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 |
80c57893 MS |
30 | #define NO_SPACE_TIMEOUT_SECS 60 |
31 | ||
32 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 33 | |
df5d2e90 MP |
34 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
35 | "A percentage of time allocated for copy on write"); | |
36 | ||
991d9fa0 JT |
37 | /* |
38 | * The block size of the device holding pool data must be | |
39 | * between 64KB and 1GB. | |
40 | */ | |
41 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
42 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
43 | ||
991d9fa0 JT |
44 | /* |
45 | * Device id is restricted to 24 bits. | |
46 | */ | |
47 | #define MAX_DEV_ID ((1 << 24) - 1) | |
48 | ||
49 | /* | |
50 | * How do we handle breaking sharing of data blocks? | |
51 | * ================================================= | |
52 | * | |
53 | * We use a standard copy-on-write btree to store the mappings for the | |
54 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
55 | * the data). When you take an internal snapshot you clone the root node | |
56 | * of the origin btree. After this there is no concept of an origin or a | |
57 | * snapshot. They are just two device trees that happen to point to the | |
58 | * same data blocks. | |
59 | * | |
60 | * When we get a write in we decide if it's to a shared data block using | |
61 | * some timestamp magic. If it is, we have to break sharing. | |
62 | * | |
63 | * Let's say we write to a shared block in what was the origin. The | |
64 | * steps are: | |
65 | * | |
66 | * i) plug io further to this physical block. (see bio_prison code). | |
67 | * | |
68 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 69 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
70 | * |
71 | * iii) copy the data block to a newly allocate block. This step can be | |
72 | * missed out if the io covers the block. (schedule_copy). | |
73 | * | |
74 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 75 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
76 | * sharing of btree nodes between the two devices. Breaking sharing only |
77 | * effects the btree of that specific device. Btrees for the other | |
78 | * devices that share the block never change. The btree for the origin | |
79 | * device as it was after the last commit is untouched, ie. we're using | |
80 | * persistent data structures in the functional programming sense. | |
81 | * | |
82 | * v) unplug io to this physical block, including the io that triggered | |
83 | * the breaking of sharing. | |
84 | * | |
85 | * Steps (ii) and (iii) occur in parallel. | |
86 | * | |
87 | * The metadata _doesn't_ need to be committed before the io continues. We | |
88 | * get away with this because the io is always written to a _new_ block. | |
89 | * If there's a crash, then: | |
90 | * | |
91 | * - The origin mapping will point to the old origin block (the shared | |
92 | * one). This will contain the data as it was before the io that triggered | |
93 | * the breaking of sharing came in. | |
94 | * | |
95 | * - The snap mapping still points to the old block. As it would after | |
96 | * the commit. | |
97 | * | |
98 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
99 | * will continue to think that data block in the snapshot device is shared | |
100 | * even after the write to the origin has broken sharing. I suspect data | |
101 | * blocks will typically be shared by many different devices, so we're | |
102 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
103 | * devices that reference this data block. At the moment I think the | |
104 | * benefits far, far outweigh the disadvantages. | |
105 | */ | |
106 | ||
107 | /*----------------------------------------------------------------*/ | |
108 | ||
991d9fa0 JT |
109 | /* |
110 | * Key building. | |
111 | */ | |
112 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 113 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
114 | { |
115 | key->virtual = 0; | |
116 | key->dev = dm_thin_dev_id(td); | |
117 | key->block = b; | |
118 | } | |
119 | ||
120 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 121 | struct dm_cell_key *key) |
991d9fa0 JT |
122 | { |
123 | key->virtual = 1; | |
124 | key->dev = dm_thin_dev_id(td); | |
125 | key->block = b; | |
126 | } | |
127 | ||
128 | /*----------------------------------------------------------------*/ | |
129 | ||
130 | /* | |
131 | * A pool device ties together a metadata device and a data device. It | |
132 | * also provides the interface for creating and destroying internal | |
133 | * devices. | |
134 | */ | |
a24c2569 | 135 | struct dm_thin_new_mapping; |
67e2e2b2 | 136 | |
e49e5829 | 137 | /* |
3e1a0699 | 138 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
139 | */ |
140 | enum pool_mode { | |
141 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 142 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
143 | PM_READ_ONLY, /* metadata may not be changed */ |
144 | PM_FAIL, /* all I/O fails */ | |
145 | }; | |
146 | ||
67e2e2b2 | 147 | struct pool_features { |
e49e5829 JT |
148 | enum pool_mode mode; |
149 | ||
9bc142dd MS |
150 | bool zero_new_blocks:1; |
151 | bool discard_enabled:1; | |
152 | bool discard_passdown:1; | |
787a996c | 153 | bool error_if_no_space:1; |
67e2e2b2 JT |
154 | }; |
155 | ||
e49e5829 JT |
156 | struct thin_c; |
157 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
158 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
159 | ||
991d9fa0 JT |
160 | struct pool { |
161 | struct list_head list; | |
162 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
163 | ||
164 | struct mapped_device *pool_md; | |
165 | struct block_device *md_dev; | |
166 | struct dm_pool_metadata *pmd; | |
167 | ||
991d9fa0 | 168 | dm_block_t low_water_blocks; |
55f2b8bd | 169 | uint32_t sectors_per_block; |
f9a8e0cd | 170 | int sectors_per_block_shift; |
991d9fa0 | 171 | |
67e2e2b2 | 172 | struct pool_features pf; |
88a6621b | 173 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 174 | |
44feb387 | 175 | struct dm_bio_prison *prison; |
991d9fa0 JT |
176 | struct dm_kcopyd_client *copier; |
177 | ||
178 | struct workqueue_struct *wq; | |
179 | struct work_struct worker; | |
905e51b3 | 180 | struct delayed_work waker; |
85ad643b | 181 | struct delayed_work no_space_timeout; |
991d9fa0 | 182 | |
905e51b3 | 183 | unsigned long last_commit_jiffies; |
55f2b8bd | 184 | unsigned ref_count; |
991d9fa0 JT |
185 | |
186 | spinlock_t lock; | |
991d9fa0 JT |
187 | struct bio_list deferred_flush_bios; |
188 | struct list_head prepared_mappings; | |
104655fd | 189 | struct list_head prepared_discards; |
c140e1c4 | 190 | struct list_head active_thins; |
991d9fa0 | 191 | |
44feb387 MS |
192 | struct dm_deferred_set *shared_read_ds; |
193 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 194 | |
a24c2569 | 195 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 196 | mempool_t *mapping_pool; |
e49e5829 JT |
197 | |
198 | process_bio_fn process_bio; | |
199 | process_bio_fn process_discard; | |
200 | ||
201 | process_mapping_fn process_prepared_mapping; | |
202 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
203 | }; |
204 | ||
e49e5829 | 205 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 206 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 207 | |
991d9fa0 JT |
208 | /* |
209 | * Target context for a pool. | |
210 | */ | |
211 | struct pool_c { | |
212 | struct dm_target *ti; | |
213 | struct pool *pool; | |
214 | struct dm_dev *data_dev; | |
215 | struct dm_dev *metadata_dev; | |
216 | struct dm_target_callbacks callbacks; | |
217 | ||
218 | dm_block_t low_water_blocks; | |
0424caa1 MS |
219 | struct pool_features requested_pf; /* Features requested during table load */ |
220 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
221 | }; |
222 | ||
223 | /* | |
224 | * Target context for a thin. | |
225 | */ | |
226 | struct thin_c { | |
c140e1c4 | 227 | struct list_head list; |
991d9fa0 | 228 | struct dm_dev *pool_dev; |
2dd9c257 | 229 | struct dm_dev *origin_dev; |
991d9fa0 JT |
230 | dm_thin_id dev_id; |
231 | ||
232 | struct pool *pool; | |
233 | struct dm_thin_device *td; | |
738211f7 | 234 | bool requeue_mode:1; |
c140e1c4 MS |
235 | spinlock_t lock; |
236 | struct bio_list deferred_bio_list; | |
237 | struct bio_list retry_on_resume_list; | |
67324ea1 | 238 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
239 | |
240 | /* | |
241 | * Ensures the thin is not destroyed until the worker has finished | |
242 | * iterating the active_thins list. | |
243 | */ | |
244 | atomic_t refcount; | |
245 | struct completion can_destroy; | |
991d9fa0 JT |
246 | }; |
247 | ||
248 | /*----------------------------------------------------------------*/ | |
249 | ||
025b9685 JT |
250 | /* |
251 | * wake_worker() is used when new work is queued and when pool_resume is | |
252 | * ready to continue deferred IO processing. | |
253 | */ | |
254 | static void wake_worker(struct pool *pool) | |
255 | { | |
256 | queue_work(pool->wq, &pool->worker); | |
257 | } | |
258 | ||
259 | /*----------------------------------------------------------------*/ | |
260 | ||
6beca5eb JT |
261 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
262 | struct dm_bio_prison_cell **cell_result) | |
263 | { | |
264 | int r; | |
265 | struct dm_bio_prison_cell *cell_prealloc; | |
266 | ||
267 | /* | |
268 | * Allocate a cell from the prison's mempool. | |
269 | * This might block but it can't fail. | |
270 | */ | |
271 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
272 | ||
273 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
274 | if (r) | |
275 | /* | |
276 | * We reused an old cell; we can get rid of | |
277 | * the new one. | |
278 | */ | |
279 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
280 | ||
281 | return r; | |
282 | } | |
283 | ||
284 | static void cell_release(struct pool *pool, | |
285 | struct dm_bio_prison_cell *cell, | |
286 | struct bio_list *bios) | |
287 | { | |
288 | dm_cell_release(pool->prison, cell, bios); | |
289 | dm_bio_prison_free_cell(pool->prison, cell); | |
290 | } | |
291 | ||
292 | static void cell_release_no_holder(struct pool *pool, | |
293 | struct dm_bio_prison_cell *cell, | |
294 | struct bio_list *bios) | |
295 | { | |
296 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
297 | dm_bio_prison_free_cell(pool->prison, cell); | |
298 | } | |
299 | ||
025b9685 JT |
300 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
301 | struct dm_bio_prison_cell *cell) | |
302 | { | |
303 | struct pool *pool = tc->pool; | |
304 | unsigned long flags; | |
305 | ||
c140e1c4 MS |
306 | spin_lock_irqsave(&tc->lock, flags); |
307 | dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list); | |
308 | spin_unlock_irqrestore(&tc->lock, flags); | |
025b9685 JT |
309 | |
310 | wake_worker(pool); | |
311 | } | |
312 | ||
6beca5eb JT |
313 | static void cell_error(struct pool *pool, |
314 | struct dm_bio_prison_cell *cell) | |
315 | { | |
316 | dm_cell_error(pool->prison, cell); | |
317 | dm_bio_prison_free_cell(pool->prison, cell); | |
318 | } | |
319 | ||
320 | /*----------------------------------------------------------------*/ | |
321 | ||
991d9fa0 JT |
322 | /* |
323 | * A global list of pools that uses a struct mapped_device as a key. | |
324 | */ | |
325 | static struct dm_thin_pool_table { | |
326 | struct mutex mutex; | |
327 | struct list_head pools; | |
328 | } dm_thin_pool_table; | |
329 | ||
330 | static void pool_table_init(void) | |
331 | { | |
332 | mutex_init(&dm_thin_pool_table.mutex); | |
333 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
334 | } | |
335 | ||
336 | static void __pool_table_insert(struct pool *pool) | |
337 | { | |
338 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
339 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
340 | } | |
341 | ||
342 | static void __pool_table_remove(struct pool *pool) | |
343 | { | |
344 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
345 | list_del(&pool->list); | |
346 | } | |
347 | ||
348 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
349 | { | |
350 | struct pool *pool = NULL, *tmp; | |
351 | ||
352 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
353 | ||
354 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
355 | if (tmp->pool_md == md) { | |
356 | pool = tmp; | |
357 | break; | |
358 | } | |
359 | } | |
360 | ||
361 | return pool; | |
362 | } | |
363 | ||
364 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
365 | { | |
366 | struct pool *pool = NULL, *tmp; | |
367 | ||
368 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
369 | ||
370 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
371 | if (tmp->md_dev == md_dev) { | |
372 | pool = tmp; | |
373 | break; | |
374 | } | |
375 | } | |
376 | ||
377 | return pool; | |
378 | } | |
379 | ||
380 | /*----------------------------------------------------------------*/ | |
381 | ||
a24c2569 | 382 | struct dm_thin_endio_hook { |
eb2aa48d | 383 | struct thin_c *tc; |
44feb387 MS |
384 | struct dm_deferred_entry *shared_read_entry; |
385 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 386 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 387 | struct rb_node rb_node; |
eb2aa48d JT |
388 | }; |
389 | ||
18adc577 | 390 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
391 | { |
392 | struct bio *bio; | |
393 | struct bio_list bios; | |
18adc577 | 394 | unsigned long flags; |
991d9fa0 JT |
395 | |
396 | bio_list_init(&bios); | |
18adc577 | 397 | |
c140e1c4 | 398 | spin_lock_irqsave(&tc->lock, flags); |
991d9fa0 JT |
399 | bio_list_merge(&bios, master); |
400 | bio_list_init(master); | |
c140e1c4 | 401 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 402 | |
c140e1c4 MS |
403 | while ((bio = bio_list_pop(&bios))) |
404 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
991d9fa0 JT |
405 | } |
406 | ||
407 | static void requeue_io(struct thin_c *tc) | |
408 | { | |
c140e1c4 MS |
409 | requeue_bio_list(tc, &tc->deferred_bio_list); |
410 | requeue_bio_list(tc, &tc->retry_on_resume_list); | |
991d9fa0 JT |
411 | } |
412 | ||
c140e1c4 | 413 | static void error_thin_retry_list(struct thin_c *tc) |
3e1a0699 JT |
414 | { |
415 | struct bio *bio; | |
416 | unsigned long flags; | |
417 | struct bio_list bios; | |
418 | ||
419 | bio_list_init(&bios); | |
420 | ||
c140e1c4 MS |
421 | spin_lock_irqsave(&tc->lock, flags); |
422 | bio_list_merge(&bios, &tc->retry_on_resume_list); | |
423 | bio_list_init(&tc->retry_on_resume_list); | |
424 | spin_unlock_irqrestore(&tc->lock, flags); | |
3e1a0699 JT |
425 | |
426 | while ((bio = bio_list_pop(&bios))) | |
427 | bio_io_error(bio); | |
428 | } | |
429 | ||
c140e1c4 MS |
430 | static void error_retry_list(struct pool *pool) |
431 | { | |
432 | struct thin_c *tc; | |
433 | ||
434 | rcu_read_lock(); | |
435 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
436 | error_thin_retry_list(tc); | |
437 | rcu_read_unlock(); | |
438 | } | |
439 | ||
991d9fa0 JT |
440 | /* |
441 | * This section of code contains the logic for processing a thin device's IO. | |
442 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
443 | * but most is exclusively called from the thin target rather than the thin-pool | |
444 | * target. | |
445 | */ | |
446 | ||
58f77a21 MS |
447 | static bool block_size_is_power_of_two(struct pool *pool) |
448 | { | |
449 | return pool->sectors_per_block_shift >= 0; | |
450 | } | |
451 | ||
991d9fa0 JT |
452 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
453 | { | |
58f77a21 | 454 | struct pool *pool = tc->pool; |
4f024f37 | 455 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 456 | |
58f77a21 MS |
457 | if (block_size_is_power_of_two(pool)) |
458 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 459 | else |
58f77a21 | 460 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
461 | |
462 | return block_nr; | |
991d9fa0 JT |
463 | } |
464 | ||
465 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
466 | { | |
467 | struct pool *pool = tc->pool; | |
4f024f37 | 468 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
469 | |
470 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 471 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
472 | bio->bi_iter.bi_sector = |
473 | (block << pool->sectors_per_block_shift) | | |
474 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 475 | else |
4f024f37 | 476 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 477 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
478 | } |
479 | ||
2dd9c257 JT |
480 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
481 | { | |
482 | bio->bi_bdev = tc->origin_dev->bdev; | |
483 | } | |
484 | ||
4afdd680 JT |
485 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
486 | { | |
487 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
488 | dm_thin_changed_this_transaction(tc->td); | |
489 | } | |
490 | ||
e8088073 JT |
491 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
492 | { | |
493 | struct dm_thin_endio_hook *h; | |
494 | ||
495 | if (bio->bi_rw & REQ_DISCARD) | |
496 | return; | |
497 | ||
59c3d2c6 | 498 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
499 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
500 | } | |
501 | ||
2dd9c257 | 502 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
503 | { |
504 | struct pool *pool = tc->pool; | |
505 | unsigned long flags; | |
506 | ||
e49e5829 JT |
507 | if (!bio_triggers_commit(tc, bio)) { |
508 | generic_make_request(bio); | |
509 | return; | |
510 | } | |
511 | ||
991d9fa0 | 512 | /* |
e49e5829 JT |
513 | * Complete bio with an error if earlier I/O caused changes to |
514 | * the metadata that can't be committed e.g, due to I/O errors | |
515 | * on the metadata device. | |
991d9fa0 | 516 | */ |
e49e5829 JT |
517 | if (dm_thin_aborted_changes(tc->td)) { |
518 | bio_io_error(bio); | |
519 | return; | |
520 | } | |
521 | ||
522 | /* | |
523 | * Batch together any bios that trigger commits and then issue a | |
524 | * single commit for them in process_deferred_bios(). | |
525 | */ | |
526 | spin_lock_irqsave(&pool->lock, flags); | |
527 | bio_list_add(&pool->deferred_flush_bios, bio); | |
528 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
529 | } |
530 | ||
2dd9c257 JT |
531 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
532 | { | |
533 | remap_to_origin(tc, bio); | |
534 | issue(tc, bio); | |
535 | } | |
536 | ||
537 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
538 | dm_block_t block) | |
539 | { | |
540 | remap(tc, bio, block); | |
541 | issue(tc, bio); | |
542 | } | |
543 | ||
991d9fa0 JT |
544 | /*----------------------------------------------------------------*/ |
545 | ||
546 | /* | |
547 | * Bio endio functions. | |
548 | */ | |
a24c2569 | 549 | struct dm_thin_new_mapping { |
991d9fa0 JT |
550 | struct list_head list; |
551 | ||
7f214665 MS |
552 | bool quiesced:1; |
553 | bool prepared:1; | |
554 | bool pass_discard:1; | |
555 | bool definitely_not_shared:1; | |
991d9fa0 | 556 | |
7f214665 | 557 | int err; |
991d9fa0 JT |
558 | struct thin_c *tc; |
559 | dm_block_t virt_block; | |
560 | dm_block_t data_block; | |
a24c2569 | 561 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
562 | |
563 | /* | |
564 | * If the bio covers the whole area of a block then we can avoid | |
565 | * zeroing or copying. Instead this bio is hooked. The bio will | |
566 | * still be in the cell, so care has to be taken to avoid issuing | |
567 | * the bio twice. | |
568 | */ | |
569 | struct bio *bio; | |
570 | bio_end_io_t *saved_bi_end_io; | |
571 | }; | |
572 | ||
a24c2569 | 573 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
574 | { |
575 | struct pool *pool = m->tc->pool; | |
576 | ||
eb2aa48d | 577 | if (m->quiesced && m->prepared) { |
daec338b | 578 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
579 | wake_worker(pool); |
580 | } | |
581 | } | |
582 | ||
583 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
584 | { | |
585 | unsigned long flags; | |
a24c2569 | 586 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
587 | struct pool *pool = m->tc->pool; |
588 | ||
589 | m->err = read_err || write_err ? -EIO : 0; | |
590 | ||
591 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 592 | m->prepared = true; |
991d9fa0 JT |
593 | __maybe_add_mapping(m); |
594 | spin_unlock_irqrestore(&pool->lock, flags); | |
595 | } | |
596 | ||
597 | static void overwrite_endio(struct bio *bio, int err) | |
598 | { | |
599 | unsigned long flags; | |
59c3d2c6 | 600 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 601 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
602 | struct pool *pool = m->tc->pool; |
603 | ||
604 | m->err = err; | |
605 | ||
606 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 607 | m->prepared = true; |
991d9fa0 JT |
608 | __maybe_add_mapping(m); |
609 | spin_unlock_irqrestore(&pool->lock, flags); | |
610 | } | |
611 | ||
991d9fa0 JT |
612 | /*----------------------------------------------------------------*/ |
613 | ||
614 | /* | |
615 | * Workqueue. | |
616 | */ | |
617 | ||
618 | /* | |
619 | * Prepared mapping jobs. | |
620 | */ | |
621 | ||
622 | /* | |
623 | * This sends the bios in the cell back to the deferred_bios list. | |
624 | */ | |
2aab3850 | 625 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
626 | { |
627 | struct pool *pool = tc->pool; | |
628 | unsigned long flags; | |
629 | ||
c140e1c4 MS |
630 | spin_lock_irqsave(&tc->lock, flags); |
631 | cell_release(pool, cell, &tc->deferred_bio_list); | |
632 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
633 | |
634 | wake_worker(pool); | |
635 | } | |
636 | ||
637 | /* | |
6beca5eb | 638 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 639 | */ |
f286ba0e | 640 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 641 | { |
991d9fa0 JT |
642 | struct pool *pool = tc->pool; |
643 | unsigned long flags; | |
644 | ||
c140e1c4 MS |
645 | spin_lock_irqsave(&tc->lock, flags); |
646 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
647 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
648 | |
649 | wake_worker(pool); | |
650 | } | |
651 | ||
e49e5829 JT |
652 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
653 | { | |
196d38bc | 654 | if (m->bio) { |
e49e5829 | 655 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
656 | atomic_inc(&m->bio->bi_remaining); |
657 | } | |
6beca5eb | 658 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
659 | list_del(&m->list); |
660 | mempool_free(m, m->tc->pool->mapping_pool); | |
661 | } | |
025b9685 | 662 | |
a24c2569 | 663 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
664 | { |
665 | struct thin_c *tc = m->tc; | |
6beca5eb | 666 | struct pool *pool = tc->pool; |
991d9fa0 JT |
667 | struct bio *bio; |
668 | int r; | |
669 | ||
670 | bio = m->bio; | |
196d38bc | 671 | if (bio) { |
991d9fa0 | 672 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
673 | atomic_inc(&bio->bi_remaining); |
674 | } | |
991d9fa0 JT |
675 | |
676 | if (m->err) { | |
6beca5eb | 677 | cell_error(pool, m->cell); |
905386f8 | 678 | goto out; |
991d9fa0 JT |
679 | } |
680 | ||
681 | /* | |
682 | * Commit the prepared block into the mapping btree. | |
683 | * Any I/O for this block arriving after this point will get | |
684 | * remapped to it directly. | |
685 | */ | |
686 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
687 | if (r) { | |
b5330655 | 688 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 689 | cell_error(pool, m->cell); |
905386f8 | 690 | goto out; |
991d9fa0 JT |
691 | } |
692 | ||
693 | /* | |
694 | * Release any bios held while the block was being provisioned. | |
695 | * If we are processing a write bio that completely covers the block, | |
696 | * we already processed it so can ignore it now when processing | |
697 | * the bios in the cell. | |
698 | */ | |
699 | if (bio) { | |
f286ba0e | 700 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
701 | bio_endio(bio, 0); |
702 | } else | |
2aab3850 | 703 | cell_defer(tc, m->cell); |
991d9fa0 | 704 | |
905386f8 | 705 | out: |
991d9fa0 | 706 | list_del(&m->list); |
6beca5eb | 707 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
708 | } |
709 | ||
e49e5829 | 710 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 711 | { |
104655fd JT |
712 | struct thin_c *tc = m->tc; |
713 | ||
e49e5829 | 714 | bio_io_error(m->bio); |
f286ba0e JT |
715 | cell_defer_no_holder(tc, m->cell); |
716 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
717 | mempool_free(m, tc->pool->mapping_pool); |
718 | } | |
719 | ||
720 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
721 | { | |
722 | struct thin_c *tc = m->tc; | |
104655fd | 723 | |
e8088073 | 724 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
725 | cell_defer_no_holder(tc, m->cell); |
726 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 727 | |
104655fd | 728 | if (m->pass_discard) |
19fa1a67 JT |
729 | if (m->definitely_not_shared) |
730 | remap_and_issue(tc, m->bio, m->data_block); | |
731 | else { | |
732 | bool used = false; | |
733 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
734 | bio_endio(m->bio, 0); | |
735 | else | |
736 | remap_and_issue(tc, m->bio, m->data_block); | |
737 | } | |
104655fd JT |
738 | else |
739 | bio_endio(m->bio, 0); | |
740 | ||
104655fd JT |
741 | mempool_free(m, tc->pool->mapping_pool); |
742 | } | |
743 | ||
e49e5829 JT |
744 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
745 | { | |
746 | int r; | |
747 | struct thin_c *tc = m->tc; | |
748 | ||
749 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
750 | if (r) | |
c397741c | 751 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
752 | |
753 | process_prepared_discard_passdown(m); | |
754 | } | |
755 | ||
104655fd | 756 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 757 | process_mapping_fn *fn) |
991d9fa0 JT |
758 | { |
759 | unsigned long flags; | |
760 | struct list_head maps; | |
a24c2569 | 761 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
762 | |
763 | INIT_LIST_HEAD(&maps); | |
764 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 765 | list_splice_init(head, &maps); |
991d9fa0 JT |
766 | spin_unlock_irqrestore(&pool->lock, flags); |
767 | ||
768 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 769 | (*fn)(m); |
991d9fa0 JT |
770 | } |
771 | ||
772 | /* | |
773 | * Deferred bio jobs. | |
774 | */ | |
104655fd | 775 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 776 | { |
4f024f37 KO |
777 | return bio->bi_iter.bi_size == |
778 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
779 | } |
780 | ||
781 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
782 | { | |
783 | return (bio_data_dir(bio) == WRITE) && | |
784 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
785 | } |
786 | ||
787 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
788 | bio_end_io_t *fn) | |
789 | { | |
790 | *save = bio->bi_end_io; | |
791 | bio->bi_end_io = fn; | |
792 | } | |
793 | ||
794 | static int ensure_next_mapping(struct pool *pool) | |
795 | { | |
796 | if (pool->next_mapping) | |
797 | return 0; | |
798 | ||
799 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
800 | ||
801 | return pool->next_mapping ? 0 : -ENOMEM; | |
802 | } | |
803 | ||
a24c2569 | 804 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 805 | { |
16961b04 | 806 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
807 | |
808 | BUG_ON(!pool->next_mapping); | |
809 | ||
16961b04 MS |
810 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
811 | INIT_LIST_HEAD(&m->list); | |
812 | m->bio = NULL; | |
813 | ||
991d9fa0 JT |
814 | pool->next_mapping = NULL; |
815 | ||
16961b04 | 816 | return m; |
991d9fa0 JT |
817 | } |
818 | ||
819 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
820 | struct dm_dev *origin, dm_block_t data_origin, |
821 | dm_block_t data_dest, | |
a24c2569 | 822 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
823 | { |
824 | int r; | |
825 | struct pool *pool = tc->pool; | |
a24c2569 | 826 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 827 | |
991d9fa0 JT |
828 | m->tc = tc; |
829 | m->virt_block = virt_block; | |
830 | m->data_block = data_dest; | |
831 | m->cell = cell; | |
991d9fa0 | 832 | |
44feb387 | 833 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 834 | m->quiesced = true; |
991d9fa0 JT |
835 | |
836 | /* | |
837 | * IO to pool_dev remaps to the pool target's data_dev. | |
838 | * | |
839 | * If the whole block of data is being overwritten, we can issue the | |
840 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
841 | */ | |
842 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 843 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 844 | |
eb2aa48d | 845 | h->overwrite_mapping = m; |
991d9fa0 JT |
846 | m->bio = bio; |
847 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 848 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
849 | remap_and_issue(tc, bio, data_dest); |
850 | } else { | |
851 | struct dm_io_region from, to; | |
852 | ||
2dd9c257 | 853 | from.bdev = origin->bdev; |
991d9fa0 JT |
854 | from.sector = data_origin * pool->sectors_per_block; |
855 | from.count = pool->sectors_per_block; | |
856 | ||
857 | to.bdev = tc->pool_dev->bdev; | |
858 | to.sector = data_dest * pool->sectors_per_block; | |
859 | to.count = pool->sectors_per_block; | |
860 | ||
861 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
862 | 0, copy_complete, m); | |
863 | if (r < 0) { | |
864 | mempool_free(m, pool->mapping_pool); | |
c397741c | 865 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 866 | cell_error(pool, cell); |
991d9fa0 JT |
867 | } |
868 | } | |
869 | } | |
870 | ||
2dd9c257 JT |
871 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
872 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 873 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
874 | { |
875 | schedule_copy(tc, virt_block, tc->pool_dev, | |
876 | data_origin, data_dest, cell, bio); | |
877 | } | |
878 | ||
879 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
880 | dm_block_t data_dest, | |
a24c2569 | 881 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
882 | { |
883 | schedule_copy(tc, virt_block, tc->origin_dev, | |
884 | virt_block, data_dest, cell, bio); | |
885 | } | |
886 | ||
991d9fa0 | 887 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 888 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
889 | struct bio *bio) |
890 | { | |
891 | struct pool *pool = tc->pool; | |
a24c2569 | 892 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 893 | |
7f214665 MS |
894 | m->quiesced = true; |
895 | m->prepared = false; | |
991d9fa0 JT |
896 | m->tc = tc; |
897 | m->virt_block = virt_block; | |
898 | m->data_block = data_block; | |
899 | m->cell = cell; | |
991d9fa0 JT |
900 | |
901 | /* | |
902 | * If the whole block of data is being overwritten or we are not | |
903 | * zeroing pre-existing data, we can issue the bio immediately. | |
904 | * Otherwise we use kcopyd to zero the data first. | |
905 | */ | |
67e2e2b2 | 906 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
907 | process_prepared_mapping(m); |
908 | ||
909 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 910 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 911 | |
eb2aa48d | 912 | h->overwrite_mapping = m; |
991d9fa0 JT |
913 | m->bio = bio; |
914 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 915 | inc_all_io_entry(pool, bio); |
991d9fa0 | 916 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
917 | } else { |
918 | int r; | |
919 | struct dm_io_region to; | |
920 | ||
921 | to.bdev = tc->pool_dev->bdev; | |
922 | to.sector = data_block * pool->sectors_per_block; | |
923 | to.count = pool->sectors_per_block; | |
924 | ||
925 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
926 | if (r < 0) { | |
927 | mempool_free(m, pool->mapping_pool); | |
c397741c | 928 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 929 | cell_error(pool, cell); |
991d9fa0 JT |
930 | } |
931 | } | |
932 | } | |
933 | ||
e49e5829 JT |
934 | /* |
935 | * A non-zero return indicates read_only or fail_io mode. | |
936 | * Many callers don't care about the return value. | |
937 | */ | |
020cc3b5 | 938 | static int commit(struct pool *pool) |
e49e5829 JT |
939 | { |
940 | int r; | |
941 | ||
8d07e8a5 | 942 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
943 | return -EINVAL; |
944 | ||
020cc3b5 | 945 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
946 | if (r) |
947 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
948 | |
949 | return r; | |
950 | } | |
951 | ||
88a6621b JT |
952 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
953 | { | |
954 | unsigned long flags; | |
955 | ||
956 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
957 | DMWARN("%s: reached low water mark for data device: sending event.", | |
958 | dm_device_name(pool->pool_md)); | |
959 | spin_lock_irqsave(&pool->lock, flags); | |
960 | pool->low_water_triggered = true; | |
961 | spin_unlock_irqrestore(&pool->lock, flags); | |
962 | dm_table_event(pool->ti->table); | |
963 | } | |
964 | } | |
965 | ||
3e1a0699 JT |
966 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
967 | ||
991d9fa0 JT |
968 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
969 | { | |
970 | int r; | |
971 | dm_block_t free_blocks; | |
991d9fa0 JT |
972 | struct pool *pool = tc->pool; |
973 | ||
3e1a0699 | 974 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
975 | return -EINVAL; |
976 | ||
991d9fa0 | 977 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
978 | if (r) { |
979 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 980 | return r; |
b5330655 | 981 | } |
991d9fa0 | 982 | |
88a6621b | 983 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
984 | |
985 | if (!free_blocks) { | |
94563bad MS |
986 | /* |
987 | * Try to commit to see if that will free up some | |
988 | * more space. | |
989 | */ | |
020cc3b5 JT |
990 | r = commit(pool); |
991 | if (r) | |
992 | return r; | |
991d9fa0 | 993 | |
94563bad | 994 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
995 | if (r) { |
996 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 997 | return r; |
b5330655 | 998 | } |
991d9fa0 | 999 | |
94563bad | 1000 | if (!free_blocks) { |
3e1a0699 | 1001 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1002 | return -ENOSPC; |
991d9fa0 JT |
1003 | } |
1004 | } | |
1005 | ||
1006 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1007 | if (r) { |
b5330655 | 1008 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1009 | return r; |
4a02b34e | 1010 | } |
991d9fa0 JT |
1011 | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * If we have run out of space, queue bios until the device is | |
1017 | * resumed, presumably after having been reloaded with more space. | |
1018 | */ | |
1019 | static void retry_on_resume(struct bio *bio) | |
1020 | { | |
59c3d2c6 | 1021 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1022 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1023 | unsigned long flags; |
1024 | ||
c140e1c4 MS |
1025 | spin_lock_irqsave(&tc->lock, flags); |
1026 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1027 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1028 | } |
1029 | ||
3e1a0699 | 1030 | static bool should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1031 | { |
3e1a0699 JT |
1032 | enum pool_mode m = get_pool_mode(pool); |
1033 | ||
1034 | switch (m) { | |
1035 | case PM_WRITE: | |
1036 | /* Shouldn't get here */ | |
1037 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
1038 | return true; | |
1039 | ||
1040 | case PM_OUT_OF_DATA_SPACE: | |
1041 | return pool->pf.error_if_no_space; | |
1042 | ||
1043 | case PM_READ_ONLY: | |
1044 | case PM_FAIL: | |
1045 | return true; | |
1046 | default: | |
1047 | /* Shouldn't get here */ | |
1048 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
1049 | return true; | |
1050 | } | |
1051 | } | |
8c0f0e8c | 1052 | |
3e1a0699 JT |
1053 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1054 | { | |
1055 | if (should_error_unserviceable_bio(pool)) | |
8c0f0e8c | 1056 | bio_io_error(bio); |
6d16202b MS |
1057 | else |
1058 | retry_on_resume(bio); | |
8c0f0e8c MS |
1059 | } |
1060 | ||
399caddf | 1061 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1062 | { |
1063 | struct bio *bio; | |
1064 | struct bio_list bios; | |
1065 | ||
3e1a0699 JT |
1066 | if (should_error_unserviceable_bio(pool)) { |
1067 | cell_error(pool, cell); | |
1068 | return; | |
1069 | } | |
1070 | ||
991d9fa0 | 1071 | bio_list_init(&bios); |
6beca5eb | 1072 | cell_release(pool, cell, &bios); |
991d9fa0 | 1073 | |
3e1a0699 JT |
1074 | if (should_error_unserviceable_bio(pool)) |
1075 | while ((bio = bio_list_pop(&bios))) | |
1076 | bio_io_error(bio); | |
1077 | else | |
1078 | while ((bio = bio_list_pop(&bios))) | |
1079 | retry_on_resume(bio); | |
991d9fa0 JT |
1080 | } |
1081 | ||
104655fd JT |
1082 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1083 | { | |
1084 | int r; | |
c3a0ce2e | 1085 | unsigned long flags; |
104655fd | 1086 | struct pool *pool = tc->pool; |
a24c2569 | 1087 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1088 | struct dm_cell_key key, key2; |
104655fd JT |
1089 | dm_block_t block = get_bio_block(tc, bio); |
1090 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1091 | struct dm_thin_new_mapping *m; |
104655fd JT |
1092 | |
1093 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1094 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1095 | return; |
1096 | ||
1097 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1098 | switch (r) { | |
1099 | case 0: | |
1100 | /* | |
1101 | * Check nobody is fiddling with this pool block. This can | |
1102 | * happen if someone's in the process of breaking sharing | |
1103 | * on this block. | |
1104 | */ | |
1105 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1106 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1107 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1108 | break; |
1109 | } | |
1110 | ||
1111 | if (io_overlaps_block(pool, bio)) { | |
1112 | /* | |
1113 | * IO may still be going to the destination block. We must | |
1114 | * quiesce before we can do the removal. | |
1115 | */ | |
1116 | m = get_next_mapping(pool); | |
1117 | m->tc = tc; | |
19fa1a67 JT |
1118 | m->pass_discard = pool->pf.discard_passdown; |
1119 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1120 | m->virt_block = block; |
1121 | m->data_block = lookup_result.block; | |
1122 | m->cell = cell; | |
1123 | m->cell2 = cell2; | |
104655fd JT |
1124 | m->bio = bio; |
1125 | ||
44feb387 | 1126 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1127 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1128 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1129 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1130 | wake_worker(pool); |
1131 | } | |
1132 | } else { | |
e8088073 | 1133 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1134 | cell_defer_no_holder(tc, cell); |
1135 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1136 | |
104655fd | 1137 | /* |
49296309 MP |
1138 | * The DM core makes sure that the discard doesn't span |
1139 | * a block boundary. So we submit the discard of a | |
1140 | * partial block appropriately. | |
104655fd | 1141 | */ |
650d2a06 MP |
1142 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1143 | remap_and_issue(tc, bio, lookup_result.block); | |
1144 | else | |
1145 | bio_endio(bio, 0); | |
104655fd JT |
1146 | } |
1147 | break; | |
1148 | ||
1149 | case -ENODATA: | |
1150 | /* | |
1151 | * It isn't provisioned, just forget it. | |
1152 | */ | |
f286ba0e | 1153 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1154 | bio_endio(bio, 0); |
1155 | break; | |
1156 | ||
1157 | default: | |
c397741c MS |
1158 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1159 | __func__, r); | |
f286ba0e | 1160 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1161 | bio_io_error(bio); |
1162 | break; | |
1163 | } | |
1164 | } | |
1165 | ||
991d9fa0 | 1166 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1167 | struct dm_cell_key *key, |
991d9fa0 | 1168 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1169 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1170 | { |
1171 | int r; | |
1172 | dm_block_t data_block; | |
d6fc2042 | 1173 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1174 | |
1175 | r = alloc_data_block(tc, &data_block); | |
1176 | switch (r) { | |
1177 | case 0: | |
2dd9c257 JT |
1178 | schedule_internal_copy(tc, block, lookup_result->block, |
1179 | data_block, cell, bio); | |
991d9fa0 JT |
1180 | break; |
1181 | ||
1182 | case -ENOSPC: | |
399caddf | 1183 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1184 | break; |
1185 | ||
1186 | default: | |
c397741c MS |
1187 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1188 | __func__, r); | |
d6fc2042 | 1189 | cell_error(pool, cell); |
991d9fa0 JT |
1190 | break; |
1191 | } | |
1192 | } | |
1193 | ||
1194 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1195 | dm_block_t block, | |
1196 | struct dm_thin_lookup_result *lookup_result) | |
1197 | { | |
a24c2569 | 1198 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1199 | struct pool *pool = tc->pool; |
44feb387 | 1200 | struct dm_cell_key key; |
991d9fa0 JT |
1201 | |
1202 | /* | |
1203 | * If cell is already occupied, then sharing is already in the process | |
1204 | * of being broken so we have nothing further to do here. | |
1205 | */ | |
1206 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1207 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1208 | return; |
1209 | ||
4f024f37 | 1210 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
991d9fa0 JT |
1211 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1212 | else { | |
59c3d2c6 | 1213 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1214 | |
44feb387 | 1215 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1216 | inc_all_io_entry(pool, bio); |
f286ba0e | 1217 | cell_defer_no_holder(tc, cell); |
e8088073 | 1218 | |
991d9fa0 JT |
1219 | remap_and_issue(tc, bio, lookup_result->block); |
1220 | } | |
1221 | } | |
1222 | ||
1223 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1224 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1225 | { |
1226 | int r; | |
1227 | dm_block_t data_block; | |
6beca5eb | 1228 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1229 | |
1230 | /* | |
1231 | * Remap empty bios (flushes) immediately, without provisioning. | |
1232 | */ | |
4f024f37 | 1233 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1234 | inc_all_io_entry(pool, bio); |
f286ba0e | 1235 | cell_defer_no_holder(tc, cell); |
e8088073 | 1236 | |
991d9fa0 JT |
1237 | remap_and_issue(tc, bio, 0); |
1238 | return; | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Fill read bios with zeroes and complete them immediately. | |
1243 | */ | |
1244 | if (bio_data_dir(bio) == READ) { | |
1245 | zero_fill_bio(bio); | |
f286ba0e | 1246 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1247 | bio_endio(bio, 0); |
1248 | return; | |
1249 | } | |
1250 | ||
1251 | r = alloc_data_block(tc, &data_block); | |
1252 | switch (r) { | |
1253 | case 0: | |
2dd9c257 JT |
1254 | if (tc->origin_dev) |
1255 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1256 | else | |
1257 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1258 | break; |
1259 | ||
1260 | case -ENOSPC: | |
399caddf | 1261 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1262 | break; |
1263 | ||
1264 | default: | |
c397741c MS |
1265 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1266 | __func__, r); | |
6beca5eb | 1267 | cell_error(pool, cell); |
991d9fa0 JT |
1268 | break; |
1269 | } | |
1270 | } | |
1271 | ||
1272 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1273 | { | |
1274 | int r; | |
6beca5eb | 1275 | struct pool *pool = tc->pool; |
991d9fa0 | 1276 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1277 | struct dm_bio_prison_cell *cell; |
44feb387 | 1278 | struct dm_cell_key key; |
991d9fa0 JT |
1279 | struct dm_thin_lookup_result lookup_result; |
1280 | ||
1281 | /* | |
1282 | * If cell is already occupied, then the block is already | |
1283 | * being provisioned so we have nothing further to do here. | |
1284 | */ | |
1285 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1286 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1287 | return; |
1288 | ||
1289 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1290 | switch (r) { | |
1291 | case 0: | |
e8088073 | 1292 | if (lookup_result.shared) { |
991d9fa0 | 1293 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1294 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1295 | } else { |
6beca5eb | 1296 | inc_all_io_entry(pool, bio); |
f286ba0e | 1297 | cell_defer_no_holder(tc, cell); |
e8088073 | 1298 | |
991d9fa0 | 1299 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1300 | } |
991d9fa0 JT |
1301 | break; |
1302 | ||
1303 | case -ENODATA: | |
2dd9c257 | 1304 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1305 | inc_all_io_entry(pool, bio); |
f286ba0e | 1306 | cell_defer_no_holder(tc, cell); |
e8088073 | 1307 | |
2dd9c257 JT |
1308 | remap_to_origin_and_issue(tc, bio); |
1309 | } else | |
1310 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1311 | break; |
1312 | ||
1313 | default: | |
c397741c MS |
1314 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1315 | __func__, r); | |
f286ba0e | 1316 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1317 | bio_io_error(bio); |
1318 | break; | |
1319 | } | |
1320 | } | |
1321 | ||
e49e5829 JT |
1322 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1323 | { | |
1324 | int r; | |
1325 | int rw = bio_data_dir(bio); | |
1326 | dm_block_t block = get_bio_block(tc, bio); | |
1327 | struct dm_thin_lookup_result lookup_result; | |
1328 | ||
1329 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1330 | switch (r) { | |
1331 | case 0: | |
4f024f37 | 1332 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
8c0f0e8c | 1333 | handle_unserviceable_bio(tc->pool, bio); |
e8088073 JT |
1334 | else { |
1335 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1336 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1337 | } |
e49e5829 JT |
1338 | break; |
1339 | ||
1340 | case -ENODATA: | |
1341 | if (rw != READ) { | |
8c0f0e8c | 1342 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1343 | break; |
1344 | } | |
1345 | ||
1346 | if (tc->origin_dev) { | |
e8088073 | 1347 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1348 | remap_to_origin_and_issue(tc, bio); |
1349 | break; | |
1350 | } | |
1351 | ||
1352 | zero_fill_bio(bio); | |
1353 | bio_endio(bio, 0); | |
1354 | break; | |
1355 | ||
1356 | default: | |
c397741c MS |
1357 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1358 | __func__, r); | |
e49e5829 JT |
1359 | bio_io_error(bio); |
1360 | break; | |
1361 | } | |
1362 | } | |
1363 | ||
3e1a0699 JT |
1364 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1365 | { | |
1366 | bio_endio(bio, 0); | |
1367 | } | |
1368 | ||
e49e5829 JT |
1369 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1370 | { | |
1371 | bio_io_error(bio); | |
1372 | } | |
1373 | ||
ac8c3f3d JT |
1374 | /* |
1375 | * FIXME: should we also commit due to size of transaction, measured in | |
1376 | * metadata blocks? | |
1377 | */ | |
905e51b3 JT |
1378 | static int need_commit_due_to_time(struct pool *pool) |
1379 | { | |
1380 | return jiffies < pool->last_commit_jiffies || | |
1381 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1382 | } | |
1383 | ||
67324ea1 MS |
1384 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1385 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1386 | ||
1387 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1388 | { | |
1389 | struct rb_node **rbp, *parent; | |
1390 | struct dm_thin_endio_hook *pbd; | |
1391 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1392 | ||
1393 | rbp = &tc->sort_bio_list.rb_node; | |
1394 | parent = NULL; | |
1395 | while (*rbp) { | |
1396 | parent = *rbp; | |
1397 | pbd = thin_pbd(parent); | |
1398 | ||
1399 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1400 | rbp = &(*rbp)->rb_left; | |
1401 | else | |
1402 | rbp = &(*rbp)->rb_right; | |
1403 | } | |
1404 | ||
1405 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1406 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1407 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1408 | } | |
1409 | ||
1410 | static void __extract_sorted_bios(struct thin_c *tc) | |
1411 | { | |
1412 | struct rb_node *node; | |
1413 | struct dm_thin_endio_hook *pbd; | |
1414 | struct bio *bio; | |
1415 | ||
1416 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1417 | pbd = thin_pbd(node); | |
1418 | bio = thin_bio(pbd); | |
1419 | ||
1420 | bio_list_add(&tc->deferred_bio_list, bio); | |
1421 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1422 | } | |
1423 | ||
1424 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1425 | } | |
1426 | ||
1427 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1428 | { | |
1429 | struct bio *bio; | |
1430 | struct bio_list bios; | |
1431 | ||
1432 | bio_list_init(&bios); | |
1433 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1434 | bio_list_init(&tc->deferred_bio_list); | |
1435 | ||
1436 | /* Sort deferred_bio_list using rb-tree */ | |
1437 | while ((bio = bio_list_pop(&bios))) | |
1438 | __thin_bio_rb_add(tc, bio); | |
1439 | ||
1440 | /* | |
1441 | * Transfer the sorted bios in sort_bio_list back to | |
1442 | * deferred_bio_list to allow lockless submission of | |
1443 | * all bios. | |
1444 | */ | |
1445 | __extract_sorted_bios(tc); | |
1446 | } | |
1447 | ||
c140e1c4 | 1448 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1449 | { |
c140e1c4 | 1450 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1451 | unsigned long flags; |
1452 | struct bio *bio; | |
1453 | struct bio_list bios; | |
67324ea1 | 1454 | struct blk_plug plug; |
991d9fa0 | 1455 | |
c140e1c4 MS |
1456 | if (tc->requeue_mode) { |
1457 | requeue_bio_list(tc, &tc->deferred_bio_list); | |
1458 | return; | |
1459 | } | |
1460 | ||
991d9fa0 JT |
1461 | bio_list_init(&bios); |
1462 | ||
c140e1c4 | 1463 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1464 | |
1465 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1466 | spin_unlock_irqrestore(&tc->lock, flags); | |
1467 | return; | |
1468 | } | |
1469 | ||
1470 | __sort_thin_deferred_bios(tc); | |
1471 | ||
c140e1c4 MS |
1472 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1473 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1474 | |
c140e1c4 | 1475 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1476 | |
67324ea1 | 1477 | blk_start_plug(&plug); |
991d9fa0 | 1478 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1479 | /* |
1480 | * If we've got no free new_mapping structs, and processing | |
1481 | * this bio might require one, we pause until there are some | |
1482 | * prepared mappings to process. | |
1483 | */ | |
1484 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1485 | spin_lock_irqsave(&tc->lock, flags); |
1486 | bio_list_add(&tc->deferred_bio_list, bio); | |
1487 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1488 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1489 | break; |
1490 | } | |
104655fd JT |
1491 | |
1492 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1493 | pool->process_discard(tc, bio); |
104655fd | 1494 | else |
e49e5829 | 1495 | pool->process_bio(tc, bio); |
991d9fa0 | 1496 | } |
67324ea1 | 1497 | blk_finish_plug(&plug); |
c140e1c4 MS |
1498 | } |
1499 | ||
b10ebd34 JT |
1500 | static void thin_get(struct thin_c *tc); |
1501 | static void thin_put(struct thin_c *tc); | |
1502 | ||
1503 | /* | |
1504 | * We can't hold rcu_read_lock() around code that can block. So we | |
1505 | * find a thin with the rcu lock held; bump a refcount; then drop | |
1506 | * the lock. | |
1507 | */ | |
1508 | static struct thin_c *get_first_thin(struct pool *pool) | |
1509 | { | |
1510 | struct thin_c *tc = NULL; | |
1511 | ||
1512 | rcu_read_lock(); | |
1513 | if (!list_empty(&pool->active_thins)) { | |
1514 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
1515 | thin_get(tc); | |
1516 | } | |
1517 | rcu_read_unlock(); | |
1518 | ||
1519 | return tc; | |
1520 | } | |
1521 | ||
1522 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
1523 | { | |
1524 | struct thin_c *old_tc = tc; | |
1525 | ||
1526 | rcu_read_lock(); | |
1527 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
1528 | thin_get(tc); | |
1529 | thin_put(old_tc); | |
1530 | rcu_read_unlock(); | |
1531 | return tc; | |
1532 | } | |
1533 | thin_put(old_tc); | |
1534 | rcu_read_unlock(); | |
1535 | ||
1536 | return NULL; | |
1537 | } | |
1538 | ||
c140e1c4 MS |
1539 | static void process_deferred_bios(struct pool *pool) |
1540 | { | |
1541 | unsigned long flags; | |
1542 | struct bio *bio; | |
1543 | struct bio_list bios; | |
1544 | struct thin_c *tc; | |
1545 | ||
b10ebd34 JT |
1546 | tc = get_first_thin(pool); |
1547 | while (tc) { | |
c140e1c4 | 1548 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
1549 | tc = get_next_thin(pool, tc); |
1550 | } | |
991d9fa0 JT |
1551 | |
1552 | /* | |
1553 | * If there are any deferred flush bios, we must commit | |
1554 | * the metadata before issuing them. | |
1555 | */ | |
1556 | bio_list_init(&bios); | |
1557 | spin_lock_irqsave(&pool->lock, flags); | |
1558 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1559 | bio_list_init(&pool->deferred_flush_bios); | |
1560 | spin_unlock_irqrestore(&pool->lock, flags); | |
1561 | ||
4d1662a3 MS |
1562 | if (bio_list_empty(&bios) && |
1563 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1564 | return; |
1565 | ||
020cc3b5 | 1566 | if (commit(pool)) { |
991d9fa0 JT |
1567 | while ((bio = bio_list_pop(&bios))) |
1568 | bio_io_error(bio); | |
1569 | return; | |
1570 | } | |
905e51b3 | 1571 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1572 | |
1573 | while ((bio = bio_list_pop(&bios))) | |
1574 | generic_make_request(bio); | |
1575 | } | |
1576 | ||
1577 | static void do_worker(struct work_struct *ws) | |
1578 | { | |
1579 | struct pool *pool = container_of(ws, struct pool, worker); | |
1580 | ||
e49e5829 JT |
1581 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1582 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1583 | process_deferred_bios(pool); |
1584 | } | |
1585 | ||
905e51b3 JT |
1586 | /* |
1587 | * We want to commit periodically so that not too much | |
1588 | * unwritten data builds up. | |
1589 | */ | |
1590 | static void do_waker(struct work_struct *ws) | |
1591 | { | |
1592 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1593 | wake_worker(pool); | |
1594 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1595 | } | |
1596 | ||
85ad643b JT |
1597 | /* |
1598 | * We're holding onto IO to allow userland time to react. After the | |
1599 | * timeout either the pool will have been resized (and thus back in | |
1600 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
1601 | */ | |
1602 | static void do_no_space_timeout(struct work_struct *ws) | |
1603 | { | |
1604 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
1605 | no_space_timeout); | |
1606 | ||
1607 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
1608 | set_pool_mode(pool, PM_READ_ONLY); | |
1609 | } | |
1610 | ||
991d9fa0 JT |
1611 | /*----------------------------------------------------------------*/ |
1612 | ||
e7a3e871 | 1613 | struct pool_work { |
738211f7 | 1614 | struct work_struct worker; |
e7a3e871 JT |
1615 | struct completion complete; |
1616 | }; | |
1617 | ||
1618 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
1619 | { | |
1620 | return container_of(ws, struct pool_work, worker); | |
1621 | } | |
1622 | ||
1623 | static void pool_work_complete(struct pool_work *pw) | |
1624 | { | |
1625 | complete(&pw->complete); | |
1626 | } | |
738211f7 | 1627 | |
e7a3e871 JT |
1628 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
1629 | void (*fn)(struct work_struct *)) | |
1630 | { | |
1631 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
1632 | init_completion(&pw->complete); | |
1633 | queue_work(pool->wq, &pw->worker); | |
1634 | wait_for_completion(&pw->complete); | |
1635 | } | |
1636 | ||
1637 | /*----------------------------------------------------------------*/ | |
1638 | ||
1639 | struct noflush_work { | |
1640 | struct pool_work pw; | |
1641 | struct thin_c *tc; | |
738211f7 JT |
1642 | }; |
1643 | ||
e7a3e871 | 1644 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 1645 | { |
e7a3e871 | 1646 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
1647 | } |
1648 | ||
1649 | static void do_noflush_start(struct work_struct *ws) | |
1650 | { | |
e7a3e871 | 1651 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
1652 | w->tc->requeue_mode = true; |
1653 | requeue_io(w->tc); | |
e7a3e871 | 1654 | pool_work_complete(&w->pw); |
738211f7 JT |
1655 | } |
1656 | ||
1657 | static void do_noflush_stop(struct work_struct *ws) | |
1658 | { | |
e7a3e871 | 1659 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 1660 | w->tc->requeue_mode = false; |
e7a3e871 | 1661 | pool_work_complete(&w->pw); |
738211f7 JT |
1662 | } |
1663 | ||
1664 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
1665 | { | |
1666 | struct noflush_work w; | |
1667 | ||
738211f7 | 1668 | w.tc = tc; |
e7a3e871 | 1669 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
1670 | } |
1671 | ||
1672 | /*----------------------------------------------------------------*/ | |
1673 | ||
e49e5829 JT |
1674 | static enum pool_mode get_pool_mode(struct pool *pool) |
1675 | { | |
1676 | return pool->pf.mode; | |
1677 | } | |
1678 | ||
3e1a0699 JT |
1679 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
1680 | { | |
1681 | dm_table_event(pool->ti->table); | |
1682 | DMINFO("%s: switching pool to %s mode", | |
1683 | dm_device_name(pool->pool_md), new_mode); | |
1684 | } | |
1685 | ||
8b64e881 | 1686 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 1687 | { |
cdc2b415 | 1688 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
1689 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
1690 | enum pool_mode old_mode = get_pool_mode(pool); | |
80c57893 | 1691 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
1692 | |
1693 | /* | |
1694 | * Never allow the pool to transition to PM_WRITE mode if user | |
1695 | * intervention is required to verify metadata and data consistency. | |
1696 | */ | |
1697 | if (new_mode == PM_WRITE && needs_check) { | |
1698 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
1699 | dm_device_name(pool->pool_md)); | |
1700 | if (old_mode != new_mode) | |
1701 | new_mode = old_mode; | |
1702 | else | |
1703 | new_mode = PM_READ_ONLY; | |
1704 | } | |
1705 | /* | |
1706 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1707 | * not going to recover without a thin_repair. So we never let the | |
1708 | * pool move out of the old mode. | |
1709 | */ | |
1710 | if (old_mode == PM_FAIL) | |
1711 | new_mode = old_mode; | |
e49e5829 | 1712 | |
8b64e881 | 1713 | switch (new_mode) { |
e49e5829 | 1714 | case PM_FAIL: |
8b64e881 | 1715 | if (old_mode != new_mode) |
3e1a0699 | 1716 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 1717 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1718 | pool->process_bio = process_bio_fail; |
1719 | pool->process_discard = process_bio_fail; | |
1720 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1721 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
1722 | |
1723 | error_retry_list(pool); | |
e49e5829 JT |
1724 | break; |
1725 | ||
1726 | case PM_READ_ONLY: | |
8b64e881 | 1727 | if (old_mode != new_mode) |
3e1a0699 JT |
1728 | notify_of_pool_mode_change(pool, "read-only"); |
1729 | dm_pool_metadata_read_only(pool->pmd); | |
1730 | pool->process_bio = process_bio_read_only; | |
1731 | pool->process_discard = process_bio_success; | |
1732 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1733 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1734 | ||
1735 | error_retry_list(pool); | |
1736 | break; | |
1737 | ||
1738 | case PM_OUT_OF_DATA_SPACE: | |
1739 | /* | |
1740 | * Ideally we'd never hit this state; the low water mark | |
1741 | * would trigger userland to extend the pool before we | |
1742 | * completely run out of data space. However, many small | |
1743 | * IOs to unprovisioned space can consume data space at an | |
1744 | * alarming rate. Adjust your low water mark if you're | |
1745 | * frequently seeing this mode. | |
1746 | */ | |
1747 | if (old_mode != new_mode) | |
1748 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
1749 | pool->process_bio = process_bio_read_only; | |
1750 | pool->process_discard = process_discard; | |
1751 | pool->process_prepared_mapping = process_prepared_mapping; | |
1752 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
85ad643b | 1753 | |
80c57893 MS |
1754 | if (!pool->pf.error_if_no_space && no_space_timeout) |
1755 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
1756 | break; |
1757 | ||
1758 | case PM_WRITE: | |
8b64e881 | 1759 | if (old_mode != new_mode) |
3e1a0699 | 1760 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 1761 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1762 | pool->process_bio = process_bio; |
1763 | pool->process_discard = process_discard; | |
1764 | pool->process_prepared_mapping = process_prepared_mapping; | |
1765 | pool->process_prepared_discard = process_prepared_discard; | |
1766 | break; | |
1767 | } | |
8b64e881 MS |
1768 | |
1769 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
1770 | /* |
1771 | * The pool mode may have changed, sync it so bind_control_target() | |
1772 | * doesn't cause an unexpected mode transition on resume. | |
1773 | */ | |
1774 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
1775 | } |
1776 | ||
07f2b6e0 | 1777 | static void abort_transaction(struct pool *pool) |
b5330655 | 1778 | { |
07f2b6e0 MS |
1779 | const char *dev_name = dm_device_name(pool->pool_md); |
1780 | ||
1781 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
1782 | if (dm_pool_abort_metadata(pool->pmd)) { | |
1783 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
1784 | set_pool_mode(pool, PM_FAIL); | |
1785 | } | |
1786 | ||
1787 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
1788 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
1789 | set_pool_mode(pool, PM_FAIL); | |
1790 | } | |
1791 | } | |
399caddf | 1792 | |
07f2b6e0 MS |
1793 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1794 | { | |
b5330655 JT |
1795 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1796 | dm_device_name(pool->pool_md), op, r); | |
1797 | ||
07f2b6e0 | 1798 | abort_transaction(pool); |
b5330655 JT |
1799 | set_pool_mode(pool, PM_READ_ONLY); |
1800 | } | |
1801 | ||
e49e5829 JT |
1802 | /*----------------------------------------------------------------*/ |
1803 | ||
991d9fa0 JT |
1804 | /* |
1805 | * Mapping functions. | |
1806 | */ | |
1807 | ||
1808 | /* | |
1809 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1810 | */ | |
1811 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1812 | { | |
1813 | unsigned long flags; | |
1814 | struct pool *pool = tc->pool; | |
1815 | ||
c140e1c4 MS |
1816 | spin_lock_irqsave(&tc->lock, flags); |
1817 | bio_list_add(&tc->deferred_bio_list, bio); | |
1818 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1819 | |
1820 | wake_worker(pool); | |
1821 | } | |
1822 | ||
59c3d2c6 | 1823 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1824 | { |
59c3d2c6 | 1825 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1826 | |
1827 | h->tc = tc; | |
1828 | h->shared_read_entry = NULL; | |
e8088073 | 1829 | h->all_io_entry = NULL; |
eb2aa48d | 1830 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1831 | } |
1832 | ||
991d9fa0 JT |
1833 | /* |
1834 | * Non-blocking function called from the thin target's map function. | |
1835 | */ | |
7de3ee57 | 1836 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1837 | { |
1838 | int r; | |
1839 | struct thin_c *tc = ti->private; | |
1840 | dm_block_t block = get_bio_block(tc, bio); | |
1841 | struct dm_thin_device *td = tc->td; | |
1842 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1843 | struct dm_bio_prison_cell cell1, cell2; |
1844 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1845 | struct dm_cell_key key; |
991d9fa0 | 1846 | |
59c3d2c6 | 1847 | thin_hook_bio(tc, bio); |
e49e5829 | 1848 | |
738211f7 JT |
1849 | if (tc->requeue_mode) { |
1850 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1851 | return DM_MAPIO_SUBMITTED; | |
1852 | } | |
1853 | ||
e49e5829 JT |
1854 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
1855 | bio_io_error(bio); | |
1856 | return DM_MAPIO_SUBMITTED; | |
1857 | } | |
1858 | ||
104655fd | 1859 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1860 | thin_defer_bio(tc, bio); |
1861 | return DM_MAPIO_SUBMITTED; | |
1862 | } | |
1863 | ||
1864 | r = dm_thin_find_block(td, block, 0, &result); | |
1865 | ||
1866 | /* | |
1867 | * Note that we defer readahead too. | |
1868 | */ | |
1869 | switch (r) { | |
1870 | case 0: | |
1871 | if (unlikely(result.shared)) { | |
1872 | /* | |
1873 | * We have a race condition here between the | |
1874 | * result.shared value returned by the lookup and | |
1875 | * snapshot creation, which may cause new | |
1876 | * sharing. | |
1877 | * | |
1878 | * To avoid this always quiesce the origin before | |
1879 | * taking the snap. You want to do this anyway to | |
1880 | * ensure a consistent application view | |
1881 | * (i.e. lockfs). | |
1882 | * | |
1883 | * More distant ancestors are irrelevant. The | |
1884 | * shared flag will be set in their case. | |
1885 | */ | |
1886 | thin_defer_bio(tc, bio); | |
e8088073 | 1887 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1888 | } |
e8088073 JT |
1889 | |
1890 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1891 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1892 | return DM_MAPIO_SUBMITTED; |
1893 | ||
1894 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1895 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1896 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1897 | return DM_MAPIO_SUBMITTED; |
1898 | } | |
1899 | ||
1900 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1901 | cell_defer_no_holder_no_free(tc, &cell2); |
1902 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1903 | |
1904 | remap(tc, bio, result.block); | |
1905 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1906 | |
1907 | case -ENODATA: | |
e49e5829 JT |
1908 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1909 | /* | |
1910 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 1911 | * of doing so. |
e49e5829 | 1912 | */ |
8c0f0e8c | 1913 | handle_unserviceable_bio(tc->pool, bio); |
2aab3850 | 1914 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1915 | } |
1916 | /* fall through */ | |
1917 | ||
1918 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1919 | /* |
1920 | * In future, the failed dm_thin_find_block above could | |
1921 | * provide the hint to load the metadata into cache. | |
1922 | */ | |
991d9fa0 | 1923 | thin_defer_bio(tc, bio); |
2aab3850 | 1924 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1925 | |
1926 | default: | |
1927 | /* | |
1928 | * Must always call bio_io_error on failure. | |
1929 | * dm_thin_find_block can fail with -EINVAL if the | |
1930 | * pool is switched to fail-io mode. | |
1931 | */ | |
1932 | bio_io_error(bio); | |
2aab3850 | 1933 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1934 | } |
991d9fa0 JT |
1935 | } |
1936 | ||
1937 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1938 | { | |
991d9fa0 | 1939 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 1940 | struct request_queue *q; |
991d9fa0 | 1941 | |
760fe67e MS |
1942 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
1943 | return 1; | |
991d9fa0 | 1944 | |
760fe67e MS |
1945 | q = bdev_get_queue(pt->data_dev->bdev); |
1946 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
1947 | } |
1948 | ||
c140e1c4 | 1949 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 1950 | { |
c140e1c4 MS |
1951 | unsigned long flags; |
1952 | struct thin_c *tc; | |
1953 | ||
1954 | rcu_read_lock(); | |
1955 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
1956 | spin_lock_irqsave(&tc->lock, flags); | |
1957 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
1958 | bio_list_init(&tc->retry_on_resume_list); | |
1959 | spin_unlock_irqrestore(&tc->lock, flags); | |
1960 | } | |
1961 | rcu_read_unlock(); | |
991d9fa0 JT |
1962 | } |
1963 | ||
1964 | /*---------------------------------------------------------------- | |
1965 | * Binding of control targets to a pool object | |
1966 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1967 | static bool data_dev_supports_discard(struct pool_c *pt) |
1968 | { | |
1969 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1970 | ||
1971 | return q && blk_queue_discard(q); | |
1972 | } | |
1973 | ||
58051b94 JT |
1974 | static bool is_factor(sector_t block_size, uint32_t n) |
1975 | { | |
1976 | return !sector_div(block_size, n); | |
1977 | } | |
1978 | ||
9bc142dd MS |
1979 | /* |
1980 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1981 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1982 | */ |
0424caa1 | 1983 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1984 | { |
0424caa1 MS |
1985 | struct pool *pool = pt->pool; |
1986 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1987 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1988 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1989 | const char *reason = NULL; | |
9bc142dd MS |
1990 | char buf[BDEVNAME_SIZE]; |
1991 | ||
0424caa1 | 1992 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1993 | return; |
1994 | ||
0424caa1 MS |
1995 | if (!data_dev_supports_discard(pt)) |
1996 | reason = "discard unsupported"; | |
1997 | ||
1998 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1999 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2000 | |
0424caa1 MS |
2001 | else if (data_limits->discard_granularity > block_size) |
2002 | reason = "discard granularity larger than a block"; | |
2003 | ||
58051b94 | 2004 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
2005 | reason = "discard granularity not a factor of block size"; |
2006 | ||
2007 | if (reason) { | |
2008 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2009 | pt->adjusted_pf.discard_passdown = false; | |
2010 | } | |
9bc142dd MS |
2011 | } |
2012 | ||
991d9fa0 JT |
2013 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2014 | { | |
2015 | struct pool_c *pt = ti->private; | |
2016 | ||
e49e5829 | 2017 | /* |
9b7aaa64 | 2018 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2019 | */ |
07f2b6e0 | 2020 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2021 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2022 | |
8b64e881 MS |
2023 | /* |
2024 | * Don't change the pool's mode until set_pool_mode() below. | |
2025 | * Otherwise the pool's process_* function pointers may | |
2026 | * not match the desired pool mode. | |
2027 | */ | |
2028 | pt->adjusted_pf.mode = old_mode; | |
2029 | ||
2030 | pool->ti = ti; | |
2031 | pool->pf = pt->adjusted_pf; | |
2032 | pool->low_water_blocks = pt->low_water_blocks; | |
2033 | ||
9bc142dd | 2034 | set_pool_mode(pool, new_mode); |
f402693d | 2035 | |
991d9fa0 JT |
2036 | return 0; |
2037 | } | |
2038 | ||
2039 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2040 | { | |
2041 | if (pool->ti == ti) | |
2042 | pool->ti = NULL; | |
2043 | } | |
2044 | ||
2045 | /*---------------------------------------------------------------- | |
2046 | * Pool creation | |
2047 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2048 | /* Initialize pool features. */ |
2049 | static void pool_features_init(struct pool_features *pf) | |
2050 | { | |
e49e5829 | 2051 | pf->mode = PM_WRITE; |
9bc142dd MS |
2052 | pf->zero_new_blocks = true; |
2053 | pf->discard_enabled = true; | |
2054 | pf->discard_passdown = true; | |
787a996c | 2055 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2056 | } |
2057 | ||
991d9fa0 JT |
2058 | static void __pool_destroy(struct pool *pool) |
2059 | { | |
2060 | __pool_table_remove(pool); | |
2061 | ||
2062 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2063 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2064 | ||
44feb387 | 2065 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2066 | dm_kcopyd_client_destroy(pool->copier); |
2067 | ||
2068 | if (pool->wq) | |
2069 | destroy_workqueue(pool->wq); | |
2070 | ||
2071 | if (pool->next_mapping) | |
2072 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2073 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2074 | dm_deferred_set_destroy(pool->shared_read_ds); |
2075 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2076 | kfree(pool); |
2077 | } | |
2078 | ||
a24c2569 | 2079 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2080 | |
991d9fa0 JT |
2081 | static struct pool *pool_create(struct mapped_device *pool_md, |
2082 | struct block_device *metadata_dev, | |
e49e5829 JT |
2083 | unsigned long block_size, |
2084 | int read_only, char **error) | |
991d9fa0 JT |
2085 | { |
2086 | int r; | |
2087 | void *err_p; | |
2088 | struct pool *pool; | |
2089 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2090 | bool format_device = read_only ? false : true; |
991d9fa0 | 2091 | |
e49e5829 | 2092 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2093 | if (IS_ERR(pmd)) { |
2094 | *error = "Error creating metadata object"; | |
2095 | return (struct pool *)pmd; | |
2096 | } | |
2097 | ||
2098 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2099 | if (!pool) { | |
2100 | *error = "Error allocating memory for pool"; | |
2101 | err_p = ERR_PTR(-ENOMEM); | |
2102 | goto bad_pool; | |
2103 | } | |
2104 | ||
2105 | pool->pmd = pmd; | |
2106 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2107 | if (block_size & (block_size - 1)) |
2108 | pool->sectors_per_block_shift = -1; | |
2109 | else | |
2110 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2111 | pool->low_water_blocks = 0; |
67e2e2b2 | 2112 | pool_features_init(&pool->pf); |
44feb387 | 2113 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
2114 | if (!pool->prison) { |
2115 | *error = "Error creating pool's bio prison"; | |
2116 | err_p = ERR_PTR(-ENOMEM); | |
2117 | goto bad_prison; | |
2118 | } | |
2119 | ||
df5d2e90 | 2120 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2121 | if (IS_ERR(pool->copier)) { |
2122 | r = PTR_ERR(pool->copier); | |
2123 | *error = "Error creating pool's kcopyd client"; | |
2124 | err_p = ERR_PTR(r); | |
2125 | goto bad_kcopyd_client; | |
2126 | } | |
2127 | ||
2128 | /* | |
2129 | * Create singlethreaded workqueue that will service all devices | |
2130 | * that use this metadata. | |
2131 | */ | |
2132 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2133 | if (!pool->wq) { | |
2134 | *error = "Error creating pool's workqueue"; | |
2135 | err_p = ERR_PTR(-ENOMEM); | |
2136 | goto bad_wq; | |
2137 | } | |
2138 | ||
2139 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 2140 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2141 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2142 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2143 | bio_list_init(&pool->deferred_flush_bios); |
2144 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2145 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2146 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2147 | pool->low_water_triggered = false; |
44feb387 MS |
2148 | |
2149 | pool->shared_read_ds = dm_deferred_set_create(); | |
2150 | if (!pool->shared_read_ds) { | |
2151 | *error = "Error creating pool's shared read deferred set"; | |
2152 | err_p = ERR_PTR(-ENOMEM); | |
2153 | goto bad_shared_read_ds; | |
2154 | } | |
2155 | ||
2156 | pool->all_io_ds = dm_deferred_set_create(); | |
2157 | if (!pool->all_io_ds) { | |
2158 | *error = "Error creating pool's all io deferred set"; | |
2159 | err_p = ERR_PTR(-ENOMEM); | |
2160 | goto bad_all_io_ds; | |
2161 | } | |
991d9fa0 JT |
2162 | |
2163 | pool->next_mapping = NULL; | |
a24c2569 MS |
2164 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2165 | _new_mapping_cache); | |
991d9fa0 JT |
2166 | if (!pool->mapping_pool) { |
2167 | *error = "Error creating pool's mapping mempool"; | |
2168 | err_p = ERR_PTR(-ENOMEM); | |
2169 | goto bad_mapping_pool; | |
2170 | } | |
2171 | ||
991d9fa0 | 2172 | pool->ref_count = 1; |
905e51b3 | 2173 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2174 | pool->pool_md = pool_md; |
2175 | pool->md_dev = metadata_dev; | |
2176 | __pool_table_insert(pool); | |
2177 | ||
2178 | return pool; | |
2179 | ||
991d9fa0 | 2180 | bad_mapping_pool: |
44feb387 MS |
2181 | dm_deferred_set_destroy(pool->all_io_ds); |
2182 | bad_all_io_ds: | |
2183 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2184 | bad_shared_read_ds: | |
991d9fa0 JT |
2185 | destroy_workqueue(pool->wq); |
2186 | bad_wq: | |
2187 | dm_kcopyd_client_destroy(pool->copier); | |
2188 | bad_kcopyd_client: | |
44feb387 | 2189 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2190 | bad_prison: |
2191 | kfree(pool); | |
2192 | bad_pool: | |
2193 | if (dm_pool_metadata_close(pmd)) | |
2194 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2195 | ||
2196 | return err_p; | |
2197 | } | |
2198 | ||
2199 | static void __pool_inc(struct pool *pool) | |
2200 | { | |
2201 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2202 | pool->ref_count++; | |
2203 | } | |
2204 | ||
2205 | static void __pool_dec(struct pool *pool) | |
2206 | { | |
2207 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2208 | BUG_ON(!pool->ref_count); | |
2209 | if (!--pool->ref_count) | |
2210 | __pool_destroy(pool); | |
2211 | } | |
2212 | ||
2213 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2214 | struct block_device *metadata_dev, | |
e49e5829 JT |
2215 | unsigned long block_size, int read_only, |
2216 | char **error, int *created) | |
991d9fa0 JT |
2217 | { |
2218 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2219 | ||
2220 | if (pool) { | |
f09996c9 MS |
2221 | if (pool->pool_md != pool_md) { |
2222 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2223 | return ERR_PTR(-EBUSY); |
f09996c9 | 2224 | } |
991d9fa0 JT |
2225 | __pool_inc(pool); |
2226 | ||
2227 | } else { | |
2228 | pool = __pool_table_lookup(pool_md); | |
2229 | if (pool) { | |
f09996c9 MS |
2230 | if (pool->md_dev != metadata_dev) { |
2231 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2232 | return ERR_PTR(-EINVAL); |
f09996c9 | 2233 | } |
991d9fa0 JT |
2234 | __pool_inc(pool); |
2235 | ||
67e2e2b2 | 2236 | } else { |
e49e5829 | 2237 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2238 | *created = 1; |
2239 | } | |
991d9fa0 JT |
2240 | } |
2241 | ||
2242 | return pool; | |
2243 | } | |
2244 | ||
2245 | /*---------------------------------------------------------------- | |
2246 | * Pool target methods | |
2247 | *--------------------------------------------------------------*/ | |
2248 | static void pool_dtr(struct dm_target *ti) | |
2249 | { | |
2250 | struct pool_c *pt = ti->private; | |
2251 | ||
2252 | mutex_lock(&dm_thin_pool_table.mutex); | |
2253 | ||
2254 | unbind_control_target(pt->pool, ti); | |
2255 | __pool_dec(pt->pool); | |
2256 | dm_put_device(ti, pt->metadata_dev); | |
2257 | dm_put_device(ti, pt->data_dev); | |
2258 | kfree(pt); | |
2259 | ||
2260 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2261 | } | |
2262 | ||
991d9fa0 JT |
2263 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2264 | struct dm_target *ti) | |
2265 | { | |
2266 | int r; | |
2267 | unsigned argc; | |
2268 | const char *arg_name; | |
2269 | ||
2270 | static struct dm_arg _args[] = { | |
74aa45c3 | 2271 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2272 | }; |
2273 | ||
2274 | /* | |
2275 | * No feature arguments supplied. | |
2276 | */ | |
2277 | if (!as->argc) | |
2278 | return 0; | |
2279 | ||
2280 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2281 | if (r) | |
2282 | return -EINVAL; | |
2283 | ||
2284 | while (argc && !r) { | |
2285 | arg_name = dm_shift_arg(as); | |
2286 | argc--; | |
2287 | ||
e49e5829 | 2288 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2289 | pf->zero_new_blocks = false; |
e49e5829 JT |
2290 | |
2291 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2292 | pf->discard_enabled = false; |
e49e5829 JT |
2293 | |
2294 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2295 | pf->discard_passdown = false; |
991d9fa0 | 2296 | |
e49e5829 JT |
2297 | else if (!strcasecmp(arg_name, "read_only")) |
2298 | pf->mode = PM_READ_ONLY; | |
2299 | ||
787a996c MS |
2300 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2301 | pf->error_if_no_space = true; | |
2302 | ||
e49e5829 JT |
2303 | else { |
2304 | ti->error = "Unrecognised pool feature requested"; | |
2305 | r = -EINVAL; | |
2306 | break; | |
2307 | } | |
991d9fa0 JT |
2308 | } |
2309 | ||
2310 | return r; | |
2311 | } | |
2312 | ||
ac8c3f3d JT |
2313 | static void metadata_low_callback(void *context) |
2314 | { | |
2315 | struct pool *pool = context; | |
2316 | ||
2317 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2318 | dm_device_name(pool->pool_md)); | |
2319 | ||
2320 | dm_table_event(pool->ti->table); | |
2321 | } | |
2322 | ||
7d48935e MS |
2323 | static sector_t get_dev_size(struct block_device *bdev) |
2324 | { | |
2325 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2326 | } | |
2327 | ||
2328 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2329 | { |
7d48935e | 2330 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2331 | char buffer[BDEVNAME_SIZE]; |
2332 | ||
7d48935e | 2333 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2334 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2335 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2336 | } |
2337 | ||
2338 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2339 | { | |
2340 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2341 | ||
2342 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2343 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2344 | |
2345 | return metadata_dev_size; | |
2346 | } | |
2347 | ||
24347e95 JT |
2348 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2349 | { | |
2350 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2351 | ||
7d48935e | 2352 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2353 | |
2354 | return metadata_dev_size; | |
2355 | } | |
2356 | ||
ac8c3f3d JT |
2357 | /* |
2358 | * When a metadata threshold is crossed a dm event is triggered, and | |
2359 | * userland should respond by growing the metadata device. We could let | |
2360 | * userland set the threshold, like we do with the data threshold, but I'm | |
2361 | * not sure they know enough to do this well. | |
2362 | */ | |
2363 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2364 | { | |
2365 | /* | |
2366 | * 4M is ample for all ops with the possible exception of thin | |
2367 | * device deletion which is harmless if it fails (just retry the | |
2368 | * delete after you've grown the device). | |
2369 | */ | |
2370 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2371 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2372 | } | |
2373 | ||
991d9fa0 JT |
2374 | /* |
2375 | * thin-pool <metadata dev> <data dev> | |
2376 | * <data block size (sectors)> | |
2377 | * <low water mark (blocks)> | |
2378 | * [<#feature args> [<arg>]*] | |
2379 | * | |
2380 | * Optional feature arguments are: | |
2381 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2382 | * ignore_discard: disable discard |
2383 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2384 | * read_only: Don't allow any changes to be made to the pool metadata. |
2385 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2386 | */ |
2387 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2388 | { | |
67e2e2b2 | 2389 | int r, pool_created = 0; |
991d9fa0 JT |
2390 | struct pool_c *pt; |
2391 | struct pool *pool; | |
2392 | struct pool_features pf; | |
2393 | struct dm_arg_set as; | |
2394 | struct dm_dev *data_dev; | |
2395 | unsigned long block_size; | |
2396 | dm_block_t low_water_blocks; | |
2397 | struct dm_dev *metadata_dev; | |
5d0db96d | 2398 | fmode_t metadata_mode; |
991d9fa0 JT |
2399 | |
2400 | /* | |
2401 | * FIXME Remove validation from scope of lock. | |
2402 | */ | |
2403 | mutex_lock(&dm_thin_pool_table.mutex); | |
2404 | ||
2405 | if (argc < 4) { | |
2406 | ti->error = "Invalid argument count"; | |
2407 | r = -EINVAL; | |
2408 | goto out_unlock; | |
2409 | } | |
5d0db96d | 2410 | |
991d9fa0 JT |
2411 | as.argc = argc; |
2412 | as.argv = argv; | |
2413 | ||
5d0db96d JT |
2414 | /* |
2415 | * Set default pool features. | |
2416 | */ | |
2417 | pool_features_init(&pf); | |
2418 | ||
2419 | dm_consume_args(&as, 4); | |
2420 | r = parse_pool_features(&as, &pf, ti); | |
2421 | if (r) | |
2422 | goto out_unlock; | |
2423 | ||
2424 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2425 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2426 | if (r) { |
2427 | ti->error = "Error opening metadata block device"; | |
2428 | goto out_unlock; | |
2429 | } | |
7d48935e | 2430 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2431 | |
2432 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2433 | if (r) { | |
2434 | ti->error = "Error getting data device"; | |
2435 | goto out_metadata; | |
2436 | } | |
2437 | ||
2438 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2439 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2440 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2441 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2442 | ti->error = "Invalid block size"; |
2443 | r = -EINVAL; | |
2444 | goto out; | |
2445 | } | |
2446 | ||
2447 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2448 | ti->error = "Invalid low water mark"; | |
2449 | r = -EINVAL; | |
2450 | goto out; | |
2451 | } | |
2452 | ||
991d9fa0 JT |
2453 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2454 | if (!pt) { | |
2455 | r = -ENOMEM; | |
2456 | goto out; | |
2457 | } | |
2458 | ||
2459 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2460 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2461 | if (IS_ERR(pool)) { |
2462 | r = PTR_ERR(pool); | |
2463 | goto out_free_pt; | |
2464 | } | |
2465 | ||
67e2e2b2 JT |
2466 | /* |
2467 | * 'pool_created' reflects whether this is the first table load. | |
2468 | * Top level discard support is not allowed to be changed after | |
2469 | * initial load. This would require a pool reload to trigger thin | |
2470 | * device changes. | |
2471 | */ | |
2472 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2473 | ti->error = "Discard support cannot be disabled once enabled"; | |
2474 | r = -EINVAL; | |
2475 | goto out_flags_changed; | |
2476 | } | |
2477 | ||
991d9fa0 JT |
2478 | pt->pool = pool; |
2479 | pt->ti = ti; | |
2480 | pt->metadata_dev = metadata_dev; | |
2481 | pt->data_dev = data_dev; | |
2482 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2483 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2484 | ti->num_flush_bios = 1; |
9bc142dd | 2485 | |
67e2e2b2 JT |
2486 | /* |
2487 | * Only need to enable discards if the pool should pass | |
2488 | * them down to the data device. The thin device's discard | |
2489 | * processing will cause mappings to be removed from the btree. | |
2490 | */ | |
b60ab990 | 2491 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2492 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2493 | ti->num_discard_bios = 1; |
9bc142dd | 2494 | |
67e2e2b2 JT |
2495 | /* |
2496 | * Setting 'discards_supported' circumvents the normal | |
2497 | * stacking of discard limits (this keeps the pool and | |
2498 | * thin devices' discard limits consistent). | |
2499 | */ | |
0ac55489 | 2500 | ti->discards_supported = true; |
67e2e2b2 | 2501 | } |
991d9fa0 JT |
2502 | ti->private = pt; |
2503 | ||
ac8c3f3d JT |
2504 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2505 | calc_metadata_threshold(pt), | |
2506 | metadata_low_callback, | |
2507 | pool); | |
2508 | if (r) | |
2509 | goto out_free_pt; | |
2510 | ||
991d9fa0 JT |
2511 | pt->callbacks.congested_fn = pool_is_congested; |
2512 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2513 | ||
2514 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2515 | ||
2516 | return 0; | |
2517 | ||
67e2e2b2 JT |
2518 | out_flags_changed: |
2519 | __pool_dec(pool); | |
991d9fa0 JT |
2520 | out_free_pt: |
2521 | kfree(pt); | |
2522 | out: | |
2523 | dm_put_device(ti, data_dev); | |
2524 | out_metadata: | |
2525 | dm_put_device(ti, metadata_dev); | |
2526 | out_unlock: | |
2527 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2528 | ||
2529 | return r; | |
2530 | } | |
2531 | ||
7de3ee57 | 2532 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2533 | { |
2534 | int r; | |
2535 | struct pool_c *pt = ti->private; | |
2536 | struct pool *pool = pt->pool; | |
2537 | unsigned long flags; | |
2538 | ||
2539 | /* | |
2540 | * As this is a singleton target, ti->begin is always zero. | |
2541 | */ | |
2542 | spin_lock_irqsave(&pool->lock, flags); | |
2543 | bio->bi_bdev = pt->data_dev->bdev; | |
2544 | r = DM_MAPIO_REMAPPED; | |
2545 | spin_unlock_irqrestore(&pool->lock, flags); | |
2546 | ||
2547 | return r; | |
2548 | } | |
2549 | ||
b17446df | 2550 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2551 | { |
2552 | int r; | |
2553 | struct pool_c *pt = ti->private; | |
2554 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2555 | sector_t data_size = ti->len; |
2556 | dm_block_t sb_data_size; | |
991d9fa0 | 2557 | |
b17446df | 2558 | *need_commit = false; |
991d9fa0 | 2559 | |
55f2b8bd MS |
2560 | (void) sector_div(data_size, pool->sectors_per_block); |
2561 | ||
991d9fa0 JT |
2562 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2563 | if (r) { | |
4fa5971a MS |
2564 | DMERR("%s: failed to retrieve data device size", |
2565 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2566 | return r; |
2567 | } | |
2568 | ||
2569 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2570 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2571 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2572 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2573 | return -EINVAL; |
2574 | ||
2575 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2576 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2577 | DMERR("%s: unable to grow the data device until repaired.", | |
2578 | dm_device_name(pool->pool_md)); | |
2579 | return 0; | |
2580 | } | |
2581 | ||
6f7f51d4 MS |
2582 | if (sb_data_size) |
2583 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2584 | dm_device_name(pool->pool_md), | |
2585 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2586 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2587 | if (r) { | |
b5330655 | 2588 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2589 | return r; |
2590 | } | |
2591 | ||
b17446df | 2592 | *need_commit = true; |
991d9fa0 JT |
2593 | } |
2594 | ||
2595 | return 0; | |
2596 | } | |
2597 | ||
24347e95 JT |
2598 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2599 | { | |
2600 | int r; | |
2601 | struct pool_c *pt = ti->private; | |
2602 | struct pool *pool = pt->pool; | |
2603 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2604 | ||
2605 | *need_commit = false; | |
2606 | ||
610bba8b | 2607 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2608 | |
2609 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2610 | if (r) { | |
4fa5971a MS |
2611 | DMERR("%s: failed to retrieve metadata device size", |
2612 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2613 | return r; |
2614 | } | |
2615 | ||
2616 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2617 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2618 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2619 | metadata_dev_size, sb_metadata_dev_size); |
2620 | return -EINVAL; | |
2621 | ||
2622 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
2623 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2624 | DMERR("%s: unable to grow the metadata device until repaired.", | |
2625 | dm_device_name(pool->pool_md)); | |
2626 | return 0; | |
2627 | } | |
2628 | ||
7d48935e | 2629 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
2630 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2631 | dm_device_name(pool->pool_md), | |
2632 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2633 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2634 | if (r) { | |
b5330655 | 2635 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2636 | return r; |
2637 | } | |
2638 | ||
2639 | *need_commit = true; | |
2640 | } | |
2641 | ||
2642 | return 0; | |
2643 | } | |
2644 | ||
b17446df JT |
2645 | /* |
2646 | * Retrieves the number of blocks of the data device from | |
2647 | * the superblock and compares it to the actual device size, | |
2648 | * thus resizing the data device in case it has grown. | |
2649 | * | |
2650 | * This both copes with opening preallocated data devices in the ctr | |
2651 | * being followed by a resume | |
2652 | * -and- | |
2653 | * calling the resume method individually after userspace has | |
2654 | * grown the data device in reaction to a table event. | |
2655 | */ | |
2656 | static int pool_preresume(struct dm_target *ti) | |
2657 | { | |
2658 | int r; | |
24347e95 | 2659 | bool need_commit1, need_commit2; |
b17446df JT |
2660 | struct pool_c *pt = ti->private; |
2661 | struct pool *pool = pt->pool; | |
2662 | ||
2663 | /* | |
2664 | * Take control of the pool object. | |
2665 | */ | |
2666 | r = bind_control_target(pool, ti); | |
2667 | if (r) | |
2668 | return r; | |
2669 | ||
2670 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2671 | if (r) | |
2672 | return r; | |
2673 | ||
24347e95 JT |
2674 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2675 | if (r) | |
2676 | return r; | |
2677 | ||
2678 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2679 | (void) commit(pool); |
b17446df JT |
2680 | |
2681 | return 0; | |
2682 | } | |
2683 | ||
991d9fa0 JT |
2684 | static void pool_resume(struct dm_target *ti) |
2685 | { | |
2686 | struct pool_c *pt = ti->private; | |
2687 | struct pool *pool = pt->pool; | |
2688 | unsigned long flags; | |
2689 | ||
2690 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 2691 | pool->low_water_triggered = false; |
991d9fa0 | 2692 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 2693 | requeue_bios(pool); |
991d9fa0 | 2694 | |
905e51b3 | 2695 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2696 | } |
2697 | ||
2698 | static void pool_postsuspend(struct dm_target *ti) | |
2699 | { | |
991d9fa0 JT |
2700 | struct pool_c *pt = ti->private; |
2701 | struct pool *pool = pt->pool; | |
2702 | ||
905e51b3 | 2703 | cancel_delayed_work(&pool->waker); |
85ad643b | 2704 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 2705 | flush_workqueue(pool->wq); |
020cc3b5 | 2706 | (void) commit(pool); |
991d9fa0 JT |
2707 | } |
2708 | ||
2709 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2710 | { | |
2711 | if (argc != args_required) { | |
2712 | DMWARN("Message received with %u arguments instead of %u.", | |
2713 | argc, args_required); | |
2714 | return -EINVAL; | |
2715 | } | |
2716 | ||
2717 | return 0; | |
2718 | } | |
2719 | ||
2720 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2721 | { | |
2722 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2723 | *dev_id <= MAX_DEV_ID) | |
2724 | return 0; | |
2725 | ||
2726 | if (warning) | |
2727 | DMWARN("Message received with invalid device id: %s", arg); | |
2728 | ||
2729 | return -EINVAL; | |
2730 | } | |
2731 | ||
2732 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2733 | { | |
2734 | dm_thin_id dev_id; | |
2735 | int r; | |
2736 | ||
2737 | r = check_arg_count(argc, 2); | |
2738 | if (r) | |
2739 | return r; | |
2740 | ||
2741 | r = read_dev_id(argv[1], &dev_id, 1); | |
2742 | if (r) | |
2743 | return r; | |
2744 | ||
2745 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2746 | if (r) { | |
2747 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2748 | argv[1]); | |
2749 | return r; | |
2750 | } | |
2751 | ||
2752 | return 0; | |
2753 | } | |
2754 | ||
2755 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2756 | { | |
2757 | dm_thin_id dev_id; | |
2758 | dm_thin_id origin_dev_id; | |
2759 | int r; | |
2760 | ||
2761 | r = check_arg_count(argc, 3); | |
2762 | if (r) | |
2763 | return r; | |
2764 | ||
2765 | r = read_dev_id(argv[1], &dev_id, 1); | |
2766 | if (r) | |
2767 | return r; | |
2768 | ||
2769 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2770 | if (r) | |
2771 | return r; | |
2772 | ||
2773 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2774 | if (r) { | |
2775 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2776 | argv[1], argv[2]); | |
2777 | return r; | |
2778 | } | |
2779 | ||
2780 | return 0; | |
2781 | } | |
2782 | ||
2783 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2784 | { | |
2785 | dm_thin_id dev_id; | |
2786 | int r; | |
2787 | ||
2788 | r = check_arg_count(argc, 2); | |
2789 | if (r) | |
2790 | return r; | |
2791 | ||
2792 | r = read_dev_id(argv[1], &dev_id, 1); | |
2793 | if (r) | |
2794 | return r; | |
2795 | ||
2796 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2797 | if (r) | |
2798 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2799 | ||
2800 | return r; | |
2801 | } | |
2802 | ||
2803 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2804 | { | |
2805 | dm_thin_id old_id, new_id; | |
2806 | int r; | |
2807 | ||
2808 | r = check_arg_count(argc, 3); | |
2809 | if (r) | |
2810 | return r; | |
2811 | ||
2812 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2813 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2814 | return -EINVAL; | |
2815 | } | |
2816 | ||
2817 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2818 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2819 | return -EINVAL; | |
2820 | } | |
2821 | ||
2822 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2823 | if (r) { | |
2824 | DMWARN("Failed to change transaction id from %s to %s.", | |
2825 | argv[1], argv[2]); | |
2826 | return r; | |
2827 | } | |
2828 | ||
2829 | return 0; | |
2830 | } | |
2831 | ||
cc8394d8 JT |
2832 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2833 | { | |
2834 | int r; | |
2835 | ||
2836 | r = check_arg_count(argc, 1); | |
2837 | if (r) | |
2838 | return r; | |
2839 | ||
020cc3b5 | 2840 | (void) commit(pool); |
0d200aef | 2841 | |
cc8394d8 JT |
2842 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2843 | if (r) | |
2844 | DMWARN("reserve_metadata_snap message failed."); | |
2845 | ||
2846 | return r; | |
2847 | } | |
2848 | ||
2849 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2850 | { | |
2851 | int r; | |
2852 | ||
2853 | r = check_arg_count(argc, 1); | |
2854 | if (r) | |
2855 | return r; | |
2856 | ||
2857 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2858 | if (r) | |
2859 | DMWARN("release_metadata_snap message failed."); | |
2860 | ||
2861 | return r; | |
2862 | } | |
2863 | ||
991d9fa0 JT |
2864 | /* |
2865 | * Messages supported: | |
2866 | * create_thin <dev_id> | |
2867 | * create_snap <dev_id> <origin_id> | |
2868 | * delete <dev_id> | |
2869 | * trim <dev_id> <new_size_in_sectors> | |
2870 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2871 | * reserve_metadata_snap |
2872 | * release_metadata_snap | |
991d9fa0 JT |
2873 | */ |
2874 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2875 | { | |
2876 | int r = -EINVAL; | |
2877 | struct pool_c *pt = ti->private; | |
2878 | struct pool *pool = pt->pool; | |
2879 | ||
2880 | if (!strcasecmp(argv[0], "create_thin")) | |
2881 | r = process_create_thin_mesg(argc, argv, pool); | |
2882 | ||
2883 | else if (!strcasecmp(argv[0], "create_snap")) | |
2884 | r = process_create_snap_mesg(argc, argv, pool); | |
2885 | ||
2886 | else if (!strcasecmp(argv[0], "delete")) | |
2887 | r = process_delete_mesg(argc, argv, pool); | |
2888 | ||
2889 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2890 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2891 | ||
cc8394d8 JT |
2892 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2893 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2894 | ||
2895 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2896 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2897 | ||
991d9fa0 JT |
2898 | else |
2899 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2900 | ||
e49e5829 | 2901 | if (!r) |
020cc3b5 | 2902 | (void) commit(pool); |
991d9fa0 JT |
2903 | |
2904 | return r; | |
2905 | } | |
2906 | ||
e49e5829 JT |
2907 | static void emit_flags(struct pool_features *pf, char *result, |
2908 | unsigned sz, unsigned maxlen) | |
2909 | { | |
2910 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
2911 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
2912 | pf->error_if_no_space; | |
e49e5829 JT |
2913 | DMEMIT("%u ", count); |
2914 | ||
2915 | if (!pf->zero_new_blocks) | |
2916 | DMEMIT("skip_block_zeroing "); | |
2917 | ||
2918 | if (!pf->discard_enabled) | |
2919 | DMEMIT("ignore_discard "); | |
2920 | ||
2921 | if (!pf->discard_passdown) | |
2922 | DMEMIT("no_discard_passdown "); | |
2923 | ||
2924 | if (pf->mode == PM_READ_ONLY) | |
2925 | DMEMIT("read_only "); | |
787a996c MS |
2926 | |
2927 | if (pf->error_if_no_space) | |
2928 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
2929 | } |
2930 | ||
991d9fa0 JT |
2931 | /* |
2932 | * Status line is: | |
2933 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2934 | * <used data sectors>/<total data sectors> <held metadata root> | |
2935 | */ | |
fd7c092e MP |
2936 | static void pool_status(struct dm_target *ti, status_type_t type, |
2937 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2938 | { |
e49e5829 | 2939 | int r; |
991d9fa0 JT |
2940 | unsigned sz = 0; |
2941 | uint64_t transaction_id; | |
2942 | dm_block_t nr_free_blocks_data; | |
2943 | dm_block_t nr_free_blocks_metadata; | |
2944 | dm_block_t nr_blocks_data; | |
2945 | dm_block_t nr_blocks_metadata; | |
2946 | dm_block_t held_root; | |
2947 | char buf[BDEVNAME_SIZE]; | |
2948 | char buf2[BDEVNAME_SIZE]; | |
2949 | struct pool_c *pt = ti->private; | |
2950 | struct pool *pool = pt->pool; | |
2951 | ||
2952 | switch (type) { | |
2953 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2954 | if (get_pool_mode(pool) == PM_FAIL) { |
2955 | DMEMIT("Fail"); | |
2956 | break; | |
2957 | } | |
2958 | ||
1f4e0ff0 AK |
2959 | /* Commit to ensure statistics aren't out-of-date */ |
2960 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2961 | (void) commit(pool); |
1f4e0ff0 | 2962 | |
fd7c092e MP |
2963 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2964 | if (r) { | |
4fa5971a MS |
2965 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2966 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2967 | goto err; |
2968 | } | |
991d9fa0 | 2969 | |
fd7c092e MP |
2970 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2971 | if (r) { | |
4fa5971a MS |
2972 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2973 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2974 | goto err; |
2975 | } | |
991d9fa0 JT |
2976 | |
2977 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2978 | if (r) { |
4fa5971a MS |
2979 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2980 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2981 | goto err; |
2982 | } | |
991d9fa0 | 2983 | |
fd7c092e MP |
2984 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2985 | if (r) { | |
4fa5971a MS |
2986 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2987 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2988 | goto err; |
2989 | } | |
991d9fa0 JT |
2990 | |
2991 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2992 | if (r) { |
4fa5971a MS |
2993 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2994 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2995 | goto err; |
2996 | } | |
991d9fa0 | 2997 | |
cc8394d8 | 2998 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2999 | if (r) { |
4fa5971a MS |
3000 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3001 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3002 | goto err; |
3003 | } | |
991d9fa0 JT |
3004 | |
3005 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3006 | (unsigned long long)transaction_id, | |
3007 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3008 | (unsigned long long)nr_blocks_metadata, | |
3009 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3010 | (unsigned long long)nr_blocks_data); | |
3011 | ||
3012 | if (held_root) | |
e49e5829 JT |
3013 | DMEMIT("%llu ", held_root); |
3014 | else | |
3015 | DMEMIT("- "); | |
3016 | ||
3e1a0699 JT |
3017 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3018 | DMEMIT("out_of_data_space "); | |
3019 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3020 | DMEMIT("ro "); |
991d9fa0 | 3021 | else |
e49e5829 JT |
3022 | DMEMIT("rw "); |
3023 | ||
018debea | 3024 | if (!pool->pf.discard_enabled) |
787a996c | 3025 | DMEMIT("ignore_discard "); |
018debea | 3026 | else if (pool->pf.discard_passdown) |
787a996c MS |
3027 | DMEMIT("discard_passdown "); |
3028 | else | |
3029 | DMEMIT("no_discard_passdown "); | |
3030 | ||
3031 | if (pool->pf.error_if_no_space) | |
3032 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3033 | else |
787a996c | 3034 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3035 | |
3036 | break; | |
3037 | ||
3038 | case STATUSTYPE_TABLE: | |
3039 | DMEMIT("%s %s %lu %llu ", | |
3040 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3041 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3042 | (unsigned long)pool->sectors_per_block, | |
3043 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3044 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3045 | break; |
3046 | } | |
fd7c092e | 3047 | return; |
991d9fa0 | 3048 | |
fd7c092e MP |
3049 | err: |
3050 | DMEMIT("Error"); | |
991d9fa0 JT |
3051 | } |
3052 | ||
3053 | static int pool_iterate_devices(struct dm_target *ti, | |
3054 | iterate_devices_callout_fn fn, void *data) | |
3055 | { | |
3056 | struct pool_c *pt = ti->private; | |
3057 | ||
3058 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3059 | } | |
3060 | ||
3061 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3062 | struct bio_vec *biovec, int max_size) | |
3063 | { | |
3064 | struct pool_c *pt = ti->private; | |
3065 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3066 | ||
3067 | if (!q->merge_bvec_fn) | |
3068 | return max_size; | |
3069 | ||
3070 | bvm->bi_bdev = pt->data_dev->bdev; | |
3071 | ||
3072 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3073 | } | |
3074 | ||
0424caa1 | 3075 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 3076 | { |
0424caa1 MS |
3077 | struct pool *pool = pt->pool; |
3078 | struct queue_limits *data_limits; | |
3079 | ||
104655fd JT |
3080 | limits->max_discard_sectors = pool->sectors_per_block; |
3081 | ||
3082 | /* | |
0424caa1 | 3083 | * discard_granularity is just a hint, and not enforced. |
104655fd | 3084 | */ |
0424caa1 MS |
3085 | if (pt->adjusted_pf.discard_passdown) { |
3086 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
3087 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 3088 | } else |
0424caa1 | 3089 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3090 | } |
3091 | ||
991d9fa0 JT |
3092 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3093 | { | |
3094 | struct pool_c *pt = ti->private; | |
3095 | struct pool *pool = pt->pool; | |
0cc67cd9 | 3096 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 3097 | |
0cc67cd9 MS |
3098 | /* |
3099 | * If the system-determined stacked limits are compatible with the | |
3100 | * pool's blocksize (io_opt is a factor) do not override them. | |
3101 | */ | |
3102 | if (io_opt_sectors < pool->sectors_per_block || | |
3103 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
3104 | blk_limits_io_min(limits, 0); | |
3105 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
3106 | } | |
0424caa1 MS |
3107 | |
3108 | /* | |
3109 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3110 | * They get transferred to the live pool in bind_control_target() | |
3111 | * called from pool_preresume(). | |
3112 | */ | |
b60ab990 MS |
3113 | if (!pt->adjusted_pf.discard_enabled) { |
3114 | /* | |
3115 | * Must explicitly disallow stacking discard limits otherwise the | |
3116 | * block layer will stack them if pool's data device has support. | |
3117 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3118 | * user to see that, so make sure to set all discard limits to 0. | |
3119 | */ | |
3120 | limits->discard_granularity = 0; | |
0424caa1 | 3121 | return; |
b60ab990 | 3122 | } |
0424caa1 MS |
3123 | |
3124 | disable_passdown_if_not_supported(pt); | |
3125 | ||
3126 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3127 | } |
3128 | ||
3129 | static struct target_type pool_target = { | |
3130 | .name = "thin-pool", | |
3131 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3132 | DM_TARGET_IMMUTABLE, | |
67324ea1 | 3133 | .version = {1, 12, 0}, |
991d9fa0 JT |
3134 | .module = THIS_MODULE, |
3135 | .ctr = pool_ctr, | |
3136 | .dtr = pool_dtr, | |
3137 | .map = pool_map, | |
3138 | .postsuspend = pool_postsuspend, | |
3139 | .preresume = pool_preresume, | |
3140 | .resume = pool_resume, | |
3141 | .message = pool_message, | |
3142 | .status = pool_status, | |
3143 | .merge = pool_merge, | |
3144 | .iterate_devices = pool_iterate_devices, | |
3145 | .io_hints = pool_io_hints, | |
3146 | }; | |
3147 | ||
3148 | /*---------------------------------------------------------------- | |
3149 | * Thin target methods | |
3150 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3151 | static void thin_get(struct thin_c *tc) |
3152 | { | |
3153 | atomic_inc(&tc->refcount); | |
3154 | } | |
3155 | ||
3156 | static void thin_put(struct thin_c *tc) | |
3157 | { | |
3158 | if (atomic_dec_and_test(&tc->refcount)) | |
3159 | complete(&tc->can_destroy); | |
3160 | } | |
3161 | ||
991d9fa0 JT |
3162 | static void thin_dtr(struct dm_target *ti) |
3163 | { | |
3164 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3165 | unsigned long flags; |
3166 | ||
b10ebd34 JT |
3167 | thin_put(tc); |
3168 | wait_for_completion(&tc->can_destroy); | |
3169 | ||
c140e1c4 MS |
3170 | spin_lock_irqsave(&tc->pool->lock, flags); |
3171 | list_del_rcu(&tc->list); | |
3172 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3173 | synchronize_rcu(); | |
991d9fa0 JT |
3174 | |
3175 | mutex_lock(&dm_thin_pool_table.mutex); | |
3176 | ||
3177 | __pool_dec(tc->pool); | |
3178 | dm_pool_close_thin_device(tc->td); | |
3179 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3180 | if (tc->origin_dev) |
3181 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3182 | kfree(tc); |
3183 | ||
3184 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3185 | } | |
3186 | ||
3187 | /* | |
3188 | * Thin target parameters: | |
3189 | * | |
2dd9c257 | 3190 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3191 | * |
3192 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3193 | * dev_id: the internal device identifier | |
2dd9c257 | 3194 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3195 | * |
3196 | * If the pool device has discards disabled, they get disabled for the thin | |
3197 | * device as well. | |
991d9fa0 JT |
3198 | */ |
3199 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3200 | { | |
3201 | int r; | |
3202 | struct thin_c *tc; | |
2dd9c257 | 3203 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3204 | struct mapped_device *pool_md; |
5e3283e2 | 3205 | unsigned long flags; |
991d9fa0 JT |
3206 | |
3207 | mutex_lock(&dm_thin_pool_table.mutex); | |
3208 | ||
2dd9c257 | 3209 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3210 | ti->error = "Invalid argument count"; |
3211 | r = -EINVAL; | |
3212 | goto out_unlock; | |
3213 | } | |
3214 | ||
3215 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3216 | if (!tc) { | |
3217 | ti->error = "Out of memory"; | |
3218 | r = -ENOMEM; | |
3219 | goto out_unlock; | |
3220 | } | |
c140e1c4 MS |
3221 | spin_lock_init(&tc->lock); |
3222 | bio_list_init(&tc->deferred_bio_list); | |
3223 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3224 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3225 | |
2dd9c257 JT |
3226 | if (argc == 3) { |
3227 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3228 | if (r) { | |
3229 | ti->error = "Error opening origin device"; | |
3230 | goto bad_origin_dev; | |
3231 | } | |
3232 | tc->origin_dev = origin_dev; | |
3233 | } | |
3234 | ||
991d9fa0 JT |
3235 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3236 | if (r) { | |
3237 | ti->error = "Error opening pool device"; | |
3238 | goto bad_pool_dev; | |
3239 | } | |
3240 | tc->pool_dev = pool_dev; | |
3241 | ||
3242 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3243 | ti->error = "Invalid device id"; | |
3244 | r = -EINVAL; | |
3245 | goto bad_common; | |
3246 | } | |
3247 | ||
3248 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3249 | if (!pool_md) { | |
3250 | ti->error = "Couldn't get pool mapped device"; | |
3251 | r = -EINVAL; | |
3252 | goto bad_common; | |
3253 | } | |
3254 | ||
3255 | tc->pool = __pool_table_lookup(pool_md); | |
3256 | if (!tc->pool) { | |
3257 | ti->error = "Couldn't find pool object"; | |
3258 | r = -EINVAL; | |
3259 | goto bad_pool_lookup; | |
3260 | } | |
3261 | __pool_inc(tc->pool); | |
3262 | ||
e49e5829 JT |
3263 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3264 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3265 | r = -EINVAL; |
e49e5829 JT |
3266 | goto bad_thin_open; |
3267 | } | |
3268 | ||
991d9fa0 JT |
3269 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3270 | if (r) { | |
3271 | ti->error = "Couldn't open thin internal device"; | |
3272 | goto bad_thin_open; | |
3273 | } | |
3274 | ||
542f9038 MS |
3275 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3276 | if (r) | |
1acacc07 | 3277 | goto bad_target_max_io_len; |
542f9038 | 3278 | |
55a62eef | 3279 | ti->num_flush_bios = 1; |
16ad3d10 | 3280 | ti->flush_supported = true; |
59c3d2c6 | 3281 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3282 | |
3283 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3284 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3285 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3286 | ti->discards_supported = true; |
55a62eef | 3287 | ti->num_discard_bios = 1; |
55a62eef AK |
3288 | /* Discard bios must be split on a block boundary */ |
3289 | ti->split_discard_bios = true; | |
67e2e2b2 | 3290 | } |
991d9fa0 JT |
3291 | |
3292 | dm_put(pool_md); | |
3293 | ||
3294 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3295 | ||
b10ebd34 JT |
3296 | atomic_set(&tc->refcount, 1); |
3297 | init_completion(&tc->can_destroy); | |
3298 | ||
5e3283e2 | 3299 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3300 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3301 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3302 | /* |
3303 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3304 | * wake_worker() call finding no bios to process (because the newly | |
3305 | * added tc isn't yet visible). So this reduces latency since we | |
3306 | * aren't then dependent on the periodic commit to wake_worker(). | |
3307 | */ | |
3308 | synchronize_rcu(); | |
3309 | ||
991d9fa0 JT |
3310 | return 0; |
3311 | ||
1acacc07 MS |
3312 | bad_target_max_io_len: |
3313 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3314 | bad_thin_open: |
3315 | __pool_dec(tc->pool); | |
3316 | bad_pool_lookup: | |
3317 | dm_put(pool_md); | |
3318 | bad_common: | |
3319 | dm_put_device(ti, tc->pool_dev); | |
3320 | bad_pool_dev: | |
2dd9c257 JT |
3321 | if (tc->origin_dev) |
3322 | dm_put_device(ti, tc->origin_dev); | |
3323 | bad_origin_dev: | |
991d9fa0 JT |
3324 | kfree(tc); |
3325 | out_unlock: | |
3326 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3327 | ||
3328 | return r; | |
3329 | } | |
3330 | ||
7de3ee57 | 3331 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3332 | { |
4f024f37 | 3333 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3334 | |
7de3ee57 | 3335 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3336 | } |
3337 | ||
7de3ee57 | 3338 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3339 | { |
3340 | unsigned long flags; | |
59c3d2c6 | 3341 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3342 | struct list_head work; |
a24c2569 | 3343 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3344 | struct pool *pool = h->tc->pool; |
3345 | ||
3346 | if (h->shared_read_entry) { | |
3347 | INIT_LIST_HEAD(&work); | |
44feb387 | 3348 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3349 | |
3350 | spin_lock_irqsave(&pool->lock, flags); | |
3351 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3352 | list_del(&m->list); | |
7f214665 | 3353 | m->quiesced = true; |
eb2aa48d JT |
3354 | __maybe_add_mapping(m); |
3355 | } | |
3356 | spin_unlock_irqrestore(&pool->lock, flags); | |
3357 | } | |
3358 | ||
104655fd JT |
3359 | if (h->all_io_entry) { |
3360 | INIT_LIST_HEAD(&work); | |
44feb387 | 3361 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3362 | if (!list_empty(&work)) { |
3363 | spin_lock_irqsave(&pool->lock, flags); | |
3364 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3365 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3366 | spin_unlock_irqrestore(&pool->lock, flags); |
3367 | wake_worker(pool); | |
3368 | } | |
104655fd JT |
3369 | } |
3370 | ||
eb2aa48d JT |
3371 | return 0; |
3372 | } | |
3373 | ||
738211f7 | 3374 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3375 | { |
738211f7 JT |
3376 | struct thin_c *tc = ti->private; |
3377 | ||
991d9fa0 | 3378 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3379 | noflush_work(tc, do_noflush_start); |
3380 | } | |
3381 | ||
3382 | static void thin_postsuspend(struct dm_target *ti) | |
3383 | { | |
3384 | struct thin_c *tc = ti->private; | |
3385 | ||
3386 | /* | |
3387 | * The dm_noflush_suspending flag has been cleared by now, so | |
3388 | * unfortunately we must always run this. | |
3389 | */ | |
3390 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3391 | } |
3392 | ||
3393 | /* | |
3394 | * <nr mapped sectors> <highest mapped sector> | |
3395 | */ | |
fd7c092e MP |
3396 | static void thin_status(struct dm_target *ti, status_type_t type, |
3397 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3398 | { |
3399 | int r; | |
3400 | ssize_t sz = 0; | |
3401 | dm_block_t mapped, highest; | |
3402 | char buf[BDEVNAME_SIZE]; | |
3403 | struct thin_c *tc = ti->private; | |
3404 | ||
e49e5829 JT |
3405 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3406 | DMEMIT("Fail"); | |
fd7c092e | 3407 | return; |
e49e5829 JT |
3408 | } |
3409 | ||
991d9fa0 JT |
3410 | if (!tc->td) |
3411 | DMEMIT("-"); | |
3412 | else { | |
3413 | switch (type) { | |
3414 | case STATUSTYPE_INFO: | |
3415 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3416 | if (r) { |
3417 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3418 | goto err; | |
3419 | } | |
991d9fa0 JT |
3420 | |
3421 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3422 | if (r < 0) { |
3423 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3424 | goto err; | |
3425 | } | |
991d9fa0 JT |
3426 | |
3427 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3428 | if (r) | |
3429 | DMEMIT("%llu", ((highest + 1) * | |
3430 | tc->pool->sectors_per_block) - 1); | |
3431 | else | |
3432 | DMEMIT("-"); | |
3433 | break; | |
3434 | ||
3435 | case STATUSTYPE_TABLE: | |
3436 | DMEMIT("%s %lu", | |
3437 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3438 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3439 | if (tc->origin_dev) |
3440 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3441 | break; |
3442 | } | |
3443 | } | |
3444 | ||
fd7c092e MP |
3445 | return; |
3446 | ||
3447 | err: | |
3448 | DMEMIT("Error"); | |
991d9fa0 JT |
3449 | } |
3450 | ||
3451 | static int thin_iterate_devices(struct dm_target *ti, | |
3452 | iterate_devices_callout_fn fn, void *data) | |
3453 | { | |
55f2b8bd | 3454 | sector_t blocks; |
991d9fa0 | 3455 | struct thin_c *tc = ti->private; |
55f2b8bd | 3456 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3457 | |
3458 | /* | |
3459 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3460 | * we follow a more convoluted path through to the pool's target. | |
3461 | */ | |
55f2b8bd | 3462 | if (!pool->ti) |
991d9fa0 JT |
3463 | return 0; /* nothing is bound */ |
3464 | ||
55f2b8bd MS |
3465 | blocks = pool->ti->len; |
3466 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3467 | if (blocks) |
55f2b8bd | 3468 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3469 | |
3470 | return 0; | |
3471 | } | |
3472 | ||
991d9fa0 JT |
3473 | static struct target_type thin_target = { |
3474 | .name = "thin", | |
67324ea1 | 3475 | .version = {1, 12, 0}, |
991d9fa0 JT |
3476 | .module = THIS_MODULE, |
3477 | .ctr = thin_ctr, | |
3478 | .dtr = thin_dtr, | |
3479 | .map = thin_map, | |
eb2aa48d | 3480 | .end_io = thin_endio, |
738211f7 | 3481 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3482 | .postsuspend = thin_postsuspend, |
3483 | .status = thin_status, | |
3484 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3485 | }; |
3486 | ||
3487 | /*----------------------------------------------------------------*/ | |
3488 | ||
3489 | static int __init dm_thin_init(void) | |
3490 | { | |
3491 | int r; | |
3492 | ||
3493 | pool_table_init(); | |
3494 | ||
3495 | r = dm_register_target(&thin_target); | |
3496 | if (r) | |
3497 | return r; | |
3498 | ||
3499 | r = dm_register_target(&pool_target); | |
3500 | if (r) | |
a24c2569 MS |
3501 | goto bad_pool_target; |
3502 | ||
3503 | r = -ENOMEM; | |
3504 | ||
a24c2569 MS |
3505 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3506 | if (!_new_mapping_cache) | |
3507 | goto bad_new_mapping_cache; | |
3508 | ||
a24c2569 MS |
3509 | return 0; |
3510 | ||
a24c2569 | 3511 | bad_new_mapping_cache: |
a24c2569 MS |
3512 | dm_unregister_target(&pool_target); |
3513 | bad_pool_target: | |
3514 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3515 | |
3516 | return r; | |
3517 | } | |
3518 | ||
3519 | static void dm_thin_exit(void) | |
3520 | { | |
3521 | dm_unregister_target(&thin_target); | |
3522 | dm_unregister_target(&pool_target); | |
a24c2569 | 3523 | |
a24c2569 | 3524 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3525 | } |
3526 | ||
3527 | module_init(dm_thin_init); | |
3528 | module_exit(dm_thin_exit); | |
3529 | ||
80c57893 MS |
3530 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
3531 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
3532 | ||
7cab8bf1 | 3533 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3534 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3535 | MODULE_LICENSE("GPL"); |