670 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
e49e5829 JT |
1671 | break; |
1672 | ||
1673 | case PM_WRITE: | |
8b64e881 | 1674 | if (old_mode != new_mode) |
3e1a0699 | 1675 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 1676 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1677 | pool->process_bio = process_bio; |
1678 | pool->process_discard = process_discard; | |
1679 | pool->process_prepared_mapping = process_prepared_mapping; | |
1680 | pool->process_prepared_discard = process_prepared_discard; | |
1681 | break; | |
1682 | } | |
8b64e881 MS |
1683 | |
1684 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
1685 | /* |
1686 | * The pool mode may have changed, sync it so bind_control_target() | |
1687 | * doesn't cause an unexpected mode transition on resume. | |
1688 | */ | |
1689 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
1690 | } |
1691 | ||
07f2b6e0 | 1692 | static void abort_transaction(struct pool *pool) |
b5330655 | 1693 | { |
07f2b6e0 MS |
1694 | const char *dev_name = dm_device_name(pool->pool_md); |
1695 | ||
1696 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
1697 | if (dm_pool_abort_metadata(pool->pmd)) { | |
1698 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
1699 | set_pool_mode(pool, PM_FAIL); | |
1700 | } | |
1701 | ||
1702 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
1703 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
1704 | set_pool_mode(pool, PM_FAIL); | |
1705 | } | |
1706 | } | |
399caddf | 1707 | |
07f2b6e0 MS |
1708 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1709 | { | |
b5330655 JT |
1710 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1711 | dm_device_name(pool->pool_md), op, r); | |
1712 | ||
07f2b6e0 | 1713 | abort_transaction(pool); |
b5330655 JT |
1714 | set_pool_mode(pool, PM_READ_ONLY); |
1715 | } | |
1716 | ||
e49e5829 JT |
1717 | /*----------------------------------------------------------------*/ |
1718 | ||
991d9fa0 JT |
1719 | /* |
1720 | * Mapping functions. | |
1721 | */ | |
1722 | ||
1723 | /* | |
1724 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1725 | */ | |
1726 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1727 | { | |
1728 | unsigned long flags; | |
1729 | struct pool *pool = tc->pool; | |
1730 | ||
c140e1c4 MS |
1731 | spin_lock_irqsave(&tc->lock, flags); |
1732 | bio_list_add(&tc->deferred_bio_list, bio); | |
1733 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1734 | |
1735 | wake_worker(pool); | |
1736 | } | |
1737 | ||
59c3d2c6 | 1738 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1739 | { |
59c3d2c6 | 1740 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1741 | |
1742 | h->tc = tc; | |
1743 | h->shared_read_entry = NULL; | |
e8088073 | 1744 | h->all_io_entry = NULL; |
eb2aa48d | 1745 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1746 | } |
1747 | ||
991d9fa0 JT |
1748 | /* |
1749 | * Non-blocking function called from the thin target's map function. | |
1750 | */ | |
7de3ee57 | 1751 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1752 | { |
1753 | int r; | |
1754 | struct thin_c *tc = ti->private; | |
1755 | dm_block_t block = get_bio_block(tc, bio); | |
1756 | struct dm_thin_device *td = tc->td; | |
1757 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1758 | struct dm_bio_prison_cell cell1, cell2; |
1759 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1760 | struct dm_cell_key key; |
991d9fa0 | 1761 | |
59c3d2c6 | 1762 | thin_hook_bio(tc, bio); |
e49e5829 | 1763 | |
738211f7 JT |
1764 | if (tc->requeue_mode) { |
1765 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1766 | return DM_MAPIO_SUBMITTED; | |
1767 | } | |
1768 | ||
e49e5829 JT |
1769 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
1770 | bio_io_error(bio); | |
1771 | return DM_MAPIO_SUBMITTED; | |
1772 | } | |
1773 | ||
104655fd | 1774 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1775 | thin_defer_bio(tc, bio); |
1776 | return DM_MAPIO_SUBMITTED; | |
1777 | } | |
1778 | ||
1779 | r = dm_thin_find_block(td, block, 0, &result); | |
1780 | ||
1781 | /* | |
1782 | * Note that we defer readahead too. | |
1783 | */ | |
1784 | switch (r) { | |
1785 | case 0: | |
1786 | if (unlikely(result.shared)) { | |
1787 | /* | |
1788 | * We have a race condition here between the | |
1789 | * result.shared value returned by the lookup and | |
1790 | * snapshot creation, which may cause new | |
1791 | * sharing. | |
1792 | * | |
1793 | * To avoid this always quiesce the origin before | |
1794 | * taking the snap. You want to do this anyway to | |
1795 | * ensure a consistent application view | |
1796 | * (i.e. lockfs). | |
1797 | * | |
1798 | * More distant ancestors are irrelevant. The | |
1799 | * shared flag will be set in their case. | |
1800 | */ | |
1801 | thin_defer_bio(tc, bio); | |
e8088073 | 1802 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1803 | } |
e8088073 JT |
1804 | |
1805 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1806 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1807 | return DM_MAPIO_SUBMITTED; |
1808 | ||
1809 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1810 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1811 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1812 | return DM_MAPIO_SUBMITTED; |
1813 | } | |
1814 | ||
1815 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1816 | cell_defer_no_holder_no_free(tc, &cell2); |
1817 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1818 | |
1819 | remap(tc, bio, result.block); | |
1820 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1821 | |
1822 | case -ENODATA: | |
e49e5829 JT |
1823 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1824 | /* | |
1825 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 1826 | * of doing so. |
e49e5829 | 1827 | */ |
8c0f0e8c | 1828 | handle_unserviceable_bio(tc->pool, bio); |
2aab3850 | 1829 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1830 | } |
1831 | /* fall through */ | |
1832 | ||
1833 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1834 | /* |
1835 | * In future, the failed dm_thin_find_block above could | |
1836 | * provide the hint to load the metadata into cache. | |
1837 | */ | |
991d9fa0 | 1838 | thin_defer_bio(tc, bio); |
2aab3850 | 1839 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1840 | |
1841 | default: | |
1842 | /* | |
1843 | * Must always call bio_io_error on failure. | |
1844 | * dm_thin_find_block can fail with -EINVAL if the | |
1845 | * pool is switched to fail-io mode. | |
1846 | */ | |
1847 | bio_io_error(bio); | |
2aab3850 | 1848 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1849 | } |
991d9fa0 JT |
1850 | } |
1851 | ||
1852 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1853 | { | |
991d9fa0 | 1854 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 1855 | struct request_queue *q; |
991d9fa0 | 1856 | |
760fe67e MS |
1857 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
1858 | return 1; | |
991d9fa0 | 1859 | |
760fe67e MS |
1860 | q = bdev_get_queue(pt->data_dev->bdev); |
1861 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
1862 | } |
1863 | ||
c140e1c4 | 1864 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 1865 | { |
c140e1c4 MS |
1866 | unsigned long flags; |
1867 | struct thin_c *tc; | |
1868 | ||
1869 | rcu_read_lock(); | |
1870 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
1871 | spin_lock_irqsave(&tc->lock, flags); | |
1872 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
1873 | bio_list_init(&tc->retry_on_resume_list); | |
1874 | spin_unlock_irqrestore(&tc->lock, flags); | |
1875 | } | |
1876 | rcu_read_unlock(); | |
991d9fa0 JT |
1877 | } |
1878 | ||
1879 | /*---------------------------------------------------------------- | |
1880 | * Binding of control targets to a pool object | |
1881 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1882 | static bool data_dev_supports_discard(struct pool_c *pt) |
1883 | { | |
1884 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1885 | ||
1886 | return q && blk_queue_discard(q); | |
1887 | } | |
1888 | ||
58051b94 JT |
1889 | static bool is_factor(sector_t block_size, uint32_t n) |
1890 | { | |
1891 | return !sector_div(block_size, n); | |
1892 | } | |
1893 | ||
9bc142dd MS |
1894 | /* |
1895 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1896 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1897 | */ |
0424caa1 | 1898 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1899 | { |
0424caa1 MS |
1900 | struct pool *pool = pt->pool; |
1901 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1902 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1903 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1904 | const char *reason = NULL; | |
9bc142dd MS |
1905 | char buf[BDEVNAME_SIZE]; |
1906 | ||
0424caa1 | 1907 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1908 | return; |
1909 | ||
0424caa1 MS |
1910 | if (!data_dev_supports_discard(pt)) |
1911 | reason = "discard unsupported"; | |
1912 | ||
1913 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1914 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1915 | |
0424caa1 MS |
1916 | else if (data_limits->discard_granularity > block_size) |
1917 | reason = "discard granularity larger than a block"; | |
1918 | ||
58051b94 | 1919 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1920 | reason = "discard granularity not a factor of block size"; |
1921 | ||
1922 | if (reason) { | |
1923 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1924 | pt->adjusted_pf.discard_passdown = false; | |
1925 | } | |
9bc142dd MS |
1926 | } |
1927 | ||
991d9fa0 JT |
1928 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1929 | { | |
1930 | struct pool_c *pt = ti->private; | |
1931 | ||
e49e5829 | 1932 | /* |
9b7aaa64 | 1933 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 1934 | */ |
07f2b6e0 | 1935 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 1936 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1937 | |
8b64e881 MS |
1938 | /* |
1939 | * Don't change the pool's mode until set_pool_mode() below. | |
1940 | * Otherwise the pool's process_* function pointers may | |
1941 | * not match the desired pool mode. | |
1942 | */ | |
1943 | pt->adjusted_pf.mode = old_mode; | |
1944 | ||
1945 | pool->ti = ti; | |
1946 | pool->pf = pt->adjusted_pf; | |
1947 | pool->low_water_blocks = pt->low_water_blocks; | |
1948 | ||
9bc142dd | 1949 | set_pool_mode(pool, new_mode); |
f402693d | 1950 | |
991d9fa0 JT |
1951 | return 0; |
1952 | } | |
1953 | ||
1954 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1955 | { | |
1956 | if (pool->ti == ti) | |
1957 | pool->ti = NULL; | |
1958 | } | |
1959 | ||
1960 | /*---------------------------------------------------------------- | |
1961 | * Pool creation | |
1962 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1963 | /* Initialize pool features. */ |
1964 | static void pool_features_init(struct pool_features *pf) | |
1965 | { | |
e49e5829 | 1966 | pf->mode = PM_WRITE; |
9bc142dd MS |
1967 | pf->zero_new_blocks = true; |
1968 | pf->discard_enabled = true; | |
1969 | pf->discard_passdown = true; | |
787a996c | 1970 | pf->error_if_no_space = false; |
67e2e2b2 JT |
1971 | } |
1972 | ||
991d9fa0 JT |
1973 | static void __pool_destroy(struct pool *pool) |
1974 | { | |
1975 | __pool_table_remove(pool); | |
1976 | ||
1977 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1978 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1979 | ||
44feb387 | 1980 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1981 | dm_kcopyd_client_destroy(pool->copier); |
1982 | ||
1983 | if (pool->wq) | |
1984 | destroy_workqueue(pool->wq); | |
1985 | ||
1986 | if (pool->next_mapping) | |
1987 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1988 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1989 | dm_deferred_set_destroy(pool->shared_read_ds); |
1990 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1991 | kfree(pool); |
1992 | } | |
1993 | ||
a24c2569 | 1994 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1995 | |
991d9fa0 JT |
1996 | static struct pool *pool_create(struct mapped_device *pool_md, |
1997 | struct block_device *metadata_dev, | |
e49e5829 JT |
1998 | unsigned long block_size, |
1999 | int read_only, char **error) | |
991d9fa0 JT |
2000 | { |
2001 | int r; | |
2002 | void *err_p; | |
2003 | struct pool *pool; | |
2004 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2005 | bool format_device = read_only ? false : true; |
991d9fa0 | 2006 | |
e49e5829 | 2007 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2008 | if (IS_ERR(pmd)) { |
2009 | *error = "Error creating metadata object"; | |
2010 | return (struct pool *)pmd; | |
2011 | } | |
2012 | ||
2013 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2014 | if (!pool) { | |
2015 | *error = "Error allocating memory for pool"; | |
2016 | err_p = ERR_PTR(-ENOMEM); | |
2017 | goto bad_pool; | |
2018 | } | |
2019 | ||
2020 | pool->pmd = pmd; | |
2021 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2022 | if (block_size & (block_size - 1)) |
2023 | pool->sectors_per_block_shift = -1; | |
2024 | else | |
2025 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2026 | pool->low_water_blocks = 0; |
67e2e2b2 | 2027 | pool_features_init(&pool->pf); |
44feb387 | 2028 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
2029 | if (!pool->prison) { |
2030 | *error = "Error creating pool's bio prison"; | |
2031 | err_p = ERR_PTR(-ENOMEM); | |
2032 | goto bad_prison; | |
2033 | } | |
2034 | ||
df5d2e90 | 2035 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2036 | if (IS_ERR(pool->copier)) { |
2037 | r = PTR_ERR(pool->copier); | |
2038 | *error = "Error creating pool's kcopyd client"; | |
2039 | err_p = ERR_PTR(r); | |
2040 | goto bad_kcopyd_client; | |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * Create singlethreaded workqueue that will service all devices | |
2045 | * that use this metadata. | |
2046 | */ | |
2047 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2048 | if (!pool->wq) { | |
2049 | *error = "Error creating pool's workqueue"; | |
2050 | err_p = ERR_PTR(-ENOMEM); | |
2051 | goto bad_wq; | |
2052 | } | |
2053 | ||
2054 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 2055 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 | 2056 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2057 | bio_list_init(&pool->deferred_flush_bios); |
2058 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2059 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2060 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2061 | pool->low_water_triggered = false; |
44feb387 MS |
2062 | |
2063 | pool->shared_read_ds = dm_deferred_set_create(); | |
2064 | if (!pool->shared_read_ds) { | |
2065 | *error = "Error creating pool's shared read deferred set"; | |
2066 | err_p = ERR_PTR(-ENOMEM); | |
2067 | goto bad_shared_read_ds; | |
2068 | } | |
2069 | ||
2070 | pool->all_io_ds = dm_deferred_set_create(); | |
2071 | if (!pool->all_io_ds) { | |
2072 | *error = "Error creating pool's all io deferred set"; | |
2073 | err_p = ERR_PTR(-ENOMEM); | |
2074 | goto bad_all_io_ds; | |
2075 | } | |
991d9fa0 JT |
2076 | |
2077 | pool->next_mapping = NULL; | |
a24c2569 MS |
2078 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2079 | _new_mapping_cache); | |
991d9fa0 JT |
2080 | if (!pool->mapping_pool) { |
2081 | *error = "Error creating pool's mapping mempool"; | |
2082 | err_p = ERR_PTR(-ENOMEM); | |
2083 | goto bad_mapping_pool; | |
2084 | } | |
2085 | ||
991d9fa0 | 2086 | pool->ref_count = 1; |
905e51b3 | 2087 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2088 | pool->pool_md = pool_md; |
2089 | pool->md_dev = metadata_dev; | |
2090 | __pool_table_insert(pool); | |
2091 | ||
2092 | return pool; | |
2093 | ||
991d9fa0 | 2094 | bad_mapping_pool: |
44feb387 MS |
2095 | dm_deferred_set_destroy(pool->all_io_ds); |
2096 | bad_all_io_ds: | |
2097 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2098 | bad_shared_read_ds: | |
991d9fa0 JT |
2099 | destroy_workqueue(pool->wq); |
2100 | bad_wq: | |
2101 | dm_kcopyd_client_destroy(pool->copier); | |
2102 | bad_kcopyd_client: | |
44feb387 | 2103 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2104 | bad_prison: |
2105 | kfree(pool); | |
2106 | bad_pool: | |
2107 | if (dm_pool_metadata_close(pmd)) | |
2108 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2109 | ||
2110 | return err_p; | |
2111 | } | |
2112 | ||
2113 | static void __pool_inc(struct pool *pool) | |
2114 | { | |
2115 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2116 | pool->ref_count++; | |
2117 | } | |
2118 | ||
2119 | static void __pool_dec(struct pool *pool) | |
2120 | { | |
2121 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2122 | BUG_ON(!pool->ref_count); | |
2123 | if (!--pool->ref_count) | |
2124 | __pool_destroy(pool); | |
2125 | } | |
2126 | ||
2127 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2128 | struct block_device *metadata_dev, | |
e49e5829 JT |
2129 | unsigned long block_size, int read_only, |
2130 | char **error, int *created) | |
991d9fa0 JT |
2131 | { |
2132 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2133 | ||
2134 | if (pool) { | |
f09996c9 MS |
2135 | if (pool->pool_md != pool_md) { |
2136 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2137 | return ERR_PTR(-EBUSY); |
f09996c9 | 2138 | } |
991d9fa0 JT |
2139 | __pool_inc(pool); |
2140 | ||
2141 | } else { | |
2142 | pool = __pool_table_lookup(pool_md); | |
2143 | if (pool) { | |
f09996c9 MS |
2144 | if (pool->md_dev != metadata_dev) { |
2145 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2146 | return ERR_PTR(-EINVAL); |
f09996c9 | 2147 | } |
991d9fa0 JT |
2148 | __pool_inc(pool); |
2149 | ||
67e2e2b2 | 2150 | } else { |
e49e5829 | 2151 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2152 | *created = 1; |
2153 | } | |
991d9fa0 JT |
2154 | } |
2155 | ||
2156 | return pool; | |
2157 | } | |
2158 | ||
2159 | /*---------------------------------------------------------------- | |
2160 | * Pool target methods | |
2161 | *--------------------------------------------------------------*/ | |
2162 | static void pool_dtr(struct dm_target *ti) | |
2163 | { | |
2164 | struct pool_c *pt = ti->private; | |
2165 | ||
2166 | mutex_lock(&dm_thin_pool_table.mutex); | |
2167 | ||
2168 | unbind_control_target(pt->pool, ti); | |
2169 | __pool_dec(pt->pool); | |
2170 | dm_put_device(ti, pt->metadata_dev); | |
2171 | dm_put_device(ti, pt->data_dev); | |
2172 | kfree(pt); | |
2173 | ||
2174 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2175 | } | |
2176 | ||
991d9fa0 JT |
2177 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2178 | struct dm_target *ti) | |
2179 | { | |
2180 | int r; | |
2181 | unsigned argc; | |
2182 | const char *arg_name; | |
2183 | ||
2184 | static struct dm_arg _args[] = { | |
74aa45c3 | 2185 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2186 | }; |
2187 | ||
2188 | /* | |
2189 | * No feature arguments supplied. | |
2190 | */ | |
2191 | if (!as->argc) | |
2192 | return 0; | |
2193 | ||
2194 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2195 | if (r) | |
2196 | return -EINVAL; | |
2197 | ||
2198 | while (argc && !r) { | |
2199 | arg_name = dm_shift_arg(as); | |
2200 | argc--; | |
2201 | ||
e49e5829 | 2202 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2203 | pf->zero_new_blocks = false; |
e49e5829 JT |
2204 | |
2205 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2206 | pf->discard_enabled = false; |
e49e5829 JT |
2207 | |
2208 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2209 | pf->discard_passdown = false; |
991d9fa0 | 2210 | |
e49e5829 JT |
2211 | else if (!strcasecmp(arg_name, "read_only")) |
2212 | pf->mode = PM_READ_ONLY; | |
2213 | ||
787a996c MS |
2214 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2215 | pf->error_if_no_space = true; | |
2216 | ||
e49e5829 JT |
2217 | else { |
2218 | ti->error = "Unrecognised pool feature requested"; | |
2219 | r = -EINVAL; | |
2220 | break; | |
2221 | } | |
991d9fa0 JT |
2222 | } |
2223 | ||
2224 | return r; | |
2225 | } | |
2226 | ||
ac8c3f3d JT |
2227 | static void metadata_low_callback(void *context) |
2228 | { | |
2229 | struct pool *pool = context; | |
2230 | ||
2231 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2232 | dm_device_name(pool->pool_md)); | |
2233 | ||
2234 | dm_table_event(pool->ti->table); | |
2235 | } | |
2236 | ||
7d48935e MS |
2237 | static sector_t get_dev_size(struct block_device *bdev) |
2238 | { | |
2239 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2240 | } | |
2241 | ||
2242 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2243 | { |
7d48935e | 2244 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2245 | char buffer[BDEVNAME_SIZE]; |
2246 | ||
7d48935e | 2247 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2248 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2249 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2250 | } |
2251 | ||
2252 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2253 | { | |
2254 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2255 | ||
2256 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2257 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2258 | |
2259 | return metadata_dev_size; | |
2260 | } | |
2261 | ||
24347e95 JT |
2262 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2263 | { | |
2264 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2265 | ||
7d48935e | 2266 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2267 | |
2268 | return metadata_dev_size; | |
2269 | } | |
2270 | ||
ac8c3f3d JT |
2271 | /* |
2272 | * When a metadata threshold is crossed a dm event is triggered, and | |
2273 | * userland should respond by growing the metadata device. We could let | |
2274 | * userland set the threshold, like we do with the data threshold, but I'm | |
2275 | * not sure they know enough to do this well. | |
2276 | */ | |
2277 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2278 | { | |
2279 | /* | |
2280 | * 4M is ample for all ops with the possible exception of thin | |
2281 | * device deletion which is harmless if it fails (just retry the | |
2282 | * delete after you've grown the device). | |
2283 | */ | |
2284 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2285 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2286 | } | |
2287 | ||
991d9fa0 JT |
2288 | /* |
2289 | * thin-pool <metadata dev> <data dev> | |
2290 | * <data block size (sectors)> | |
2291 | * <low water mark (blocks)> | |
2292 | * [<#feature args> [<arg>]*] | |
2293 | * | |
2294 | * Optional feature arguments are: | |
2295 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2296 | * ignore_discard: disable discard |
2297 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2298 | * read_only: Don't allow any changes to be made to the pool metadata. |
2299 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2300 | */ |
2301 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2302 | { | |
67e2e2b2 | 2303 | int r, pool_created = 0; |
991d9fa0 JT |
2304 | struct pool_c *pt; |
2305 | struct pool *pool; | |
2306 | struct pool_features pf; | |
2307 | struct dm_arg_set as; | |
2308 | struct dm_dev *data_dev; | |
2309 | unsigned long block_size; | |
2310 | dm_block_t low_water_blocks; | |
2311 | struct dm_dev *metadata_dev; | |
5d0db96d | 2312 | fmode_t metadata_mode; |
991d9fa0 JT |
2313 | |
2314 | /* | |
2315 | * FIXME Remove validation from scope of lock. | |
2316 | */ | |
2317 | mutex_lock(&dm_thin_pool_table.mutex); | |
2318 | ||
2319 | if (argc < 4) { | |
2320 | ti->error = "Invalid argument count"; | |
2321 | r = -EINVAL; | |
2322 | goto out_unlock; | |
2323 | } | |
5d0db96d | 2324 | |
991d9fa0 JT |
2325 | as.argc = argc; |
2326 | as.argv = argv; | |
2327 | ||
5d0db96d JT |
2328 | /* |
2329 | * Set default pool features. | |
2330 | */ | |
2331 | pool_features_init(&pf); | |
2332 | ||
2333 | dm_consume_args(&as, 4); | |
2334 | r = parse_pool_features(&as, &pf, ti); | |
2335 | if (r) | |
2336 | goto out_unlock; | |
2337 | ||
2338 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2339 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2340 | if (r) { |
2341 | ti->error = "Error opening metadata block device"; | |
2342 | goto out_unlock; | |
2343 | } | |
7d48935e | 2344 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2345 | |
2346 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2347 | if (r) { | |
2348 | ti->error = "Error getting data device"; | |
2349 | goto out_metadata; | |
2350 | } | |
2351 | ||
2352 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2353 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2354 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2355 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2356 | ti->error = "Invalid block size"; |
2357 | r = -EINVAL; | |
2358 | goto out; | |
2359 | } | |
2360 | ||
2361 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2362 | ti->error = "Invalid low water mark"; | |
2363 | r = -EINVAL; | |
2364 | goto out; | |
2365 | } | |
2366 | ||
991d9fa0 JT |
2367 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2368 | if (!pt) { | |
2369 | r = -ENOMEM; | |
2370 | goto out; | |
2371 | } | |
2372 | ||
2373 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2374 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2375 | if (IS_ERR(pool)) { |
2376 | r = PTR_ERR(pool); | |
2377 | goto out_free_pt; | |
2378 | } | |
2379 | ||
67e2e2b2 JT |
2380 | /* |
2381 | * 'pool_created' reflects whether this is the first table load. | |
2382 | * Top level discard support is not allowed to be changed after | |
2383 | * initial load. This would require a pool reload to trigger thin | |
2384 | * device changes. | |
2385 | */ | |
2386 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2387 | ti->error = "Discard support cannot be disabled once enabled"; | |
2388 | r = -EINVAL; | |
2389 | goto out_flags_changed; | |
2390 | } | |
2391 | ||
991d9fa0 JT |
2392 | pt->pool = pool; |
2393 | pt->ti = ti; | |
2394 | pt->metadata_dev = metadata_dev; | |
2395 | pt->data_dev = data_dev; | |
2396 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2397 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2398 | ti->num_flush_bios = 1; |
9bc142dd | 2399 | |
67e2e2b2 JT |
2400 | /* |
2401 | * Only need to enable discards if the pool should pass | |
2402 | * them down to the data device. The thin device's discard | |
2403 | * processing will cause mappings to be removed from the btree. | |
2404 | */ | |
b60ab990 | 2405 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2406 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2407 | ti->num_discard_bios = 1; |
9bc142dd | 2408 | |
67e2e2b2 JT |
2409 | /* |
2410 | * Setting 'discards_supported' circumvents the normal | |
2411 | * stacking of discard limits (this keeps the pool and | |
2412 | * thin devices' discard limits consistent). | |
2413 | */ | |
0ac55489 | 2414 | ti->discards_supported = true; |
67e2e2b2 | 2415 | } |
991d9fa0 JT |
2416 | ti->private = pt; |
2417 | ||
ac8c3f3d JT |
2418 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2419 | calc_metadata_threshold(pt), | |
2420 | metadata_low_callback, | |
2421 | pool); | |
2422 | if (r) | |
2423 | goto out_free_pt; | |
2424 | ||
991d9fa0 JT |
2425 | pt->callbacks.congested_fn = pool_is_congested; |
2426 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2427 | ||
2428 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2429 | ||
2430 | return 0; | |
2431 | ||
67e2e2b2 JT |
2432 | out_flags_changed: |
2433 | __pool_dec(pool); | |
991d9fa0 JT |
2434 | out_free_pt: |
2435 | kfree(pt); | |
2436 | out: | |
2437 | dm_put_device(ti, data_dev); | |
2438 | out_metadata: | |
2439 | dm_put_device(ti, metadata_dev); | |
2440 | out_unlock: | |
2441 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2442 | ||
2443 | return r; | |
2444 | } | |
2445 | ||
7de3ee57 | 2446 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2447 | { |
2448 | int r; | |
2449 | struct pool_c *pt = ti->private; | |
2450 | struct pool *pool = pt->pool; | |
2451 | unsigned long flags; | |
2452 | ||
2453 | /* | |
2454 | * As this is a singleton target, ti->begin is always zero. | |
2455 | */ | |
2456 | spin_lock_irqsave(&pool->lock, flags); | |
2457 | bio->bi_bdev = pt->data_dev->bdev; | |
2458 | r = DM_MAPIO_REMAPPED; | |
2459 | spin_unlock_irqrestore(&pool->lock, flags); | |
2460 | ||
2461 | return r; | |
2462 | } | |
2463 | ||
b17446df | 2464 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2465 | { |
2466 | int r; | |
2467 | struct pool_c *pt = ti->private; | |
2468 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2469 | sector_t data_size = ti->len; |
2470 | dm_block_t sb_data_size; | |
991d9fa0 | 2471 | |
b17446df | 2472 | *need_commit = false; |
991d9fa0 | 2473 | |
55f2b8bd MS |
2474 | (void) sector_div(data_size, pool->sectors_per_block); |
2475 | ||
991d9fa0 JT |
2476 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2477 | if (r) { | |
4fa5971a MS |
2478 | DMERR("%s: failed to retrieve data device size", |
2479 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2480 | return r; |
2481 | } | |
2482 | ||
2483 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2484 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2485 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2486 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2487 | return -EINVAL; |
2488 | ||
2489 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2490 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2491 | DMERR("%s: unable to grow the data device until repaired.", | |
2492 | dm_device_name(pool->pool_md)); | |
2493 | return 0; | |
2494 | } | |
2495 | ||
6f7f51d4 MS |
2496 | if (sb_data_size) |
2497 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2498 | dm_device_name(pool->pool_md), | |
2499 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2500 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2501 | if (r) { | |
b5330655 | 2502 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2503 | return r; |
2504 | } | |
2505 | ||
b17446df | 2506 | *need_commit = true; |
991d9fa0 JT |
2507 | } |
2508 | ||
2509 | return 0; | |
2510 | } | |
2511 | ||
24347e95 JT |
2512 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2513 | { | |
2514 | int r; | |
2515 | struct pool_c *pt = ti->private; | |
2516 | struct pool *pool = pt->pool; | |
2517 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2518 | ||
2519 | *need_commit = false; | |
2520 | ||
610bba8b | 2521 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2522 | |
2523 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2524 | if (r) { | |
4fa5971a MS |
2525 | DMERR("%s: failed to retrieve metadata device size", |
2526 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2527 | return r; |
2528 | } | |
2529 | ||
2530 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2531 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2532 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2533 | metadata_dev_size, sb_metadata_dev_size); |
2534 | return -EINVAL; | |
2535 | ||
2536 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
2537 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2538 | DMERR("%s: unable to grow the metadata device until repaired.", | |
2539 | dm_device_name(pool->pool_md)); | |
2540 | return 0; | |
2541 | } | |
2542 | ||
7d48935e | 2543 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
2544 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2545 | dm_device_name(pool->pool_md), | |
2546 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2547 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2548 | if (r) { | |
b5330655 | 2549 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2550 | return r; |
2551 | } | |
2552 | ||
2553 | *need_commit = true; | |
2554 | } | |
2555 | ||
2556 | return 0; | |
2557 | } | |
2558 | ||
b17446df JT |
2559 | /* |
2560 | * Retrieves the number of blocks of the data device from | |
2561 | * the superblock and compares it to the actual device size, | |
2562 | * thus resizing the data device in case it has grown. | |
2563 | * | |
2564 | * This both copes with opening preallocated data devices in the ctr | |
2565 | * being followed by a resume | |
2566 | * -and- | |
2567 | * calling the resume method individually after userspace has | |
2568 | * grown the data device in reaction to a table event. | |
2569 | */ | |
2570 | static int pool_preresume(struct dm_target *ti) | |
2571 | { | |
2572 | int r; | |
24347e95 | 2573 | bool need_commit1, need_commit2; |
b17446df JT |
2574 | struct pool_c *pt = ti->private; |
2575 | struct pool *pool = pt->pool; | |
2576 | ||
2577 | /* | |
2578 | * Take control of the pool object. | |
2579 | */ | |
2580 | r = bind_control_target(pool, ti); | |
2581 | if (r) | |
2582 | return r; | |
2583 | ||
2584 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2585 | if (r) | |
2586 | return r; | |
2587 | ||
24347e95 JT |
2588 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2589 | if (r) | |
2590 | return r; | |
2591 | ||
2592 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2593 | (void) commit(pool); |
b17446df JT |
2594 | |
2595 | return 0; | |
2596 | } | |
2597 | ||
991d9fa0 JT |
2598 | static void pool_resume(struct dm_target *ti) |
2599 | { | |
2600 | struct pool_c *pt = ti->private; | |
2601 | struct pool *pool = pt->pool; | |
2602 | unsigned long flags; | |
2603 | ||
2604 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 2605 | pool->low_water_triggered = false; |
991d9fa0 | 2606 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 2607 | requeue_bios(pool); |
991d9fa0 | 2608 | |
905e51b3 | 2609 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2610 | } |
2611 | ||
2612 | static void pool_postsuspend(struct dm_target *ti) | |
2613 | { | |
991d9fa0 JT |
2614 | struct pool_c *pt = ti->private; |
2615 | struct pool *pool = pt->pool; | |
2616 | ||
905e51b3 | 2617 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2618 | flush_workqueue(pool->wq); |
020cc3b5 | 2619 | (void) commit(pool); |
991d9fa0 JT |
2620 | } |
2621 | ||
2622 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2623 | { | |
2624 | if (argc != args_required) { | |
2625 | DMWARN("Message received with %u arguments instead of %u.", | |
2626 | argc, args_required); | |
2627 | return -EINVAL; | |
2628 | } | |
2629 | ||
2630 | return 0; | |
2631 | } | |
2632 | ||
2633 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2634 | { | |
2635 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2636 | *dev_id <= MAX_DEV_ID) | |
2637 | return 0; | |
2638 | ||
2639 | if (warning) | |
2640 | DMWARN("Message received with invalid device id: %s", arg); | |
2641 | ||
2642 | return -EINVAL; | |
2643 | } | |
2644 | ||
2645 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2646 | { | |
2647 | dm_thin_id dev_id; | |
2648 | int r; | |
2649 | ||
2650 | r = check_arg_count(argc, 2); | |
2651 | if (r) | |
2652 | return r; | |
2653 | ||
2654 | r = read_dev_id(argv[1], &dev_id, 1); | |
2655 | if (r) | |
2656 | return r; | |
2657 | ||
2658 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2659 | if (r) { | |
2660 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2661 | argv[1]); | |
2662 | return r; | |
2663 | } | |
2664 | ||
2665 | return 0; | |
2666 | } | |
2667 | ||
2668 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2669 | { | |
2670 | dm_thin_id dev_id; | |
2671 | dm_thin_id origin_dev_id; | |
2672 | int r; | |
2673 | ||
2674 | r = check_arg_count(argc, 3); | |
2675 | if (r) | |
2676 | return r; | |
2677 | ||
2678 | r = read_dev_id(argv[1], &dev_id, 1); | |
2679 | if (r) | |
2680 | return r; | |
2681 | ||
2682 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2683 | if (r) | |
2684 | return r; | |
2685 | ||
2686 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2687 | if (r) { | |
2688 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2689 | argv[1], argv[2]); | |
2690 | return r; | |
2691 | } | |
2692 | ||
2693 | return 0; | |
2694 | } | |
2695 | ||
2696 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2697 | { | |
2698 | dm_thin_id dev_id; | |
2699 | int r; | |
2700 | ||
2701 | r = check_arg_count(argc, 2); | |
2702 | if (r) | |
2703 | return r; | |
2704 | ||
2705 | r = read_dev_id(argv[1], &dev_id, 1); | |
2706 | if (r) | |
2707 | return r; | |
2708 | ||
2709 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2710 | if (r) | |
2711 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2712 | ||
2713 | return r; | |
2714 | } | |
2715 | ||
2716 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2717 | { | |
2718 | dm_thin_id old_id, new_id; | |
2719 | int r; | |
2720 | ||
2721 | r = check_arg_count(argc, 3); | |
2722 | if (r) | |
2723 | return r; | |
2724 | ||
2725 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2726 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2727 | return -EINVAL; | |
2728 | } | |
2729 | ||
2730 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2731 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2732 | return -EINVAL; | |
2733 | } | |
2734 | ||
2735 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2736 | if (r) { | |
2737 | DMWARN("Failed to change transaction id from %s to %s.", | |
2738 | argv[1], argv[2]); | |
2739 | return r; | |
2740 | } | |
2741 | ||
2742 | return 0; | |
2743 | } | |
2744 | ||
cc8394d8 JT |
2745 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2746 | { | |
2747 | int r; | |
2748 | ||
2749 | r = check_arg_count(argc, 1); | |
2750 | if (r) | |
2751 | return r; | |
2752 | ||
020cc3b5 | 2753 | (void) commit(pool); |
0d200aef | 2754 | |
cc8394d8 JT |
2755 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2756 | if (r) | |
2757 | DMWARN("reserve_metadata_snap message failed."); | |
2758 | ||
2759 | return r; | |
2760 | } | |
2761 | ||
2762 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2763 | { | |
2764 | int r; | |
2765 | ||
2766 | r = check_arg_count(argc, 1); | |
2767 | if (r) | |
2768 | return r; | |
2769 | ||
2770 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2771 | if (r) | |
2772 | DMWARN("release_metadata_snap message failed."); | |
2773 | ||
2774 | return r; | |
2775 | } | |
2776 | ||
991d9fa0 JT |
2777 | /* |
2778 | * Messages supported: | |
2779 | * create_thin <dev_id> | |
2780 | * create_snap <dev_id> <origin_id> | |
2781 | * delete <dev_id> | |
2782 | * trim <dev_id> <new_size_in_sectors> | |
2783 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2784 | * reserve_metadata_snap |
2785 | * release_metadata_snap | |
991d9fa0 JT |
2786 | */ |
2787 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2788 | { | |
2789 | int r = -EINVAL; | |
2790 | struct pool_c *pt = ti->private; | |
2791 | struct pool *pool = pt->pool; | |
2792 | ||
2793 | if (!strcasecmp(argv[0], "create_thin")) | |
2794 | r = process_create_thin_mesg(argc, argv, pool); | |
2795 | ||
2796 | else if (!strcasecmp(argv[0], "create_snap")) | |
2797 | r = process_create_snap_mesg(argc, argv, pool); | |
2798 | ||
2799 | else if (!strcasecmp(argv[0], "delete")) | |
2800 | r = process_delete_mesg(argc, argv, pool); | |
2801 | ||
2802 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2803 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2804 | ||
cc8394d8 JT |
2805 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2806 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2807 | ||
2808 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2809 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2810 | ||
991d9fa0 JT |
2811 | else |
2812 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2813 | ||
e49e5829 | 2814 | if (!r) |
020cc3b5 | 2815 | (void) commit(pool); |
991d9fa0 JT |
2816 | |
2817 | return r; | |
2818 | } | |
2819 | ||
e49e5829 JT |
2820 | static void emit_flags(struct pool_features *pf, char *result, |
2821 | unsigned sz, unsigned maxlen) | |
2822 | { | |
2823 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
2824 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
2825 | pf->error_if_no_space; | |
e49e5829 JT |
2826 | DMEMIT("%u ", count); |
2827 | ||
2828 | if (!pf->zero_new_blocks) | |
2829 | DMEMIT("skip_block_zeroing "); | |
2830 | ||
2831 | if (!pf->discard_enabled) | |
2832 | DMEMIT("ignore_discard "); | |
2833 | ||
2834 | if (!pf->discard_passdown) | |
2835 | DMEMIT("no_discard_passdown "); | |
2836 | ||
2837 | if (pf->mode == PM_READ_ONLY) | |
2838 | DMEMIT("read_only "); | |
787a996c MS |
2839 | |
2840 | if (pf->error_if_no_space) | |
2841 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
2842 | } |
2843 | ||
991d9fa0 JT |
2844 | /* |
2845 | * Status line is: | |
2846 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2847 | * <used data sectors>/<total data sectors> <held metadata root> | |
2848 | */ | |
fd7c092e MP |
2849 | static void pool_status(struct dm_target *ti, status_type_t type, |
2850 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2851 | { |
e49e5829 | 2852 | int r; |
991d9fa0 JT |
2853 | unsigned sz = 0; |
2854 | uint64_t transaction_id; | |
2855 | dm_block_t nr_free_blocks_data; | |
2856 | dm_block_t nr_free_blocks_metadata; | |
2857 | dm_block_t nr_blocks_data; | |
2858 | dm_block_t nr_blocks_metadata; | |
2859 | dm_block_t held_root; | |
2860 | char buf[BDEVNAME_SIZE]; | |
2861 | char buf2[BDEVNAME_SIZE]; | |
2862 | struct pool_c *pt = ti->private; | |
2863 | struct pool *pool = pt->pool; | |
2864 | ||
2865 | switch (type) { | |
2866 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2867 | if (get_pool_mode(pool) == PM_FAIL) { |
2868 | DMEMIT("Fail"); | |
2869 | break; | |
2870 | } | |
2871 | ||
1f4e0ff0 AK |
2872 | /* Commit to ensure statistics aren't out-of-date */ |
2873 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2874 | (void) commit(pool); |
1f4e0ff0 | 2875 | |
fd7c092e MP |
2876 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2877 | if (r) { | |
4fa5971a MS |
2878 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2879 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2880 | goto err; |
2881 | } | |
991d9fa0 | 2882 | |
fd7c092e MP |
2883 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2884 | if (r) { | |
4fa5971a MS |
2885 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2886 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2887 | goto err; |
2888 | } | |
991d9fa0 JT |
2889 | |
2890 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2891 | if (r) { |
4fa5971a MS |
2892 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2893 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2894 | goto err; |
2895 | } | |
991d9fa0 | 2896 | |
fd7c092e MP |
2897 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2898 | if (r) { | |
4fa5971a MS |
2899 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2900 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2901 | goto err; |
2902 | } | |
991d9fa0 JT |
2903 | |
2904 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2905 | if (r) { |
4fa5971a MS |
2906 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2907 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2908 | goto err; |
2909 | } | |
991d9fa0 | 2910 | |
cc8394d8 | 2911 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2912 | if (r) { |
4fa5971a MS |
2913 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2914 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2915 | goto err; |
2916 | } | |
991d9fa0 JT |
2917 | |
2918 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2919 | (unsigned long long)transaction_id, | |
2920 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2921 | (unsigned long long)nr_blocks_metadata, | |
2922 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2923 | (unsigned long long)nr_blocks_data); | |
2924 | ||
2925 | if (held_root) | |
e49e5829 JT |
2926 | DMEMIT("%llu ", held_root); |
2927 | else | |
2928 | DMEMIT("- "); | |
2929 | ||
3e1a0699 JT |
2930 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
2931 | DMEMIT("out_of_data_space "); | |
2932 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 2933 | DMEMIT("ro "); |
991d9fa0 | 2934 | else |
e49e5829 JT |
2935 | DMEMIT("rw "); |
2936 | ||
018debea | 2937 | if (!pool->pf.discard_enabled) |
787a996c | 2938 | DMEMIT("ignore_discard "); |
018debea | 2939 | else if (pool->pf.discard_passdown) |
787a996c MS |
2940 | DMEMIT("discard_passdown "); |
2941 | else | |
2942 | DMEMIT("no_discard_passdown "); | |
2943 | ||
2944 | if (pool->pf.error_if_no_space) | |
2945 | DMEMIT("error_if_no_space "); | |
e49e5829 | 2946 | else |
787a996c | 2947 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
2948 | |
2949 | break; | |
2950 | ||
2951 | case STATUSTYPE_TABLE: | |
2952 | DMEMIT("%s %s %lu %llu ", | |
2953 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2954 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2955 | (unsigned long)pool->sectors_per_block, | |
2956 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2957 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2958 | break; |
2959 | } | |
fd7c092e | 2960 | return; |
991d9fa0 | 2961 | |
fd7c092e MP |
2962 | err: |
2963 | DMEMIT("Error"); | |
991d9fa0 JT |
2964 | } |
2965 | ||
2966 | static int pool_iterate_devices(struct dm_target *ti, | |
2967 | iterate_devices_callout_fn fn, void *data) | |
2968 | { | |
2969 | struct pool_c *pt = ti->private; | |
2970 | ||
2971 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2972 | } | |
2973 | ||
2974 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2975 | struct bio_vec *biovec, int max_size) | |
2976 | { | |
2977 | struct pool_c *pt = ti->private; | |
2978 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2979 | ||
2980 | if (!q->merge_bvec_fn) | |
2981 | return max_size; | |
2982 | ||
2983 | bvm->bi_bdev = pt->data_dev->bdev; | |
2984 | ||
2985 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2986 | } | |
2987 | ||
0424caa1 | 2988 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2989 | { |
0424caa1 MS |
2990 | struct pool *pool = pt->pool; |
2991 | struct queue_limits *data_limits; | |
2992 | ||
104655fd JT |
2993 | limits->max_discard_sectors = pool->sectors_per_block; |
2994 | ||
2995 | /* | |
0424caa1 | 2996 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2997 | */ |
0424caa1 MS |
2998 | if (pt->adjusted_pf.discard_passdown) { |
2999 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
3000 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 3001 | } else |
0424caa1 | 3002 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3003 | } |
3004 | ||
991d9fa0 JT |
3005 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3006 | { | |
3007 | struct pool_c *pt = ti->private; | |
3008 | struct pool *pool = pt->pool; | |
0cc67cd9 | 3009 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 3010 | |
0cc67cd9 MS |
3011 | /* |
3012 | * If the system-determined stacked limits are compatible with the | |
3013 | * pool's blocksize (io_opt is a factor) do not override them. | |
3014 | */ | |
3015 | if (io_opt_sectors < pool->sectors_per_block || | |
3016 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
3017 | blk_limits_io_min(limits, 0); | |
3018 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
3019 | } | |
0424caa1 MS |
3020 | |
3021 | /* | |
3022 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3023 | * They get transferred to the live pool in bind_control_target() | |
3024 | * called from pool_preresume(). | |
3025 | */ | |
b60ab990 MS |
3026 | if (!pt->adjusted_pf.discard_enabled) { |
3027 | /* | |
3028 | * Must explicitly disallow stacking discard limits otherwise the | |
3029 | * block layer will stack them if pool's data device has support. | |
3030 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3031 | * user to see that, so make sure to set all discard limits to 0. | |
3032 | */ | |
3033 | limits->discard_granularity = 0; | |
0424caa1 | 3034 | return; |
b60ab990 | 3035 | } |
0424caa1 MS |
3036 | |
3037 | disable_passdown_if_not_supported(pt); | |
3038 | ||
3039 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3040 | } |
3041 | ||
3042 | static struct target_type pool_target = { | |
3043 | .name = "thin-pool", | |
3044 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3045 | DM_TARGET_IMMUTABLE, | |
67324ea1 | 3046 | .version = {1, 12, 0}, |
991d9fa0 JT |
3047 | .module = THIS_MODULE, |
3048 | .ctr = pool_ctr, | |
3049 | .dtr = pool_dtr, | |
3050 | .map = pool_map, | |
3051 | .postsuspend = pool_postsuspend, | |
3052 | .preresume = pool_preresume, | |
3053 | .resume = pool_resume, | |
3054 | .message = pool_message, | |
3055 | .status = pool_status, | |
3056 | .merge = pool_merge, | |
3057 | .iterate_devices = pool_iterate_devices, | |
3058 | .io_hints = pool_io_hints, | |
3059 | }; | |
3060 | ||
3061 | /*---------------------------------------------------------------- | |
3062 | * Thin target methods | |
3063 | *--------------------------------------------------------------*/ | |
3064 | static void thin_dtr(struct dm_target *ti) | |
3065 | { | |
3066 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3067 | unsigned long flags; |
3068 | ||
3069 | spin_lock_irqsave(&tc->pool->lock, flags); | |
3070 | list_del_rcu(&tc->list); | |
3071 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3072 | synchronize_rcu(); | |
991d9fa0 JT |
3073 | |
3074 | mutex_lock(&dm_thin_pool_table.mutex); | |
3075 | ||
3076 | __pool_dec(tc->pool); | |
3077 | dm_pool_close_thin_device(tc->td); | |
3078 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3079 | if (tc->origin_dev) |
3080 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3081 | kfree(tc); |
3082 | ||
3083 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3084 | } | |
3085 | ||
3086 | /* | |
3087 | * Thin target parameters: | |
3088 | * | |
2dd9c257 | 3089 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3090 | * |
3091 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3092 | * dev_id: the internal device identifier | |
2dd9c257 | 3093 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3094 | * |
3095 | * If the pool device has discards disabled, they get disabled for the thin | |
3096 | * device as well. | |
991d9fa0 JT |
3097 | */ |
3098 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3099 | { | |
3100 | int r; | |
3101 | struct thin_c *tc; | |
2dd9c257 | 3102 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3103 | struct mapped_device *pool_md; |
5e3283e2 | 3104 | unsigned long flags; |
991d9fa0 JT |
3105 | |
3106 | mutex_lock(&dm_thin_pool_table.mutex); | |
3107 | ||
2dd9c257 | 3108 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3109 | ti->error = "Invalid argument count"; |
3110 | r = -EINVAL; | |
3111 | goto out_unlock; | |
3112 | } | |
3113 | ||
3114 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3115 | if (!tc) { | |
3116 | ti->error = "Out of memory"; | |
3117 | r = -ENOMEM; | |
3118 | goto out_unlock; | |
3119 | } | |
c140e1c4 MS |
3120 | spin_lock_init(&tc->lock); |
3121 | bio_list_init(&tc->deferred_bio_list); | |
3122 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3123 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3124 | |
2dd9c257 JT |
3125 | if (argc == 3) { |
3126 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3127 | if (r) { | |
3128 | ti->error = "Error opening origin device"; | |
3129 | goto bad_origin_dev; | |
3130 | } | |
3131 | tc->origin_dev = origin_dev; | |
3132 | } | |
3133 | ||
991d9fa0 JT |
3134 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3135 | if (r) { | |
3136 | ti->error = "Error opening pool device"; | |
3137 | goto bad_pool_dev; | |
3138 | } | |
3139 | tc->pool_dev = pool_dev; | |
3140 | ||
3141 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3142 | ti->error = "Invalid device id"; | |
3143 | r = -EINVAL; | |
3144 | goto bad_common; | |
3145 | } | |
3146 | ||
3147 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3148 | if (!pool_md) { | |
3149 | ti->error = "Couldn't get pool mapped device"; | |
3150 | r = -EINVAL; | |
3151 | goto bad_common; | |
3152 | } | |
3153 | ||
3154 | tc->pool = __pool_table_lookup(pool_md); | |
3155 | if (!tc->pool) { | |
3156 | ti->error = "Couldn't find pool object"; | |
3157 | r = -EINVAL; | |
3158 | goto bad_pool_lookup; | |
3159 | } | |
3160 | __pool_inc(tc->pool); | |
3161 | ||
e49e5829 JT |
3162 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3163 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3164 | r = -EINVAL; |
e49e5829 JT |
3165 | goto bad_thin_open; |
3166 | } | |
3167 | ||
991d9fa0 JT |
3168 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3169 | if (r) { | |
3170 | ti->error = "Couldn't open thin internal device"; | |
3171 | goto bad_thin_open; | |
3172 | } | |
3173 | ||
542f9038 MS |
3174 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3175 | if (r) | |
1acacc07 | 3176 | goto bad_target_max_io_len; |
542f9038 | 3177 | |
55a62eef | 3178 | ti->num_flush_bios = 1; |
16ad3d10 | 3179 | ti->flush_supported = true; |
59c3d2c6 | 3180 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3181 | |
3182 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3183 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3184 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3185 | ti->discards_supported = true; |
55a62eef | 3186 | ti->num_discard_bios = 1; |
55a62eef AK |
3187 | /* Discard bios must be split on a block boundary */ |
3188 | ti->split_discard_bios = true; | |
67e2e2b2 | 3189 | } |
991d9fa0 JT |
3190 | |
3191 | dm_put(pool_md); | |
3192 | ||
3193 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3194 | ||
5e3283e2 | 3195 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3196 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3197 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3198 | /* |
3199 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3200 | * wake_worker() call finding no bios to process (because the newly | |
3201 | * added tc isn't yet visible). So this reduces latency since we | |
3202 | * aren't then dependent on the periodic commit to wake_worker(). | |
3203 | */ | |
3204 | synchronize_rcu(); | |
3205 | ||
991d9fa0 JT |
3206 | return 0; |
3207 | ||
1acacc07 MS |
3208 | bad_target_max_io_len: |
3209 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3210 | bad_thin_open: |
3211 | __pool_dec(tc->pool); | |
3212 | bad_pool_lookup: | |
3213 | dm_put(pool_md); | |
3214 | bad_common: | |
3215 | dm_put_device(ti, tc->pool_dev); | |
3216 | bad_pool_dev: | |
2dd9c257 JT |
3217 | if (tc->origin_dev) |
3218 | dm_put_device(ti, tc->origin_dev); | |
3219 | bad_origin_dev: | |
991d9fa0 JT |
3220 | kfree(tc); |
3221 | out_unlock: | |
3222 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3223 | ||
3224 | return r; | |
3225 | } | |
3226 | ||
7de3ee57 | 3227 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3228 | { |
4f024f37 | 3229 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3230 | |
7de3ee57 | 3231 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3232 | } |
3233 | ||
7de3ee57 | 3234 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3235 | { |
3236 | unsigned long flags; | |
59c3d2c6 | 3237 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3238 | struct list_head work; |
a24c2569 | 3239 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3240 | struct pool *pool = h->tc->pool; |
3241 | ||
3242 | if (h->shared_read_entry) { | |
3243 | INIT_LIST_HEAD(&work); | |
44feb387 | 3244 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3245 | |
3246 | spin_lock_irqsave(&pool->lock, flags); | |
3247 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3248 | list_del(&m->list); | |
7f214665 | 3249 | m->quiesced = true; |
eb2aa48d JT |
3250 | __maybe_add_mapping(m); |
3251 | } | |
3252 | spin_unlock_irqrestore(&pool->lock, flags); | |
3253 | } | |
3254 | ||
104655fd JT |
3255 | if (h->all_io_entry) { |
3256 | INIT_LIST_HEAD(&work); | |
44feb387 | 3257 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3258 | if (!list_empty(&work)) { |
3259 | spin_lock_irqsave(&pool->lock, flags); | |
3260 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3261 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3262 | spin_unlock_irqrestore(&pool->lock, flags); |
3263 | wake_worker(pool); | |
3264 | } | |
104655fd JT |
3265 | } |
3266 | ||
eb2aa48d JT |
3267 | return 0; |
3268 | } | |
3269 | ||
738211f7 | 3270 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3271 | { |
738211f7 JT |
3272 | struct thin_c *tc = ti->private; |
3273 | ||
991d9fa0 | 3274 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3275 | noflush_work(tc, do_noflush_start); |
3276 | } | |
3277 | ||
3278 | static void thin_postsuspend(struct dm_target *ti) | |
3279 | { | |
3280 | struct thin_c *tc = ti->private; | |
3281 | ||
3282 | /* | |
3283 | * The dm_noflush_suspending flag has been cleared by now, so | |
3284 | * unfortunately we must always run this. | |
3285 | */ | |
3286 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3287 | } |
3288 | ||
3289 | /* | |
3290 | * <nr mapped sectors> <highest mapped sector> | |
3291 | */ | |
fd7c092e MP |
3292 | static void thin_status(struct dm_target *ti, status_type_t type, |
3293 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3294 | { |
3295 | int r; | |
3296 | ssize_t sz = 0; | |
3297 | dm_block_t mapped, highest; | |
3298 | char buf[BDEVNAME_SIZE]; | |
3299 | struct thin_c *tc = ti->private; | |
3300 | ||
e49e5829 JT |
3301 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3302 | DMEMIT("Fail"); | |
fd7c092e | 3303 | return; |
e49e5829 JT |
3304 | } |
3305 | ||
991d9fa0 JT |
3306 | if (!tc->td) |
3307 | DMEMIT("-"); | |
3308 | else { | |
3309 | switch (type) { | |
3310 | case STATUSTYPE_INFO: | |
3311 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3312 | if (r) { |
3313 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3314 | goto err; | |
3315 | } | |
991d9fa0 JT |
3316 | |
3317 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3318 | if (r < 0) { |
3319 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3320 | goto err; | |
3321 | } | |
991d9fa0 JT |
3322 | |
3323 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3324 | if (r) | |
3325 | DMEMIT("%llu", ((highest + 1) * | |
3326 | tc->pool->sectors_per_block) - 1); | |
3327 | else | |
3328 | DMEMIT("-"); | |
3329 | break; | |
3330 | ||
3331 | case STATUSTYPE_TABLE: | |
3332 | DMEMIT("%s %lu", | |
3333 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3334 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3335 | if (tc->origin_dev) |
3336 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3337 | break; |
3338 | } | |
3339 | } | |
3340 | ||
fd7c092e MP |
3341 | return; |
3342 | ||
3343 | err: | |
3344 | DMEMIT("Error"); | |
991d9fa0 JT |
3345 | } |
3346 | ||
3347 | static int thin_iterate_devices(struct dm_target *ti, | |
3348 | iterate_devices_callout_fn fn, void *data) | |
3349 | { | |
55f2b8bd | 3350 | sector_t blocks; |
991d9fa0 | 3351 | struct thin_c *tc = ti->private; |
55f2b8bd | 3352 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3353 | |
3354 | /* | |
3355 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3356 | * we follow a more convoluted path through to the pool's target. | |
3357 | */ | |
55f2b8bd | 3358 | if (!pool->ti) |
991d9fa0 JT |
3359 | return 0; /* nothing is bound */ |
3360 | ||
55f2b8bd MS |
3361 | blocks = pool->ti->len; |
3362 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3363 | if (blocks) |
55f2b8bd | 3364 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3365 | |
3366 | return 0; | |
3367 | } | |
3368 | ||
991d9fa0 JT |
3369 | static struct target_type thin_target = { |
3370 | .name = "thin", | |
67324ea1 | 3371 | .version = {1, 12, 0}, |
991d9fa0 JT |
3372 | .module = THIS_MODULE, |
3373 | .ctr = thin_ctr, | |
3374 | .dtr = thin_dtr, | |
3375 | .map = thin_map, | |
eb2aa48d | 3376 | .end_io = thin_endio, |
738211f7 | 3377 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3378 | .postsuspend = thin_postsuspend, |
3379 | .status = thin_status, | |
3380 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3381 | }; |
3382 | ||
3383 | /*----------------------------------------------------------------*/ | |
3384 | ||
3385 | static int __init dm_thin_init(void) | |
3386 | { | |
3387 | int r; | |
3388 | ||
3389 | pool_table_init(); | |
3390 | ||
3391 | r = dm_register_target(&thin_target); | |
3392 | if (r) | |
3393 | return r; | |
3394 | ||
3395 | r = dm_register_target(&pool_target); | |
3396 | if (r) | |
a24c2569 MS |
3397 | goto bad_pool_target; |
3398 | ||
3399 | r = -ENOMEM; | |
3400 | ||
a24c2569 MS |
3401 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3402 | if (!_new_mapping_cache) | |
3403 | goto bad_new_mapping_cache; | |
3404 | ||
a24c2569 MS |
3405 | return 0; |
3406 | ||
a24c2569 | 3407 | bad_new_mapping_cache: |
a24c2569 MS |
3408 | dm_unregister_target(&pool_target); |
3409 | bad_pool_target: | |
3410 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3411 | |
3412 | return r; | |
3413 | } | |
3414 | ||
3415 | static void dm_thin_exit(void) | |
3416 | { | |
3417 | dm_unregister_target(&thin_target); | |
3418 | dm_unregister_target(&pool_target); | |
a24c2569 | 3419 | |
a24c2569 | 3420 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3421 | } |
3422 | ||
3423 | module_init(dm_thin_init); | |
3424 | module_exit(dm_thin_exit); | |
3425 | ||
7cab8bf1 | 3426 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3427 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3428 | MODULE_LICENSE("GPL"); |