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" | |
742c8fdc | 8 | #include "dm-bio-prison-v1.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> | |
0f30af98 | 14 | #include <linux/jiffies.h> |
604ea906 | 15 | #include <linux/log2.h> |
991d9fa0 | 16 | #include <linux/list.h> |
c140e1c4 | 17 | #include <linux/rculist.h> |
991d9fa0 JT |
18 | #include <linux/init.h> |
19 | #include <linux/module.h> | |
20 | #include <linux/slab.h> | |
a822c83e | 21 | #include <linux/vmalloc.h> |
ac4c3f34 | 22 | #include <linux/sort.h> |
67324ea1 | 23 | #include <linux/rbtree.h> |
991d9fa0 JT |
24 | |
25 | #define DM_MSG_PREFIX "thin" | |
26 | ||
27 | /* | |
28 | * Tunable constants | |
29 | */ | |
7768ed33 | 30 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 31 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 32 | #define COMMIT_PERIOD HZ |
80c57893 MS |
33 | #define NO_SPACE_TIMEOUT_SECS 60 |
34 | ||
35 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 36 | |
df5d2e90 MP |
37 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
38 | "A percentage of time allocated for copy on write"); | |
39 | ||
991d9fa0 JT |
40 | /* |
41 | * The block size of the device holding pool data must be | |
42 | * between 64KB and 1GB. | |
43 | */ | |
44 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
45 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
46 | ||
991d9fa0 JT |
47 | /* |
48 | * Device id is restricted to 24 bits. | |
49 | */ | |
50 | #define MAX_DEV_ID ((1 << 24) - 1) | |
51 | ||
52 | /* | |
53 | * How do we handle breaking sharing of data blocks? | |
54 | * ================================================= | |
55 | * | |
56 | * We use a standard copy-on-write btree to store the mappings for the | |
57 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
58 | * the data). When you take an internal snapshot you clone the root node | |
59 | * of the origin btree. After this there is no concept of an origin or a | |
60 | * snapshot. They are just two device trees that happen to point to the | |
61 | * same data blocks. | |
62 | * | |
63 | * When we get a write in we decide if it's to a shared data block using | |
64 | * some timestamp magic. If it is, we have to break sharing. | |
65 | * | |
66 | * Let's say we write to a shared block in what was the origin. The | |
67 | * steps are: | |
68 | * | |
69 | * i) plug io further to this physical block. (see bio_prison code). | |
70 | * | |
71 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 72 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
73 | * |
74 | * iii) copy the data block to a newly allocate block. This step can be | |
75 | * missed out if the io covers the block. (schedule_copy). | |
76 | * | |
77 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 78 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
79 | * sharing of btree nodes between the two devices. Breaking sharing only |
80 | * effects the btree of that specific device. Btrees for the other | |
81 | * devices that share the block never change. The btree for the origin | |
82 | * device as it was after the last commit is untouched, ie. we're using | |
83 | * persistent data structures in the functional programming sense. | |
84 | * | |
85 | * v) unplug io to this physical block, including the io that triggered | |
86 | * the breaking of sharing. | |
87 | * | |
88 | * Steps (ii) and (iii) occur in parallel. | |
89 | * | |
90 | * The metadata _doesn't_ need to be committed before the io continues. We | |
91 | * get away with this because the io is always written to a _new_ block. | |
92 | * If there's a crash, then: | |
93 | * | |
94 | * - The origin mapping will point to the old origin block (the shared | |
95 | * one). This will contain the data as it was before the io that triggered | |
96 | * the breaking of sharing came in. | |
97 | * | |
98 | * - The snap mapping still points to the old block. As it would after | |
99 | * the commit. | |
100 | * | |
101 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
102 | * will continue to think that data block in the snapshot device is shared | |
103 | * even after the write to the origin has broken sharing. I suspect data | |
104 | * blocks will typically be shared by many different devices, so we're | |
105 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
106 | * devices that reference this data block. At the moment I think the | |
107 | * benefits far, far outweigh the disadvantages. | |
108 | */ | |
109 | ||
110 | /*----------------------------------------------------------------*/ | |
111 | ||
991d9fa0 JT |
112 | /* |
113 | * Key building. | |
114 | */ | |
34fbcf62 JT |
115 | enum lock_space { |
116 | VIRTUAL, | |
117 | PHYSICAL | |
118 | }; | |
119 | ||
120 | static void build_key(struct dm_thin_device *td, enum lock_space ls, | |
121 | dm_block_t b, dm_block_t e, struct dm_cell_key *key) | |
991d9fa0 | 122 | { |
34fbcf62 | 123 | key->virtual = (ls == VIRTUAL); |
991d9fa0 | 124 | key->dev = dm_thin_dev_id(td); |
5f274d88 | 125 | key->block_begin = b; |
34fbcf62 JT |
126 | key->block_end = e; |
127 | } | |
128 | ||
129 | static void build_data_key(struct dm_thin_device *td, dm_block_t b, | |
130 | struct dm_cell_key *key) | |
131 | { | |
132 | build_key(td, PHYSICAL, b, b + 1llu, key); | |
991d9fa0 JT |
133 | } |
134 | ||
135 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 136 | struct dm_cell_key *key) |
991d9fa0 | 137 | { |
34fbcf62 | 138 | build_key(td, VIRTUAL, b, b + 1llu, key); |
991d9fa0 JT |
139 | } |
140 | ||
141 | /*----------------------------------------------------------------*/ | |
142 | ||
7d327fe0 JT |
143 | #define THROTTLE_THRESHOLD (1 * HZ) |
144 | ||
145 | struct throttle { | |
146 | struct rw_semaphore lock; | |
147 | unsigned long threshold; | |
148 | bool throttle_applied; | |
149 | }; | |
150 | ||
151 | static void throttle_init(struct throttle *t) | |
152 | { | |
153 | init_rwsem(&t->lock); | |
154 | t->throttle_applied = false; | |
155 | } | |
156 | ||
157 | static void throttle_work_start(struct throttle *t) | |
158 | { | |
159 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
160 | } | |
161 | ||
162 | static void throttle_work_update(struct throttle *t) | |
163 | { | |
164 | if (!t->throttle_applied && jiffies > t->threshold) { | |
165 | down_write(&t->lock); | |
166 | t->throttle_applied = true; | |
167 | } | |
168 | } | |
169 | ||
170 | static void throttle_work_complete(struct throttle *t) | |
171 | { | |
172 | if (t->throttle_applied) { | |
173 | t->throttle_applied = false; | |
174 | up_write(&t->lock); | |
175 | } | |
176 | } | |
177 | ||
178 | static void throttle_lock(struct throttle *t) | |
179 | { | |
180 | down_read(&t->lock); | |
181 | } | |
182 | ||
183 | static void throttle_unlock(struct throttle *t) | |
184 | { | |
185 | up_read(&t->lock); | |
186 | } | |
187 | ||
188 | /*----------------------------------------------------------------*/ | |
189 | ||
991d9fa0 JT |
190 | /* |
191 | * A pool device ties together a metadata device and a data device. It | |
192 | * also provides the interface for creating and destroying internal | |
193 | * devices. | |
194 | */ | |
a24c2569 | 195 | struct dm_thin_new_mapping; |
67e2e2b2 | 196 | |
e49e5829 | 197 | /* |
3e1a0699 | 198 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
199 | */ |
200 | enum pool_mode { | |
201 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 202 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
203 | PM_READ_ONLY, /* metadata may not be changed */ |
204 | PM_FAIL, /* all I/O fails */ | |
205 | }; | |
206 | ||
67e2e2b2 | 207 | struct pool_features { |
e49e5829 JT |
208 | enum pool_mode mode; |
209 | ||
9bc142dd MS |
210 | bool zero_new_blocks:1; |
211 | bool discard_enabled:1; | |
212 | bool discard_passdown:1; | |
787a996c | 213 | bool error_if_no_space:1; |
67e2e2b2 JT |
214 | }; |
215 | ||
e49e5829 JT |
216 | struct thin_c; |
217 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 218 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
219 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
220 | ||
ac4c3f34 JT |
221 | #define CELL_SORT_ARRAY_SIZE 8192 |
222 | ||
991d9fa0 JT |
223 | struct pool { |
224 | struct list_head list; | |
225 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
226 | ||
227 | struct mapped_device *pool_md; | |
228 | struct block_device *md_dev; | |
229 | struct dm_pool_metadata *pmd; | |
230 | ||
991d9fa0 | 231 | dm_block_t low_water_blocks; |
55f2b8bd | 232 | uint32_t sectors_per_block; |
f9a8e0cd | 233 | int sectors_per_block_shift; |
991d9fa0 | 234 | |
67e2e2b2 | 235 | struct pool_features pf; |
88a6621b | 236 | bool low_water_triggered:1; /* A dm event has been sent */ |
80e96c54 | 237 | bool suspended:1; |
c3667cc6 | 238 | bool out_of_data_space:1; |
991d9fa0 | 239 | |
44feb387 | 240 | struct dm_bio_prison *prison; |
991d9fa0 JT |
241 | struct dm_kcopyd_client *copier; |
242 | ||
72d711c8 | 243 | struct work_struct worker; |
991d9fa0 | 244 | struct workqueue_struct *wq; |
7d327fe0 | 245 | struct throttle throttle; |
905e51b3 | 246 | struct delayed_work waker; |
85ad643b | 247 | struct delayed_work no_space_timeout; |
991d9fa0 | 248 | |
905e51b3 | 249 | unsigned long last_commit_jiffies; |
55f2b8bd | 250 | unsigned ref_count; |
991d9fa0 JT |
251 | |
252 | spinlock_t lock; | |
991d9fa0 JT |
253 | struct bio_list deferred_flush_bios; |
254 | struct list_head prepared_mappings; | |
104655fd | 255 | struct list_head prepared_discards; |
2a0fbffb | 256 | struct list_head prepared_discards_pt2; |
c140e1c4 | 257 | struct list_head active_thins; |
991d9fa0 | 258 | |
44feb387 MS |
259 | struct dm_deferred_set *shared_read_ds; |
260 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 261 | |
a24c2569 | 262 | struct dm_thin_new_mapping *next_mapping; |
e49e5829 JT |
263 | |
264 | process_bio_fn process_bio; | |
265 | process_bio_fn process_discard; | |
266 | ||
a374bb21 JT |
267 | process_cell_fn process_cell; |
268 | process_cell_fn process_discard_cell; | |
269 | ||
e49e5829 JT |
270 | process_mapping_fn process_prepared_mapping; |
271 | process_mapping_fn process_prepared_discard; | |
2a0fbffb | 272 | process_mapping_fn process_prepared_discard_pt2; |
ac4c3f34 | 273 | |
a822c83e | 274 | struct dm_bio_prison_cell **cell_sort_array; |
72d711c8 MS |
275 | |
276 | mempool_t mapping_pool; | |
991d9fa0 JT |
277 | }; |
278 | ||
e49e5829 | 279 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 280 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 281 | |
991d9fa0 JT |
282 | /* |
283 | * Target context for a pool. | |
284 | */ | |
285 | struct pool_c { | |
286 | struct dm_target *ti; | |
287 | struct pool *pool; | |
288 | struct dm_dev *data_dev; | |
289 | struct dm_dev *metadata_dev; | |
290 | struct dm_target_callbacks callbacks; | |
291 | ||
292 | dm_block_t low_water_blocks; | |
0424caa1 MS |
293 | struct pool_features requested_pf; /* Features requested during table load */ |
294 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
295 | }; |
296 | ||
297 | /* | |
298 | * Target context for a thin. | |
299 | */ | |
300 | struct thin_c { | |
c140e1c4 | 301 | struct list_head list; |
991d9fa0 | 302 | struct dm_dev *pool_dev; |
2dd9c257 | 303 | struct dm_dev *origin_dev; |
e5aea7b4 | 304 | sector_t origin_size; |
991d9fa0 JT |
305 | dm_thin_id dev_id; |
306 | ||
307 | struct pool *pool; | |
308 | struct dm_thin_device *td; | |
583024d2 MS |
309 | struct mapped_device *thin_md; |
310 | ||
738211f7 | 311 | bool requeue_mode:1; |
c140e1c4 | 312 | spinlock_t lock; |
a374bb21 | 313 | struct list_head deferred_cells; |
c140e1c4 MS |
314 | struct bio_list deferred_bio_list; |
315 | struct bio_list retry_on_resume_list; | |
67324ea1 | 316 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
317 | |
318 | /* | |
319 | * Ensures the thin is not destroyed until the worker has finished | |
320 | * iterating the active_thins list. | |
321 | */ | |
322 | atomic_t refcount; | |
323 | struct completion can_destroy; | |
991d9fa0 JT |
324 | }; |
325 | ||
326 | /*----------------------------------------------------------------*/ | |
327 | ||
34fbcf62 JT |
328 | static bool block_size_is_power_of_two(struct pool *pool) |
329 | { | |
330 | return pool->sectors_per_block_shift >= 0; | |
331 | } | |
332 | ||
333 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) | |
334 | { | |
335 | return block_size_is_power_of_two(pool) ? | |
336 | (b << pool->sectors_per_block_shift) : | |
337 | (b * pool->sectors_per_block); | |
338 | } | |
339 | ||
202bae52 JT |
340 | /*----------------------------------------------------------------*/ |
341 | ||
342 | struct discard_op { | |
343 | struct thin_c *tc; | |
344 | struct blk_plug plug; | |
345 | struct bio *parent_bio; | |
346 | struct bio *bio; | |
347 | }; | |
348 | ||
349 | static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent) | |
350 | { | |
351 | BUG_ON(!parent); | |
352 | ||
353 | op->tc = tc; | |
354 | blk_start_plug(&op->plug); | |
355 | op->parent_bio = parent; | |
356 | op->bio = NULL; | |
357 | } | |
358 | ||
359 | static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e) | |
34fbcf62 | 360 | { |
202bae52 | 361 | struct thin_c *tc = op->tc; |
34fbcf62 JT |
362 | sector_t s = block_to_sectors(tc->pool, data_b); |
363 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); | |
3dba53a9 | 364 | |
202bae52 | 365 | return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, |
469e3216 | 366 | GFP_NOWAIT, 0, &op->bio); |
202bae52 JT |
367 | } |
368 | ||
369 | static void end_discard(struct discard_op *op, int r) | |
370 | { | |
371 | if (op->bio) { | |
372 | /* | |
373 | * Even if one of the calls to issue_discard failed, we | |
374 | * need to wait for the chain to complete. | |
375 | */ | |
376 | bio_chain(op->bio, op->parent_bio); | |
e6047149 | 377 | bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0); |
4e49ea4a | 378 | submit_bio(op->bio); |
3dba53a9 | 379 | } |
34fbcf62 | 380 | |
202bae52 JT |
381 | blk_finish_plug(&op->plug); |
382 | ||
383 | /* | |
384 | * Even if r is set, there could be sub discards in flight that we | |
385 | * need to wait for. | |
386 | */ | |
4e4cbee9 CH |
387 | if (r && !op->parent_bio->bi_status) |
388 | op->parent_bio->bi_status = errno_to_blk_status(r); | |
202bae52 | 389 | bio_endio(op->parent_bio); |
34fbcf62 JT |
390 | } |
391 | ||
392 | /*----------------------------------------------------------------*/ | |
393 | ||
025b9685 JT |
394 | /* |
395 | * wake_worker() is used when new work is queued and when pool_resume is | |
396 | * ready to continue deferred IO processing. | |
397 | */ | |
398 | static void wake_worker(struct pool *pool) | |
399 | { | |
400 | queue_work(pool->wq, &pool->worker); | |
401 | } | |
402 | ||
403 | /*----------------------------------------------------------------*/ | |
404 | ||
6beca5eb JT |
405 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
406 | struct dm_bio_prison_cell **cell_result) | |
407 | { | |
408 | int r; | |
409 | struct dm_bio_prison_cell *cell_prealloc; | |
410 | ||
411 | /* | |
412 | * Allocate a cell from the prison's mempool. | |
413 | * This might block but it can't fail. | |
414 | */ | |
415 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
416 | ||
417 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
418 | if (r) | |
419 | /* | |
420 | * We reused an old cell; we can get rid of | |
421 | * the new one. | |
422 | */ | |
423 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
424 | ||
425 | return r; | |
426 | } | |
427 | ||
428 | static void cell_release(struct pool *pool, | |
429 | struct dm_bio_prison_cell *cell, | |
430 | struct bio_list *bios) | |
431 | { | |
432 | dm_cell_release(pool->prison, cell, bios); | |
433 | dm_bio_prison_free_cell(pool->prison, cell); | |
434 | } | |
435 | ||
2d759a46 JT |
436 | static void cell_visit_release(struct pool *pool, |
437 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
438 | void *context, | |
439 | struct dm_bio_prison_cell *cell) | |
440 | { | |
441 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
442 | dm_bio_prison_free_cell(pool->prison, cell); | |
443 | } | |
444 | ||
6beca5eb JT |
445 | static void cell_release_no_holder(struct pool *pool, |
446 | struct dm_bio_prison_cell *cell, | |
447 | struct bio_list *bios) | |
448 | { | |
449 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
450 | dm_bio_prison_free_cell(pool->prison, cell); | |
451 | } | |
452 | ||
af91805a | 453 | static void cell_error_with_code(struct pool *pool, |
4e4cbee9 | 454 | struct dm_bio_prison_cell *cell, blk_status_t error_code) |
6beca5eb | 455 | { |
af91805a | 456 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
457 | dm_bio_prison_free_cell(pool->prison, cell); |
458 | } | |
459 | ||
4e4cbee9 | 460 | static blk_status_t get_pool_io_error_code(struct pool *pool) |
c3667cc6 | 461 | { |
4e4cbee9 | 462 | return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR; |
c3667cc6 MS |
463 | } |
464 | ||
af91805a MS |
465 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
466 | { | |
4e4cbee9 | 467 | cell_error_with_code(pool, cell, get_pool_io_error_code(pool)); |
af91805a MS |
468 | } |
469 | ||
a374bb21 JT |
470 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
471 | { | |
472 | cell_error_with_code(pool, cell, 0); | |
473 | } | |
474 | ||
475 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
476 | { | |
4e4cbee9 | 477 | cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE); |
a374bb21 JT |
478 | } |
479 | ||
6beca5eb JT |
480 | /*----------------------------------------------------------------*/ |
481 | ||
991d9fa0 JT |
482 | /* |
483 | * A global list of pools that uses a struct mapped_device as a key. | |
484 | */ | |
485 | static struct dm_thin_pool_table { | |
486 | struct mutex mutex; | |
487 | struct list_head pools; | |
488 | } dm_thin_pool_table; | |
489 | ||
490 | static void pool_table_init(void) | |
491 | { | |
492 | mutex_init(&dm_thin_pool_table.mutex); | |
493 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
494 | } | |
495 | ||
d5ffebdd MS |
496 | static void pool_table_exit(void) |
497 | { | |
498 | mutex_destroy(&dm_thin_pool_table.mutex); | |
499 | } | |
500 | ||
991d9fa0 JT |
501 | static void __pool_table_insert(struct pool *pool) |
502 | { | |
503 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
504 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
505 | } | |
506 | ||
507 | static void __pool_table_remove(struct pool *pool) | |
508 | { | |
509 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
510 | list_del(&pool->list); | |
511 | } | |
512 | ||
513 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
514 | { | |
515 | struct pool *pool = NULL, *tmp; | |
516 | ||
517 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
518 | ||
519 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
520 | if (tmp->pool_md == md) { | |
521 | pool = tmp; | |
522 | break; | |
523 | } | |
524 | } | |
525 | ||
526 | return pool; | |
527 | } | |
528 | ||
529 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
530 | { | |
531 | struct pool *pool = NULL, *tmp; | |
532 | ||
533 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
534 | ||
535 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
536 | if (tmp->md_dev == md_dev) { | |
537 | pool = tmp; | |
538 | break; | |
539 | } | |
540 | } | |
541 | ||
542 | return pool; | |
543 | } | |
544 | ||
545 | /*----------------------------------------------------------------*/ | |
546 | ||
a24c2569 | 547 | struct dm_thin_endio_hook { |
eb2aa48d | 548 | struct thin_c *tc; |
44feb387 MS |
549 | struct dm_deferred_entry *shared_read_entry; |
550 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 551 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 552 | struct rb_node rb_node; |
34fbcf62 | 553 | struct dm_bio_prison_cell *cell; |
eb2aa48d JT |
554 | }; |
555 | ||
42d6a8ce MS |
556 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
557 | { | |
558 | bio_list_merge(bios, master); | |
559 | bio_list_init(master); | |
560 | } | |
561 | ||
4e4cbee9 | 562 | static void error_bio_list(struct bio_list *bios, blk_status_t error) |
991d9fa0 JT |
563 | { |
564 | struct bio *bio; | |
42d6a8ce | 565 | |
4246a0b6 | 566 | while ((bio = bio_list_pop(bios))) { |
4e4cbee9 | 567 | bio->bi_status = error; |
4246a0b6 CH |
568 | bio_endio(bio); |
569 | } | |
42d6a8ce MS |
570 | } |
571 | ||
4e4cbee9 CH |
572 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, |
573 | blk_status_t error) | |
42d6a8ce | 574 | { |
991d9fa0 | 575 | struct bio_list bios; |
18adc577 | 576 | unsigned long flags; |
991d9fa0 JT |
577 | |
578 | bio_list_init(&bios); | |
18adc577 | 579 | |
c140e1c4 | 580 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce | 581 | __merge_bio_list(&bios, master); |
c140e1c4 | 582 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 583 | |
42d6a8ce | 584 | error_bio_list(&bios, error); |
991d9fa0 JT |
585 | } |
586 | ||
a374bb21 JT |
587 | static void requeue_deferred_cells(struct thin_c *tc) |
588 | { | |
589 | struct pool *pool = tc->pool; | |
590 | unsigned long flags; | |
591 | struct list_head cells; | |
592 | struct dm_bio_prison_cell *cell, *tmp; | |
593 | ||
594 | INIT_LIST_HEAD(&cells); | |
595 | ||
596 | spin_lock_irqsave(&tc->lock, flags); | |
597 | list_splice_init(&tc->deferred_cells, &cells); | |
598 | spin_unlock_irqrestore(&tc->lock, flags); | |
599 | ||
600 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
601 | cell_requeue(pool, cell); | |
602 | } | |
603 | ||
991d9fa0 JT |
604 | static void requeue_io(struct thin_c *tc) |
605 | { | |
3e1a0699 | 606 | struct bio_list bios; |
42d6a8ce | 607 | unsigned long flags; |
3e1a0699 JT |
608 | |
609 | bio_list_init(&bios); | |
610 | ||
c140e1c4 | 611 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce MS |
612 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
613 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
c140e1c4 | 614 | spin_unlock_irqrestore(&tc->lock, flags); |
3e1a0699 | 615 | |
4e4cbee9 | 616 | error_bio_list(&bios, BLK_STS_DM_REQUEUE); |
42d6a8ce | 617 | requeue_deferred_cells(tc); |
3e1a0699 JT |
618 | } |
619 | ||
4e4cbee9 | 620 | static void error_retry_list_with_code(struct pool *pool, blk_status_t error) |
c140e1c4 MS |
621 | { |
622 | struct thin_c *tc; | |
623 | ||
624 | rcu_read_lock(); | |
625 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
0a927c2f | 626 | error_thin_bio_list(tc, &tc->retry_on_resume_list, error); |
c140e1c4 MS |
627 | rcu_read_unlock(); |
628 | } | |
629 | ||
0a927c2f MS |
630 | static void error_retry_list(struct pool *pool) |
631 | { | |
4e4cbee9 | 632 | error_retry_list_with_code(pool, get_pool_io_error_code(pool)); |
0a927c2f MS |
633 | } |
634 | ||
991d9fa0 JT |
635 | /* |
636 | * This section of code contains the logic for processing a thin device's IO. | |
637 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
638 | * but most is exclusively called from the thin target rather than the thin-pool | |
639 | * target. | |
640 | */ | |
641 | ||
642 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
643 | { | |
58f77a21 | 644 | struct pool *pool = tc->pool; |
4f024f37 | 645 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 646 | |
58f77a21 MS |
647 | if (block_size_is_power_of_two(pool)) |
648 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 649 | else |
58f77a21 | 650 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
651 | |
652 | return block_nr; | |
991d9fa0 JT |
653 | } |
654 | ||
34fbcf62 JT |
655 | /* |
656 | * Returns the _complete_ blocks that this bio covers. | |
657 | */ | |
658 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, | |
659 | dm_block_t *begin, dm_block_t *end) | |
660 | { | |
661 | struct pool *pool = tc->pool; | |
662 | sector_t b = bio->bi_iter.bi_sector; | |
663 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
664 | ||
665 | b += pool->sectors_per_block - 1ull; /* so we round up */ | |
666 | ||
667 | if (block_size_is_power_of_two(pool)) { | |
668 | b >>= pool->sectors_per_block_shift; | |
669 | e >>= pool->sectors_per_block_shift; | |
670 | } else { | |
671 | (void) sector_div(b, pool->sectors_per_block); | |
672 | (void) sector_div(e, pool->sectors_per_block); | |
673 | } | |
674 | ||
675 | if (e < b) | |
676 | /* Can happen if the bio is within a single block. */ | |
677 | e = b; | |
678 | ||
679 | *begin = b; | |
680 | *end = e; | |
681 | } | |
682 | ||
991d9fa0 JT |
683 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
684 | { | |
685 | struct pool *pool = tc->pool; | |
4f024f37 | 686 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 | 687 | |
74d46992 | 688 | bio_set_dev(bio, tc->pool_dev->bdev); |
58f77a21 | 689 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
690 | bio->bi_iter.bi_sector = |
691 | (block << pool->sectors_per_block_shift) | | |
692 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 693 | else |
4f024f37 | 694 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 695 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
696 | } |
697 | ||
2dd9c257 JT |
698 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
699 | { | |
74d46992 | 700 | bio_set_dev(bio, tc->origin_dev->bdev); |
2dd9c257 JT |
701 | } |
702 | ||
4afdd680 JT |
703 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
704 | { | |
f73f44eb | 705 | return op_is_flush(bio->bi_opf) && |
4afdd680 JT |
706 | dm_thin_changed_this_transaction(tc->td); |
707 | } | |
708 | ||
e8088073 JT |
709 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
710 | { | |
711 | struct dm_thin_endio_hook *h; | |
712 | ||
e6047149 | 713 | if (bio_op(bio) == REQ_OP_DISCARD) |
e8088073 JT |
714 | return; |
715 | ||
59c3d2c6 | 716 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
717 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
718 | } | |
719 | ||
2dd9c257 | 720 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
721 | { |
722 | struct pool *pool = tc->pool; | |
723 | unsigned long flags; | |
724 | ||
e49e5829 JT |
725 | if (!bio_triggers_commit(tc, bio)) { |
726 | generic_make_request(bio); | |
727 | return; | |
728 | } | |
729 | ||
991d9fa0 | 730 | /* |
e49e5829 JT |
731 | * Complete bio with an error if earlier I/O caused changes to |
732 | * the metadata that can't be committed e.g, due to I/O errors | |
733 | * on the metadata device. | |
991d9fa0 | 734 | */ |
e49e5829 JT |
735 | if (dm_thin_aborted_changes(tc->td)) { |
736 | bio_io_error(bio); | |
737 | return; | |
738 | } | |
739 | ||
740 | /* | |
741 | * Batch together any bios that trigger commits and then issue a | |
742 | * single commit for them in process_deferred_bios(). | |
743 | */ | |
744 | spin_lock_irqsave(&pool->lock, flags); | |
745 | bio_list_add(&pool->deferred_flush_bios, bio); | |
746 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
747 | } |
748 | ||
2dd9c257 JT |
749 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
750 | { | |
751 | remap_to_origin(tc, bio); | |
752 | issue(tc, bio); | |
753 | } | |
754 | ||
755 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
756 | dm_block_t block) | |
757 | { | |
758 | remap(tc, bio, block); | |
759 | issue(tc, bio); | |
760 | } | |
761 | ||
991d9fa0 JT |
762 | /*----------------------------------------------------------------*/ |
763 | ||
764 | /* | |
765 | * Bio endio functions. | |
766 | */ | |
a24c2569 | 767 | struct dm_thin_new_mapping { |
991d9fa0 JT |
768 | struct list_head list; |
769 | ||
7f214665 | 770 | bool pass_discard:1; |
34fbcf62 | 771 | bool maybe_shared:1; |
991d9fa0 | 772 | |
50f3c3ef JT |
773 | /* |
774 | * Track quiescing, copying and zeroing preparation actions. When this | |
775 | * counter hits zero the block is prepared and can be inserted into the | |
776 | * btree. | |
777 | */ | |
778 | atomic_t prepare_actions; | |
779 | ||
4e4cbee9 | 780 | blk_status_t status; |
991d9fa0 | 781 | struct thin_c *tc; |
34fbcf62 | 782 | dm_block_t virt_begin, virt_end; |
991d9fa0 | 783 | dm_block_t data_block; |
34fbcf62 | 784 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
785 | |
786 | /* | |
787 | * If the bio covers the whole area of a block then we can avoid | |
788 | * zeroing or copying. Instead this bio is hooked. The bio will | |
789 | * still be in the cell, so care has to be taken to avoid issuing | |
790 | * the bio twice. | |
791 | */ | |
792 | struct bio *bio; | |
793 | bio_end_io_t *saved_bi_end_io; | |
794 | }; | |
795 | ||
50f3c3ef | 796 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
797 | { |
798 | struct pool *pool = m->tc->pool; | |
799 | ||
50f3c3ef | 800 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 801 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
802 | wake_worker(pool); |
803 | } | |
804 | } | |
805 | ||
e5aea7b4 | 806 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
807 | { |
808 | unsigned long flags; | |
991d9fa0 JT |
809 | struct pool *pool = m->tc->pool; |
810 | ||
991d9fa0 | 811 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 812 | __complete_mapping_preparation(m); |
991d9fa0 JT |
813 | spin_unlock_irqrestore(&pool->lock, flags); |
814 | } | |
815 | ||
e5aea7b4 JT |
816 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
817 | { | |
818 | struct dm_thin_new_mapping *m = context; | |
819 | ||
4e4cbee9 | 820 | m->status = read_err || write_err ? BLK_STS_IOERR : 0; |
e5aea7b4 JT |
821 | complete_mapping_preparation(m); |
822 | } | |
823 | ||
4246a0b6 | 824 | static void overwrite_endio(struct bio *bio) |
991d9fa0 | 825 | { |
59c3d2c6 | 826 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 827 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 | 828 | |
8b908f8e MS |
829 | bio->bi_end_io = m->saved_bi_end_io; |
830 | ||
4e4cbee9 | 831 | m->status = bio->bi_status; |
e5aea7b4 | 832 | complete_mapping_preparation(m); |
991d9fa0 JT |
833 | } |
834 | ||
991d9fa0 JT |
835 | /*----------------------------------------------------------------*/ |
836 | ||
837 | /* | |
838 | * Workqueue. | |
839 | */ | |
840 | ||
841 | /* | |
842 | * Prepared mapping jobs. | |
843 | */ | |
844 | ||
845 | /* | |
2d759a46 JT |
846 | * This sends the bios in the cell, except the original holder, back |
847 | * to the deferred_bios list. | |
991d9fa0 | 848 | */ |
f286ba0e | 849 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 850 | { |
991d9fa0 JT |
851 | struct pool *pool = tc->pool; |
852 | unsigned long flags; | |
853 | ||
c140e1c4 MS |
854 | spin_lock_irqsave(&tc->lock, flags); |
855 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
856 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
857 | |
858 | wake_worker(pool); | |
859 | } | |
860 | ||
a374bb21 JT |
861 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
862 | ||
2d759a46 JT |
863 | struct remap_info { |
864 | struct thin_c *tc; | |
865 | struct bio_list defer_bios; | |
866 | struct bio_list issue_bios; | |
867 | }; | |
868 | ||
869 | static void __inc_remap_and_issue_cell(void *context, | |
870 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 871 | { |
2d759a46 | 872 | struct remap_info *info = context; |
a374bb21 | 873 | struct bio *bio; |
a374bb21 | 874 | |
2d759a46 | 875 | while ((bio = bio_list_pop(&cell->bios))) { |
f73f44eb | 876 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) |
2d759a46 | 877 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 878 | else { |
2d759a46 JT |
879 | inc_all_io_entry(info->tc->pool, bio); |
880 | ||
881 | /* | |
882 | * We can't issue the bios with the bio prison lock | |
883 | * held, so we add them to a list to issue on | |
884 | * return from this function. | |
885 | */ | |
886 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
887 | } |
888 | } | |
889 | } | |
890 | ||
2d759a46 JT |
891 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
892 | struct dm_bio_prison_cell *cell, | |
893 | dm_block_t block) | |
894 | { | |
895 | struct bio *bio; | |
896 | struct remap_info info; | |
897 | ||
898 | info.tc = tc; | |
899 | bio_list_init(&info.defer_bios); | |
900 | bio_list_init(&info.issue_bios); | |
901 | ||
902 | /* | |
903 | * We have to be careful to inc any bios we're about to issue | |
904 | * before the cell is released, and avoid a race with new bios | |
905 | * being added to the cell. | |
906 | */ | |
907 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
908 | &info, cell); | |
909 | ||
910 | while ((bio = bio_list_pop(&info.defer_bios))) | |
911 | thin_defer_bio(tc, bio); | |
912 | ||
913 | while ((bio = bio_list_pop(&info.issue_bios))) | |
914 | remap_and_issue(info.tc, bio, block); | |
915 | } | |
916 | ||
e49e5829 JT |
917 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
918 | { | |
6beca5eb | 919 | cell_error(m->tc->pool, m->cell); |
e49e5829 | 920 | list_del(&m->list); |
6f1c819c | 921 | mempool_free(m, &m->tc->pool->mapping_pool); |
e49e5829 | 922 | } |
025b9685 | 923 | |
a24c2569 | 924 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
925 | { |
926 | struct thin_c *tc = m->tc; | |
6beca5eb | 927 | struct pool *pool = tc->pool; |
8b908f8e | 928 | struct bio *bio = m->bio; |
991d9fa0 JT |
929 | int r; |
930 | ||
4e4cbee9 | 931 | if (m->status) { |
6beca5eb | 932 | cell_error(pool, m->cell); |
905386f8 | 933 | goto out; |
991d9fa0 JT |
934 | } |
935 | ||
936 | /* | |
937 | * Commit the prepared block into the mapping btree. | |
938 | * Any I/O for this block arriving after this point will get | |
939 | * remapped to it directly. | |
940 | */ | |
34fbcf62 | 941 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
991d9fa0 | 942 | if (r) { |
b5330655 | 943 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 944 | cell_error(pool, m->cell); |
905386f8 | 945 | goto out; |
991d9fa0 JT |
946 | } |
947 | ||
948 | /* | |
949 | * Release any bios held while the block was being provisioned. | |
950 | * If we are processing a write bio that completely covers the block, | |
951 | * we already processed it so can ignore it now when processing | |
952 | * the bios in the cell. | |
953 | */ | |
954 | if (bio) { | |
2d759a46 | 955 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
4246a0b6 | 956 | bio_endio(bio); |
2d759a46 JT |
957 | } else { |
958 | inc_all_io_entry(tc->pool, m->cell->holder); | |
959 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
960 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
961 | } | |
991d9fa0 | 962 | |
905386f8 | 963 | out: |
991d9fa0 | 964 | list_del(&m->list); |
6f1c819c | 965 | mempool_free(m, &pool->mapping_pool); |
991d9fa0 JT |
966 | } |
967 | ||
34fbcf62 JT |
968 | /*----------------------------------------------------------------*/ |
969 | ||
970 | static void free_discard_mapping(struct dm_thin_new_mapping *m) | |
104655fd | 971 | { |
104655fd | 972 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
973 | if (m->cell) |
974 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 975 | mempool_free(m, &tc->pool->mapping_pool); |
34fbcf62 | 976 | } |
104655fd | 977 | |
34fbcf62 JT |
978 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
979 | { | |
e49e5829 | 980 | bio_io_error(m->bio); |
34fbcf62 JT |
981 | free_discard_mapping(m); |
982 | } | |
983 | ||
984 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) | |
985 | { | |
4246a0b6 | 986 | bio_endio(m->bio); |
34fbcf62 JT |
987 | free_discard_mapping(m); |
988 | } | |
989 | ||
990 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) | |
991 | { | |
992 | int r; | |
993 | struct thin_c *tc = m->tc; | |
994 | ||
995 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); | |
996 | if (r) { | |
997 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); | |
998 | bio_io_error(m->bio); | |
999 | } else | |
4246a0b6 | 1000 | bio_endio(m->bio); |
34fbcf62 | 1001 | |
f286ba0e | 1002 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1003 | mempool_free(m, &tc->pool->mapping_pool); |
e49e5829 JT |
1004 | } |
1005 | ||
202bae52 JT |
1006 | /*----------------------------------------------------------------*/ |
1007 | ||
2a0fbffb JT |
1008 | static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m, |
1009 | struct bio *discard_parent) | |
e49e5829 | 1010 | { |
34fbcf62 JT |
1011 | /* |
1012 | * We've already unmapped this range of blocks, but before we | |
1013 | * passdown we have to check that these blocks are now unused. | |
1014 | */ | |
202bae52 | 1015 | int r = 0; |
34fbcf62 | 1016 | bool used = true; |
e49e5829 | 1017 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1018 | struct pool *pool = tc->pool; |
1019 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; | |
202bae52 | 1020 | struct discard_op op; |
104655fd | 1021 | |
2a0fbffb | 1022 | begin_discard(&op, tc, discard_parent); |
34fbcf62 JT |
1023 | while (b != end) { |
1024 | /* find start of unmapped run */ | |
1025 | for (; b < end; b++) { | |
1026 | r = dm_pool_block_is_used(pool->pmd, b, &used); | |
1027 | if (r) | |
202bae52 | 1028 | goto out; |
e8088073 | 1029 | |
34fbcf62 JT |
1030 | if (!used) |
1031 | break; | |
19fa1a67 | 1032 | } |
104655fd | 1033 | |
34fbcf62 JT |
1034 | if (b == end) |
1035 | break; | |
1036 | ||
1037 | /* find end of run */ | |
1038 | for (e = b + 1; e != end; e++) { | |
1039 | r = dm_pool_block_is_used(pool->pmd, e, &used); | |
1040 | if (r) | |
202bae52 | 1041 | goto out; |
34fbcf62 JT |
1042 | |
1043 | if (used) | |
1044 | break; | |
1045 | } | |
1046 | ||
202bae52 | 1047 | r = issue_discard(&op, b, e); |
34fbcf62 | 1048 | if (r) |
202bae52 | 1049 | goto out; |
34fbcf62 JT |
1050 | |
1051 | b = e; | |
1052 | } | |
202bae52 JT |
1053 | out: |
1054 | end_discard(&op, r); | |
104655fd JT |
1055 | } |
1056 | ||
2a0fbffb JT |
1057 | static void queue_passdown_pt2(struct dm_thin_new_mapping *m) |
1058 | { | |
1059 | unsigned long flags; | |
1060 | struct pool *pool = m->tc->pool; | |
1061 | ||
1062 | spin_lock_irqsave(&pool->lock, flags); | |
1063 | list_add_tail(&m->list, &pool->prepared_discards_pt2); | |
1064 | spin_unlock_irqrestore(&pool->lock, flags); | |
1065 | wake_worker(pool); | |
1066 | } | |
1067 | ||
1068 | static void passdown_endio(struct bio *bio) | |
1069 | { | |
1070 | /* | |
1071 | * It doesn't matter if the passdown discard failed, we still want | |
1072 | * to unmap (we ignore err). | |
1073 | */ | |
1074 | queue_passdown_pt2(bio->bi_private); | |
948f581a | 1075 | bio_put(bio); |
2a0fbffb JT |
1076 | } |
1077 | ||
1078 | static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m) | |
e49e5829 JT |
1079 | { |
1080 | int r; | |
1081 | struct thin_c *tc = m->tc; | |
34fbcf62 | 1082 | struct pool *pool = tc->pool; |
2a0fbffb JT |
1083 | struct bio *discard_parent; |
1084 | dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin); | |
e49e5829 | 1085 | |
2a0fbffb JT |
1086 | /* |
1087 | * Only this thread allocates blocks, so we can be sure that the | |
1088 | * newly unmapped blocks will not be allocated before the end of | |
1089 | * the function. | |
1090 | */ | |
34fbcf62 | 1091 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
202bae52 | 1092 | if (r) { |
34fbcf62 | 1093 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
202bae52 | 1094 | bio_io_error(m->bio); |
2a0fbffb | 1095 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1096 | mempool_free(m, &pool->mapping_pool); |
2a0fbffb JT |
1097 | return; |
1098 | } | |
34fbcf62 | 1099 | |
00a0ea33 VV |
1100 | /* |
1101 | * Increment the unmapped blocks. This prevents a race between the | |
1102 | * passdown io and reallocation of freed blocks. | |
1103 | */ | |
1104 | r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end); | |
1105 | if (r) { | |
1106 | metadata_operation_failed(pool, "dm_pool_inc_data_range", r); | |
1107 | bio_io_error(m->bio); | |
1108 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1109 | mempool_free(m, &pool->mapping_pool); |
00a0ea33 VV |
1110 | return; |
1111 | } | |
1112 | ||
2a0fbffb JT |
1113 | discard_parent = bio_alloc(GFP_NOIO, 1); |
1114 | if (!discard_parent) { | |
1115 | DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.", | |
1116 | dm_device_name(tc->pool->pool_md)); | |
1117 | queue_passdown_pt2(m); | |
202bae52 JT |
1118 | |
1119 | } else { | |
2a0fbffb JT |
1120 | discard_parent->bi_end_io = passdown_endio; |
1121 | discard_parent->bi_private = m; | |
1122 | ||
1123 | if (m->maybe_shared) | |
1124 | passdown_double_checking_shared_status(m, discard_parent); | |
1125 | else { | |
1126 | struct discard_op op; | |
1127 | ||
1128 | begin_discard(&op, tc, discard_parent); | |
1129 | r = issue_discard(&op, m->data_block, data_end); | |
1130 | end_discard(&op, r); | |
1131 | } | |
202bae52 | 1132 | } |
2a0fbffb JT |
1133 | } |
1134 | ||
1135 | static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m) | |
1136 | { | |
1137 | int r; | |
1138 | struct thin_c *tc = m->tc; | |
1139 | struct pool *pool = tc->pool; | |
1140 | ||
1141 | /* | |
1142 | * The passdown has completed, so now we can decrement all those | |
1143 | * unmapped blocks. | |
1144 | */ | |
1145 | r = dm_pool_dec_data_range(pool->pmd, m->data_block, | |
1146 | m->data_block + (m->virt_end - m->virt_begin)); | |
1147 | if (r) { | |
1148 | metadata_operation_failed(pool, "dm_pool_dec_data_range", r); | |
1149 | bio_io_error(m->bio); | |
1150 | } else | |
1151 | bio_endio(m->bio); | |
1152 | ||
34fbcf62 | 1153 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1154 | mempool_free(m, &pool->mapping_pool); |
e49e5829 JT |
1155 | } |
1156 | ||
104655fd | 1157 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 1158 | process_mapping_fn *fn) |
991d9fa0 JT |
1159 | { |
1160 | unsigned long flags; | |
1161 | struct list_head maps; | |
a24c2569 | 1162 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
1163 | |
1164 | INIT_LIST_HEAD(&maps); | |
1165 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 1166 | list_splice_init(head, &maps); |
991d9fa0 JT |
1167 | spin_unlock_irqrestore(&pool->lock, flags); |
1168 | ||
1169 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1170 | (*fn)(m); |
991d9fa0 JT |
1171 | } |
1172 | ||
1173 | /* | |
1174 | * Deferred bio jobs. | |
1175 | */ | |
104655fd | 1176 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1177 | { |
4f024f37 KO |
1178 | return bio->bi_iter.bi_size == |
1179 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
1180 | } |
1181 | ||
1182 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1183 | { | |
1184 | return (bio_data_dir(bio) == WRITE) && | |
1185 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1186 | } |
1187 | ||
1188 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1189 | bio_end_io_t *fn) | |
1190 | { | |
1191 | *save = bio->bi_end_io; | |
1192 | bio->bi_end_io = fn; | |
1193 | } | |
1194 | ||
1195 | static int ensure_next_mapping(struct pool *pool) | |
1196 | { | |
1197 | if (pool->next_mapping) | |
1198 | return 0; | |
1199 | ||
6f1c819c | 1200 | pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC); |
991d9fa0 JT |
1201 | |
1202 | return pool->next_mapping ? 0 : -ENOMEM; | |
1203 | } | |
1204 | ||
a24c2569 | 1205 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1206 | { |
16961b04 | 1207 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
1208 | |
1209 | BUG_ON(!pool->next_mapping); | |
1210 | ||
16961b04 MS |
1211 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
1212 | INIT_LIST_HEAD(&m->list); | |
1213 | m->bio = NULL; | |
1214 | ||
991d9fa0 JT |
1215 | pool->next_mapping = NULL; |
1216 | ||
16961b04 | 1217 | return m; |
991d9fa0 JT |
1218 | } |
1219 | ||
e5aea7b4 JT |
1220 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
1221 | sector_t begin, sector_t end) | |
1222 | { | |
1223 | int r; | |
1224 | struct dm_io_region to; | |
1225 | ||
1226 | to.bdev = tc->pool_dev->bdev; | |
1227 | to.sector = begin; | |
1228 | to.count = end - begin; | |
1229 | ||
1230 | r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); | |
1231 | if (r < 0) { | |
1232 | DMERR_LIMIT("dm_kcopyd_zero() failed"); | |
1233 | copy_complete(1, 1, m); | |
1234 | } | |
1235 | } | |
1236 | ||
452d7a62 | 1237 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
34fbcf62 | 1238 | dm_block_t data_begin, |
452d7a62 MS |
1239 | struct dm_thin_new_mapping *m) |
1240 | { | |
1241 | struct pool *pool = tc->pool; | |
1242 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1243 | ||
1244 | h->overwrite_mapping = m; | |
1245 | m->bio = bio; | |
1246 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
1247 | inc_all_io_entry(pool, bio); | |
34fbcf62 | 1248 | remap_and_issue(tc, bio, data_begin); |
452d7a62 MS |
1249 | } |
1250 | ||
e5aea7b4 JT |
1251 | /* |
1252 | * A partial copy also needs to zero the uncopied region. | |
1253 | */ | |
991d9fa0 | 1254 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
1255 | struct dm_dev *origin, dm_block_t data_origin, |
1256 | dm_block_t data_dest, | |
e5aea7b4 JT |
1257 | struct dm_bio_prison_cell *cell, struct bio *bio, |
1258 | sector_t len) | |
991d9fa0 JT |
1259 | { |
1260 | int r; | |
1261 | struct pool *pool = tc->pool; | |
a24c2569 | 1262 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1263 | |
991d9fa0 | 1264 | m->tc = tc; |
34fbcf62 JT |
1265 | m->virt_begin = virt_block; |
1266 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1267 | m->data_block = data_dest; |
1268 | m->cell = cell; | |
991d9fa0 | 1269 | |
e5aea7b4 JT |
1270 | /* |
1271 | * quiesce action + copy action + an extra reference held for the | |
1272 | * duration of this function (we may need to inc later for a | |
1273 | * partial zero). | |
1274 | */ | |
1275 | atomic_set(&m->prepare_actions, 3); | |
1276 | ||
44feb387 | 1277 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1278 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1279 | |
1280 | /* | |
1281 | * IO to pool_dev remaps to the pool target's data_dev. | |
1282 | * | |
1283 | * If the whole block of data is being overwritten, we can issue the | |
1284 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1285 | */ | |
452d7a62 MS |
1286 | if (io_overwrites_block(pool, bio)) |
1287 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1288 | else { | |
991d9fa0 JT |
1289 | struct dm_io_region from, to; |
1290 | ||
2dd9c257 | 1291 | from.bdev = origin->bdev; |
991d9fa0 | 1292 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1293 | from.count = len; |
991d9fa0 JT |
1294 | |
1295 | to.bdev = tc->pool_dev->bdev; | |
1296 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1297 | to.count = len; |
991d9fa0 JT |
1298 | |
1299 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1300 | 0, copy_complete, m); | |
1301 | if (r < 0) { | |
c397741c | 1302 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
e5aea7b4 JT |
1303 | copy_complete(1, 1, m); |
1304 | ||
1305 | /* | |
1306 | * We allow the zero to be issued, to simplify the | |
1307 | * error path. Otherwise we'd need to start | |
1308 | * worrying about decrementing the prepare_actions | |
1309 | * counter. | |
1310 | */ | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Do we need to zero a tail region? | |
1315 | */ | |
1316 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1317 | atomic_inc(&m->prepare_actions); | |
1318 | ll_zero(tc, m, | |
1319 | data_dest * pool->sectors_per_block + len, | |
1320 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1321 | } |
1322 | } | |
e5aea7b4 JT |
1323 | |
1324 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1325 | } |
1326 | ||
2dd9c257 JT |
1327 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1328 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1329 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1330 | { |
1331 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1332 | data_origin, data_dest, cell, bio, |
1333 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1334 | } |
1335 | ||
991d9fa0 | 1336 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1337 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1338 | struct bio *bio) |
1339 | { | |
1340 | struct pool *pool = tc->pool; | |
a24c2569 | 1341 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1342 | |
50f3c3ef | 1343 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 | 1344 | m->tc = tc; |
34fbcf62 JT |
1345 | m->virt_begin = virt_block; |
1346 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1347 | m->data_block = data_block; |
1348 | m->cell = cell; | |
991d9fa0 JT |
1349 | |
1350 | /* | |
1351 | * If the whole block of data is being overwritten or we are not | |
1352 | * zeroing pre-existing data, we can issue the bio immediately. | |
1353 | * Otherwise we use kcopyd to zero the data first. | |
1354 | */ | |
f8ae7525 MS |
1355 | if (pool->pf.zero_new_blocks) { |
1356 | if (io_overwrites_block(pool, bio)) | |
1357 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
1358 | else | |
1359 | ll_zero(tc, m, data_block * pool->sectors_per_block, | |
1360 | (data_block + 1) * pool->sectors_per_block); | |
1361 | } else | |
991d9fa0 | 1362 | process_prepared_mapping(m); |
e5aea7b4 | 1363 | } |
991d9fa0 | 1364 | |
e5aea7b4 JT |
1365 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1366 | dm_block_t data_dest, | |
1367 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1368 | { | |
1369 | struct pool *pool = tc->pool; | |
1370 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1371 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1372 | ||
1373 | if (virt_block_end <= tc->origin_size) | |
1374 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1375 | virt_block, data_dest, cell, bio, | |
1376 | pool->sectors_per_block); | |
1377 | ||
1378 | else if (virt_block_begin < tc->origin_size) | |
1379 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1380 | virt_block, data_dest, cell, bio, | |
1381 | tc->origin_size - virt_block_begin); | |
1382 | ||
1383 | else | |
1384 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1385 | } |
1386 | ||
2c43fd26 JT |
1387 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1388 | ||
1389 | static void check_for_space(struct pool *pool) | |
1390 | { | |
1391 | int r; | |
1392 | dm_block_t nr_free; | |
1393 | ||
1394 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) | |
1395 | return; | |
1396 | ||
1397 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); | |
1398 | if (r) | |
1399 | return; | |
1400 | ||
1401 | if (nr_free) | |
1402 | set_pool_mode(pool, PM_WRITE); | |
1403 | } | |
1404 | ||
e49e5829 JT |
1405 | /* |
1406 | * A non-zero return indicates read_only or fail_io mode. | |
1407 | * Many callers don't care about the return value. | |
1408 | */ | |
020cc3b5 | 1409 | static int commit(struct pool *pool) |
e49e5829 JT |
1410 | { |
1411 | int r; | |
1412 | ||
8d07e8a5 | 1413 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
1414 | return -EINVAL; |
1415 | ||
020cc3b5 | 1416 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1417 | if (r) |
1418 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
2c43fd26 JT |
1419 | else |
1420 | check_for_space(pool); | |
e49e5829 JT |
1421 | |
1422 | return r; | |
1423 | } | |
1424 | ||
88a6621b JT |
1425 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1426 | { | |
1427 | unsigned long flags; | |
1428 | ||
1429 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1430 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1431 | dm_device_name(pool->pool_md)); | |
1432 | spin_lock_irqsave(&pool->lock, flags); | |
1433 | pool->low_water_triggered = true; | |
1434 | spin_unlock_irqrestore(&pool->lock, flags); | |
1435 | dm_table_event(pool->ti->table); | |
1436 | } | |
1437 | } | |
1438 | ||
991d9fa0 JT |
1439 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1440 | { | |
1441 | int r; | |
1442 | dm_block_t free_blocks; | |
991d9fa0 JT |
1443 | struct pool *pool = tc->pool; |
1444 | ||
3e1a0699 | 1445 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1446 | return -EINVAL; |
1447 | ||
991d9fa0 | 1448 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1449 | if (r) { |
1450 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1451 | return r; |
b5330655 | 1452 | } |
991d9fa0 | 1453 | |
88a6621b | 1454 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1455 | |
1456 | if (!free_blocks) { | |
94563bad MS |
1457 | /* |
1458 | * Try to commit to see if that will free up some | |
1459 | * more space. | |
1460 | */ | |
020cc3b5 JT |
1461 | r = commit(pool); |
1462 | if (r) | |
1463 | return r; | |
991d9fa0 | 1464 | |
94563bad | 1465 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1466 | if (r) { |
1467 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1468 | return r; |
b5330655 | 1469 | } |
991d9fa0 | 1470 | |
94563bad | 1471 | if (!free_blocks) { |
3e1a0699 | 1472 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1473 | return -ENOSPC; |
991d9fa0 JT |
1474 | } |
1475 | } | |
1476 | ||
1477 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1478 | if (r) { |
b5330655 | 1479 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1480 | return r; |
4a02b34e | 1481 | } |
991d9fa0 JT |
1482 | |
1483 | return 0; | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * If we have run out of space, queue bios until the device is | |
1488 | * resumed, presumably after having been reloaded with more space. | |
1489 | */ | |
1490 | static void retry_on_resume(struct bio *bio) | |
1491 | { | |
59c3d2c6 | 1492 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1493 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1494 | unsigned long flags; |
1495 | ||
c140e1c4 MS |
1496 | spin_lock_irqsave(&tc->lock, flags); |
1497 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1498 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1499 | } |
1500 | ||
4e4cbee9 | 1501 | static blk_status_t should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1502 | { |
3e1a0699 JT |
1503 | enum pool_mode m = get_pool_mode(pool); |
1504 | ||
1505 | switch (m) { | |
1506 | case PM_WRITE: | |
1507 | /* Shouldn't get here */ | |
1508 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
4e4cbee9 | 1509 | return BLK_STS_IOERR; |
3e1a0699 JT |
1510 | |
1511 | case PM_OUT_OF_DATA_SPACE: | |
4e4cbee9 | 1512 | return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0; |
3e1a0699 JT |
1513 | |
1514 | case PM_READ_ONLY: | |
1515 | case PM_FAIL: | |
4e4cbee9 | 1516 | return BLK_STS_IOERR; |
3e1a0699 JT |
1517 | default: |
1518 | /* Shouldn't get here */ | |
1519 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
4e4cbee9 | 1520 | return BLK_STS_IOERR; |
3e1a0699 JT |
1521 | } |
1522 | } | |
8c0f0e8c | 1523 | |
3e1a0699 JT |
1524 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1525 | { | |
4e4cbee9 | 1526 | blk_status_t error = should_error_unserviceable_bio(pool); |
af91805a | 1527 | |
4246a0b6 | 1528 | if (error) { |
4e4cbee9 | 1529 | bio->bi_status = error; |
4246a0b6 CH |
1530 | bio_endio(bio); |
1531 | } else | |
6d16202b | 1532 | retry_on_resume(bio); |
8c0f0e8c MS |
1533 | } |
1534 | ||
399caddf | 1535 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1536 | { |
1537 | struct bio *bio; | |
1538 | struct bio_list bios; | |
4e4cbee9 | 1539 | blk_status_t error; |
991d9fa0 | 1540 | |
af91805a MS |
1541 | error = should_error_unserviceable_bio(pool); |
1542 | if (error) { | |
1543 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1544 | return; |
1545 | } | |
1546 | ||
991d9fa0 | 1547 | bio_list_init(&bios); |
6beca5eb | 1548 | cell_release(pool, cell, &bios); |
991d9fa0 | 1549 | |
9d094eeb MS |
1550 | while ((bio = bio_list_pop(&bios))) |
1551 | retry_on_resume(bio); | |
991d9fa0 JT |
1552 | } |
1553 | ||
34fbcf62 JT |
1554 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
1555 | struct dm_bio_prison_cell *virt_cell) | |
104655fd | 1556 | { |
104655fd | 1557 | struct pool *pool = tc->pool; |
34fbcf62 | 1558 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
104655fd | 1559 | |
34fbcf62 JT |
1560 | /* |
1561 | * We don't need to lock the data blocks, since there's no | |
1562 | * passdown. We only lock data blocks for allocation and breaking sharing. | |
1563 | */ | |
1564 | m->tc = tc; | |
1565 | m->virt_begin = virt_cell->key.block_begin; | |
1566 | m->virt_end = virt_cell->key.block_end; | |
1567 | m->cell = virt_cell; | |
1568 | m->bio = virt_cell->holder; | |
104655fd | 1569 | |
34fbcf62 JT |
1570 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1571 | pool->process_prepared_discard(m); | |
1572 | } | |
104655fd | 1573 | |
34fbcf62 JT |
1574 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
1575 | struct bio *bio) | |
1576 | { | |
1577 | struct pool *pool = tc->pool; | |
1578 | ||
1579 | int r; | |
1580 | bool maybe_shared; | |
1581 | struct dm_cell_key data_key; | |
1582 | struct dm_bio_prison_cell *data_cell; | |
1583 | struct dm_thin_new_mapping *m; | |
1584 | dm_block_t virt_begin, virt_end, data_begin; | |
1585 | ||
1586 | while (begin != end) { | |
1587 | r = ensure_next_mapping(pool); | |
1588 | if (r) | |
1589 | /* we did our best */ | |
1590 | return; | |
e8088073 | 1591 | |
34fbcf62 JT |
1592 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
1593 | &data_begin, &maybe_shared); | |
1594 | if (r) | |
104655fd | 1595 | /* |
34fbcf62 JT |
1596 | * Silently fail, letting any mappings we've |
1597 | * created complete. | |
104655fd | 1598 | */ |
34fbcf62 JT |
1599 | break; |
1600 | ||
1601 | build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); | |
1602 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { | |
1603 | /* contention, we'll give up with this range */ | |
1604 | begin = virt_end; | |
1605 | continue; | |
104655fd | 1606 | } |
104655fd | 1607 | |
104655fd | 1608 | /* |
34fbcf62 JT |
1609 | * IO may still be going to the destination block. We must |
1610 | * quiesce before we can do the removal. | |
104655fd | 1611 | */ |
34fbcf62 JT |
1612 | m = get_next_mapping(pool); |
1613 | m->tc = tc; | |
1614 | m->maybe_shared = maybe_shared; | |
1615 | m->virt_begin = virt_begin; | |
1616 | m->virt_end = virt_end; | |
1617 | m->data_block = data_begin; | |
1618 | m->cell = data_cell; | |
1619 | m->bio = bio; | |
104655fd | 1620 | |
34fbcf62 JT |
1621 | /* |
1622 | * The parent bio must not complete before sub discard bios are | |
202bae52 | 1623 | * chained to it (see end_discard's bio_chain)! |
34fbcf62 JT |
1624 | * |
1625 | * This per-mapping bi_remaining increment is paired with | |
1626 | * the implicit decrement that occurs via bio_endio() in | |
202bae52 | 1627 | * end_discard(). |
34fbcf62 | 1628 | */ |
13e4f8a6 | 1629 | bio_inc_remaining(bio); |
34fbcf62 JT |
1630 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1631 | pool->process_prepared_discard(m); | |
1632 | ||
1633 | begin = virt_end; | |
104655fd JT |
1634 | } |
1635 | } | |
1636 | ||
34fbcf62 JT |
1637 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
1638 | { | |
1639 | struct bio *bio = virt_cell->holder; | |
1640 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1641 | ||
1642 | /* | |
1643 | * The virt_cell will only get freed once the origin bio completes. | |
1644 | * This means it will remain locked while all the individual | |
1645 | * passdown bios are in flight. | |
1646 | */ | |
1647 | h->cell = virt_cell; | |
1648 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); | |
1649 | ||
1650 | /* | |
1651 | * We complete the bio now, knowing that the bi_remaining field | |
1652 | * will prevent completion until the sub range discards have | |
1653 | * completed. | |
1654 | */ | |
4246a0b6 | 1655 | bio_endio(bio); |
34fbcf62 JT |
1656 | } |
1657 | ||
a374bb21 JT |
1658 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1659 | { | |
34fbcf62 JT |
1660 | dm_block_t begin, end; |
1661 | struct dm_cell_key virt_key; | |
1662 | struct dm_bio_prison_cell *virt_cell; | |
a374bb21 | 1663 | |
34fbcf62 JT |
1664 | get_bio_block_range(tc, bio, &begin, &end); |
1665 | if (begin == end) { | |
1666 | /* | |
1667 | * The discard covers less than a block. | |
1668 | */ | |
4246a0b6 | 1669 | bio_endio(bio); |
a374bb21 | 1670 | return; |
34fbcf62 | 1671 | } |
a374bb21 | 1672 | |
34fbcf62 JT |
1673 | build_key(tc->td, VIRTUAL, begin, end, &virt_key); |
1674 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) | |
1675 | /* | |
1676 | * Potential starvation issue: We're relying on the | |
1677 | * fs/application being well behaved, and not trying to | |
1678 | * send IO to a region at the same time as discarding it. | |
1679 | * If they do this persistently then it's possible this | |
1680 | * cell will never be granted. | |
1681 | */ | |
1682 | return; | |
1683 | ||
1684 | tc->pool->process_discard_cell(tc, virt_cell); | |
a374bb21 JT |
1685 | } |
1686 | ||
991d9fa0 | 1687 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1688 | struct dm_cell_key *key, |
991d9fa0 | 1689 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1690 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1691 | { |
1692 | int r; | |
1693 | dm_block_t data_block; | |
d6fc2042 | 1694 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1695 | |
1696 | r = alloc_data_block(tc, &data_block); | |
1697 | switch (r) { | |
1698 | case 0: | |
2dd9c257 JT |
1699 | schedule_internal_copy(tc, block, lookup_result->block, |
1700 | data_block, cell, bio); | |
991d9fa0 JT |
1701 | break; |
1702 | ||
1703 | case -ENOSPC: | |
399caddf | 1704 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1705 | break; |
1706 | ||
1707 | default: | |
c397741c MS |
1708 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1709 | __func__, r); | |
d6fc2042 | 1710 | cell_error(pool, cell); |
991d9fa0 JT |
1711 | break; |
1712 | } | |
1713 | } | |
1714 | ||
23ca2bb6 JT |
1715 | static void __remap_and_issue_shared_cell(void *context, |
1716 | struct dm_bio_prison_cell *cell) | |
1717 | { | |
1718 | struct remap_info *info = context; | |
1719 | struct bio *bio; | |
1720 | ||
1721 | while ((bio = bio_list_pop(&cell->bios))) { | |
f73f44eb CH |
1722 | if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) || |
1723 | bio_op(bio) == REQ_OP_DISCARD) | |
23ca2bb6 JT |
1724 | bio_list_add(&info->defer_bios, bio); |
1725 | else { | |
bd6d1e0a | 1726 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
23ca2bb6 JT |
1727 | |
1728 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1729 | inc_all_io_entry(info->tc->pool, bio); | |
1730 | bio_list_add(&info->issue_bios, bio); | |
1731 | } | |
1732 | } | |
1733 | } | |
1734 | ||
1735 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1736 | struct dm_bio_prison_cell *cell, | |
1737 | dm_block_t block) | |
1738 | { | |
1739 | struct bio *bio; | |
1740 | struct remap_info info; | |
1741 | ||
1742 | info.tc = tc; | |
1743 | bio_list_init(&info.defer_bios); | |
1744 | bio_list_init(&info.issue_bios); | |
1745 | ||
1746 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1747 | &info, cell); | |
1748 | ||
1749 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1750 | thin_defer_bio(tc, bio); | |
1751 | ||
1752 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1753 | remap_and_issue(tc, bio, block); | |
1754 | } | |
1755 | ||
991d9fa0 JT |
1756 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1757 | dm_block_t block, | |
23ca2bb6 JT |
1758 | struct dm_thin_lookup_result *lookup_result, |
1759 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1760 | { |
23ca2bb6 | 1761 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1762 | struct pool *pool = tc->pool; |
44feb387 | 1763 | struct dm_cell_key key; |
991d9fa0 JT |
1764 | |
1765 | /* | |
1766 | * If cell is already occupied, then sharing is already in the process | |
1767 | * of being broken so we have nothing further to do here. | |
1768 | */ | |
1769 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1770 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1771 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1772 | return; |
23ca2bb6 | 1773 | } |
991d9fa0 | 1774 | |
23ca2bb6 JT |
1775 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1776 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1777 | cell_defer_no_holder(tc, virt_cell); | |
1778 | } else { | |
59c3d2c6 | 1779 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1780 | |
44feb387 | 1781 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1782 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1783 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1784 | |
1785 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1786 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1787 | } |
1788 | } | |
1789 | ||
1790 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1791 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1792 | { |
1793 | int r; | |
1794 | dm_block_t data_block; | |
6beca5eb | 1795 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1796 | |
1797 | /* | |
1798 | * Remap empty bios (flushes) immediately, without provisioning. | |
1799 | */ | |
4f024f37 | 1800 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1801 | inc_all_io_entry(pool, bio); |
f286ba0e | 1802 | cell_defer_no_holder(tc, cell); |
e8088073 | 1803 | |
991d9fa0 JT |
1804 | remap_and_issue(tc, bio, 0); |
1805 | return; | |
1806 | } | |
1807 | ||
1808 | /* | |
1809 | * Fill read bios with zeroes and complete them immediately. | |
1810 | */ | |
1811 | if (bio_data_dir(bio) == READ) { | |
1812 | zero_fill_bio(bio); | |
f286ba0e | 1813 | cell_defer_no_holder(tc, cell); |
4246a0b6 | 1814 | bio_endio(bio); |
991d9fa0 JT |
1815 | return; |
1816 | } | |
1817 | ||
1818 | r = alloc_data_block(tc, &data_block); | |
1819 | switch (r) { | |
1820 | case 0: | |
2dd9c257 JT |
1821 | if (tc->origin_dev) |
1822 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1823 | else | |
1824 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1825 | break; |
1826 | ||
1827 | case -ENOSPC: | |
399caddf | 1828 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1829 | break; |
1830 | ||
1831 | default: | |
c397741c MS |
1832 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1833 | __func__, r); | |
6beca5eb | 1834 | cell_error(pool, cell); |
991d9fa0 JT |
1835 | break; |
1836 | } | |
1837 | } | |
1838 | ||
a374bb21 | 1839 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1840 | { |
1841 | int r; | |
6beca5eb | 1842 | struct pool *pool = tc->pool; |
a374bb21 | 1843 | struct bio *bio = cell->holder; |
991d9fa0 | 1844 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1845 | struct dm_thin_lookup_result lookup_result; |
1846 | ||
a374bb21 JT |
1847 | if (tc->requeue_mode) { |
1848 | cell_requeue(pool, cell); | |
991d9fa0 | 1849 | return; |
a374bb21 | 1850 | } |
991d9fa0 JT |
1851 | |
1852 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1853 | switch (r) { | |
1854 | case 0: | |
23ca2bb6 JT |
1855 | if (lookup_result.shared) |
1856 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1857 | else { | |
6beca5eb | 1858 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1859 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1860 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1861 | } |
991d9fa0 JT |
1862 | break; |
1863 | ||
1864 | case -ENODATA: | |
2dd9c257 | 1865 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1866 | inc_all_io_entry(pool, bio); |
f286ba0e | 1867 | cell_defer_no_holder(tc, cell); |
e8088073 | 1868 | |
e5aea7b4 JT |
1869 | if (bio_end_sector(bio) <= tc->origin_size) |
1870 | remap_to_origin_and_issue(tc, bio); | |
1871 | ||
1872 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1873 | zero_fill_bio(bio); | |
1874 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1875 | remap_to_origin_and_issue(tc, bio); | |
1876 | ||
1877 | } else { | |
1878 | zero_fill_bio(bio); | |
4246a0b6 | 1879 | bio_endio(bio); |
e5aea7b4 | 1880 | } |
2dd9c257 JT |
1881 | } else |
1882 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1883 | break; |
1884 | ||
1885 | default: | |
c397741c MS |
1886 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1887 | __func__, r); | |
f286ba0e | 1888 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1889 | bio_io_error(bio); |
1890 | break; | |
1891 | } | |
1892 | } | |
1893 | ||
a374bb21 JT |
1894 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1895 | { | |
1896 | struct pool *pool = tc->pool; | |
1897 | dm_block_t block = get_bio_block(tc, bio); | |
1898 | struct dm_bio_prison_cell *cell; | |
1899 | struct dm_cell_key key; | |
1900 | ||
1901 | /* | |
1902 | * If cell is already occupied, then the block is already | |
1903 | * being provisioned so we have nothing further to do here. | |
1904 | */ | |
1905 | build_virtual_key(tc->td, block, &key); | |
1906 | if (bio_detain(pool, &key, bio, &cell)) | |
1907 | return; | |
1908 | ||
1909 | process_cell(tc, cell); | |
1910 | } | |
1911 | ||
1912 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
1913 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
1914 | { |
1915 | int r; | |
1916 | int rw = bio_data_dir(bio); | |
1917 | dm_block_t block = get_bio_block(tc, bio); | |
1918 | struct dm_thin_lookup_result lookup_result; | |
1919 | ||
1920 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1921 | switch (r) { | |
1922 | case 0: | |
a374bb21 | 1923 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 1924 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
1925 | if (cell) |
1926 | cell_defer_no_holder(tc, cell); | |
1927 | } else { | |
e8088073 | 1928 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 1929 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
1930 | if (cell) |
1931 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 1932 | } |
e49e5829 JT |
1933 | break; |
1934 | ||
1935 | case -ENODATA: | |
a374bb21 JT |
1936 | if (cell) |
1937 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 1938 | if (rw != READ) { |
8c0f0e8c | 1939 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1940 | break; |
1941 | } | |
1942 | ||
1943 | if (tc->origin_dev) { | |
e8088073 | 1944 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1945 | remap_to_origin_and_issue(tc, bio); |
1946 | break; | |
1947 | } | |
1948 | ||
1949 | zero_fill_bio(bio); | |
4246a0b6 | 1950 | bio_endio(bio); |
e49e5829 JT |
1951 | break; |
1952 | ||
1953 | default: | |
c397741c MS |
1954 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1955 | __func__, r); | |
a374bb21 JT |
1956 | if (cell) |
1957 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
1958 | bio_io_error(bio); |
1959 | break; | |
1960 | } | |
1961 | } | |
1962 | ||
a374bb21 JT |
1963 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1964 | { | |
1965 | __process_bio_read_only(tc, bio, NULL); | |
1966 | } | |
1967 | ||
1968 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1969 | { | |
1970 | __process_bio_read_only(tc, cell->holder, cell); | |
1971 | } | |
1972 | ||
3e1a0699 JT |
1973 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1974 | { | |
4246a0b6 | 1975 | bio_endio(bio); |
3e1a0699 JT |
1976 | } |
1977 | ||
e49e5829 JT |
1978 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1979 | { | |
1980 | bio_io_error(bio); | |
1981 | } | |
1982 | ||
a374bb21 JT |
1983 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
1984 | { | |
1985 | cell_success(tc->pool, cell); | |
1986 | } | |
1987 | ||
1988 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1989 | { | |
1990 | cell_error(tc->pool, cell); | |
1991 | } | |
1992 | ||
ac8c3f3d JT |
1993 | /* |
1994 | * FIXME: should we also commit due to size of transaction, measured in | |
1995 | * metadata blocks? | |
1996 | */ | |
905e51b3 JT |
1997 | static int need_commit_due_to_time(struct pool *pool) |
1998 | { | |
0f30af98 MS |
1999 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
2000 | pool->last_commit_jiffies + COMMIT_PERIOD); | |
905e51b3 JT |
2001 | } |
2002 | ||
67324ea1 MS |
2003 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
2004 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
2005 | ||
2006 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
2007 | { | |
2008 | struct rb_node **rbp, *parent; | |
2009 | struct dm_thin_endio_hook *pbd; | |
2010 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
2011 | ||
2012 | rbp = &tc->sort_bio_list.rb_node; | |
2013 | parent = NULL; | |
2014 | while (*rbp) { | |
2015 | parent = *rbp; | |
2016 | pbd = thin_pbd(parent); | |
2017 | ||
2018 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
2019 | rbp = &(*rbp)->rb_left; | |
2020 | else | |
2021 | rbp = &(*rbp)->rb_right; | |
2022 | } | |
2023 | ||
2024 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
2025 | rb_link_node(&pbd->rb_node, parent, rbp); | |
2026 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
2027 | } | |
2028 | ||
2029 | static void __extract_sorted_bios(struct thin_c *tc) | |
2030 | { | |
2031 | struct rb_node *node; | |
2032 | struct dm_thin_endio_hook *pbd; | |
2033 | struct bio *bio; | |
2034 | ||
2035 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
2036 | pbd = thin_pbd(node); | |
2037 | bio = thin_bio(pbd); | |
2038 | ||
2039 | bio_list_add(&tc->deferred_bio_list, bio); | |
2040 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
2041 | } | |
2042 | ||
2043 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
2044 | } | |
2045 | ||
2046 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
2047 | { | |
2048 | struct bio *bio; | |
2049 | struct bio_list bios; | |
2050 | ||
2051 | bio_list_init(&bios); | |
2052 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
2053 | bio_list_init(&tc->deferred_bio_list); | |
2054 | ||
2055 | /* Sort deferred_bio_list using rb-tree */ | |
2056 | while ((bio = bio_list_pop(&bios))) | |
2057 | __thin_bio_rb_add(tc, bio); | |
2058 | ||
2059 | /* | |
2060 | * Transfer the sorted bios in sort_bio_list back to | |
2061 | * deferred_bio_list to allow lockless submission of | |
2062 | * all bios. | |
2063 | */ | |
2064 | __extract_sorted_bios(tc); | |
2065 | } | |
2066 | ||
c140e1c4 | 2067 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 2068 | { |
c140e1c4 | 2069 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2070 | unsigned long flags; |
2071 | struct bio *bio; | |
2072 | struct bio_list bios; | |
67324ea1 | 2073 | struct blk_plug plug; |
8a01a6af | 2074 | unsigned count = 0; |
991d9fa0 | 2075 | |
c140e1c4 | 2076 | if (tc->requeue_mode) { |
4e4cbee9 CH |
2077 | error_thin_bio_list(tc, &tc->deferred_bio_list, |
2078 | BLK_STS_DM_REQUEUE); | |
c140e1c4 MS |
2079 | return; |
2080 | } | |
2081 | ||
991d9fa0 JT |
2082 | bio_list_init(&bios); |
2083 | ||
c140e1c4 | 2084 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
2085 | |
2086 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
2087 | spin_unlock_irqrestore(&tc->lock, flags); | |
2088 | return; | |
2089 | } | |
2090 | ||
2091 | __sort_thin_deferred_bios(tc); | |
2092 | ||
c140e1c4 MS |
2093 | bio_list_merge(&bios, &tc->deferred_bio_list); |
2094 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 2095 | |
c140e1c4 | 2096 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 2097 | |
67324ea1 | 2098 | blk_start_plug(&plug); |
991d9fa0 | 2099 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
2100 | /* |
2101 | * If we've got no free new_mapping structs, and processing | |
2102 | * this bio might require one, we pause until there are some | |
2103 | * prepared mappings to process. | |
2104 | */ | |
2105 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
2106 | spin_lock_irqsave(&tc->lock, flags); |
2107 | bio_list_add(&tc->deferred_bio_list, bio); | |
2108 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
2109 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2110 | break; |
2111 | } | |
104655fd | 2112 | |
e6047149 | 2113 | if (bio_op(bio) == REQ_OP_DISCARD) |
e49e5829 | 2114 | pool->process_discard(tc, bio); |
104655fd | 2115 | else |
e49e5829 | 2116 | pool->process_bio(tc, bio); |
8a01a6af JT |
2117 | |
2118 | if ((count++ & 127) == 0) { | |
7d327fe0 | 2119 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
2120 | dm_pool_issue_prefetches(pool->pmd); |
2121 | } | |
991d9fa0 | 2122 | } |
67324ea1 | 2123 | blk_finish_plug(&plug); |
c140e1c4 MS |
2124 | } |
2125 | ||
ac4c3f34 JT |
2126 | static int cmp_cells(const void *lhs, const void *rhs) |
2127 | { | |
2128 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
2129 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
2130 | ||
2131 | BUG_ON(!lhs_cell->holder); | |
2132 | BUG_ON(!rhs_cell->holder); | |
2133 | ||
2134 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
2135 | return -1; | |
2136 | ||
2137 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
2138 | return 1; | |
2139 | ||
2140 | return 0; | |
2141 | } | |
2142 | ||
2143 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) | |
2144 | { | |
2145 | unsigned count = 0; | |
2146 | struct dm_bio_prison_cell *cell, *tmp; | |
2147 | ||
2148 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
2149 | if (count >= CELL_SORT_ARRAY_SIZE) | |
2150 | break; | |
2151 | ||
2152 | pool->cell_sort_array[count++] = cell; | |
2153 | list_del(&cell->user_list); | |
2154 | } | |
2155 | ||
2156 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
2157 | ||
2158 | return count; | |
2159 | } | |
2160 | ||
a374bb21 JT |
2161 | static void process_thin_deferred_cells(struct thin_c *tc) |
2162 | { | |
2163 | struct pool *pool = tc->pool; | |
2164 | unsigned long flags; | |
2165 | struct list_head cells; | |
ac4c3f34 JT |
2166 | struct dm_bio_prison_cell *cell; |
2167 | unsigned i, j, count; | |
a374bb21 JT |
2168 | |
2169 | INIT_LIST_HEAD(&cells); | |
2170 | ||
2171 | spin_lock_irqsave(&tc->lock, flags); | |
2172 | list_splice_init(&tc->deferred_cells, &cells); | |
2173 | spin_unlock_irqrestore(&tc->lock, flags); | |
2174 | ||
2175 | if (list_empty(&cells)) | |
2176 | return; | |
2177 | ||
ac4c3f34 JT |
2178 | do { |
2179 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 2180 | |
ac4c3f34 JT |
2181 | for (i = 0; i < count; i++) { |
2182 | cell = pool->cell_sort_array[i]; | |
2183 | BUG_ON(!cell->holder); | |
a374bb21 | 2184 | |
ac4c3f34 JT |
2185 | /* |
2186 | * If we've got no free new_mapping structs, and processing | |
2187 | * this bio might require one, we pause until there are some | |
2188 | * prepared mappings to process. | |
2189 | */ | |
2190 | if (ensure_next_mapping(pool)) { | |
2191 | for (j = i; j < count; j++) | |
2192 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
2193 | ||
2194 | spin_lock_irqsave(&tc->lock, flags); | |
2195 | list_splice(&cells, &tc->deferred_cells); | |
2196 | spin_unlock_irqrestore(&tc->lock, flags); | |
2197 | return; | |
2198 | } | |
2199 | ||
e6047149 | 2200 | if (bio_op(cell->holder) == REQ_OP_DISCARD) |
ac4c3f34 JT |
2201 | pool->process_discard_cell(tc, cell); |
2202 | else | |
2203 | pool->process_cell(tc, cell); | |
2204 | } | |
2205 | } while (!list_empty(&cells)); | |
a374bb21 JT |
2206 | } |
2207 | ||
b10ebd34 JT |
2208 | static void thin_get(struct thin_c *tc); |
2209 | static void thin_put(struct thin_c *tc); | |
2210 | ||
2211 | /* | |
2212 | * We can't hold rcu_read_lock() around code that can block. So we | |
2213 | * find a thin with the rcu lock held; bump a refcount; then drop | |
2214 | * the lock. | |
2215 | */ | |
2216 | static struct thin_c *get_first_thin(struct pool *pool) | |
2217 | { | |
2218 | struct thin_c *tc = NULL; | |
2219 | ||
2220 | rcu_read_lock(); | |
2221 | if (!list_empty(&pool->active_thins)) { | |
2222 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
2223 | thin_get(tc); | |
2224 | } | |
2225 | rcu_read_unlock(); | |
2226 | ||
2227 | return tc; | |
2228 | } | |
2229 | ||
2230 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
2231 | { | |
2232 | struct thin_c *old_tc = tc; | |
2233 | ||
2234 | rcu_read_lock(); | |
2235 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
2236 | thin_get(tc); | |
2237 | thin_put(old_tc); | |
2238 | rcu_read_unlock(); | |
2239 | return tc; | |
2240 | } | |
2241 | thin_put(old_tc); | |
2242 | rcu_read_unlock(); | |
2243 | ||
2244 | return NULL; | |
2245 | } | |
2246 | ||
c140e1c4 MS |
2247 | static void process_deferred_bios(struct pool *pool) |
2248 | { | |
2249 | unsigned long flags; | |
2250 | struct bio *bio; | |
2251 | struct bio_list bios; | |
2252 | struct thin_c *tc; | |
2253 | ||
b10ebd34 JT |
2254 | tc = get_first_thin(pool); |
2255 | while (tc) { | |
a374bb21 | 2256 | process_thin_deferred_cells(tc); |
c140e1c4 | 2257 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
2258 | tc = get_next_thin(pool, tc); |
2259 | } | |
991d9fa0 JT |
2260 | |
2261 | /* | |
2262 | * If there are any deferred flush bios, we must commit | |
2263 | * the metadata before issuing them. | |
2264 | */ | |
2265 | bio_list_init(&bios); | |
2266 | spin_lock_irqsave(&pool->lock, flags); | |
2267 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
2268 | bio_list_init(&pool->deferred_flush_bios); | |
2269 | spin_unlock_irqrestore(&pool->lock, flags); | |
2270 | ||
4d1662a3 MS |
2271 | if (bio_list_empty(&bios) && |
2272 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
2273 | return; |
2274 | ||
020cc3b5 | 2275 | if (commit(pool)) { |
991d9fa0 JT |
2276 | while ((bio = bio_list_pop(&bios))) |
2277 | bio_io_error(bio); | |
2278 | return; | |
2279 | } | |
905e51b3 | 2280 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2281 | |
2282 | while ((bio = bio_list_pop(&bios))) | |
2283 | generic_make_request(bio); | |
2284 | } | |
2285 | ||
2286 | static void do_worker(struct work_struct *ws) | |
2287 | { | |
2288 | struct pool *pool = container_of(ws, struct pool, worker); | |
2289 | ||
7d327fe0 | 2290 | throttle_work_start(&pool->throttle); |
8a01a6af | 2291 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 2292 | throttle_work_update(&pool->throttle); |
e49e5829 | 2293 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 2294 | throttle_work_update(&pool->throttle); |
e49e5829 | 2295 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 2296 | throttle_work_update(&pool->throttle); |
2a0fbffb JT |
2297 | process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2); |
2298 | throttle_work_update(&pool->throttle); | |
991d9fa0 | 2299 | process_deferred_bios(pool); |
7d327fe0 | 2300 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
2301 | } |
2302 | ||
905e51b3 JT |
2303 | /* |
2304 | * We want to commit periodically so that not too much | |
2305 | * unwritten data builds up. | |
2306 | */ | |
2307 | static void do_waker(struct work_struct *ws) | |
2308 | { | |
2309 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
2310 | wake_worker(pool); | |
2311 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
2312 | } | |
2313 | ||
bcc696fa MS |
2314 | static void notify_of_pool_mode_change_to_oods(struct pool *pool); |
2315 | ||
85ad643b JT |
2316 | /* |
2317 | * We're holding onto IO to allow userland time to react. After the | |
2318 | * timeout either the pool will have been resized (and thus back in | |
bcc696fa | 2319 | * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space. |
85ad643b JT |
2320 | */ |
2321 | static void do_no_space_timeout(struct work_struct *ws) | |
2322 | { | |
2323 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2324 | no_space_timeout); | |
2325 | ||
bcc696fa MS |
2326 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) { |
2327 | pool->pf.error_if_no_space = true; | |
2328 | notify_of_pool_mode_change_to_oods(pool); | |
4e4cbee9 | 2329 | error_retry_list_with_code(pool, BLK_STS_NOSPC); |
bcc696fa | 2330 | } |
85ad643b JT |
2331 | } |
2332 | ||
991d9fa0 JT |
2333 | /*----------------------------------------------------------------*/ |
2334 | ||
e7a3e871 | 2335 | struct pool_work { |
738211f7 | 2336 | struct work_struct worker; |
e7a3e871 JT |
2337 | struct completion complete; |
2338 | }; | |
2339 | ||
2340 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2341 | { | |
2342 | return container_of(ws, struct pool_work, worker); | |
2343 | } | |
2344 | ||
2345 | static void pool_work_complete(struct pool_work *pw) | |
2346 | { | |
2347 | complete(&pw->complete); | |
2348 | } | |
738211f7 | 2349 | |
e7a3e871 JT |
2350 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2351 | void (*fn)(struct work_struct *)) | |
2352 | { | |
2353 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2354 | init_completion(&pw->complete); | |
2355 | queue_work(pool->wq, &pw->worker); | |
2356 | wait_for_completion(&pw->complete); | |
2357 | } | |
2358 | ||
2359 | /*----------------------------------------------------------------*/ | |
2360 | ||
2361 | struct noflush_work { | |
2362 | struct pool_work pw; | |
2363 | struct thin_c *tc; | |
738211f7 JT |
2364 | }; |
2365 | ||
e7a3e871 | 2366 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2367 | { |
e7a3e871 | 2368 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2369 | } |
2370 | ||
2371 | static void do_noflush_start(struct work_struct *ws) | |
2372 | { | |
e7a3e871 | 2373 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2374 | w->tc->requeue_mode = true; |
2375 | requeue_io(w->tc); | |
e7a3e871 | 2376 | pool_work_complete(&w->pw); |
738211f7 JT |
2377 | } |
2378 | ||
2379 | static void do_noflush_stop(struct work_struct *ws) | |
2380 | { | |
e7a3e871 | 2381 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2382 | w->tc->requeue_mode = false; |
e7a3e871 | 2383 | pool_work_complete(&w->pw); |
738211f7 JT |
2384 | } |
2385 | ||
2386 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2387 | { | |
2388 | struct noflush_work w; | |
2389 | ||
738211f7 | 2390 | w.tc = tc; |
e7a3e871 | 2391 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2392 | } |
2393 | ||
2394 | /*----------------------------------------------------------------*/ | |
2395 | ||
e49e5829 JT |
2396 | static enum pool_mode get_pool_mode(struct pool *pool) |
2397 | { | |
2398 | return pool->pf.mode; | |
2399 | } | |
2400 | ||
3e1a0699 JT |
2401 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
2402 | { | |
2403 | dm_table_event(pool->ti->table); | |
2404 | DMINFO("%s: switching pool to %s mode", | |
2405 | dm_device_name(pool->pool_md), new_mode); | |
2406 | } | |
2407 | ||
bcc696fa MS |
2408 | static void notify_of_pool_mode_change_to_oods(struct pool *pool) |
2409 | { | |
2410 | if (!pool->pf.error_if_no_space) | |
2411 | notify_of_pool_mode_change(pool, "out-of-data-space (queue IO)"); | |
2412 | else | |
2413 | notify_of_pool_mode_change(pool, "out-of-data-space (error IO)"); | |
2414 | } | |
2415 | ||
34fbcf62 JT |
2416 | static bool passdown_enabled(struct pool_c *pt) |
2417 | { | |
2418 | return pt->adjusted_pf.discard_passdown; | |
2419 | } | |
2420 | ||
2421 | static void set_discard_callbacks(struct pool *pool) | |
2422 | { | |
2423 | struct pool_c *pt = pool->ti->private; | |
2424 | ||
2425 | if (passdown_enabled(pt)) { | |
2426 | pool->process_discard_cell = process_discard_cell_passdown; | |
2a0fbffb JT |
2427 | pool->process_prepared_discard = process_prepared_discard_passdown_pt1; |
2428 | pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2; | |
34fbcf62 JT |
2429 | } else { |
2430 | pool->process_discard_cell = process_discard_cell_no_passdown; | |
2431 | pool->process_prepared_discard = process_prepared_discard_no_passdown; | |
2432 | } | |
2433 | } | |
2434 | ||
8b64e881 | 2435 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2436 | { |
cdc2b415 | 2437 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2438 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2439 | enum pool_mode old_mode = get_pool_mode(pool); | |
6aa7de05 | 2440 | unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2441 | |
2442 | /* | |
2443 | * Never allow the pool to transition to PM_WRITE mode if user | |
2444 | * intervention is required to verify metadata and data consistency. | |
2445 | */ | |
2446 | if (new_mode == PM_WRITE && needs_check) { | |
2447 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2448 | dm_device_name(pool->pool_md)); | |
2449 | if (old_mode != new_mode) | |
2450 | new_mode = old_mode; | |
2451 | else | |
2452 | new_mode = PM_READ_ONLY; | |
2453 | } | |
2454 | /* | |
2455 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2456 | * not going to recover without a thin_repair. So we never let the | |
2457 | * pool move out of the old mode. | |
2458 | */ | |
2459 | if (old_mode == PM_FAIL) | |
2460 | new_mode = old_mode; | |
e49e5829 | 2461 | |
8b64e881 | 2462 | switch (new_mode) { |
e49e5829 | 2463 | case PM_FAIL: |
8b64e881 | 2464 | if (old_mode != new_mode) |
3e1a0699 | 2465 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 2466 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2467 | pool->process_bio = process_bio_fail; |
2468 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2469 | pool->process_cell = process_cell_fail; |
2470 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2471 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2472 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2473 | |
2474 | error_retry_list(pool); | |
e49e5829 JT |
2475 | break; |
2476 | ||
2477 | case PM_READ_ONLY: | |
8b64e881 | 2478 | if (old_mode != new_mode) |
3e1a0699 JT |
2479 | notify_of_pool_mode_change(pool, "read-only"); |
2480 | dm_pool_metadata_read_only(pool->pmd); | |
2481 | pool->process_bio = process_bio_read_only; | |
2482 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2483 | pool->process_cell = process_cell_read_only; |
2484 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 | 2485 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
34fbcf62 | 2486 | pool->process_prepared_discard = process_prepared_discard_success; |
3e1a0699 JT |
2487 | |
2488 | error_retry_list(pool); | |
2489 | break; | |
2490 | ||
2491 | case PM_OUT_OF_DATA_SPACE: | |
2492 | /* | |
2493 | * Ideally we'd never hit this state; the low water mark | |
2494 | * would trigger userland to extend the pool before we | |
2495 | * completely run out of data space. However, many small | |
2496 | * IOs to unprovisioned space can consume data space at an | |
2497 | * alarming rate. Adjust your low water mark if you're | |
2498 | * frequently seeing this mode. | |
2499 | */ | |
2500 | if (old_mode != new_mode) | |
bcc696fa | 2501 | notify_of_pool_mode_change_to_oods(pool); |
c3667cc6 | 2502 | pool->out_of_data_space = true; |
3e1a0699 | 2503 | pool->process_bio = process_bio_read_only; |
a374bb21 JT |
2504 | pool->process_discard = process_discard_bio; |
2505 | pool->process_cell = process_cell_read_only; | |
3e1a0699 | 2506 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2507 | set_discard_callbacks(pool); |
85ad643b | 2508 | |
80c57893 MS |
2509 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2510 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2511 | break; |
2512 | ||
2513 | case PM_WRITE: | |
8b64e881 | 2514 | if (old_mode != new_mode) |
3e1a0699 | 2515 | notify_of_pool_mode_change(pool, "write"); |
c3667cc6 | 2516 | pool->out_of_data_space = false; |
172c2386 | 2517 | pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space; |
9b7aaa64 | 2518 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2519 | pool->process_bio = process_bio; |
a374bb21 JT |
2520 | pool->process_discard = process_discard_bio; |
2521 | pool->process_cell = process_cell; | |
e49e5829 | 2522 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2523 | set_discard_callbacks(pool); |
e49e5829 JT |
2524 | break; |
2525 | } | |
8b64e881 MS |
2526 | |
2527 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2528 | /* |
2529 | * The pool mode may have changed, sync it so bind_control_target() | |
2530 | * doesn't cause an unexpected mode transition on resume. | |
2531 | */ | |
2532 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
2533 | } |
2534 | ||
07f2b6e0 | 2535 | static void abort_transaction(struct pool *pool) |
b5330655 | 2536 | { |
07f2b6e0 MS |
2537 | const char *dev_name = dm_device_name(pool->pool_md); |
2538 | ||
2539 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2540 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2541 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2542 | set_pool_mode(pool, PM_FAIL); | |
2543 | } | |
2544 | ||
2545 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2546 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2547 | set_pool_mode(pool, PM_FAIL); | |
2548 | } | |
2549 | } | |
399caddf | 2550 | |
07f2b6e0 MS |
2551 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2552 | { | |
b5330655 JT |
2553 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2554 | dm_device_name(pool->pool_md), op, r); | |
2555 | ||
07f2b6e0 | 2556 | abort_transaction(pool); |
b5330655 JT |
2557 | set_pool_mode(pool, PM_READ_ONLY); |
2558 | } | |
2559 | ||
e49e5829 JT |
2560 | /*----------------------------------------------------------------*/ |
2561 | ||
991d9fa0 JT |
2562 | /* |
2563 | * Mapping functions. | |
2564 | */ | |
2565 | ||
2566 | /* | |
2567 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2568 | */ | |
2569 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2570 | { | |
2571 | unsigned long flags; | |
2572 | struct pool *pool = tc->pool; | |
2573 | ||
c140e1c4 MS |
2574 | spin_lock_irqsave(&tc->lock, flags); |
2575 | bio_list_add(&tc->deferred_bio_list, bio); | |
2576 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2577 | |
2578 | wake_worker(pool); | |
2579 | } | |
2580 | ||
7d327fe0 JT |
2581 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2582 | { | |
2583 | struct pool *pool = tc->pool; | |
2584 | ||
2585 | throttle_lock(&pool->throttle); | |
2586 | thin_defer_bio(tc, bio); | |
2587 | throttle_unlock(&pool->throttle); | |
2588 | } | |
2589 | ||
a374bb21 JT |
2590 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2591 | { | |
2592 | unsigned long flags; | |
2593 | struct pool *pool = tc->pool; | |
2594 | ||
2595 | throttle_lock(&pool->throttle); | |
2596 | spin_lock_irqsave(&tc->lock, flags); | |
2597 | list_add_tail(&cell->user_list, &tc->deferred_cells); | |
2598 | spin_unlock_irqrestore(&tc->lock, flags); | |
2599 | throttle_unlock(&pool->throttle); | |
2600 | ||
2601 | wake_worker(pool); | |
2602 | } | |
2603 | ||
59c3d2c6 | 2604 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2605 | { |
59c3d2c6 | 2606 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2607 | |
2608 | h->tc = tc; | |
2609 | h->shared_read_entry = NULL; | |
e8088073 | 2610 | h->all_io_entry = NULL; |
eb2aa48d | 2611 | h->overwrite_mapping = NULL; |
34fbcf62 | 2612 | h->cell = NULL; |
eb2aa48d JT |
2613 | } |
2614 | ||
991d9fa0 JT |
2615 | /* |
2616 | * Non-blocking function called from the thin target's map function. | |
2617 | */ | |
7de3ee57 | 2618 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2619 | { |
2620 | int r; | |
2621 | struct thin_c *tc = ti->private; | |
2622 | dm_block_t block = get_bio_block(tc, bio); | |
2623 | struct dm_thin_device *td = tc->td; | |
2624 | struct dm_thin_lookup_result result; | |
a374bb21 | 2625 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2626 | struct dm_cell_key key; |
991d9fa0 | 2627 | |
59c3d2c6 | 2628 | thin_hook_bio(tc, bio); |
e49e5829 | 2629 | |
738211f7 | 2630 | if (tc->requeue_mode) { |
4e4cbee9 | 2631 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 | 2632 | bio_endio(bio); |
738211f7 JT |
2633 | return DM_MAPIO_SUBMITTED; |
2634 | } | |
2635 | ||
e49e5829 JT |
2636 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2637 | bio_io_error(bio); | |
2638 | return DM_MAPIO_SUBMITTED; | |
2639 | } | |
2640 | ||
f73f44eb | 2641 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) { |
7d327fe0 | 2642 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2643 | return DM_MAPIO_SUBMITTED; |
2644 | } | |
2645 | ||
c822ed96 JT |
2646 | /* |
2647 | * We must hold the virtual cell before doing the lookup, otherwise | |
2648 | * there's a race with discard. | |
2649 | */ | |
2650 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2651 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2652 | return DM_MAPIO_SUBMITTED; |
2653 | ||
991d9fa0 JT |
2654 | r = dm_thin_find_block(td, block, 0, &result); |
2655 | ||
2656 | /* | |
2657 | * Note that we defer readahead too. | |
2658 | */ | |
2659 | switch (r) { | |
2660 | case 0: | |
2661 | if (unlikely(result.shared)) { | |
2662 | /* | |
2663 | * We have a race condition here between the | |
2664 | * result.shared value returned by the lookup and | |
2665 | * snapshot creation, which may cause new | |
2666 | * sharing. | |
2667 | * | |
2668 | * To avoid this always quiesce the origin before | |
2669 | * taking the snap. You want to do this anyway to | |
2670 | * ensure a consistent application view | |
2671 | * (i.e. lockfs). | |
2672 | * | |
2673 | * More distant ancestors are irrelevant. The | |
2674 | * shared flag will be set in their case. | |
2675 | */ | |
a374bb21 | 2676 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2677 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2678 | } |
e8088073 | 2679 | |
e8088073 | 2680 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2681 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2682 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2683 | return DM_MAPIO_SUBMITTED; |
2684 | } | |
2685 | ||
2686 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2687 | cell_defer_no_holder(tc, data_cell); |
2688 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2689 | |
2690 | remap(tc, bio, result.block); | |
2691 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2692 | |
2693 | case -ENODATA: | |
e49e5829 | 2694 | case -EWOULDBLOCK: |
a374bb21 | 2695 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2696 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2697 | |
2698 | default: | |
2699 | /* | |
2700 | * Must always call bio_io_error on failure. | |
2701 | * dm_thin_find_block can fail with -EINVAL if the | |
2702 | * pool is switched to fail-io mode. | |
2703 | */ | |
2704 | bio_io_error(bio); | |
a374bb21 | 2705 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2706 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2707 | } |
991d9fa0 JT |
2708 | } |
2709 | ||
2710 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2711 | { | |
991d9fa0 | 2712 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 2713 | struct request_queue *q; |
991d9fa0 | 2714 | |
760fe67e MS |
2715 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
2716 | return 1; | |
991d9fa0 | 2717 | |
760fe67e | 2718 | q = bdev_get_queue(pt->data_dev->bdev); |
dc3b17cc | 2719 | return bdi_congested(q->backing_dev_info, bdi_bits); |
991d9fa0 JT |
2720 | } |
2721 | ||
c140e1c4 | 2722 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2723 | { |
c140e1c4 MS |
2724 | unsigned long flags; |
2725 | struct thin_c *tc; | |
2726 | ||
2727 | rcu_read_lock(); | |
2728 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
2729 | spin_lock_irqsave(&tc->lock, flags); | |
2730 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
2731 | bio_list_init(&tc->retry_on_resume_list); | |
2732 | spin_unlock_irqrestore(&tc->lock, flags); | |
2733 | } | |
2734 | rcu_read_unlock(); | |
991d9fa0 JT |
2735 | } |
2736 | ||
2737 | /*---------------------------------------------------------------- | |
2738 | * Binding of control targets to a pool object | |
2739 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2740 | static bool data_dev_supports_discard(struct pool_c *pt) |
2741 | { | |
2742 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2743 | ||
2744 | return q && blk_queue_discard(q); | |
2745 | } | |
2746 | ||
58051b94 JT |
2747 | static bool is_factor(sector_t block_size, uint32_t n) |
2748 | { | |
2749 | return !sector_div(block_size, n); | |
2750 | } | |
2751 | ||
9bc142dd MS |
2752 | /* |
2753 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2754 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2755 | */ |
0424caa1 | 2756 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2757 | { |
0424caa1 MS |
2758 | struct pool *pool = pt->pool; |
2759 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2760 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
0424caa1 | 2761 | const char *reason = NULL; |
9bc142dd MS |
2762 | char buf[BDEVNAME_SIZE]; |
2763 | ||
0424caa1 | 2764 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2765 | return; |
2766 | ||
0424caa1 MS |
2767 | if (!data_dev_supports_discard(pt)) |
2768 | reason = "discard unsupported"; | |
2769 | ||
2770 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2771 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2772 | |
0424caa1 MS |
2773 | if (reason) { |
2774 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2775 | pt->adjusted_pf.discard_passdown = false; | |
2776 | } | |
9bc142dd MS |
2777 | } |
2778 | ||
991d9fa0 JT |
2779 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2780 | { | |
2781 | struct pool_c *pt = ti->private; | |
2782 | ||
e49e5829 | 2783 | /* |
9b7aaa64 | 2784 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2785 | */ |
07f2b6e0 | 2786 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2787 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2788 | |
8b64e881 MS |
2789 | /* |
2790 | * Don't change the pool's mode until set_pool_mode() below. | |
2791 | * Otherwise the pool's process_* function pointers may | |
2792 | * not match the desired pool mode. | |
2793 | */ | |
2794 | pt->adjusted_pf.mode = old_mode; | |
2795 | ||
2796 | pool->ti = ti; | |
2797 | pool->pf = pt->adjusted_pf; | |
2798 | pool->low_water_blocks = pt->low_water_blocks; | |
2799 | ||
9bc142dd | 2800 | set_pool_mode(pool, new_mode); |
f402693d | 2801 | |
991d9fa0 JT |
2802 | return 0; |
2803 | } | |
2804 | ||
2805 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2806 | { | |
2807 | if (pool->ti == ti) | |
2808 | pool->ti = NULL; | |
2809 | } | |
2810 | ||
2811 | /*---------------------------------------------------------------- | |
2812 | * Pool creation | |
2813 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2814 | /* Initialize pool features. */ |
2815 | static void pool_features_init(struct pool_features *pf) | |
2816 | { | |
e49e5829 | 2817 | pf->mode = PM_WRITE; |
9bc142dd MS |
2818 | pf->zero_new_blocks = true; |
2819 | pf->discard_enabled = true; | |
2820 | pf->discard_passdown = true; | |
787a996c | 2821 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2822 | } |
2823 | ||
991d9fa0 JT |
2824 | static void __pool_destroy(struct pool *pool) |
2825 | { | |
2826 | __pool_table_remove(pool); | |
2827 | ||
a822c83e | 2828 | vfree(pool->cell_sort_array); |
991d9fa0 JT |
2829 | if (dm_pool_metadata_close(pool->pmd) < 0) |
2830 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2831 | ||
44feb387 | 2832 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2833 | dm_kcopyd_client_destroy(pool->copier); |
2834 | ||
2835 | if (pool->wq) | |
2836 | destroy_workqueue(pool->wq); | |
2837 | ||
2838 | if (pool->next_mapping) | |
6f1c819c KO |
2839 | mempool_free(pool->next_mapping, &pool->mapping_pool); |
2840 | mempool_exit(&pool->mapping_pool); | |
44feb387 MS |
2841 | dm_deferred_set_destroy(pool->shared_read_ds); |
2842 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2843 | kfree(pool); |
2844 | } | |
2845 | ||
a24c2569 | 2846 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2847 | |
991d9fa0 JT |
2848 | static struct pool *pool_create(struct mapped_device *pool_md, |
2849 | struct block_device *metadata_dev, | |
e49e5829 JT |
2850 | unsigned long block_size, |
2851 | int read_only, char **error) | |
991d9fa0 JT |
2852 | { |
2853 | int r; | |
2854 | void *err_p; | |
2855 | struct pool *pool; | |
2856 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2857 | bool format_device = read_only ? false : true; |
991d9fa0 | 2858 | |
e49e5829 | 2859 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2860 | if (IS_ERR(pmd)) { |
2861 | *error = "Error creating metadata object"; | |
2862 | return (struct pool *)pmd; | |
2863 | } | |
2864 | ||
d3775354 | 2865 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
991d9fa0 JT |
2866 | if (!pool) { |
2867 | *error = "Error allocating memory for pool"; | |
2868 | err_p = ERR_PTR(-ENOMEM); | |
2869 | goto bad_pool; | |
2870 | } | |
2871 | ||
2872 | pool->pmd = pmd; | |
2873 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2874 | if (block_size & (block_size - 1)) |
2875 | pool->sectors_per_block_shift = -1; | |
2876 | else | |
2877 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2878 | pool->low_water_blocks = 0; |
67e2e2b2 | 2879 | pool_features_init(&pool->pf); |
a195db2d | 2880 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2881 | if (!pool->prison) { |
2882 | *error = "Error creating pool's bio prison"; | |
2883 | err_p = ERR_PTR(-ENOMEM); | |
2884 | goto bad_prison; | |
2885 | } | |
2886 | ||
df5d2e90 | 2887 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2888 | if (IS_ERR(pool->copier)) { |
2889 | r = PTR_ERR(pool->copier); | |
2890 | *error = "Error creating pool's kcopyd client"; | |
2891 | err_p = ERR_PTR(r); | |
2892 | goto bad_kcopyd_client; | |
2893 | } | |
2894 | ||
2895 | /* | |
2896 | * Create singlethreaded workqueue that will service all devices | |
2897 | * that use this metadata. | |
2898 | */ | |
2899 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2900 | if (!pool->wq) { | |
2901 | *error = "Error creating pool's workqueue"; | |
2902 | err_p = ERR_PTR(-ENOMEM); | |
2903 | goto bad_wq; | |
2904 | } | |
2905 | ||
7d327fe0 | 2906 | throttle_init(&pool->throttle); |
991d9fa0 | 2907 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 2908 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2909 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2910 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2911 | bio_list_init(&pool->deferred_flush_bios); |
2912 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2913 | INIT_LIST_HEAD(&pool->prepared_discards); |
2a0fbffb | 2914 | INIT_LIST_HEAD(&pool->prepared_discards_pt2); |
c140e1c4 | 2915 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2916 | pool->low_water_triggered = false; |
80e96c54 | 2917 | pool->suspended = true; |
c3667cc6 | 2918 | pool->out_of_data_space = false; |
44feb387 MS |
2919 | |
2920 | pool->shared_read_ds = dm_deferred_set_create(); | |
2921 | if (!pool->shared_read_ds) { | |
2922 | *error = "Error creating pool's shared read deferred set"; | |
2923 | err_p = ERR_PTR(-ENOMEM); | |
2924 | goto bad_shared_read_ds; | |
2925 | } | |
2926 | ||
2927 | pool->all_io_ds = dm_deferred_set_create(); | |
2928 | if (!pool->all_io_ds) { | |
2929 | *error = "Error creating pool's all io deferred set"; | |
2930 | err_p = ERR_PTR(-ENOMEM); | |
2931 | goto bad_all_io_ds; | |
2932 | } | |
991d9fa0 JT |
2933 | |
2934 | pool->next_mapping = NULL; | |
6f1c819c KO |
2935 | r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE, |
2936 | _new_mapping_cache); | |
2937 | if (r) { | |
991d9fa0 | 2938 | *error = "Error creating pool's mapping mempool"; |
6f1c819c | 2939 | err_p = ERR_PTR(r); |
991d9fa0 JT |
2940 | goto bad_mapping_pool; |
2941 | } | |
2942 | ||
42bc47b3 KC |
2943 | pool->cell_sort_array = |
2944 | vmalloc(array_size(CELL_SORT_ARRAY_SIZE, | |
2945 | sizeof(*pool->cell_sort_array))); | |
a822c83e JT |
2946 | if (!pool->cell_sort_array) { |
2947 | *error = "Error allocating cell sort array"; | |
2948 | err_p = ERR_PTR(-ENOMEM); | |
2949 | goto bad_sort_array; | |
2950 | } | |
2951 | ||
991d9fa0 | 2952 | pool->ref_count = 1; |
905e51b3 | 2953 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2954 | pool->pool_md = pool_md; |
2955 | pool->md_dev = metadata_dev; | |
2956 | __pool_table_insert(pool); | |
2957 | ||
2958 | return pool; | |
2959 | ||
a822c83e | 2960 | bad_sort_array: |
6f1c819c | 2961 | mempool_exit(&pool->mapping_pool); |
991d9fa0 | 2962 | bad_mapping_pool: |
44feb387 MS |
2963 | dm_deferred_set_destroy(pool->all_io_ds); |
2964 | bad_all_io_ds: | |
2965 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2966 | bad_shared_read_ds: | |
991d9fa0 JT |
2967 | destroy_workqueue(pool->wq); |
2968 | bad_wq: | |
2969 | dm_kcopyd_client_destroy(pool->copier); | |
2970 | bad_kcopyd_client: | |
44feb387 | 2971 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2972 | bad_prison: |
2973 | kfree(pool); | |
2974 | bad_pool: | |
2975 | if (dm_pool_metadata_close(pmd)) | |
2976 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2977 | ||
2978 | return err_p; | |
2979 | } | |
2980 | ||
2981 | static void __pool_inc(struct pool *pool) | |
2982 | { | |
2983 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2984 | pool->ref_count++; | |
2985 | } | |
2986 | ||
2987 | static void __pool_dec(struct pool *pool) | |
2988 | { | |
2989 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2990 | BUG_ON(!pool->ref_count); | |
2991 | if (!--pool->ref_count) | |
2992 | __pool_destroy(pool); | |
2993 | } | |
2994 | ||
2995 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2996 | struct block_device *metadata_dev, | |
e49e5829 JT |
2997 | unsigned long block_size, int read_only, |
2998 | char **error, int *created) | |
991d9fa0 JT |
2999 | { |
3000 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
3001 | ||
3002 | if (pool) { | |
f09996c9 MS |
3003 | if (pool->pool_md != pool_md) { |
3004 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 3005 | return ERR_PTR(-EBUSY); |
f09996c9 | 3006 | } |
991d9fa0 JT |
3007 | __pool_inc(pool); |
3008 | ||
3009 | } else { | |
3010 | pool = __pool_table_lookup(pool_md); | |
3011 | if (pool) { | |
f09996c9 MS |
3012 | if (pool->md_dev != metadata_dev) { |
3013 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 3014 | return ERR_PTR(-EINVAL); |
f09996c9 | 3015 | } |
991d9fa0 JT |
3016 | __pool_inc(pool); |
3017 | ||
67e2e2b2 | 3018 | } else { |
e49e5829 | 3019 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
3020 | *created = 1; |
3021 | } | |
991d9fa0 JT |
3022 | } |
3023 | ||
3024 | return pool; | |
3025 | } | |
3026 | ||
3027 | /*---------------------------------------------------------------- | |
3028 | * Pool target methods | |
3029 | *--------------------------------------------------------------*/ | |
3030 | static void pool_dtr(struct dm_target *ti) | |
3031 | { | |
3032 | struct pool_c *pt = ti->private; | |
3033 | ||
3034 | mutex_lock(&dm_thin_pool_table.mutex); | |
3035 | ||
3036 | unbind_control_target(pt->pool, ti); | |
3037 | __pool_dec(pt->pool); | |
3038 | dm_put_device(ti, pt->metadata_dev); | |
3039 | dm_put_device(ti, pt->data_dev); | |
3040 | kfree(pt); | |
3041 | ||
3042 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3043 | } | |
3044 | ||
991d9fa0 JT |
3045 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
3046 | struct dm_target *ti) | |
3047 | { | |
3048 | int r; | |
3049 | unsigned argc; | |
3050 | const char *arg_name; | |
3051 | ||
5916a22b | 3052 | static const struct dm_arg _args[] = { |
74aa45c3 | 3053 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
3054 | }; |
3055 | ||
3056 | /* | |
3057 | * No feature arguments supplied. | |
3058 | */ | |
3059 | if (!as->argc) | |
3060 | return 0; | |
3061 | ||
3062 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
3063 | if (r) | |
3064 | return -EINVAL; | |
3065 | ||
3066 | while (argc && !r) { | |
3067 | arg_name = dm_shift_arg(as); | |
3068 | argc--; | |
3069 | ||
e49e5829 | 3070 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 3071 | pf->zero_new_blocks = false; |
e49e5829 JT |
3072 | |
3073 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 3074 | pf->discard_enabled = false; |
e49e5829 JT |
3075 | |
3076 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 3077 | pf->discard_passdown = false; |
991d9fa0 | 3078 | |
e49e5829 JT |
3079 | else if (!strcasecmp(arg_name, "read_only")) |
3080 | pf->mode = PM_READ_ONLY; | |
3081 | ||
787a996c MS |
3082 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
3083 | pf->error_if_no_space = true; | |
3084 | ||
e49e5829 JT |
3085 | else { |
3086 | ti->error = "Unrecognised pool feature requested"; | |
3087 | r = -EINVAL; | |
3088 | break; | |
3089 | } | |
991d9fa0 JT |
3090 | } |
3091 | ||
3092 | return r; | |
3093 | } | |
3094 | ||
ac8c3f3d JT |
3095 | static void metadata_low_callback(void *context) |
3096 | { | |
3097 | struct pool *pool = context; | |
3098 | ||
3099 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
3100 | dm_device_name(pool->pool_md)); | |
3101 | ||
3102 | dm_table_event(pool->ti->table); | |
3103 | } | |
3104 | ||
7d48935e MS |
3105 | static sector_t get_dev_size(struct block_device *bdev) |
3106 | { | |
3107 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
3108 | } | |
3109 | ||
3110 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 3111 | { |
7d48935e | 3112 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
3113 | char buffer[BDEVNAME_SIZE]; |
3114 | ||
7d48935e | 3115 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
3116 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
3117 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
3118 | } |
3119 | ||
3120 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
3121 | { | |
3122 | sector_t metadata_dev_size = get_dev_size(bdev); | |
3123 | ||
3124 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
3125 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
3126 | |
3127 | return metadata_dev_size; | |
3128 | } | |
3129 | ||
24347e95 JT |
3130 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
3131 | { | |
3132 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
3133 | ||
7d48935e | 3134 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
3135 | |
3136 | return metadata_dev_size; | |
3137 | } | |
3138 | ||
ac8c3f3d JT |
3139 | /* |
3140 | * When a metadata threshold is crossed a dm event is triggered, and | |
3141 | * userland should respond by growing the metadata device. We could let | |
3142 | * userland set the threshold, like we do with the data threshold, but I'm | |
3143 | * not sure they know enough to do this well. | |
3144 | */ | |
3145 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
3146 | { | |
3147 | /* | |
3148 | * 4M is ample for all ops with the possible exception of thin | |
3149 | * device deletion which is harmless if it fails (just retry the | |
3150 | * delete after you've grown the device). | |
3151 | */ | |
3152 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
3153 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
3154 | } | |
3155 | ||
991d9fa0 JT |
3156 | /* |
3157 | * thin-pool <metadata dev> <data dev> | |
3158 | * <data block size (sectors)> | |
3159 | * <low water mark (blocks)> | |
3160 | * [<#feature args> [<arg>]*] | |
3161 | * | |
3162 | * Optional feature arguments are: | |
3163 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
3164 | * ignore_discard: disable discard |
3165 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
3166 | * read_only: Don't allow any changes to be made to the pool metadata. |
3167 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
3168 | */ |
3169 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3170 | { | |
67e2e2b2 | 3171 | int r, pool_created = 0; |
991d9fa0 JT |
3172 | struct pool_c *pt; |
3173 | struct pool *pool; | |
3174 | struct pool_features pf; | |
3175 | struct dm_arg_set as; | |
3176 | struct dm_dev *data_dev; | |
3177 | unsigned long block_size; | |
3178 | dm_block_t low_water_blocks; | |
3179 | struct dm_dev *metadata_dev; | |
5d0db96d | 3180 | fmode_t metadata_mode; |
991d9fa0 JT |
3181 | |
3182 | /* | |
3183 | * FIXME Remove validation from scope of lock. | |
3184 | */ | |
3185 | mutex_lock(&dm_thin_pool_table.mutex); | |
3186 | ||
3187 | if (argc < 4) { | |
3188 | ti->error = "Invalid argument count"; | |
3189 | r = -EINVAL; | |
3190 | goto out_unlock; | |
3191 | } | |
5d0db96d | 3192 | |
991d9fa0 JT |
3193 | as.argc = argc; |
3194 | as.argv = argv; | |
3195 | ||
5d0db96d JT |
3196 | /* |
3197 | * Set default pool features. | |
3198 | */ | |
3199 | pool_features_init(&pf); | |
3200 | ||
3201 | dm_consume_args(&as, 4); | |
3202 | r = parse_pool_features(&as, &pf, ti); | |
3203 | if (r) | |
3204 | goto out_unlock; | |
3205 | ||
3206 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
3207 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
3208 | if (r) { |
3209 | ti->error = "Error opening metadata block device"; | |
3210 | goto out_unlock; | |
3211 | } | |
7d48935e | 3212 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
3213 | |
3214 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
3215 | if (r) { | |
3216 | ti->error = "Error getting data device"; | |
3217 | goto out_metadata; | |
3218 | } | |
3219 | ||
3220 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
3221 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
3222 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 3223 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
3224 | ti->error = "Invalid block size"; |
3225 | r = -EINVAL; | |
3226 | goto out; | |
3227 | } | |
3228 | ||
3229 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
3230 | ti->error = "Invalid low water mark"; | |
3231 | r = -EINVAL; | |
3232 | goto out; | |
3233 | } | |
3234 | ||
991d9fa0 JT |
3235 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
3236 | if (!pt) { | |
3237 | r = -ENOMEM; | |
3238 | goto out; | |
3239 | } | |
3240 | ||
3241 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 3242 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
3243 | if (IS_ERR(pool)) { |
3244 | r = PTR_ERR(pool); | |
3245 | goto out_free_pt; | |
3246 | } | |
3247 | ||
67e2e2b2 JT |
3248 | /* |
3249 | * 'pool_created' reflects whether this is the first table load. | |
3250 | * Top level discard support is not allowed to be changed after | |
3251 | * initial load. This would require a pool reload to trigger thin | |
3252 | * device changes. | |
3253 | */ | |
3254 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
3255 | ti->error = "Discard support cannot be disabled once enabled"; | |
3256 | r = -EINVAL; | |
3257 | goto out_flags_changed; | |
3258 | } | |
3259 | ||
991d9fa0 JT |
3260 | pt->pool = pool; |
3261 | pt->ti = ti; | |
3262 | pt->metadata_dev = metadata_dev; | |
3263 | pt->data_dev = data_dev; | |
3264 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 3265 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 3266 | ti->num_flush_bios = 1; |
9bc142dd | 3267 | |
67e2e2b2 JT |
3268 | /* |
3269 | * Only need to enable discards if the pool should pass | |
3270 | * them down to the data device. The thin device's discard | |
3271 | * processing will cause mappings to be removed from the btree. | |
3272 | */ | |
3273 | if (pf.discard_enabled && pf.discard_passdown) { | |
55a62eef | 3274 | ti->num_discard_bios = 1; |
9bc142dd | 3275 | |
67e2e2b2 JT |
3276 | /* |
3277 | * Setting 'discards_supported' circumvents the normal | |
3278 | * stacking of discard limits (this keeps the pool and | |
3279 | * thin devices' discard limits consistent). | |
3280 | */ | |
0ac55489 | 3281 | ti->discards_supported = true; |
67e2e2b2 | 3282 | } |
991d9fa0 JT |
3283 | ti->private = pt; |
3284 | ||
ac8c3f3d JT |
3285 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
3286 | calc_metadata_threshold(pt), | |
3287 | metadata_low_callback, | |
3288 | pool); | |
3289 | if (r) | |
ba30670f | 3290 | goto out_flags_changed; |
ac8c3f3d | 3291 | |
991d9fa0 JT |
3292 | pt->callbacks.congested_fn = pool_is_congested; |
3293 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
3294 | ||
3295 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3296 | ||
3297 | return 0; | |
3298 | ||
67e2e2b2 JT |
3299 | out_flags_changed: |
3300 | __pool_dec(pool); | |
991d9fa0 JT |
3301 | out_free_pt: |
3302 | kfree(pt); | |
3303 | out: | |
3304 | dm_put_device(ti, data_dev); | |
3305 | out_metadata: | |
3306 | dm_put_device(ti, metadata_dev); | |
3307 | out_unlock: | |
3308 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3309 | ||
3310 | return r; | |
3311 | } | |
3312 | ||
7de3ee57 | 3313 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
3314 | { |
3315 | int r; | |
3316 | struct pool_c *pt = ti->private; | |
3317 | struct pool *pool = pt->pool; | |
3318 | unsigned long flags; | |
3319 | ||
3320 | /* | |
3321 | * As this is a singleton target, ti->begin is always zero. | |
3322 | */ | |
3323 | spin_lock_irqsave(&pool->lock, flags); | |
74d46992 | 3324 | bio_set_dev(bio, pt->data_dev->bdev); |
991d9fa0 JT |
3325 | r = DM_MAPIO_REMAPPED; |
3326 | spin_unlock_irqrestore(&pool->lock, flags); | |
3327 | ||
3328 | return r; | |
3329 | } | |
3330 | ||
b17446df | 3331 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
3332 | { |
3333 | int r; | |
3334 | struct pool_c *pt = ti->private; | |
3335 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
3336 | sector_t data_size = ti->len; |
3337 | dm_block_t sb_data_size; | |
991d9fa0 | 3338 | |
b17446df | 3339 | *need_commit = false; |
991d9fa0 | 3340 | |
55f2b8bd MS |
3341 | (void) sector_div(data_size, pool->sectors_per_block); |
3342 | ||
991d9fa0 JT |
3343 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
3344 | if (r) { | |
4fa5971a MS |
3345 | DMERR("%s: failed to retrieve data device size", |
3346 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
3347 | return r; |
3348 | } | |
3349 | ||
3350 | if (data_size < sb_data_size) { | |
4fa5971a MS |
3351 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
3352 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3353 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3354 | return -EINVAL; |
3355 | ||
3356 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3357 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3358 | DMERR("%s: unable to grow the data device until repaired.", | |
3359 | dm_device_name(pool->pool_md)); | |
3360 | return 0; | |
3361 | } | |
3362 | ||
6f7f51d4 MS |
3363 | if (sb_data_size) |
3364 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3365 | dm_device_name(pool->pool_md), | |
3366 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3367 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3368 | if (r) { | |
b5330655 | 3369 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3370 | return r; |
3371 | } | |
3372 | ||
b17446df | 3373 | *need_commit = true; |
991d9fa0 JT |
3374 | } |
3375 | ||
3376 | return 0; | |
3377 | } | |
3378 | ||
24347e95 JT |
3379 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3380 | { | |
3381 | int r; | |
3382 | struct pool_c *pt = ti->private; | |
3383 | struct pool *pool = pt->pool; | |
3384 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3385 | ||
3386 | *need_commit = false; | |
3387 | ||
610bba8b | 3388 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3389 | |
3390 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3391 | if (r) { | |
4fa5971a MS |
3392 | DMERR("%s: failed to retrieve metadata device size", |
3393 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3394 | return r; |
3395 | } | |
3396 | ||
3397 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3398 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3399 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3400 | metadata_dev_size, sb_metadata_dev_size); |
3401 | return -EINVAL; | |
3402 | ||
3403 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3404 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3405 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3406 | dm_device_name(pool->pool_md)); | |
3407 | return 0; | |
3408 | } | |
3409 | ||
7d48935e | 3410 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3411 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3412 | dm_device_name(pool->pool_md), | |
3413 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
3414 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3415 | if (r) { | |
b5330655 | 3416 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3417 | return r; |
3418 | } | |
3419 | ||
3420 | *need_commit = true; | |
3421 | } | |
3422 | ||
3423 | return 0; | |
3424 | } | |
3425 | ||
b17446df JT |
3426 | /* |
3427 | * Retrieves the number of blocks of the data device from | |
3428 | * the superblock and compares it to the actual device size, | |
3429 | * thus resizing the data device in case it has grown. | |
3430 | * | |
3431 | * This both copes with opening preallocated data devices in the ctr | |
3432 | * being followed by a resume | |
3433 | * -and- | |
3434 | * calling the resume method individually after userspace has | |
3435 | * grown the data device in reaction to a table event. | |
3436 | */ | |
3437 | static int pool_preresume(struct dm_target *ti) | |
3438 | { | |
3439 | int r; | |
24347e95 | 3440 | bool need_commit1, need_commit2; |
b17446df JT |
3441 | struct pool_c *pt = ti->private; |
3442 | struct pool *pool = pt->pool; | |
3443 | ||
3444 | /* | |
3445 | * Take control of the pool object. | |
3446 | */ | |
3447 | r = bind_control_target(pool, ti); | |
3448 | if (r) | |
3449 | return r; | |
3450 | ||
3451 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3452 | if (r) | |
3453 | return r; | |
3454 | ||
24347e95 JT |
3455 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3456 | if (r) | |
3457 | return r; | |
3458 | ||
3459 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3460 | (void) commit(pool); |
b17446df JT |
3461 | |
3462 | return 0; | |
3463 | } | |
3464 | ||
583024d2 MS |
3465 | static void pool_suspend_active_thins(struct pool *pool) |
3466 | { | |
3467 | struct thin_c *tc; | |
3468 | ||
3469 | /* Suspend all active thin devices */ | |
3470 | tc = get_first_thin(pool); | |
3471 | while (tc) { | |
3472 | dm_internal_suspend_noflush(tc->thin_md); | |
3473 | tc = get_next_thin(pool, tc); | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | static void pool_resume_active_thins(struct pool *pool) | |
3478 | { | |
3479 | struct thin_c *tc; | |
3480 | ||
3481 | /* Resume all active thin devices */ | |
3482 | tc = get_first_thin(pool); | |
3483 | while (tc) { | |
3484 | dm_internal_resume(tc->thin_md); | |
3485 | tc = get_next_thin(pool, tc); | |
3486 | } | |
3487 | } | |
3488 | ||
991d9fa0 JT |
3489 | static void pool_resume(struct dm_target *ti) |
3490 | { | |
3491 | struct pool_c *pt = ti->private; | |
3492 | struct pool *pool = pt->pool; | |
3493 | unsigned long flags; | |
3494 | ||
583024d2 MS |
3495 | /* |
3496 | * Must requeue active_thins' bios and then resume | |
3497 | * active_thins _before_ clearing 'suspend' flag. | |
3498 | */ | |
3499 | requeue_bios(pool); | |
3500 | pool_resume_active_thins(pool); | |
3501 | ||
991d9fa0 | 3502 | spin_lock_irqsave(&pool->lock, flags); |
88a6621b | 3503 | pool->low_water_triggered = false; |
80e96c54 | 3504 | pool->suspended = false; |
991d9fa0 | 3505 | spin_unlock_irqrestore(&pool->lock, flags); |
80e96c54 | 3506 | |
905e51b3 | 3507 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3508 | } |
3509 | ||
80e96c54 MS |
3510 | static void pool_presuspend(struct dm_target *ti) |
3511 | { | |
3512 | struct pool_c *pt = ti->private; | |
3513 | struct pool *pool = pt->pool; | |
3514 | unsigned long flags; | |
3515 | ||
3516 | spin_lock_irqsave(&pool->lock, flags); | |
3517 | pool->suspended = true; | |
3518 | spin_unlock_irqrestore(&pool->lock, flags); | |
583024d2 MS |
3519 | |
3520 | pool_suspend_active_thins(pool); | |
80e96c54 MS |
3521 | } |
3522 | ||
3523 | static void pool_presuspend_undo(struct dm_target *ti) | |
3524 | { | |
3525 | struct pool_c *pt = ti->private; | |
3526 | struct pool *pool = pt->pool; | |
3527 | unsigned long flags; | |
3528 | ||
583024d2 MS |
3529 | pool_resume_active_thins(pool); |
3530 | ||
80e96c54 MS |
3531 | spin_lock_irqsave(&pool->lock, flags); |
3532 | pool->suspended = false; | |
3533 | spin_unlock_irqrestore(&pool->lock, flags); | |
3534 | } | |
3535 | ||
991d9fa0 JT |
3536 | static void pool_postsuspend(struct dm_target *ti) |
3537 | { | |
991d9fa0 JT |
3538 | struct pool_c *pt = ti->private; |
3539 | struct pool *pool = pt->pool; | |
3540 | ||
18d03e8c NB |
3541 | cancel_delayed_work_sync(&pool->waker); |
3542 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
991d9fa0 | 3543 | flush_workqueue(pool->wq); |
020cc3b5 | 3544 | (void) commit(pool); |
991d9fa0 JT |
3545 | } |
3546 | ||
3547 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3548 | { | |
3549 | if (argc != args_required) { | |
3550 | DMWARN("Message received with %u arguments instead of %u.", | |
3551 | argc, args_required); | |
3552 | return -EINVAL; | |
3553 | } | |
3554 | ||
3555 | return 0; | |
3556 | } | |
3557 | ||
3558 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3559 | { | |
3560 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3561 | *dev_id <= MAX_DEV_ID) | |
3562 | return 0; | |
3563 | ||
3564 | if (warning) | |
3565 | DMWARN("Message received with invalid device id: %s", arg); | |
3566 | ||
3567 | return -EINVAL; | |
3568 | } | |
3569 | ||
3570 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3571 | { | |
3572 | dm_thin_id dev_id; | |
3573 | int r; | |
3574 | ||
3575 | r = check_arg_count(argc, 2); | |
3576 | if (r) | |
3577 | return r; | |
3578 | ||
3579 | r = read_dev_id(argv[1], &dev_id, 1); | |
3580 | if (r) | |
3581 | return r; | |
3582 | ||
3583 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3584 | if (r) { | |
3585 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3586 | argv[1]); | |
3587 | return r; | |
3588 | } | |
3589 | ||
3590 | return 0; | |
3591 | } | |
3592 | ||
3593 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3594 | { | |
3595 | dm_thin_id dev_id; | |
3596 | dm_thin_id origin_dev_id; | |
3597 | int r; | |
3598 | ||
3599 | r = check_arg_count(argc, 3); | |
3600 | if (r) | |
3601 | return r; | |
3602 | ||
3603 | r = read_dev_id(argv[1], &dev_id, 1); | |
3604 | if (r) | |
3605 | return r; | |
3606 | ||
3607 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3608 | if (r) | |
3609 | return r; | |
3610 | ||
3611 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3612 | if (r) { | |
3613 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3614 | argv[1], argv[2]); | |
3615 | return r; | |
3616 | } | |
3617 | ||
3618 | return 0; | |
3619 | } | |
3620 | ||
3621 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3622 | { | |
3623 | dm_thin_id dev_id; | |
3624 | int r; | |
3625 | ||
3626 | r = check_arg_count(argc, 2); | |
3627 | if (r) | |
3628 | return r; | |
3629 | ||
3630 | r = read_dev_id(argv[1], &dev_id, 1); | |
3631 | if (r) | |
3632 | return r; | |
3633 | ||
3634 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3635 | if (r) | |
3636 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3637 | ||
3638 | return r; | |
3639 | } | |
3640 | ||
3641 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3642 | { | |
3643 | dm_thin_id old_id, new_id; | |
3644 | int r; | |
3645 | ||
3646 | r = check_arg_count(argc, 3); | |
3647 | if (r) | |
3648 | return r; | |
3649 | ||
3650 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3651 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3652 | return -EINVAL; | |
3653 | } | |
3654 | ||
3655 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3656 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3657 | return -EINVAL; | |
3658 | } | |
3659 | ||
3660 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3661 | if (r) { | |
3662 | DMWARN("Failed to change transaction id from %s to %s.", | |
3663 | argv[1], argv[2]); | |
3664 | return r; | |
3665 | } | |
3666 | ||
3667 | return 0; | |
3668 | } | |
3669 | ||
cc8394d8 JT |
3670 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3671 | { | |
3672 | int r; | |
3673 | ||
3674 | r = check_arg_count(argc, 1); | |
3675 | if (r) | |
3676 | return r; | |
3677 | ||
020cc3b5 | 3678 | (void) commit(pool); |
0d200aef | 3679 | |
cc8394d8 JT |
3680 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3681 | if (r) | |
3682 | DMWARN("reserve_metadata_snap message failed."); | |
3683 | ||
3684 | return r; | |
3685 | } | |
3686 | ||
3687 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3688 | { | |
3689 | int r; | |
3690 | ||
3691 | r = check_arg_count(argc, 1); | |
3692 | if (r) | |
3693 | return r; | |
3694 | ||
3695 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3696 | if (r) | |
3697 | DMWARN("release_metadata_snap message failed."); | |
3698 | ||
3699 | return r; | |
3700 | } | |
3701 | ||
991d9fa0 JT |
3702 | /* |
3703 | * Messages supported: | |
3704 | * create_thin <dev_id> | |
3705 | * create_snap <dev_id> <origin_id> | |
3706 | * delete <dev_id> | |
991d9fa0 | 3707 | * set_transaction_id <current_trans_id> <new_trans_id> |
cc8394d8 JT |
3708 | * reserve_metadata_snap |
3709 | * release_metadata_snap | |
991d9fa0 | 3710 | */ |
1eb5fa84 MS |
3711 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv, |
3712 | char *result, unsigned maxlen) | |
991d9fa0 JT |
3713 | { |
3714 | int r = -EINVAL; | |
3715 | struct pool_c *pt = ti->private; | |
3716 | struct pool *pool = pt->pool; | |
3717 | ||
2a7eaea0 JT |
3718 | if (get_pool_mode(pool) >= PM_READ_ONLY) { |
3719 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", | |
3720 | dm_device_name(pool->pool_md)); | |
fd467696 | 3721 | return -EOPNOTSUPP; |
2a7eaea0 JT |
3722 | } |
3723 | ||
991d9fa0 JT |
3724 | if (!strcasecmp(argv[0], "create_thin")) |
3725 | r = process_create_thin_mesg(argc, argv, pool); | |
3726 | ||
3727 | else if (!strcasecmp(argv[0], "create_snap")) | |
3728 | r = process_create_snap_mesg(argc, argv, pool); | |
3729 | ||
3730 | else if (!strcasecmp(argv[0], "delete")) | |
3731 | r = process_delete_mesg(argc, argv, pool); | |
3732 | ||
3733 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3734 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3735 | ||
cc8394d8 JT |
3736 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3737 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3738 | ||
3739 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3740 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3741 | ||
991d9fa0 JT |
3742 | else |
3743 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3744 | ||
e49e5829 | 3745 | if (!r) |
020cc3b5 | 3746 | (void) commit(pool); |
991d9fa0 JT |
3747 | |
3748 | return r; | |
3749 | } | |
3750 | ||
e49e5829 JT |
3751 | static void emit_flags(struct pool_features *pf, char *result, |
3752 | unsigned sz, unsigned maxlen) | |
3753 | { | |
3754 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3755 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3756 | pf->error_if_no_space; | |
e49e5829 JT |
3757 | DMEMIT("%u ", count); |
3758 | ||
3759 | if (!pf->zero_new_blocks) | |
3760 | DMEMIT("skip_block_zeroing "); | |
3761 | ||
3762 | if (!pf->discard_enabled) | |
3763 | DMEMIT("ignore_discard "); | |
3764 | ||
3765 | if (!pf->discard_passdown) | |
3766 | DMEMIT("no_discard_passdown "); | |
3767 | ||
3768 | if (pf->mode == PM_READ_ONLY) | |
3769 | DMEMIT("read_only "); | |
787a996c MS |
3770 | |
3771 | if (pf->error_if_no_space) | |
3772 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3773 | } |
3774 | ||
991d9fa0 JT |
3775 | /* |
3776 | * Status line is: | |
3777 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3778 | * <used data sectors>/<total data sectors> <held metadata root> | |
e4c78e21 | 3779 | * <pool mode> <discard config> <no space config> <needs_check> |
991d9fa0 | 3780 | */ |
fd7c092e MP |
3781 | static void pool_status(struct dm_target *ti, status_type_t type, |
3782 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3783 | { |
e49e5829 | 3784 | int r; |
991d9fa0 JT |
3785 | unsigned sz = 0; |
3786 | uint64_t transaction_id; | |
3787 | dm_block_t nr_free_blocks_data; | |
3788 | dm_block_t nr_free_blocks_metadata; | |
3789 | dm_block_t nr_blocks_data; | |
3790 | dm_block_t nr_blocks_metadata; | |
3791 | dm_block_t held_root; | |
3792 | char buf[BDEVNAME_SIZE]; | |
3793 | char buf2[BDEVNAME_SIZE]; | |
3794 | struct pool_c *pt = ti->private; | |
3795 | struct pool *pool = pt->pool; | |
3796 | ||
3797 | switch (type) { | |
3798 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3799 | if (get_pool_mode(pool) == PM_FAIL) { |
3800 | DMEMIT("Fail"); | |
3801 | break; | |
3802 | } | |
3803 | ||
1f4e0ff0 AK |
3804 | /* Commit to ensure statistics aren't out-of-date */ |
3805 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3806 | (void) commit(pool); |
1f4e0ff0 | 3807 | |
fd7c092e MP |
3808 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3809 | if (r) { | |
4fa5971a MS |
3810 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3811 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3812 | goto err; |
3813 | } | |
991d9fa0 | 3814 | |
fd7c092e MP |
3815 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3816 | if (r) { | |
4fa5971a MS |
3817 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3818 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3819 | goto err; |
3820 | } | |
991d9fa0 JT |
3821 | |
3822 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3823 | if (r) { |
4fa5971a MS |
3824 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3825 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3826 | goto err; |
3827 | } | |
991d9fa0 | 3828 | |
fd7c092e MP |
3829 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3830 | if (r) { | |
4fa5971a MS |
3831 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3832 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3833 | goto err; |
3834 | } | |
991d9fa0 JT |
3835 | |
3836 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3837 | if (r) { |
4fa5971a MS |
3838 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3839 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3840 | goto err; |
3841 | } | |
991d9fa0 | 3842 | |
cc8394d8 | 3843 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3844 | if (r) { |
4fa5971a MS |
3845 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3846 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3847 | goto err; |
3848 | } | |
991d9fa0 JT |
3849 | |
3850 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3851 | (unsigned long long)transaction_id, | |
3852 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3853 | (unsigned long long)nr_blocks_metadata, | |
3854 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3855 | (unsigned long long)nr_blocks_data); | |
3856 | ||
3857 | if (held_root) | |
e49e5829 JT |
3858 | DMEMIT("%llu ", held_root); |
3859 | else | |
3860 | DMEMIT("- "); | |
3861 | ||
3e1a0699 JT |
3862 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3863 | DMEMIT("out_of_data_space "); | |
3864 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3865 | DMEMIT("ro "); |
991d9fa0 | 3866 | else |
e49e5829 JT |
3867 | DMEMIT("rw "); |
3868 | ||
018debea | 3869 | if (!pool->pf.discard_enabled) |
787a996c | 3870 | DMEMIT("ignore_discard "); |
018debea | 3871 | else if (pool->pf.discard_passdown) |
787a996c MS |
3872 | DMEMIT("discard_passdown "); |
3873 | else | |
3874 | DMEMIT("no_discard_passdown "); | |
3875 | ||
3876 | if (pool->pf.error_if_no_space) | |
3877 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3878 | else |
787a996c | 3879 | DMEMIT("queue_if_no_space "); |
991d9fa0 | 3880 | |
e4c78e21 MS |
3881 | if (dm_pool_metadata_needs_check(pool->pmd)) |
3882 | DMEMIT("needs_check "); | |
3883 | else | |
3884 | DMEMIT("- "); | |
3885 | ||
991d9fa0 JT |
3886 | break; |
3887 | ||
3888 | case STATUSTYPE_TABLE: | |
3889 | DMEMIT("%s %s %lu %llu ", | |
3890 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3891 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3892 | (unsigned long)pool->sectors_per_block, | |
3893 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3894 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3895 | break; |
3896 | } | |
fd7c092e | 3897 | return; |
991d9fa0 | 3898 | |
fd7c092e MP |
3899 | err: |
3900 | DMEMIT("Error"); | |
991d9fa0 JT |
3901 | } |
3902 | ||
3903 | static int pool_iterate_devices(struct dm_target *ti, | |
3904 | iterate_devices_callout_fn fn, void *data) | |
3905 | { | |
3906 | struct pool_c *pt = ti->private; | |
3907 | ||
3908 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3909 | } | |
3910 | ||
991d9fa0 JT |
3911 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3912 | { | |
3913 | struct pool_c *pt = ti->private; | |
3914 | struct pool *pool = pt->pool; | |
604ea906 MS |
3915 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
3916 | ||
3917 | /* | |
d200c30e MS |
3918 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
3919 | * to the highest possible power-of-2 factor of pool->sectors_per_block. | |
3920 | * This is especially beneficial when the pool's data device is a RAID | |
3921 | * device that has a full stripe width that matches pool->sectors_per_block | |
3922 | * -- because even though partial RAID stripe-sized IOs will be issued to a | |
3923 | * single RAID stripe; when aggregated they will end on a full RAID stripe | |
3924 | * boundary.. which avoids additional partial RAID stripe writes cascading | |
604ea906 | 3925 | */ |
604ea906 MS |
3926 | if (limits->max_sectors < pool->sectors_per_block) { |
3927 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
3928 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3929 | limits->max_sectors--; | |
3930 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3931 | } | |
604ea906 | 3932 | } |
991d9fa0 | 3933 | |
0cc67cd9 MS |
3934 | /* |
3935 | * If the system-determined stacked limits are compatible with the | |
3936 | * pool's blocksize (io_opt is a factor) do not override them. | |
3937 | */ | |
3938 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
3939 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
3940 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
3941 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
3942 | else | |
3943 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
3944 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
3945 | } | |
0424caa1 MS |
3946 | |
3947 | /* | |
3948 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3949 | * They get transferred to the live pool in bind_control_target() | |
3950 | * called from pool_preresume(). | |
3951 | */ | |
b60ab990 MS |
3952 | if (!pt->adjusted_pf.discard_enabled) { |
3953 | /* | |
3954 | * Must explicitly disallow stacking discard limits otherwise the | |
3955 | * block layer will stack them if pool's data device has support. | |
3956 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3957 | * user to see that, so make sure to set all discard limits to 0. | |
3958 | */ | |
3959 | limits->discard_granularity = 0; | |
0424caa1 | 3960 | return; |
b60ab990 | 3961 | } |
0424caa1 MS |
3962 | |
3963 | disable_passdown_if_not_supported(pt); | |
3964 | ||
34fbcf62 JT |
3965 | /* |
3966 | * The pool uses the same discard limits as the underlying data | |
3967 | * device. DM core has already set this up. | |
3968 | */ | |
991d9fa0 JT |
3969 | } |
3970 | ||
3971 | static struct target_type pool_target = { | |
3972 | .name = "thin-pool", | |
3973 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3974 | DM_TARGET_IMMUTABLE, | |
202bae52 | 3975 | .version = {1, 19, 0}, |
991d9fa0 JT |
3976 | .module = THIS_MODULE, |
3977 | .ctr = pool_ctr, | |
3978 | .dtr = pool_dtr, | |
3979 | .map = pool_map, | |
80e96c54 MS |
3980 | .presuspend = pool_presuspend, |
3981 | .presuspend_undo = pool_presuspend_undo, | |
991d9fa0 JT |
3982 | .postsuspend = pool_postsuspend, |
3983 | .preresume = pool_preresume, | |
3984 | .resume = pool_resume, | |
3985 | .message = pool_message, | |
3986 | .status = pool_status, | |
991d9fa0 JT |
3987 | .iterate_devices = pool_iterate_devices, |
3988 | .io_hints = pool_io_hints, | |
3989 | }; | |
3990 | ||
3991 | /*---------------------------------------------------------------- | |
3992 | * Thin target methods | |
3993 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3994 | static void thin_get(struct thin_c *tc) |
3995 | { | |
3996 | atomic_inc(&tc->refcount); | |
3997 | } | |
3998 | ||
3999 | static void thin_put(struct thin_c *tc) | |
4000 | { | |
4001 | if (atomic_dec_and_test(&tc->refcount)) | |
4002 | complete(&tc->can_destroy); | |
4003 | } | |
4004 | ||
991d9fa0 JT |
4005 | static void thin_dtr(struct dm_target *ti) |
4006 | { | |
4007 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
4008 | unsigned long flags; |
4009 | ||
4010 | spin_lock_irqsave(&tc->pool->lock, flags); | |
4011 | list_del_rcu(&tc->list); | |
4012 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
4013 | synchronize_rcu(); | |
991d9fa0 | 4014 | |
17181fb7 MP |
4015 | thin_put(tc); |
4016 | wait_for_completion(&tc->can_destroy); | |
4017 | ||
991d9fa0 JT |
4018 | mutex_lock(&dm_thin_pool_table.mutex); |
4019 | ||
4020 | __pool_dec(tc->pool); | |
4021 | dm_pool_close_thin_device(tc->td); | |
4022 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
4023 | if (tc->origin_dev) |
4024 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
4025 | kfree(tc); |
4026 | ||
4027 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4028 | } | |
4029 | ||
4030 | /* | |
4031 | * Thin target parameters: | |
4032 | * | |
2dd9c257 | 4033 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
4034 | * |
4035 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
4036 | * dev_id: the internal device identifier | |
2dd9c257 | 4037 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
4038 | * |
4039 | * If the pool device has discards disabled, they get disabled for the thin | |
4040 | * device as well. | |
991d9fa0 JT |
4041 | */ |
4042 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
4043 | { | |
4044 | int r; | |
4045 | struct thin_c *tc; | |
2dd9c257 | 4046 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 4047 | struct mapped_device *pool_md; |
5e3283e2 | 4048 | unsigned long flags; |
991d9fa0 JT |
4049 | |
4050 | mutex_lock(&dm_thin_pool_table.mutex); | |
4051 | ||
2dd9c257 | 4052 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
4053 | ti->error = "Invalid argument count"; |
4054 | r = -EINVAL; | |
4055 | goto out_unlock; | |
4056 | } | |
4057 | ||
4058 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
4059 | if (!tc) { | |
4060 | ti->error = "Out of memory"; | |
4061 | r = -ENOMEM; | |
4062 | goto out_unlock; | |
4063 | } | |
583024d2 | 4064 | tc->thin_md = dm_table_get_md(ti->table); |
c140e1c4 | 4065 | spin_lock_init(&tc->lock); |
a374bb21 | 4066 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
4067 | bio_list_init(&tc->deferred_bio_list); |
4068 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 4069 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 4070 | |
2dd9c257 JT |
4071 | if (argc == 3) { |
4072 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
4073 | if (r) { | |
4074 | ti->error = "Error opening origin device"; | |
4075 | goto bad_origin_dev; | |
4076 | } | |
4077 | tc->origin_dev = origin_dev; | |
4078 | } | |
4079 | ||
991d9fa0 JT |
4080 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
4081 | if (r) { | |
4082 | ti->error = "Error opening pool device"; | |
4083 | goto bad_pool_dev; | |
4084 | } | |
4085 | tc->pool_dev = pool_dev; | |
4086 | ||
4087 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
4088 | ti->error = "Invalid device id"; | |
4089 | r = -EINVAL; | |
4090 | goto bad_common; | |
4091 | } | |
4092 | ||
4093 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
4094 | if (!pool_md) { | |
4095 | ti->error = "Couldn't get pool mapped device"; | |
4096 | r = -EINVAL; | |
4097 | goto bad_common; | |
4098 | } | |
4099 | ||
4100 | tc->pool = __pool_table_lookup(pool_md); | |
4101 | if (!tc->pool) { | |
4102 | ti->error = "Couldn't find pool object"; | |
4103 | r = -EINVAL; | |
4104 | goto bad_pool_lookup; | |
4105 | } | |
4106 | __pool_inc(tc->pool); | |
4107 | ||
e49e5829 JT |
4108 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4109 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 4110 | r = -EINVAL; |
80e96c54 | 4111 | goto bad_pool; |
e49e5829 JT |
4112 | } |
4113 | ||
991d9fa0 JT |
4114 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
4115 | if (r) { | |
4116 | ti->error = "Couldn't open thin internal device"; | |
80e96c54 | 4117 | goto bad_pool; |
991d9fa0 JT |
4118 | } |
4119 | ||
542f9038 MS |
4120 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
4121 | if (r) | |
80e96c54 | 4122 | goto bad; |
542f9038 | 4123 | |
55a62eef | 4124 | ti->num_flush_bios = 1; |
16ad3d10 | 4125 | ti->flush_supported = true; |
30187e1d | 4126 | ti->per_io_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
4127 | |
4128 | /* In case the pool supports discards, pass them on. */ | |
4129 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 4130 | ti->discards_supported = true; |
55a62eef | 4131 | ti->num_discard_bios = 1; |
34fbcf62 | 4132 | ti->split_discard_bios = false; |
67e2e2b2 | 4133 | } |
991d9fa0 | 4134 | |
991d9fa0 JT |
4135 | mutex_unlock(&dm_thin_pool_table.mutex); |
4136 | ||
5e3283e2 | 4137 | spin_lock_irqsave(&tc->pool->lock, flags); |
80e96c54 MS |
4138 | if (tc->pool->suspended) { |
4139 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
4140 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ | |
4141 | ti->error = "Unable to activate thin device while pool is suspended"; | |
4142 | r = -EINVAL; | |
4143 | goto bad; | |
4144 | } | |
2b94e896 MD |
4145 | atomic_set(&tc->refcount, 1); |
4146 | init_completion(&tc->can_destroy); | |
c140e1c4 | 4147 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 4148 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
4149 | /* |
4150 | * This synchronize_rcu() call is needed here otherwise we risk a | |
4151 | * wake_worker() call finding no bios to process (because the newly | |
4152 | * added tc isn't yet visible). So this reduces latency since we | |
4153 | * aren't then dependent on the periodic commit to wake_worker(). | |
4154 | */ | |
4155 | synchronize_rcu(); | |
4156 | ||
80e96c54 MS |
4157 | dm_put(pool_md); |
4158 | ||
991d9fa0 JT |
4159 | return 0; |
4160 | ||
80e96c54 | 4161 | bad: |
1acacc07 | 4162 | dm_pool_close_thin_device(tc->td); |
80e96c54 | 4163 | bad_pool: |
991d9fa0 JT |
4164 | __pool_dec(tc->pool); |
4165 | bad_pool_lookup: | |
4166 | dm_put(pool_md); | |
4167 | bad_common: | |
4168 | dm_put_device(ti, tc->pool_dev); | |
4169 | bad_pool_dev: | |
2dd9c257 JT |
4170 | if (tc->origin_dev) |
4171 | dm_put_device(ti, tc->origin_dev); | |
4172 | bad_origin_dev: | |
991d9fa0 JT |
4173 | kfree(tc); |
4174 | out_unlock: | |
4175 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4176 | ||
4177 | return r; | |
4178 | } | |
4179 | ||
7de3ee57 | 4180 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 4181 | { |
4f024f37 | 4182 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 4183 | |
7de3ee57 | 4184 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
4185 | } |
4186 | ||
4e4cbee9 CH |
4187 | static int thin_endio(struct dm_target *ti, struct bio *bio, |
4188 | blk_status_t *err) | |
eb2aa48d JT |
4189 | { |
4190 | unsigned long flags; | |
59c3d2c6 | 4191 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 4192 | struct list_head work; |
a24c2569 | 4193 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
4194 | struct pool *pool = h->tc->pool; |
4195 | ||
4196 | if (h->shared_read_entry) { | |
4197 | INIT_LIST_HEAD(&work); | |
44feb387 | 4198 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
4199 | |
4200 | spin_lock_irqsave(&pool->lock, flags); | |
4201 | list_for_each_entry_safe(m, tmp, &work, list) { | |
4202 | list_del(&m->list); | |
50f3c3ef | 4203 | __complete_mapping_preparation(m); |
eb2aa48d JT |
4204 | } |
4205 | spin_unlock_irqrestore(&pool->lock, flags); | |
4206 | } | |
4207 | ||
104655fd JT |
4208 | if (h->all_io_entry) { |
4209 | INIT_LIST_HEAD(&work); | |
44feb387 | 4210 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
4211 | if (!list_empty(&work)) { |
4212 | spin_lock_irqsave(&pool->lock, flags); | |
4213 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 4214 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
4215 | spin_unlock_irqrestore(&pool->lock, flags); |
4216 | wake_worker(pool); | |
4217 | } | |
104655fd JT |
4218 | } |
4219 | ||
34fbcf62 JT |
4220 | if (h->cell) |
4221 | cell_defer_no_holder(h->tc, h->cell); | |
4222 | ||
1be56909 | 4223 | return DM_ENDIO_DONE; |
eb2aa48d JT |
4224 | } |
4225 | ||
738211f7 | 4226 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 4227 | { |
738211f7 JT |
4228 | struct thin_c *tc = ti->private; |
4229 | ||
991d9fa0 | 4230 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
4231 | noflush_work(tc, do_noflush_start); |
4232 | } | |
4233 | ||
4234 | static void thin_postsuspend(struct dm_target *ti) | |
4235 | { | |
4236 | struct thin_c *tc = ti->private; | |
4237 | ||
4238 | /* | |
4239 | * The dm_noflush_suspending flag has been cleared by now, so | |
4240 | * unfortunately we must always run this. | |
4241 | */ | |
4242 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
4243 | } |
4244 | ||
e5aea7b4 JT |
4245 | static int thin_preresume(struct dm_target *ti) |
4246 | { | |
4247 | struct thin_c *tc = ti->private; | |
4248 | ||
4249 | if (tc->origin_dev) | |
4250 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
4251 | ||
4252 | return 0; | |
4253 | } | |
4254 | ||
991d9fa0 JT |
4255 | /* |
4256 | * <nr mapped sectors> <highest mapped sector> | |
4257 | */ | |
fd7c092e MP |
4258 | static void thin_status(struct dm_target *ti, status_type_t type, |
4259 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
4260 | { |
4261 | int r; | |
4262 | ssize_t sz = 0; | |
4263 | dm_block_t mapped, highest; | |
4264 | char buf[BDEVNAME_SIZE]; | |
4265 | struct thin_c *tc = ti->private; | |
4266 | ||
e49e5829 JT |
4267 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4268 | DMEMIT("Fail"); | |
fd7c092e | 4269 | return; |
e49e5829 JT |
4270 | } |
4271 | ||
991d9fa0 JT |
4272 | if (!tc->td) |
4273 | DMEMIT("-"); | |
4274 | else { | |
4275 | switch (type) { | |
4276 | case STATUSTYPE_INFO: | |
4277 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
4278 | if (r) { |
4279 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
4280 | goto err; | |
4281 | } | |
991d9fa0 JT |
4282 | |
4283 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
4284 | if (r < 0) { |
4285 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
4286 | goto err; | |
4287 | } | |
991d9fa0 JT |
4288 | |
4289 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
4290 | if (r) | |
4291 | DMEMIT("%llu", ((highest + 1) * | |
4292 | tc->pool->sectors_per_block) - 1); | |
4293 | else | |
4294 | DMEMIT("-"); | |
4295 | break; | |
4296 | ||
4297 | case STATUSTYPE_TABLE: | |
4298 | DMEMIT("%s %lu", | |
4299 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
4300 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
4301 | if (tc->origin_dev) |
4302 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
4303 | break; |
4304 | } | |
4305 | } | |
4306 | ||
fd7c092e MP |
4307 | return; |
4308 | ||
4309 | err: | |
4310 | DMEMIT("Error"); | |
991d9fa0 JT |
4311 | } |
4312 | ||
4313 | static int thin_iterate_devices(struct dm_target *ti, | |
4314 | iterate_devices_callout_fn fn, void *data) | |
4315 | { | |
55f2b8bd | 4316 | sector_t blocks; |
991d9fa0 | 4317 | struct thin_c *tc = ti->private; |
55f2b8bd | 4318 | struct pool *pool = tc->pool; |
991d9fa0 JT |
4319 | |
4320 | /* | |
4321 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
4322 | * we follow a more convoluted path through to the pool's target. | |
4323 | */ | |
55f2b8bd | 4324 | if (!pool->ti) |
991d9fa0 JT |
4325 | return 0; /* nothing is bound */ |
4326 | ||
55f2b8bd MS |
4327 | blocks = pool->ti->len; |
4328 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 4329 | if (blocks) |
55f2b8bd | 4330 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
4331 | |
4332 | return 0; | |
4333 | } | |
4334 | ||
34fbcf62 JT |
4335 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4336 | { | |
4337 | struct thin_c *tc = ti->private; | |
4338 | struct pool *pool = tc->pool; | |
21607670 | 4339 | |
0fcb04d5 MS |
4340 | if (!pool->pf.discard_enabled) |
4341 | return; | |
34fbcf62 JT |
4342 | |
4343 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
4344 | limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ | |
4345 | } | |
4346 | ||
991d9fa0 JT |
4347 | static struct target_type thin_target = { |
4348 | .name = "thin", | |
202bae52 | 4349 | .version = {1, 19, 0}, |
991d9fa0 JT |
4350 | .module = THIS_MODULE, |
4351 | .ctr = thin_ctr, | |
4352 | .dtr = thin_dtr, | |
4353 | .map = thin_map, | |
eb2aa48d | 4354 | .end_io = thin_endio, |
e5aea7b4 | 4355 | .preresume = thin_preresume, |
738211f7 | 4356 | .presuspend = thin_presuspend, |
991d9fa0 JT |
4357 | .postsuspend = thin_postsuspend, |
4358 | .status = thin_status, | |
4359 | .iterate_devices = thin_iterate_devices, | |
34fbcf62 | 4360 | .io_hints = thin_io_hints, |
991d9fa0 JT |
4361 | }; |
4362 | ||
4363 | /*----------------------------------------------------------------*/ | |
4364 | ||
4365 | static int __init dm_thin_init(void) | |
4366 | { | |
7e6358d2 | 4367 | int r = -ENOMEM; |
991d9fa0 JT |
4368 | |
4369 | pool_table_init(); | |
4370 | ||
7e6358d2 | 4371 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
4372 | if (!_new_mapping_cache) | |
4373 | return r; | |
4374 | ||
991d9fa0 JT |
4375 | r = dm_register_target(&thin_target); |
4376 | if (r) | |
7e6358d2 | 4377 | goto bad_new_mapping_cache; |
991d9fa0 JT |
4378 | |
4379 | r = dm_register_target(&pool_target); | |
4380 | if (r) | |
7e6358d2 | 4381 | goto bad_thin_target; |
a24c2569 | 4382 | |
a24c2569 MS |
4383 | return 0; |
4384 | ||
7e6358d2 | 4385 | bad_thin_target: |
a24c2569 | 4386 | dm_unregister_target(&thin_target); |
7e6358d2 | 4387 | bad_new_mapping_cache: |
4388 | kmem_cache_destroy(_new_mapping_cache); | |
991d9fa0 JT |
4389 | |
4390 | return r; | |
4391 | } | |
4392 | ||
4393 | static void dm_thin_exit(void) | |
4394 | { | |
4395 | dm_unregister_target(&thin_target); | |
4396 | dm_unregister_target(&pool_target); | |
a24c2569 | 4397 | |
a24c2569 | 4398 | kmem_cache_destroy(_new_mapping_cache); |
d5ffebdd MS |
4399 | |
4400 | pool_table_exit(); | |
991d9fa0 JT |
4401 | } |
4402 | ||
4403 | module_init(dm_thin_init); | |
4404 | module_exit(dm_thin_exit); | |
4405 | ||
80c57893 MS |
4406 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
4407 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
4408 | ||
7cab8bf1 | 4409 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4410 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4411 | MODULE_LICENSE("GPL"); |