]>
Commit | Line | Data |
---|---|---|
991d9fa0 | 1 | /* |
e49e5829 | 2 | * Copyright (C) 2011-2012 Red Hat UK. |
991d9fa0 JT |
3 | * |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-thin-metadata.h" | |
4f81a417 | 8 | #include "dm-bio-prison.h" |
1f4e0ff0 | 9 | #include "dm.h" |
991d9fa0 JT |
10 | |
11 | #include <linux/device-mapper.h> | |
12 | #include <linux/dm-io.h> | |
13 | #include <linux/dm-kcopyd.h> | |
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> | |
ac4c3f34 | 21 | #include <linux/sort.h> |
67324ea1 | 22 | #include <linux/rbtree.h> |
991d9fa0 JT |
23 | |
24 | #define DM_MSG_PREFIX "thin" | |
25 | ||
26 | /* | |
27 | * Tunable constants | |
28 | */ | |
7768ed33 | 29 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 30 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 31 | #define COMMIT_PERIOD HZ |
80c57893 MS |
32 | #define NO_SPACE_TIMEOUT_SECS 60 |
33 | ||
34 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 35 | |
df5d2e90 MP |
36 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
37 | "A percentage of time allocated for copy on write"); | |
38 | ||
991d9fa0 JT |
39 | /* |
40 | * The block size of the device holding pool data must be | |
41 | * between 64KB and 1GB. | |
42 | */ | |
43 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
44 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
45 | ||
991d9fa0 JT |
46 | /* |
47 | * Device id is restricted to 24 bits. | |
48 | */ | |
49 | #define MAX_DEV_ID ((1 << 24) - 1) | |
50 | ||
51 | /* | |
52 | * How do we handle breaking sharing of data blocks? | |
53 | * ================================================= | |
54 | * | |
55 | * We use a standard copy-on-write btree to store the mappings for the | |
56 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
57 | * the data). When you take an internal snapshot you clone the root node | |
58 | * of the origin btree. After this there is no concept of an origin or a | |
59 | * snapshot. They are just two device trees that happen to point to the | |
60 | * same data blocks. | |
61 | * | |
62 | * When we get a write in we decide if it's to a shared data block using | |
63 | * some timestamp magic. If it is, we have to break sharing. | |
64 | * | |
65 | * Let's say we write to a shared block in what was the origin. The | |
66 | * steps are: | |
67 | * | |
68 | * i) plug io further to this physical block. (see bio_prison code). | |
69 | * | |
70 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 71 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
72 | * |
73 | * iii) copy the data block to a newly allocate block. This step can be | |
74 | * missed out if the io covers the block. (schedule_copy). | |
75 | * | |
76 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 77 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
78 | * sharing of btree nodes between the two devices. Breaking sharing only |
79 | * effects the btree of that specific device. Btrees for the other | |
80 | * devices that share the block never change. The btree for the origin | |
81 | * device as it was after the last commit is untouched, ie. we're using | |
82 | * persistent data structures in the functional programming sense. | |
83 | * | |
84 | * v) unplug io to this physical block, including the io that triggered | |
85 | * the breaking of sharing. | |
86 | * | |
87 | * Steps (ii) and (iii) occur in parallel. | |
88 | * | |
89 | * The metadata _doesn't_ need to be committed before the io continues. We | |
90 | * get away with this because the io is always written to a _new_ block. | |
91 | * If there's a crash, then: | |
92 | * | |
93 | * - The origin mapping will point to the old origin block (the shared | |
94 | * one). This will contain the data as it was before the io that triggered | |
95 | * the breaking of sharing came in. | |
96 | * | |
97 | * - The snap mapping still points to the old block. As it would after | |
98 | * the commit. | |
99 | * | |
100 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
101 | * will continue to think that data block in the snapshot device is shared | |
102 | * even after the write to the origin has broken sharing. I suspect data | |
103 | * blocks will typically be shared by many different devices, so we're | |
104 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
105 | * devices that reference this data block. At the moment I think the | |
106 | * benefits far, far outweigh the disadvantages. | |
107 | */ | |
108 | ||
109 | /*----------------------------------------------------------------*/ | |
110 | ||
991d9fa0 JT |
111 | /* |
112 | * Key building. | |
113 | */ | |
34fbcf62 JT |
114 | enum lock_space { |
115 | VIRTUAL, | |
116 | PHYSICAL | |
117 | }; | |
118 | ||
119 | static void build_key(struct dm_thin_device *td, enum lock_space ls, | |
120 | dm_block_t b, dm_block_t e, struct dm_cell_key *key) | |
991d9fa0 | 121 | { |
34fbcf62 | 122 | key->virtual = (ls == VIRTUAL); |
991d9fa0 | 123 | key->dev = dm_thin_dev_id(td); |
5f274d88 | 124 | key->block_begin = b; |
34fbcf62 JT |
125 | key->block_end = e; |
126 | } | |
127 | ||
128 | static void build_data_key(struct dm_thin_device *td, dm_block_t b, | |
129 | struct dm_cell_key *key) | |
130 | { | |
131 | build_key(td, PHYSICAL, b, b + 1llu, key); | |
991d9fa0 JT |
132 | } |
133 | ||
134 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 135 | struct dm_cell_key *key) |
991d9fa0 | 136 | { |
34fbcf62 | 137 | build_key(td, VIRTUAL, b, b + 1llu, key); |
991d9fa0 JT |
138 | } |
139 | ||
140 | /*----------------------------------------------------------------*/ | |
141 | ||
7d327fe0 JT |
142 | #define THROTTLE_THRESHOLD (1 * HZ) |
143 | ||
144 | struct throttle { | |
145 | struct rw_semaphore lock; | |
146 | unsigned long threshold; | |
147 | bool throttle_applied; | |
148 | }; | |
149 | ||
150 | static void throttle_init(struct throttle *t) | |
151 | { | |
152 | init_rwsem(&t->lock); | |
153 | t->throttle_applied = false; | |
154 | } | |
155 | ||
156 | static void throttle_work_start(struct throttle *t) | |
157 | { | |
158 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
159 | } | |
160 | ||
161 | static void throttle_work_update(struct throttle *t) | |
162 | { | |
163 | if (!t->throttle_applied && jiffies > t->threshold) { | |
164 | down_write(&t->lock); | |
165 | t->throttle_applied = true; | |
166 | } | |
167 | } | |
168 | ||
169 | static void throttle_work_complete(struct throttle *t) | |
170 | { | |
171 | if (t->throttle_applied) { | |
172 | t->throttle_applied = false; | |
173 | up_write(&t->lock); | |
174 | } | |
175 | } | |
176 | ||
177 | static void throttle_lock(struct throttle *t) | |
178 | { | |
179 | down_read(&t->lock); | |
180 | } | |
181 | ||
182 | static void throttle_unlock(struct throttle *t) | |
183 | { | |
184 | up_read(&t->lock); | |
185 | } | |
186 | ||
187 | /*----------------------------------------------------------------*/ | |
188 | ||
991d9fa0 JT |
189 | /* |
190 | * A pool device ties together a metadata device and a data device. It | |
191 | * also provides the interface for creating and destroying internal | |
192 | * devices. | |
193 | */ | |
a24c2569 | 194 | struct dm_thin_new_mapping; |
67e2e2b2 | 195 | |
e49e5829 | 196 | /* |
3e1a0699 | 197 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
198 | */ |
199 | enum pool_mode { | |
200 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 201 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
202 | PM_READ_ONLY, /* metadata may not be changed */ |
203 | PM_FAIL, /* all I/O fails */ | |
204 | }; | |
205 | ||
67e2e2b2 | 206 | struct pool_features { |
e49e5829 JT |
207 | enum pool_mode mode; |
208 | ||
9bc142dd MS |
209 | bool zero_new_blocks:1; |
210 | bool discard_enabled:1; | |
211 | bool discard_passdown:1; | |
787a996c | 212 | bool error_if_no_space:1; |
67e2e2b2 JT |
213 | }; |
214 | ||
e49e5829 JT |
215 | struct thin_c; |
216 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 217 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
218 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
219 | ||
ac4c3f34 JT |
220 | #define CELL_SORT_ARRAY_SIZE 8192 |
221 | ||
991d9fa0 JT |
222 | struct pool { |
223 | struct list_head list; | |
224 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
225 | ||
226 | struct mapped_device *pool_md; | |
227 | struct block_device *md_dev; | |
228 | struct dm_pool_metadata *pmd; | |
229 | ||
991d9fa0 | 230 | dm_block_t low_water_blocks; |
55f2b8bd | 231 | uint32_t sectors_per_block; |
f9a8e0cd | 232 | int sectors_per_block_shift; |
991d9fa0 | 233 | |
67e2e2b2 | 234 | struct pool_features pf; |
88a6621b | 235 | bool low_water_triggered:1; /* A dm event has been sent */ |
80e96c54 | 236 | bool suspended:1; |
991d9fa0 | 237 | |
44feb387 | 238 | struct dm_bio_prison *prison; |
991d9fa0 JT |
239 | struct dm_kcopyd_client *copier; |
240 | ||
241 | struct workqueue_struct *wq; | |
7d327fe0 | 242 | struct throttle throttle; |
991d9fa0 | 243 | struct work_struct worker; |
905e51b3 | 244 | struct delayed_work waker; |
85ad643b | 245 | struct delayed_work no_space_timeout; |
991d9fa0 | 246 | |
905e51b3 | 247 | unsigned long last_commit_jiffies; |
55f2b8bd | 248 | unsigned ref_count; |
991d9fa0 JT |
249 | |
250 | spinlock_t lock; | |
991d9fa0 JT |
251 | struct bio_list deferred_flush_bios; |
252 | struct list_head prepared_mappings; | |
104655fd | 253 | struct list_head prepared_discards; |
c140e1c4 | 254 | struct list_head active_thins; |
991d9fa0 | 255 | |
44feb387 MS |
256 | struct dm_deferred_set *shared_read_ds; |
257 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 258 | |
a24c2569 | 259 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 260 | mempool_t *mapping_pool; |
e49e5829 JT |
261 | |
262 | process_bio_fn process_bio; | |
263 | process_bio_fn process_discard; | |
264 | ||
a374bb21 JT |
265 | process_cell_fn process_cell; |
266 | process_cell_fn process_discard_cell; | |
267 | ||
e49e5829 JT |
268 | process_mapping_fn process_prepared_mapping; |
269 | process_mapping_fn process_prepared_discard; | |
ac4c3f34 JT |
270 | |
271 | struct dm_bio_prison_cell *cell_sort_array[CELL_SORT_ARRAY_SIZE]; | |
991d9fa0 JT |
272 | }; |
273 | ||
e49e5829 | 274 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 275 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 276 | |
991d9fa0 JT |
277 | /* |
278 | * Target context for a pool. | |
279 | */ | |
280 | struct pool_c { | |
281 | struct dm_target *ti; | |
282 | struct pool *pool; | |
283 | struct dm_dev *data_dev; | |
284 | struct dm_dev *metadata_dev; | |
285 | struct dm_target_callbacks callbacks; | |
286 | ||
287 | dm_block_t low_water_blocks; | |
0424caa1 MS |
288 | struct pool_features requested_pf; /* Features requested during table load */ |
289 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
290 | }; |
291 | ||
292 | /* | |
293 | * Target context for a thin. | |
294 | */ | |
295 | struct thin_c { | |
c140e1c4 | 296 | struct list_head list; |
991d9fa0 | 297 | struct dm_dev *pool_dev; |
2dd9c257 | 298 | struct dm_dev *origin_dev; |
e5aea7b4 | 299 | sector_t origin_size; |
991d9fa0 JT |
300 | dm_thin_id dev_id; |
301 | ||
302 | struct pool *pool; | |
303 | struct dm_thin_device *td; | |
583024d2 MS |
304 | struct mapped_device *thin_md; |
305 | ||
738211f7 | 306 | bool requeue_mode:1; |
c140e1c4 | 307 | spinlock_t lock; |
a374bb21 | 308 | struct list_head deferred_cells; |
c140e1c4 MS |
309 | struct bio_list deferred_bio_list; |
310 | struct bio_list retry_on_resume_list; | |
67324ea1 | 311 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
312 | |
313 | /* | |
314 | * Ensures the thin is not destroyed until the worker has finished | |
315 | * iterating the active_thins list. | |
316 | */ | |
317 | atomic_t refcount; | |
318 | struct completion can_destroy; | |
991d9fa0 JT |
319 | }; |
320 | ||
321 | /*----------------------------------------------------------------*/ | |
322 | ||
34fbcf62 JT |
323 | /** |
324 | * __blkdev_issue_discard_async - queue a discard with async completion | |
325 | * @bdev: blockdev to issue discard for | |
326 | * @sector: start sector | |
327 | * @nr_sects: number of sectors to discard | |
328 | * @gfp_mask: memory allocation flags (for bio_alloc) | |
329 | * @flags: BLKDEV_IFL_* flags to control behaviour | |
330 | * @parent_bio: parent discard bio that all sub discards get chained to | |
331 | * | |
332 | * Description: | |
333 | * Asynchronously issue a discard request for the sectors in question. | |
334 | * NOTE: this variant of blk-core's blkdev_issue_discard() is a stop-gap | |
335 | * that is being kept local to DM thinp until the block changes to allow | |
336 | * late bio splitting land upstream. | |
337 | */ | |
338 | static int __blkdev_issue_discard_async(struct block_device *bdev, sector_t sector, | |
339 | sector_t nr_sects, gfp_t gfp_mask, unsigned long flags, | |
340 | struct bio *parent_bio) | |
341 | { | |
342 | struct request_queue *q = bdev_get_queue(bdev); | |
343 | int type = REQ_WRITE | REQ_DISCARD; | |
344 | unsigned int max_discard_sectors, granularity; | |
345 | int alignment; | |
346 | struct bio *bio; | |
347 | int ret = 0; | |
348 | struct blk_plug plug; | |
349 | ||
350 | if (!q) | |
351 | return -ENXIO; | |
352 | ||
353 | if (!blk_queue_discard(q)) | |
354 | return -EOPNOTSUPP; | |
355 | ||
356 | /* Zero-sector (unknown) and one-sector granularities are the same. */ | |
357 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
358 | alignment = (bdev_discard_alignment(bdev) >> 9) % granularity; | |
359 | ||
360 | /* | |
361 | * Ensure that max_discard_sectors is of the proper | |
362 | * granularity, so that requests stay aligned after a split. | |
363 | */ | |
364 | max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9); | |
365 | max_discard_sectors -= max_discard_sectors % granularity; | |
366 | if (unlikely(!max_discard_sectors)) { | |
367 | /* Avoid infinite loop below. Being cautious never hurts. */ | |
368 | return -EOPNOTSUPP; | |
369 | } | |
370 | ||
371 | if (flags & BLKDEV_DISCARD_SECURE) { | |
372 | if (!blk_queue_secdiscard(q)) | |
373 | return -EOPNOTSUPP; | |
374 | type |= REQ_SECURE; | |
375 | } | |
376 | ||
377 | blk_start_plug(&plug); | |
378 | while (nr_sects) { | |
379 | unsigned int req_sects; | |
380 | sector_t end_sect, tmp; | |
381 | ||
382 | /* | |
383 | * Required bio_put occurs in bio_endio thanks to bio_chain below | |
384 | */ | |
385 | bio = bio_alloc(gfp_mask, 1); | |
386 | if (!bio) { | |
387 | ret = -ENOMEM; | |
388 | break; | |
389 | } | |
390 | ||
391 | req_sects = min_t(sector_t, nr_sects, max_discard_sectors); | |
392 | ||
393 | /* | |
394 | * If splitting a request, and the next starting sector would be | |
395 | * misaligned, stop the discard at the previous aligned sector. | |
396 | */ | |
397 | end_sect = sector + req_sects; | |
398 | tmp = end_sect; | |
399 | if (req_sects < nr_sects && | |
400 | sector_div(tmp, granularity) != alignment) { | |
401 | end_sect = end_sect - alignment; | |
402 | sector_div(end_sect, granularity); | |
403 | end_sect = end_sect * granularity + alignment; | |
404 | req_sects = end_sect - sector; | |
405 | } | |
406 | ||
407 | bio_chain(bio, parent_bio); | |
408 | ||
409 | bio->bi_iter.bi_sector = sector; | |
410 | bio->bi_bdev = bdev; | |
411 | ||
412 | bio->bi_iter.bi_size = req_sects << 9; | |
413 | nr_sects -= req_sects; | |
414 | sector = end_sect; | |
415 | ||
416 | submit_bio(type, bio); | |
417 | ||
418 | /* | |
419 | * We can loop for a long time in here, if someone does | |
420 | * full device discards (like mkfs). Be nice and allow | |
421 | * us to schedule out to avoid softlocking if preempt | |
422 | * is disabled. | |
423 | */ | |
424 | cond_resched(); | |
425 | } | |
426 | blk_finish_plug(&plug); | |
427 | ||
428 | return ret; | |
429 | } | |
430 | ||
431 | static bool block_size_is_power_of_two(struct pool *pool) | |
432 | { | |
433 | return pool->sectors_per_block_shift >= 0; | |
434 | } | |
435 | ||
436 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) | |
437 | { | |
438 | return block_size_is_power_of_two(pool) ? | |
439 | (b << pool->sectors_per_block_shift) : | |
440 | (b * pool->sectors_per_block); | |
441 | } | |
442 | ||
443 | static int issue_discard(struct thin_c *tc, dm_block_t data_b, dm_block_t data_e, | |
444 | struct bio *parent_bio) | |
445 | { | |
446 | sector_t s = block_to_sectors(tc->pool, data_b); | |
447 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); | |
448 | ||
449 | return __blkdev_issue_discard_async(tc->pool_dev->bdev, s, len, | |
450 | GFP_NOWAIT, 0, parent_bio); | |
451 | } | |
452 | ||
453 | /*----------------------------------------------------------------*/ | |
454 | ||
025b9685 JT |
455 | /* |
456 | * wake_worker() is used when new work is queued and when pool_resume is | |
457 | * ready to continue deferred IO processing. | |
458 | */ | |
459 | static void wake_worker(struct pool *pool) | |
460 | { | |
461 | queue_work(pool->wq, &pool->worker); | |
462 | } | |
463 | ||
464 | /*----------------------------------------------------------------*/ | |
465 | ||
6beca5eb JT |
466 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
467 | struct dm_bio_prison_cell **cell_result) | |
468 | { | |
469 | int r; | |
470 | struct dm_bio_prison_cell *cell_prealloc; | |
471 | ||
472 | /* | |
473 | * Allocate a cell from the prison's mempool. | |
474 | * This might block but it can't fail. | |
475 | */ | |
476 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
477 | ||
478 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
479 | if (r) | |
480 | /* | |
481 | * We reused an old cell; we can get rid of | |
482 | * the new one. | |
483 | */ | |
484 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
485 | ||
486 | return r; | |
487 | } | |
488 | ||
489 | static void cell_release(struct pool *pool, | |
490 | struct dm_bio_prison_cell *cell, | |
491 | struct bio_list *bios) | |
492 | { | |
493 | dm_cell_release(pool->prison, cell, bios); | |
494 | dm_bio_prison_free_cell(pool->prison, cell); | |
495 | } | |
496 | ||
2d759a46 JT |
497 | static void cell_visit_release(struct pool *pool, |
498 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
499 | void *context, | |
500 | struct dm_bio_prison_cell *cell) | |
501 | { | |
502 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
503 | dm_bio_prison_free_cell(pool->prison, cell); | |
504 | } | |
505 | ||
6beca5eb JT |
506 | static void cell_release_no_holder(struct pool *pool, |
507 | struct dm_bio_prison_cell *cell, | |
508 | struct bio_list *bios) | |
509 | { | |
510 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
511 | dm_bio_prison_free_cell(pool->prison, cell); | |
512 | } | |
513 | ||
af91805a MS |
514 | static void cell_error_with_code(struct pool *pool, |
515 | struct dm_bio_prison_cell *cell, int error_code) | |
6beca5eb | 516 | { |
af91805a | 517 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
518 | dm_bio_prison_free_cell(pool->prison, cell); |
519 | } | |
520 | ||
af91805a MS |
521 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
522 | { | |
523 | cell_error_with_code(pool, cell, -EIO); | |
524 | } | |
525 | ||
a374bb21 JT |
526 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
527 | { | |
528 | cell_error_with_code(pool, cell, 0); | |
529 | } | |
530 | ||
531 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
532 | { | |
533 | cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE); | |
534 | } | |
535 | ||
6beca5eb JT |
536 | /*----------------------------------------------------------------*/ |
537 | ||
991d9fa0 JT |
538 | /* |
539 | * A global list of pools that uses a struct mapped_device as a key. | |
540 | */ | |
541 | static struct dm_thin_pool_table { | |
542 | struct mutex mutex; | |
543 | struct list_head pools; | |
544 | } dm_thin_pool_table; | |
545 | ||
546 | static void pool_table_init(void) | |
547 | { | |
548 | mutex_init(&dm_thin_pool_table.mutex); | |
549 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
550 | } | |
551 | ||
552 | static void __pool_table_insert(struct pool *pool) | |
553 | { | |
554 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
555 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
556 | } | |
557 | ||
558 | static void __pool_table_remove(struct pool *pool) | |
559 | { | |
560 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
561 | list_del(&pool->list); | |
562 | } | |
563 | ||
564 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
565 | { | |
566 | struct pool *pool = NULL, *tmp; | |
567 | ||
568 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
569 | ||
570 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
571 | if (tmp->pool_md == md) { | |
572 | pool = tmp; | |
573 | break; | |
574 | } | |
575 | } | |
576 | ||
577 | return pool; | |
578 | } | |
579 | ||
580 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
581 | { | |
582 | struct pool *pool = NULL, *tmp; | |
583 | ||
584 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
585 | ||
586 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
587 | if (tmp->md_dev == md_dev) { | |
588 | pool = tmp; | |
589 | break; | |
590 | } | |
591 | } | |
592 | ||
593 | return pool; | |
594 | } | |
595 | ||
596 | /*----------------------------------------------------------------*/ | |
597 | ||
a24c2569 | 598 | struct dm_thin_endio_hook { |
eb2aa48d | 599 | struct thin_c *tc; |
44feb387 MS |
600 | struct dm_deferred_entry *shared_read_entry; |
601 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 602 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 603 | struct rb_node rb_node; |
34fbcf62 | 604 | struct dm_bio_prison_cell *cell; |
eb2aa48d JT |
605 | }; |
606 | ||
42d6a8ce MS |
607 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
608 | { | |
609 | bio_list_merge(bios, master); | |
610 | bio_list_init(master); | |
611 | } | |
612 | ||
613 | static void error_bio_list(struct bio_list *bios, int error) | |
991d9fa0 JT |
614 | { |
615 | struct bio *bio; | |
42d6a8ce MS |
616 | |
617 | while ((bio = bio_list_pop(bios))) | |
618 | bio_endio(bio, error); | |
619 | } | |
620 | ||
621 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error) | |
622 | { | |
991d9fa0 | 623 | struct bio_list bios; |
18adc577 | 624 | unsigned long flags; |
991d9fa0 JT |
625 | |
626 | bio_list_init(&bios); | |
18adc577 | 627 | |
c140e1c4 | 628 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce | 629 | __merge_bio_list(&bios, master); |
c140e1c4 | 630 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 631 | |
42d6a8ce | 632 | error_bio_list(&bios, error); |
991d9fa0 JT |
633 | } |
634 | ||
a374bb21 JT |
635 | static void requeue_deferred_cells(struct thin_c *tc) |
636 | { | |
637 | struct pool *pool = tc->pool; | |
638 | unsigned long flags; | |
639 | struct list_head cells; | |
640 | struct dm_bio_prison_cell *cell, *tmp; | |
641 | ||
642 | INIT_LIST_HEAD(&cells); | |
643 | ||
644 | spin_lock_irqsave(&tc->lock, flags); | |
645 | list_splice_init(&tc->deferred_cells, &cells); | |
646 | spin_unlock_irqrestore(&tc->lock, flags); | |
647 | ||
648 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
649 | cell_requeue(pool, cell); | |
650 | } | |
651 | ||
991d9fa0 JT |
652 | static void requeue_io(struct thin_c *tc) |
653 | { | |
3e1a0699 | 654 | struct bio_list bios; |
42d6a8ce | 655 | unsigned long flags; |
3e1a0699 JT |
656 | |
657 | bio_list_init(&bios); | |
658 | ||
c140e1c4 | 659 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce MS |
660 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
661 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
c140e1c4 | 662 | spin_unlock_irqrestore(&tc->lock, flags); |
3e1a0699 | 663 | |
42d6a8ce MS |
664 | error_bio_list(&bios, DM_ENDIO_REQUEUE); |
665 | requeue_deferred_cells(tc); | |
3e1a0699 JT |
666 | } |
667 | ||
c140e1c4 MS |
668 | static void error_retry_list(struct pool *pool) |
669 | { | |
670 | struct thin_c *tc; | |
671 | ||
672 | rcu_read_lock(); | |
673 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
42d6a8ce | 674 | error_thin_bio_list(tc, &tc->retry_on_resume_list, -EIO); |
c140e1c4 MS |
675 | rcu_read_unlock(); |
676 | } | |
677 | ||
991d9fa0 JT |
678 | /* |
679 | * This section of code contains the logic for processing a thin device's IO. | |
680 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
681 | * but most is exclusively called from the thin target rather than the thin-pool | |
682 | * target. | |
683 | */ | |
684 | ||
685 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
686 | { | |
58f77a21 | 687 | struct pool *pool = tc->pool; |
4f024f37 | 688 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 689 | |
58f77a21 MS |
690 | if (block_size_is_power_of_two(pool)) |
691 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 692 | else |
58f77a21 | 693 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
694 | |
695 | return block_nr; | |
991d9fa0 JT |
696 | } |
697 | ||
34fbcf62 JT |
698 | /* |
699 | * Returns the _complete_ blocks that this bio covers. | |
700 | */ | |
701 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, | |
702 | dm_block_t *begin, dm_block_t *end) | |
703 | { | |
704 | struct pool *pool = tc->pool; | |
705 | sector_t b = bio->bi_iter.bi_sector; | |
706 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
707 | ||
708 | b += pool->sectors_per_block - 1ull; /* so we round up */ | |
709 | ||
710 | if (block_size_is_power_of_two(pool)) { | |
711 | b >>= pool->sectors_per_block_shift; | |
712 | e >>= pool->sectors_per_block_shift; | |
713 | } else { | |
714 | (void) sector_div(b, pool->sectors_per_block); | |
715 | (void) sector_div(e, pool->sectors_per_block); | |
716 | } | |
717 | ||
718 | if (e < b) | |
719 | /* Can happen if the bio is within a single block. */ | |
720 | e = b; | |
721 | ||
722 | *begin = b; | |
723 | *end = e; | |
724 | } | |
725 | ||
991d9fa0 JT |
726 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
727 | { | |
728 | struct pool *pool = tc->pool; | |
4f024f37 | 729 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
730 | |
731 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 732 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
733 | bio->bi_iter.bi_sector = |
734 | (block << pool->sectors_per_block_shift) | | |
735 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 736 | else |
4f024f37 | 737 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 738 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
739 | } |
740 | ||
2dd9c257 JT |
741 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
742 | { | |
743 | bio->bi_bdev = tc->origin_dev->bdev; | |
744 | } | |
745 | ||
4afdd680 JT |
746 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
747 | { | |
748 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
749 | dm_thin_changed_this_transaction(tc->td); | |
750 | } | |
751 | ||
e8088073 JT |
752 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
753 | { | |
754 | struct dm_thin_endio_hook *h; | |
755 | ||
756 | if (bio->bi_rw & REQ_DISCARD) | |
757 | return; | |
758 | ||
59c3d2c6 | 759 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
760 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
761 | } | |
762 | ||
2dd9c257 | 763 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
764 | { |
765 | struct pool *pool = tc->pool; | |
766 | unsigned long flags; | |
767 | ||
e49e5829 JT |
768 | if (!bio_triggers_commit(tc, bio)) { |
769 | generic_make_request(bio); | |
770 | return; | |
771 | } | |
772 | ||
991d9fa0 | 773 | /* |
e49e5829 JT |
774 | * Complete bio with an error if earlier I/O caused changes to |
775 | * the metadata that can't be committed e.g, due to I/O errors | |
776 | * on the metadata device. | |
991d9fa0 | 777 | */ |
e49e5829 JT |
778 | if (dm_thin_aborted_changes(tc->td)) { |
779 | bio_io_error(bio); | |
780 | return; | |
781 | } | |
782 | ||
783 | /* | |
784 | * Batch together any bios that trigger commits and then issue a | |
785 | * single commit for them in process_deferred_bios(). | |
786 | */ | |
787 | spin_lock_irqsave(&pool->lock, flags); | |
788 | bio_list_add(&pool->deferred_flush_bios, bio); | |
789 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
790 | } |
791 | ||
2dd9c257 JT |
792 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
793 | { | |
794 | remap_to_origin(tc, bio); | |
795 | issue(tc, bio); | |
796 | } | |
797 | ||
798 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
799 | dm_block_t block) | |
800 | { | |
801 | remap(tc, bio, block); | |
802 | issue(tc, bio); | |
803 | } | |
804 | ||
991d9fa0 JT |
805 | /*----------------------------------------------------------------*/ |
806 | ||
807 | /* | |
808 | * Bio endio functions. | |
809 | */ | |
a24c2569 | 810 | struct dm_thin_new_mapping { |
991d9fa0 JT |
811 | struct list_head list; |
812 | ||
7f214665 | 813 | bool pass_discard:1; |
34fbcf62 | 814 | bool maybe_shared:1; |
991d9fa0 | 815 | |
50f3c3ef JT |
816 | /* |
817 | * Track quiescing, copying and zeroing preparation actions. When this | |
818 | * counter hits zero the block is prepared and can be inserted into the | |
819 | * btree. | |
820 | */ | |
821 | atomic_t prepare_actions; | |
822 | ||
7f214665 | 823 | int err; |
991d9fa0 | 824 | struct thin_c *tc; |
34fbcf62 | 825 | dm_block_t virt_begin, virt_end; |
991d9fa0 | 826 | dm_block_t data_block; |
34fbcf62 | 827 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
828 | |
829 | /* | |
830 | * If the bio covers the whole area of a block then we can avoid | |
831 | * zeroing or copying. Instead this bio is hooked. The bio will | |
832 | * still be in the cell, so care has to be taken to avoid issuing | |
833 | * the bio twice. | |
834 | */ | |
835 | struct bio *bio; | |
836 | bio_end_io_t *saved_bi_end_io; | |
837 | }; | |
838 | ||
50f3c3ef | 839 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
840 | { |
841 | struct pool *pool = m->tc->pool; | |
842 | ||
50f3c3ef | 843 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 844 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
845 | wake_worker(pool); |
846 | } | |
847 | } | |
848 | ||
e5aea7b4 | 849 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
850 | { |
851 | unsigned long flags; | |
991d9fa0 JT |
852 | struct pool *pool = m->tc->pool; |
853 | ||
991d9fa0 | 854 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 855 | __complete_mapping_preparation(m); |
991d9fa0 JT |
856 | spin_unlock_irqrestore(&pool->lock, flags); |
857 | } | |
858 | ||
e5aea7b4 JT |
859 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
860 | { | |
861 | struct dm_thin_new_mapping *m = context; | |
862 | ||
863 | m->err = read_err || write_err ? -EIO : 0; | |
864 | complete_mapping_preparation(m); | |
865 | } | |
866 | ||
991d9fa0 JT |
867 | static void overwrite_endio(struct bio *bio, int err) |
868 | { | |
59c3d2c6 | 869 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 870 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 | 871 | |
8b908f8e MS |
872 | bio->bi_end_io = m->saved_bi_end_io; |
873 | ||
991d9fa0 | 874 | m->err = err; |
e5aea7b4 | 875 | complete_mapping_preparation(m); |
991d9fa0 JT |
876 | } |
877 | ||
991d9fa0 JT |
878 | /*----------------------------------------------------------------*/ |
879 | ||
880 | /* | |
881 | * Workqueue. | |
882 | */ | |
883 | ||
884 | /* | |
885 | * Prepared mapping jobs. | |
886 | */ | |
887 | ||
888 | /* | |
2d759a46 JT |
889 | * This sends the bios in the cell, except the original holder, back |
890 | * to the deferred_bios list. | |
991d9fa0 | 891 | */ |
f286ba0e | 892 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 893 | { |
991d9fa0 JT |
894 | struct pool *pool = tc->pool; |
895 | unsigned long flags; | |
896 | ||
c140e1c4 MS |
897 | spin_lock_irqsave(&tc->lock, flags); |
898 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
899 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
900 | |
901 | wake_worker(pool); | |
902 | } | |
903 | ||
a374bb21 JT |
904 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
905 | ||
2d759a46 JT |
906 | struct remap_info { |
907 | struct thin_c *tc; | |
908 | struct bio_list defer_bios; | |
909 | struct bio_list issue_bios; | |
910 | }; | |
911 | ||
912 | static void __inc_remap_and_issue_cell(void *context, | |
913 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 914 | { |
2d759a46 | 915 | struct remap_info *info = context; |
a374bb21 | 916 | struct bio *bio; |
a374bb21 | 917 | |
2d759a46 | 918 | while ((bio = bio_list_pop(&cell->bios))) { |
a374bb21 | 919 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) |
2d759a46 | 920 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 921 | else { |
2d759a46 JT |
922 | inc_all_io_entry(info->tc->pool, bio); |
923 | ||
924 | /* | |
925 | * We can't issue the bios with the bio prison lock | |
926 | * held, so we add them to a list to issue on | |
927 | * return from this function. | |
928 | */ | |
929 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
930 | } |
931 | } | |
932 | } | |
933 | ||
2d759a46 JT |
934 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
935 | struct dm_bio_prison_cell *cell, | |
936 | dm_block_t block) | |
937 | { | |
938 | struct bio *bio; | |
939 | struct remap_info info; | |
940 | ||
941 | info.tc = tc; | |
942 | bio_list_init(&info.defer_bios); | |
943 | bio_list_init(&info.issue_bios); | |
944 | ||
945 | /* | |
946 | * We have to be careful to inc any bios we're about to issue | |
947 | * before the cell is released, and avoid a race with new bios | |
948 | * being added to the cell. | |
949 | */ | |
950 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
951 | &info, cell); | |
952 | ||
953 | while ((bio = bio_list_pop(&info.defer_bios))) | |
954 | thin_defer_bio(tc, bio); | |
955 | ||
956 | while ((bio = bio_list_pop(&info.issue_bios))) | |
957 | remap_and_issue(info.tc, bio, block); | |
958 | } | |
959 | ||
e49e5829 JT |
960 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
961 | { | |
6beca5eb | 962 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
963 | list_del(&m->list); |
964 | mempool_free(m, m->tc->pool->mapping_pool); | |
965 | } | |
025b9685 | 966 | |
a24c2569 | 967 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
968 | { |
969 | struct thin_c *tc = m->tc; | |
6beca5eb | 970 | struct pool *pool = tc->pool; |
8b908f8e | 971 | struct bio *bio = m->bio; |
991d9fa0 JT |
972 | int r; |
973 | ||
991d9fa0 | 974 | if (m->err) { |
6beca5eb | 975 | cell_error(pool, m->cell); |
905386f8 | 976 | goto out; |
991d9fa0 JT |
977 | } |
978 | ||
979 | /* | |
980 | * Commit the prepared block into the mapping btree. | |
981 | * Any I/O for this block arriving after this point will get | |
982 | * remapped to it directly. | |
983 | */ | |
34fbcf62 | 984 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
991d9fa0 | 985 | if (r) { |
b5330655 | 986 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 987 | cell_error(pool, m->cell); |
905386f8 | 988 | goto out; |
991d9fa0 JT |
989 | } |
990 | ||
991 | /* | |
992 | * Release any bios held while the block was being provisioned. | |
993 | * If we are processing a write bio that completely covers the block, | |
994 | * we already processed it so can ignore it now when processing | |
995 | * the bios in the cell. | |
996 | */ | |
997 | if (bio) { | |
2d759a46 | 998 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
991d9fa0 | 999 | bio_endio(bio, 0); |
2d759a46 JT |
1000 | } else { |
1001 | inc_all_io_entry(tc->pool, m->cell->holder); | |
1002 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
1003 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
1004 | } | |
991d9fa0 | 1005 | |
905386f8 | 1006 | out: |
991d9fa0 | 1007 | list_del(&m->list); |
6beca5eb | 1008 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
1009 | } |
1010 | ||
34fbcf62 JT |
1011 | /*----------------------------------------------------------------*/ |
1012 | ||
1013 | static void free_discard_mapping(struct dm_thin_new_mapping *m) | |
104655fd | 1014 | { |
104655fd | 1015 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1016 | if (m->cell) |
1017 | cell_defer_no_holder(tc, m->cell); | |
1018 | mempool_free(m, tc->pool->mapping_pool); | |
1019 | } | |
104655fd | 1020 | |
34fbcf62 JT |
1021 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
1022 | { | |
e49e5829 | 1023 | bio_io_error(m->bio); |
34fbcf62 JT |
1024 | free_discard_mapping(m); |
1025 | } | |
1026 | ||
1027 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) | |
1028 | { | |
1029 | bio_endio(m->bio, 0); | |
1030 | free_discard_mapping(m); | |
1031 | } | |
1032 | ||
1033 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) | |
1034 | { | |
1035 | int r; | |
1036 | struct thin_c *tc = m->tc; | |
1037 | ||
1038 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); | |
1039 | if (r) { | |
1040 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); | |
1041 | bio_io_error(m->bio); | |
1042 | } else | |
1043 | bio_endio(m->bio, 0); | |
1044 | ||
f286ba0e | 1045 | cell_defer_no_holder(tc, m->cell); |
e49e5829 JT |
1046 | mempool_free(m, tc->pool->mapping_pool); |
1047 | } | |
1048 | ||
34fbcf62 | 1049 | static int passdown_double_checking_shared_status(struct dm_thin_new_mapping *m) |
e49e5829 | 1050 | { |
34fbcf62 JT |
1051 | /* |
1052 | * We've already unmapped this range of blocks, but before we | |
1053 | * passdown we have to check that these blocks are now unused. | |
1054 | */ | |
1055 | int r; | |
1056 | bool used = true; | |
e49e5829 | 1057 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1058 | struct pool *pool = tc->pool; |
1059 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; | |
104655fd | 1060 | |
34fbcf62 JT |
1061 | while (b != end) { |
1062 | /* find start of unmapped run */ | |
1063 | for (; b < end; b++) { | |
1064 | r = dm_pool_block_is_used(pool->pmd, b, &used); | |
1065 | if (r) | |
1066 | return r; | |
e8088073 | 1067 | |
34fbcf62 JT |
1068 | if (!used) |
1069 | break; | |
19fa1a67 | 1070 | } |
104655fd | 1071 | |
34fbcf62 JT |
1072 | if (b == end) |
1073 | break; | |
1074 | ||
1075 | /* find end of run */ | |
1076 | for (e = b + 1; e != end; e++) { | |
1077 | r = dm_pool_block_is_used(pool->pmd, e, &used); | |
1078 | if (r) | |
1079 | return r; | |
1080 | ||
1081 | if (used) | |
1082 | break; | |
1083 | } | |
1084 | ||
1085 | r = issue_discard(tc, b, e, m->bio); | |
1086 | if (r) | |
1087 | return r; | |
1088 | ||
1089 | b = e; | |
1090 | } | |
1091 | ||
1092 | return 0; | |
104655fd JT |
1093 | } |
1094 | ||
34fbcf62 | 1095 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) |
e49e5829 JT |
1096 | { |
1097 | int r; | |
1098 | struct thin_c *tc = m->tc; | |
34fbcf62 | 1099 | struct pool *pool = tc->pool; |
e49e5829 | 1100 | |
34fbcf62 | 1101 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
e49e5829 | 1102 | if (r) |
34fbcf62 JT |
1103 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
1104 | ||
1105 | else if (m->maybe_shared) | |
1106 | r = passdown_double_checking_shared_status(m); | |
1107 | else | |
1108 | r = issue_discard(tc, m->data_block, m->data_block + (m->virt_end - m->virt_begin), m->bio); | |
e49e5829 | 1109 | |
34fbcf62 JT |
1110 | /* |
1111 | * Even if r is set, there could be sub discards in flight that we | |
1112 | * need to wait for. | |
1113 | */ | |
1114 | bio_endio(m->bio, r); | |
1115 | cell_defer_no_holder(tc, m->cell); | |
1116 | mempool_free(m, pool->mapping_pool); | |
e49e5829 JT |
1117 | } |
1118 | ||
104655fd | 1119 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 1120 | process_mapping_fn *fn) |
991d9fa0 JT |
1121 | { |
1122 | unsigned long flags; | |
1123 | struct list_head maps; | |
a24c2569 | 1124 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
1125 | |
1126 | INIT_LIST_HEAD(&maps); | |
1127 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 1128 | list_splice_init(head, &maps); |
991d9fa0 JT |
1129 | spin_unlock_irqrestore(&pool->lock, flags); |
1130 | ||
1131 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1132 | (*fn)(m); |
991d9fa0 JT |
1133 | } |
1134 | ||
1135 | /* | |
1136 | * Deferred bio jobs. | |
1137 | */ | |
104655fd | 1138 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1139 | { |
4f024f37 KO |
1140 | return bio->bi_iter.bi_size == |
1141 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
1142 | } |
1143 | ||
1144 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1145 | { | |
1146 | return (bio_data_dir(bio) == WRITE) && | |
1147 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1148 | } |
1149 | ||
1150 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1151 | bio_end_io_t *fn) | |
1152 | { | |
1153 | *save = bio->bi_end_io; | |
1154 | bio->bi_end_io = fn; | |
1155 | } | |
1156 | ||
1157 | static int ensure_next_mapping(struct pool *pool) | |
1158 | { | |
1159 | if (pool->next_mapping) | |
1160 | return 0; | |
1161 | ||
1162 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
1163 | ||
1164 | return pool->next_mapping ? 0 : -ENOMEM; | |
1165 | } | |
1166 | ||
a24c2569 | 1167 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1168 | { |
16961b04 | 1169 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
1170 | |
1171 | BUG_ON(!pool->next_mapping); | |
1172 | ||
16961b04 MS |
1173 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
1174 | INIT_LIST_HEAD(&m->list); | |
1175 | m->bio = NULL; | |
1176 | ||
991d9fa0 JT |
1177 | pool->next_mapping = NULL; |
1178 | ||
16961b04 | 1179 | return m; |
991d9fa0 JT |
1180 | } |
1181 | ||
e5aea7b4 JT |
1182 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
1183 | sector_t begin, sector_t end) | |
1184 | { | |
1185 | int r; | |
1186 | struct dm_io_region to; | |
1187 | ||
1188 | to.bdev = tc->pool_dev->bdev; | |
1189 | to.sector = begin; | |
1190 | to.count = end - begin; | |
1191 | ||
1192 | r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); | |
1193 | if (r < 0) { | |
1194 | DMERR_LIMIT("dm_kcopyd_zero() failed"); | |
1195 | copy_complete(1, 1, m); | |
1196 | } | |
1197 | } | |
1198 | ||
452d7a62 | 1199 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
34fbcf62 | 1200 | dm_block_t data_begin, |
452d7a62 MS |
1201 | struct dm_thin_new_mapping *m) |
1202 | { | |
1203 | struct pool *pool = tc->pool; | |
1204 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1205 | ||
1206 | h->overwrite_mapping = m; | |
1207 | m->bio = bio; | |
1208 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
1209 | inc_all_io_entry(pool, bio); | |
34fbcf62 | 1210 | remap_and_issue(tc, bio, data_begin); |
452d7a62 MS |
1211 | } |
1212 | ||
e5aea7b4 JT |
1213 | /* |
1214 | * A partial copy also needs to zero the uncopied region. | |
1215 | */ | |
991d9fa0 | 1216 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
1217 | struct dm_dev *origin, dm_block_t data_origin, |
1218 | dm_block_t data_dest, | |
e5aea7b4 JT |
1219 | struct dm_bio_prison_cell *cell, struct bio *bio, |
1220 | sector_t len) | |
991d9fa0 JT |
1221 | { |
1222 | int r; | |
1223 | struct pool *pool = tc->pool; | |
a24c2569 | 1224 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1225 | |
991d9fa0 | 1226 | m->tc = tc; |
34fbcf62 JT |
1227 | m->virt_begin = virt_block; |
1228 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1229 | m->data_block = data_dest; |
1230 | m->cell = cell; | |
991d9fa0 | 1231 | |
e5aea7b4 JT |
1232 | /* |
1233 | * quiesce action + copy action + an extra reference held for the | |
1234 | * duration of this function (we may need to inc later for a | |
1235 | * partial zero). | |
1236 | */ | |
1237 | atomic_set(&m->prepare_actions, 3); | |
1238 | ||
44feb387 | 1239 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1240 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1241 | |
1242 | /* | |
1243 | * IO to pool_dev remaps to the pool target's data_dev. | |
1244 | * | |
1245 | * If the whole block of data is being overwritten, we can issue the | |
1246 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1247 | */ | |
452d7a62 MS |
1248 | if (io_overwrites_block(pool, bio)) |
1249 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1250 | else { | |
991d9fa0 JT |
1251 | struct dm_io_region from, to; |
1252 | ||
2dd9c257 | 1253 | from.bdev = origin->bdev; |
991d9fa0 | 1254 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1255 | from.count = len; |
991d9fa0 JT |
1256 | |
1257 | to.bdev = tc->pool_dev->bdev; | |
1258 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1259 | to.count = len; |
991d9fa0 JT |
1260 | |
1261 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1262 | 0, copy_complete, m); | |
1263 | if (r < 0) { | |
c397741c | 1264 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
e5aea7b4 JT |
1265 | copy_complete(1, 1, m); |
1266 | ||
1267 | /* | |
1268 | * We allow the zero to be issued, to simplify the | |
1269 | * error path. Otherwise we'd need to start | |
1270 | * worrying about decrementing the prepare_actions | |
1271 | * counter. | |
1272 | */ | |
1273 | } | |
1274 | ||
1275 | /* | |
1276 | * Do we need to zero a tail region? | |
1277 | */ | |
1278 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1279 | atomic_inc(&m->prepare_actions); | |
1280 | ll_zero(tc, m, | |
1281 | data_dest * pool->sectors_per_block + len, | |
1282 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1283 | } |
1284 | } | |
e5aea7b4 JT |
1285 | |
1286 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1287 | } |
1288 | ||
2dd9c257 JT |
1289 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1290 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1291 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1292 | { |
1293 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1294 | data_origin, data_dest, cell, bio, |
1295 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1296 | } |
1297 | ||
991d9fa0 | 1298 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1299 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1300 | struct bio *bio) |
1301 | { | |
1302 | struct pool *pool = tc->pool; | |
a24c2569 | 1303 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1304 | |
50f3c3ef | 1305 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 | 1306 | m->tc = tc; |
34fbcf62 JT |
1307 | m->virt_begin = virt_block; |
1308 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1309 | m->data_block = data_block; |
1310 | m->cell = cell; | |
991d9fa0 JT |
1311 | |
1312 | /* | |
1313 | * If the whole block of data is being overwritten or we are not | |
1314 | * zeroing pre-existing data, we can issue the bio immediately. | |
1315 | * Otherwise we use kcopyd to zero the data first. | |
1316 | */ | |
f8ae7525 MS |
1317 | if (pool->pf.zero_new_blocks) { |
1318 | if (io_overwrites_block(pool, bio)) | |
1319 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
1320 | else | |
1321 | ll_zero(tc, m, data_block * pool->sectors_per_block, | |
1322 | (data_block + 1) * pool->sectors_per_block); | |
1323 | } else | |
991d9fa0 | 1324 | process_prepared_mapping(m); |
e5aea7b4 | 1325 | } |
991d9fa0 | 1326 | |
e5aea7b4 JT |
1327 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1328 | dm_block_t data_dest, | |
1329 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1330 | { | |
1331 | struct pool *pool = tc->pool; | |
1332 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1333 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1334 | ||
1335 | if (virt_block_end <= tc->origin_size) | |
1336 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1337 | virt_block, data_dest, cell, bio, | |
1338 | pool->sectors_per_block); | |
1339 | ||
1340 | else if (virt_block_begin < tc->origin_size) | |
1341 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1342 | virt_block, data_dest, cell, bio, | |
1343 | tc->origin_size - virt_block_begin); | |
1344 | ||
1345 | else | |
1346 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1347 | } |
1348 | ||
2c43fd26 JT |
1349 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1350 | ||
1351 | static void check_for_space(struct pool *pool) | |
1352 | { | |
1353 | int r; | |
1354 | dm_block_t nr_free; | |
1355 | ||
1356 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) | |
1357 | return; | |
1358 | ||
1359 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); | |
1360 | if (r) | |
1361 | return; | |
1362 | ||
1363 | if (nr_free) | |
1364 | set_pool_mode(pool, PM_WRITE); | |
1365 | } | |
1366 | ||
e49e5829 JT |
1367 | /* |
1368 | * A non-zero return indicates read_only or fail_io mode. | |
1369 | * Many callers don't care about the return value. | |
1370 | */ | |
020cc3b5 | 1371 | static int commit(struct pool *pool) |
e49e5829 JT |
1372 | { |
1373 | int r; | |
1374 | ||
8d07e8a5 | 1375 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
1376 | return -EINVAL; |
1377 | ||
020cc3b5 | 1378 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1379 | if (r) |
1380 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
2c43fd26 JT |
1381 | else |
1382 | check_for_space(pool); | |
e49e5829 JT |
1383 | |
1384 | return r; | |
1385 | } | |
1386 | ||
88a6621b JT |
1387 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1388 | { | |
1389 | unsigned long flags; | |
1390 | ||
1391 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1392 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1393 | dm_device_name(pool->pool_md)); | |
1394 | spin_lock_irqsave(&pool->lock, flags); | |
1395 | pool->low_water_triggered = true; | |
1396 | spin_unlock_irqrestore(&pool->lock, flags); | |
1397 | dm_table_event(pool->ti->table); | |
1398 | } | |
1399 | } | |
1400 | ||
991d9fa0 JT |
1401 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1402 | { | |
1403 | int r; | |
1404 | dm_block_t free_blocks; | |
991d9fa0 JT |
1405 | struct pool *pool = tc->pool; |
1406 | ||
3e1a0699 | 1407 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1408 | return -EINVAL; |
1409 | ||
991d9fa0 | 1410 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1411 | if (r) { |
1412 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1413 | return r; |
b5330655 | 1414 | } |
991d9fa0 | 1415 | |
88a6621b | 1416 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1417 | |
1418 | if (!free_blocks) { | |
94563bad MS |
1419 | /* |
1420 | * Try to commit to see if that will free up some | |
1421 | * more space. | |
1422 | */ | |
020cc3b5 JT |
1423 | r = commit(pool); |
1424 | if (r) | |
1425 | return r; | |
991d9fa0 | 1426 | |
94563bad | 1427 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1428 | if (r) { |
1429 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1430 | return r; |
b5330655 | 1431 | } |
991d9fa0 | 1432 | |
94563bad | 1433 | if (!free_blocks) { |
3e1a0699 | 1434 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1435 | return -ENOSPC; |
991d9fa0 JT |
1436 | } |
1437 | } | |
1438 | ||
1439 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1440 | if (r) { |
b5330655 | 1441 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1442 | return r; |
4a02b34e | 1443 | } |
991d9fa0 JT |
1444 | |
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * If we have run out of space, queue bios until the device is | |
1450 | * resumed, presumably after having been reloaded with more space. | |
1451 | */ | |
1452 | static void retry_on_resume(struct bio *bio) | |
1453 | { | |
59c3d2c6 | 1454 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1455 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1456 | unsigned long flags; |
1457 | ||
c140e1c4 MS |
1458 | spin_lock_irqsave(&tc->lock, flags); |
1459 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1460 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1461 | } |
1462 | ||
af91805a | 1463 | static int should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1464 | { |
3e1a0699 JT |
1465 | enum pool_mode m = get_pool_mode(pool); |
1466 | ||
1467 | switch (m) { | |
1468 | case PM_WRITE: | |
1469 | /* Shouldn't get here */ | |
1470 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
af91805a | 1471 | return -EIO; |
3e1a0699 JT |
1472 | |
1473 | case PM_OUT_OF_DATA_SPACE: | |
af91805a | 1474 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
3e1a0699 JT |
1475 | |
1476 | case PM_READ_ONLY: | |
1477 | case PM_FAIL: | |
af91805a | 1478 | return -EIO; |
3e1a0699 JT |
1479 | default: |
1480 | /* Shouldn't get here */ | |
1481 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
af91805a | 1482 | return -EIO; |
3e1a0699 JT |
1483 | } |
1484 | } | |
8c0f0e8c | 1485 | |
3e1a0699 JT |
1486 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1487 | { | |
af91805a MS |
1488 | int error = should_error_unserviceable_bio(pool); |
1489 | ||
1490 | if (error) | |
1491 | bio_endio(bio, error); | |
6d16202b MS |
1492 | else |
1493 | retry_on_resume(bio); | |
8c0f0e8c MS |
1494 | } |
1495 | ||
399caddf | 1496 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1497 | { |
1498 | struct bio *bio; | |
1499 | struct bio_list bios; | |
af91805a | 1500 | int error; |
991d9fa0 | 1501 | |
af91805a MS |
1502 | error = should_error_unserviceable_bio(pool); |
1503 | if (error) { | |
1504 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1505 | return; |
1506 | } | |
1507 | ||
991d9fa0 | 1508 | bio_list_init(&bios); |
6beca5eb | 1509 | cell_release(pool, cell, &bios); |
991d9fa0 | 1510 | |
9d094eeb MS |
1511 | while ((bio = bio_list_pop(&bios))) |
1512 | retry_on_resume(bio); | |
991d9fa0 JT |
1513 | } |
1514 | ||
34fbcf62 JT |
1515 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
1516 | struct dm_bio_prison_cell *virt_cell) | |
104655fd | 1517 | { |
104655fd | 1518 | struct pool *pool = tc->pool; |
34fbcf62 | 1519 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
104655fd | 1520 | |
34fbcf62 JT |
1521 | /* |
1522 | * We don't need to lock the data blocks, since there's no | |
1523 | * passdown. We only lock data blocks for allocation and breaking sharing. | |
1524 | */ | |
1525 | m->tc = tc; | |
1526 | m->virt_begin = virt_cell->key.block_begin; | |
1527 | m->virt_end = virt_cell->key.block_end; | |
1528 | m->cell = virt_cell; | |
1529 | m->bio = virt_cell->holder; | |
104655fd | 1530 | |
34fbcf62 JT |
1531 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1532 | pool->process_prepared_discard(m); | |
1533 | } | |
104655fd | 1534 | |
34fbcf62 JT |
1535 | /* |
1536 | * FIXME: DM local hack to defer parent bios's end_io until we | |
1537 | * _know_ all chained sub range discard bios have completed. | |
1538 | * Will go away once late bio splitting lands upstream! | |
1539 | */ | |
1540 | static inline void __bio_inc_remaining(struct bio *bio) | |
1541 | { | |
1542 | bio->bi_flags |= (1 << BIO_CHAIN); | |
1543 | smp_mb__before_atomic(); | |
1544 | atomic_inc(&bio->__bi_remaining); | |
1545 | } | |
7a7e97ca | 1546 | |
34fbcf62 JT |
1547 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
1548 | struct bio *bio) | |
1549 | { | |
1550 | struct pool *pool = tc->pool; | |
1551 | ||
1552 | int r; | |
1553 | bool maybe_shared; | |
1554 | struct dm_cell_key data_key; | |
1555 | struct dm_bio_prison_cell *data_cell; | |
1556 | struct dm_thin_new_mapping *m; | |
1557 | dm_block_t virt_begin, virt_end, data_begin; | |
1558 | ||
1559 | while (begin != end) { | |
1560 | r = ensure_next_mapping(pool); | |
1561 | if (r) | |
1562 | /* we did our best */ | |
1563 | return; | |
e8088073 | 1564 | |
34fbcf62 JT |
1565 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
1566 | &data_begin, &maybe_shared); | |
1567 | if (r) | |
104655fd | 1568 | /* |
34fbcf62 JT |
1569 | * Silently fail, letting any mappings we've |
1570 | * created complete. | |
104655fd | 1571 | */ |
34fbcf62 JT |
1572 | break; |
1573 | ||
1574 | build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); | |
1575 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { | |
1576 | /* contention, we'll give up with this range */ | |
1577 | begin = virt_end; | |
1578 | continue; | |
104655fd | 1579 | } |
104655fd | 1580 | |
104655fd | 1581 | /* |
34fbcf62 JT |
1582 | * IO may still be going to the destination block. We must |
1583 | * quiesce before we can do the removal. | |
104655fd | 1584 | */ |
34fbcf62 JT |
1585 | m = get_next_mapping(pool); |
1586 | m->tc = tc; | |
1587 | m->maybe_shared = maybe_shared; | |
1588 | m->virt_begin = virt_begin; | |
1589 | m->virt_end = virt_end; | |
1590 | m->data_block = data_begin; | |
1591 | m->cell = data_cell; | |
1592 | m->bio = bio; | |
104655fd | 1593 | |
34fbcf62 JT |
1594 | /* |
1595 | * The parent bio must not complete before sub discard bios are | |
1596 | * chained to it (see __blkdev_issue_discard_async's bio_chain)! | |
1597 | * | |
1598 | * This per-mapping bi_remaining increment is paired with | |
1599 | * the implicit decrement that occurs via bio_endio() in | |
1600 | * process_prepared_discard_{passdown,no_passdown}. | |
1601 | */ | |
1602 | __bio_inc_remaining(bio); | |
1603 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) | |
1604 | pool->process_prepared_discard(m); | |
1605 | ||
1606 | begin = virt_end; | |
104655fd JT |
1607 | } |
1608 | } | |
1609 | ||
34fbcf62 JT |
1610 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
1611 | { | |
1612 | struct bio *bio = virt_cell->holder; | |
1613 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1614 | ||
1615 | /* | |
1616 | * The virt_cell will only get freed once the origin bio completes. | |
1617 | * This means it will remain locked while all the individual | |
1618 | * passdown bios are in flight. | |
1619 | */ | |
1620 | h->cell = virt_cell; | |
1621 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); | |
1622 | ||
1623 | /* | |
1624 | * We complete the bio now, knowing that the bi_remaining field | |
1625 | * will prevent completion until the sub range discards have | |
1626 | * completed. | |
1627 | */ | |
1628 | bio_endio(bio, 0); | |
1629 | } | |
1630 | ||
a374bb21 JT |
1631 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1632 | { | |
34fbcf62 JT |
1633 | dm_block_t begin, end; |
1634 | struct dm_cell_key virt_key; | |
1635 | struct dm_bio_prison_cell *virt_cell; | |
a374bb21 | 1636 | |
34fbcf62 JT |
1637 | get_bio_block_range(tc, bio, &begin, &end); |
1638 | if (begin == end) { | |
1639 | /* | |
1640 | * The discard covers less than a block. | |
1641 | */ | |
1642 | bio_endio(bio, 0); | |
a374bb21 | 1643 | return; |
34fbcf62 | 1644 | } |
a374bb21 | 1645 | |
34fbcf62 JT |
1646 | build_key(tc->td, VIRTUAL, begin, end, &virt_key); |
1647 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) | |
1648 | /* | |
1649 | * Potential starvation issue: We're relying on the | |
1650 | * fs/application being well behaved, and not trying to | |
1651 | * send IO to a region at the same time as discarding it. | |
1652 | * If they do this persistently then it's possible this | |
1653 | * cell will never be granted. | |
1654 | */ | |
1655 | return; | |
1656 | ||
1657 | tc->pool->process_discard_cell(tc, virt_cell); | |
a374bb21 JT |
1658 | } |
1659 | ||
991d9fa0 | 1660 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1661 | struct dm_cell_key *key, |
991d9fa0 | 1662 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1663 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1664 | { |
1665 | int r; | |
1666 | dm_block_t data_block; | |
d6fc2042 | 1667 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1668 | |
1669 | r = alloc_data_block(tc, &data_block); | |
1670 | switch (r) { | |
1671 | case 0: | |
2dd9c257 JT |
1672 | schedule_internal_copy(tc, block, lookup_result->block, |
1673 | data_block, cell, bio); | |
991d9fa0 JT |
1674 | break; |
1675 | ||
1676 | case -ENOSPC: | |
399caddf | 1677 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1678 | break; |
1679 | ||
1680 | default: | |
c397741c MS |
1681 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1682 | __func__, r); | |
d6fc2042 | 1683 | cell_error(pool, cell); |
991d9fa0 JT |
1684 | break; |
1685 | } | |
1686 | } | |
1687 | ||
23ca2bb6 JT |
1688 | static void __remap_and_issue_shared_cell(void *context, |
1689 | struct dm_bio_prison_cell *cell) | |
1690 | { | |
1691 | struct remap_info *info = context; | |
1692 | struct bio *bio; | |
1693 | ||
1694 | while ((bio = bio_list_pop(&cell->bios))) { | |
1695 | if ((bio_data_dir(bio) == WRITE) || | |
1696 | (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))) | |
1697 | bio_list_add(&info->defer_bios, bio); | |
1698 | else { | |
1699 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));; | |
1700 | ||
1701 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1702 | inc_all_io_entry(info->tc->pool, bio); | |
1703 | bio_list_add(&info->issue_bios, bio); | |
1704 | } | |
1705 | } | |
1706 | } | |
1707 | ||
1708 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1709 | struct dm_bio_prison_cell *cell, | |
1710 | dm_block_t block) | |
1711 | { | |
1712 | struct bio *bio; | |
1713 | struct remap_info info; | |
1714 | ||
1715 | info.tc = tc; | |
1716 | bio_list_init(&info.defer_bios); | |
1717 | bio_list_init(&info.issue_bios); | |
1718 | ||
1719 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1720 | &info, cell); | |
1721 | ||
1722 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1723 | thin_defer_bio(tc, bio); | |
1724 | ||
1725 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1726 | remap_and_issue(tc, bio, block); | |
1727 | } | |
1728 | ||
991d9fa0 JT |
1729 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1730 | dm_block_t block, | |
23ca2bb6 JT |
1731 | struct dm_thin_lookup_result *lookup_result, |
1732 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1733 | { |
23ca2bb6 | 1734 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1735 | struct pool *pool = tc->pool; |
44feb387 | 1736 | struct dm_cell_key key; |
991d9fa0 JT |
1737 | |
1738 | /* | |
1739 | * If cell is already occupied, then sharing is already in the process | |
1740 | * of being broken so we have nothing further to do here. | |
1741 | */ | |
1742 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1743 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1744 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1745 | return; |
23ca2bb6 | 1746 | } |
991d9fa0 | 1747 | |
23ca2bb6 JT |
1748 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1749 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1750 | cell_defer_no_holder(tc, virt_cell); | |
1751 | } else { | |
59c3d2c6 | 1752 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1753 | |
44feb387 | 1754 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1755 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1756 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1757 | |
1758 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1759 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1760 | } |
1761 | } | |
1762 | ||
1763 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1764 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1765 | { |
1766 | int r; | |
1767 | dm_block_t data_block; | |
6beca5eb | 1768 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1769 | |
1770 | /* | |
1771 | * Remap empty bios (flushes) immediately, without provisioning. | |
1772 | */ | |
4f024f37 | 1773 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1774 | inc_all_io_entry(pool, bio); |
f286ba0e | 1775 | cell_defer_no_holder(tc, cell); |
e8088073 | 1776 | |
991d9fa0 JT |
1777 | remap_and_issue(tc, bio, 0); |
1778 | return; | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Fill read bios with zeroes and complete them immediately. | |
1783 | */ | |
1784 | if (bio_data_dir(bio) == READ) { | |
1785 | zero_fill_bio(bio); | |
f286ba0e | 1786 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1787 | bio_endio(bio, 0); |
1788 | return; | |
1789 | } | |
1790 | ||
1791 | r = alloc_data_block(tc, &data_block); | |
1792 | switch (r) { | |
1793 | case 0: | |
2dd9c257 JT |
1794 | if (tc->origin_dev) |
1795 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1796 | else | |
1797 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1798 | break; |
1799 | ||
1800 | case -ENOSPC: | |
399caddf | 1801 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1802 | break; |
1803 | ||
1804 | default: | |
c397741c MS |
1805 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1806 | __func__, r); | |
6beca5eb | 1807 | cell_error(pool, cell); |
991d9fa0 JT |
1808 | break; |
1809 | } | |
1810 | } | |
1811 | ||
a374bb21 | 1812 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1813 | { |
1814 | int r; | |
6beca5eb | 1815 | struct pool *pool = tc->pool; |
a374bb21 | 1816 | struct bio *bio = cell->holder; |
991d9fa0 | 1817 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1818 | struct dm_thin_lookup_result lookup_result; |
1819 | ||
a374bb21 JT |
1820 | if (tc->requeue_mode) { |
1821 | cell_requeue(pool, cell); | |
991d9fa0 | 1822 | return; |
a374bb21 | 1823 | } |
991d9fa0 JT |
1824 | |
1825 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1826 | switch (r) { | |
1827 | case 0: | |
23ca2bb6 JT |
1828 | if (lookup_result.shared) |
1829 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1830 | else { | |
6beca5eb | 1831 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1832 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1833 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1834 | } |
991d9fa0 JT |
1835 | break; |
1836 | ||
1837 | case -ENODATA: | |
2dd9c257 | 1838 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1839 | inc_all_io_entry(pool, bio); |
f286ba0e | 1840 | cell_defer_no_holder(tc, cell); |
e8088073 | 1841 | |
e5aea7b4 JT |
1842 | if (bio_end_sector(bio) <= tc->origin_size) |
1843 | remap_to_origin_and_issue(tc, bio); | |
1844 | ||
1845 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1846 | zero_fill_bio(bio); | |
1847 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1848 | remap_to_origin_and_issue(tc, bio); | |
1849 | ||
1850 | } else { | |
1851 | zero_fill_bio(bio); | |
1852 | bio_endio(bio, 0); | |
1853 | } | |
2dd9c257 JT |
1854 | } else |
1855 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1856 | break; |
1857 | ||
1858 | default: | |
c397741c MS |
1859 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1860 | __func__, r); | |
f286ba0e | 1861 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1862 | bio_io_error(bio); |
1863 | break; | |
1864 | } | |
1865 | } | |
1866 | ||
a374bb21 JT |
1867 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1868 | { | |
1869 | struct pool *pool = tc->pool; | |
1870 | dm_block_t block = get_bio_block(tc, bio); | |
1871 | struct dm_bio_prison_cell *cell; | |
1872 | struct dm_cell_key key; | |
1873 | ||
1874 | /* | |
1875 | * If cell is already occupied, then the block is already | |
1876 | * being provisioned so we have nothing further to do here. | |
1877 | */ | |
1878 | build_virtual_key(tc->td, block, &key); | |
1879 | if (bio_detain(pool, &key, bio, &cell)) | |
1880 | return; | |
1881 | ||
1882 | process_cell(tc, cell); | |
1883 | } | |
1884 | ||
1885 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
1886 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
1887 | { |
1888 | int r; | |
1889 | int rw = bio_data_dir(bio); | |
1890 | dm_block_t block = get_bio_block(tc, bio); | |
1891 | struct dm_thin_lookup_result lookup_result; | |
1892 | ||
1893 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1894 | switch (r) { | |
1895 | case 0: | |
a374bb21 | 1896 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 1897 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
1898 | if (cell) |
1899 | cell_defer_no_holder(tc, cell); | |
1900 | } else { | |
e8088073 | 1901 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 1902 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
1903 | if (cell) |
1904 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 1905 | } |
e49e5829 JT |
1906 | break; |
1907 | ||
1908 | case -ENODATA: | |
a374bb21 JT |
1909 | if (cell) |
1910 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 1911 | if (rw != READ) { |
8c0f0e8c | 1912 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1913 | break; |
1914 | } | |
1915 | ||
1916 | if (tc->origin_dev) { | |
e8088073 | 1917 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1918 | remap_to_origin_and_issue(tc, bio); |
1919 | break; | |
1920 | } | |
1921 | ||
1922 | zero_fill_bio(bio); | |
1923 | bio_endio(bio, 0); | |
1924 | break; | |
1925 | ||
1926 | default: | |
c397741c MS |
1927 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1928 | __func__, r); | |
a374bb21 JT |
1929 | if (cell) |
1930 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
1931 | bio_io_error(bio); |
1932 | break; | |
1933 | } | |
1934 | } | |
1935 | ||
a374bb21 JT |
1936 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1937 | { | |
1938 | __process_bio_read_only(tc, bio, NULL); | |
1939 | } | |
1940 | ||
1941 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1942 | { | |
1943 | __process_bio_read_only(tc, cell->holder, cell); | |
1944 | } | |
1945 | ||
3e1a0699 JT |
1946 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1947 | { | |
1948 | bio_endio(bio, 0); | |
1949 | } | |
1950 | ||
e49e5829 JT |
1951 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1952 | { | |
1953 | bio_io_error(bio); | |
1954 | } | |
1955 | ||
a374bb21 JT |
1956 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
1957 | { | |
1958 | cell_success(tc->pool, cell); | |
1959 | } | |
1960 | ||
1961 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
1962 | { | |
1963 | cell_error(tc->pool, cell); | |
1964 | } | |
1965 | ||
ac8c3f3d JT |
1966 | /* |
1967 | * FIXME: should we also commit due to size of transaction, measured in | |
1968 | * metadata blocks? | |
1969 | */ | |
905e51b3 JT |
1970 | static int need_commit_due_to_time(struct pool *pool) |
1971 | { | |
0f30af98 MS |
1972 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
1973 | pool->last_commit_jiffies + COMMIT_PERIOD); | |
905e51b3 JT |
1974 | } |
1975 | ||
67324ea1 MS |
1976 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1977 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1978 | ||
1979 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1980 | { | |
1981 | struct rb_node **rbp, *parent; | |
1982 | struct dm_thin_endio_hook *pbd; | |
1983 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1984 | ||
1985 | rbp = &tc->sort_bio_list.rb_node; | |
1986 | parent = NULL; | |
1987 | while (*rbp) { | |
1988 | parent = *rbp; | |
1989 | pbd = thin_pbd(parent); | |
1990 | ||
1991 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1992 | rbp = &(*rbp)->rb_left; | |
1993 | else | |
1994 | rbp = &(*rbp)->rb_right; | |
1995 | } | |
1996 | ||
1997 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1998 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1999 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
2000 | } | |
2001 | ||
2002 | static void __extract_sorted_bios(struct thin_c *tc) | |
2003 | { | |
2004 | struct rb_node *node; | |
2005 | struct dm_thin_endio_hook *pbd; | |
2006 | struct bio *bio; | |
2007 | ||
2008 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
2009 | pbd = thin_pbd(node); | |
2010 | bio = thin_bio(pbd); | |
2011 | ||
2012 | bio_list_add(&tc->deferred_bio_list, bio); | |
2013 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
2014 | } | |
2015 | ||
2016 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
2017 | } | |
2018 | ||
2019 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
2020 | { | |
2021 | struct bio *bio; | |
2022 | struct bio_list bios; | |
2023 | ||
2024 | bio_list_init(&bios); | |
2025 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
2026 | bio_list_init(&tc->deferred_bio_list); | |
2027 | ||
2028 | /* Sort deferred_bio_list using rb-tree */ | |
2029 | while ((bio = bio_list_pop(&bios))) | |
2030 | __thin_bio_rb_add(tc, bio); | |
2031 | ||
2032 | /* | |
2033 | * Transfer the sorted bios in sort_bio_list back to | |
2034 | * deferred_bio_list to allow lockless submission of | |
2035 | * all bios. | |
2036 | */ | |
2037 | __extract_sorted_bios(tc); | |
2038 | } | |
2039 | ||
c140e1c4 | 2040 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 2041 | { |
c140e1c4 | 2042 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2043 | unsigned long flags; |
2044 | struct bio *bio; | |
2045 | struct bio_list bios; | |
67324ea1 | 2046 | struct blk_plug plug; |
8a01a6af | 2047 | unsigned count = 0; |
991d9fa0 | 2048 | |
c140e1c4 | 2049 | if (tc->requeue_mode) { |
42d6a8ce | 2050 | error_thin_bio_list(tc, &tc->deferred_bio_list, DM_ENDIO_REQUEUE); |
c140e1c4 MS |
2051 | return; |
2052 | } | |
2053 | ||
991d9fa0 JT |
2054 | bio_list_init(&bios); |
2055 | ||
c140e1c4 | 2056 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
2057 | |
2058 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
2059 | spin_unlock_irqrestore(&tc->lock, flags); | |
2060 | return; | |
2061 | } | |
2062 | ||
2063 | __sort_thin_deferred_bios(tc); | |
2064 | ||
c140e1c4 MS |
2065 | bio_list_merge(&bios, &tc->deferred_bio_list); |
2066 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 2067 | |
c140e1c4 | 2068 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 2069 | |
67324ea1 | 2070 | blk_start_plug(&plug); |
991d9fa0 | 2071 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
2072 | /* |
2073 | * If we've got no free new_mapping structs, and processing | |
2074 | * this bio might require one, we pause until there are some | |
2075 | * prepared mappings to process. | |
2076 | */ | |
2077 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
2078 | spin_lock_irqsave(&tc->lock, flags); |
2079 | bio_list_add(&tc->deferred_bio_list, bio); | |
2080 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
2081 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2082 | break; |
2083 | } | |
104655fd JT |
2084 | |
2085 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 2086 | pool->process_discard(tc, bio); |
104655fd | 2087 | else |
e49e5829 | 2088 | pool->process_bio(tc, bio); |
8a01a6af JT |
2089 | |
2090 | if ((count++ & 127) == 0) { | |
7d327fe0 | 2091 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
2092 | dm_pool_issue_prefetches(pool->pmd); |
2093 | } | |
991d9fa0 | 2094 | } |
67324ea1 | 2095 | blk_finish_plug(&plug); |
c140e1c4 MS |
2096 | } |
2097 | ||
ac4c3f34 JT |
2098 | static int cmp_cells(const void *lhs, const void *rhs) |
2099 | { | |
2100 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
2101 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
2102 | ||
2103 | BUG_ON(!lhs_cell->holder); | |
2104 | BUG_ON(!rhs_cell->holder); | |
2105 | ||
2106 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
2107 | return -1; | |
2108 | ||
2109 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
2110 | return 1; | |
2111 | ||
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) | |
2116 | { | |
2117 | unsigned count = 0; | |
2118 | struct dm_bio_prison_cell *cell, *tmp; | |
2119 | ||
2120 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
2121 | if (count >= CELL_SORT_ARRAY_SIZE) | |
2122 | break; | |
2123 | ||
2124 | pool->cell_sort_array[count++] = cell; | |
2125 | list_del(&cell->user_list); | |
2126 | } | |
2127 | ||
2128 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
2129 | ||
2130 | return count; | |
2131 | } | |
2132 | ||
a374bb21 JT |
2133 | static void process_thin_deferred_cells(struct thin_c *tc) |
2134 | { | |
2135 | struct pool *pool = tc->pool; | |
2136 | unsigned long flags; | |
2137 | struct list_head cells; | |
ac4c3f34 JT |
2138 | struct dm_bio_prison_cell *cell; |
2139 | unsigned i, j, count; | |
a374bb21 JT |
2140 | |
2141 | INIT_LIST_HEAD(&cells); | |
2142 | ||
2143 | spin_lock_irqsave(&tc->lock, flags); | |
2144 | list_splice_init(&tc->deferred_cells, &cells); | |
2145 | spin_unlock_irqrestore(&tc->lock, flags); | |
2146 | ||
2147 | if (list_empty(&cells)) | |
2148 | return; | |
2149 | ||
ac4c3f34 JT |
2150 | do { |
2151 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 2152 | |
ac4c3f34 JT |
2153 | for (i = 0; i < count; i++) { |
2154 | cell = pool->cell_sort_array[i]; | |
2155 | BUG_ON(!cell->holder); | |
a374bb21 | 2156 | |
ac4c3f34 JT |
2157 | /* |
2158 | * If we've got no free new_mapping structs, and processing | |
2159 | * this bio might require one, we pause until there are some | |
2160 | * prepared mappings to process. | |
2161 | */ | |
2162 | if (ensure_next_mapping(pool)) { | |
2163 | for (j = i; j < count; j++) | |
2164 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
2165 | ||
2166 | spin_lock_irqsave(&tc->lock, flags); | |
2167 | list_splice(&cells, &tc->deferred_cells); | |
2168 | spin_unlock_irqrestore(&tc->lock, flags); | |
2169 | return; | |
2170 | } | |
2171 | ||
2172 | if (cell->holder->bi_rw & REQ_DISCARD) | |
2173 | pool->process_discard_cell(tc, cell); | |
2174 | else | |
2175 | pool->process_cell(tc, cell); | |
2176 | } | |
2177 | } while (!list_empty(&cells)); | |
a374bb21 JT |
2178 | } |
2179 | ||
b10ebd34 JT |
2180 | static void thin_get(struct thin_c *tc); |
2181 | static void thin_put(struct thin_c *tc); | |
2182 | ||
2183 | /* | |
2184 | * We can't hold rcu_read_lock() around code that can block. So we | |
2185 | * find a thin with the rcu lock held; bump a refcount; then drop | |
2186 | * the lock. | |
2187 | */ | |
2188 | static struct thin_c *get_first_thin(struct pool *pool) | |
2189 | { | |
2190 | struct thin_c *tc = NULL; | |
2191 | ||
2192 | rcu_read_lock(); | |
2193 | if (!list_empty(&pool->active_thins)) { | |
2194 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
2195 | thin_get(tc); | |
2196 | } | |
2197 | rcu_read_unlock(); | |
2198 | ||
2199 | return tc; | |
2200 | } | |
2201 | ||
2202 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
2203 | { | |
2204 | struct thin_c *old_tc = tc; | |
2205 | ||
2206 | rcu_read_lock(); | |
2207 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
2208 | thin_get(tc); | |
2209 | thin_put(old_tc); | |
2210 | rcu_read_unlock(); | |
2211 | return tc; | |
2212 | } | |
2213 | thin_put(old_tc); | |
2214 | rcu_read_unlock(); | |
2215 | ||
2216 | return NULL; | |
2217 | } | |
2218 | ||
c140e1c4 MS |
2219 | static void process_deferred_bios(struct pool *pool) |
2220 | { | |
2221 | unsigned long flags; | |
2222 | struct bio *bio; | |
2223 | struct bio_list bios; | |
2224 | struct thin_c *tc; | |
2225 | ||
b10ebd34 JT |
2226 | tc = get_first_thin(pool); |
2227 | while (tc) { | |
a374bb21 | 2228 | process_thin_deferred_cells(tc); |
c140e1c4 | 2229 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
2230 | tc = get_next_thin(pool, tc); |
2231 | } | |
991d9fa0 JT |
2232 | |
2233 | /* | |
2234 | * If there are any deferred flush bios, we must commit | |
2235 | * the metadata before issuing them. | |
2236 | */ | |
2237 | bio_list_init(&bios); | |
2238 | spin_lock_irqsave(&pool->lock, flags); | |
2239 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
2240 | bio_list_init(&pool->deferred_flush_bios); | |
2241 | spin_unlock_irqrestore(&pool->lock, flags); | |
2242 | ||
4d1662a3 MS |
2243 | if (bio_list_empty(&bios) && |
2244 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
2245 | return; |
2246 | ||
020cc3b5 | 2247 | if (commit(pool)) { |
991d9fa0 JT |
2248 | while ((bio = bio_list_pop(&bios))) |
2249 | bio_io_error(bio); | |
2250 | return; | |
2251 | } | |
905e51b3 | 2252 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2253 | |
2254 | while ((bio = bio_list_pop(&bios))) | |
2255 | generic_make_request(bio); | |
2256 | } | |
2257 | ||
2258 | static void do_worker(struct work_struct *ws) | |
2259 | { | |
2260 | struct pool *pool = container_of(ws, struct pool, worker); | |
2261 | ||
7d327fe0 | 2262 | throttle_work_start(&pool->throttle); |
8a01a6af | 2263 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 2264 | throttle_work_update(&pool->throttle); |
e49e5829 | 2265 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 2266 | throttle_work_update(&pool->throttle); |
e49e5829 | 2267 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 2268 | throttle_work_update(&pool->throttle); |
991d9fa0 | 2269 | process_deferred_bios(pool); |
7d327fe0 | 2270 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
2271 | } |
2272 | ||
905e51b3 JT |
2273 | /* |
2274 | * We want to commit periodically so that not too much | |
2275 | * unwritten data builds up. | |
2276 | */ | |
2277 | static void do_waker(struct work_struct *ws) | |
2278 | { | |
2279 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
2280 | wake_worker(pool); | |
2281 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
2282 | } | |
2283 | ||
85ad643b JT |
2284 | /* |
2285 | * We're holding onto IO to allow userland time to react. After the | |
2286 | * timeout either the pool will have been resized (and thus back in | |
2287 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
2288 | */ | |
2289 | static void do_no_space_timeout(struct work_struct *ws) | |
2290 | { | |
2291 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2292 | no_space_timeout); | |
2293 | ||
2294 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
2295 | set_pool_mode(pool, PM_READ_ONLY); | |
2296 | } | |
2297 | ||
991d9fa0 JT |
2298 | /*----------------------------------------------------------------*/ |
2299 | ||
e7a3e871 | 2300 | struct pool_work { |
738211f7 | 2301 | struct work_struct worker; |
e7a3e871 JT |
2302 | struct completion complete; |
2303 | }; | |
2304 | ||
2305 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2306 | { | |
2307 | return container_of(ws, struct pool_work, worker); | |
2308 | } | |
2309 | ||
2310 | static void pool_work_complete(struct pool_work *pw) | |
2311 | { | |
2312 | complete(&pw->complete); | |
2313 | } | |
738211f7 | 2314 | |
e7a3e871 JT |
2315 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2316 | void (*fn)(struct work_struct *)) | |
2317 | { | |
2318 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2319 | init_completion(&pw->complete); | |
2320 | queue_work(pool->wq, &pw->worker); | |
2321 | wait_for_completion(&pw->complete); | |
2322 | } | |
2323 | ||
2324 | /*----------------------------------------------------------------*/ | |
2325 | ||
2326 | struct noflush_work { | |
2327 | struct pool_work pw; | |
2328 | struct thin_c *tc; | |
738211f7 JT |
2329 | }; |
2330 | ||
e7a3e871 | 2331 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2332 | { |
e7a3e871 | 2333 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2334 | } |
2335 | ||
2336 | static void do_noflush_start(struct work_struct *ws) | |
2337 | { | |
e7a3e871 | 2338 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2339 | w->tc->requeue_mode = true; |
2340 | requeue_io(w->tc); | |
e7a3e871 | 2341 | pool_work_complete(&w->pw); |
738211f7 JT |
2342 | } |
2343 | ||
2344 | static void do_noflush_stop(struct work_struct *ws) | |
2345 | { | |
e7a3e871 | 2346 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2347 | w->tc->requeue_mode = false; |
e7a3e871 | 2348 | pool_work_complete(&w->pw); |
738211f7 JT |
2349 | } |
2350 | ||
2351 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2352 | { | |
2353 | struct noflush_work w; | |
2354 | ||
738211f7 | 2355 | w.tc = tc; |
e7a3e871 | 2356 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2357 | } |
2358 | ||
2359 | /*----------------------------------------------------------------*/ | |
2360 | ||
e49e5829 JT |
2361 | static enum pool_mode get_pool_mode(struct pool *pool) |
2362 | { | |
2363 | return pool->pf.mode; | |
2364 | } | |
2365 | ||
3e1a0699 JT |
2366 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
2367 | { | |
2368 | dm_table_event(pool->ti->table); | |
2369 | DMINFO("%s: switching pool to %s mode", | |
2370 | dm_device_name(pool->pool_md), new_mode); | |
2371 | } | |
2372 | ||
34fbcf62 JT |
2373 | static bool passdown_enabled(struct pool_c *pt) |
2374 | { | |
2375 | return pt->adjusted_pf.discard_passdown; | |
2376 | } | |
2377 | ||
2378 | static void set_discard_callbacks(struct pool *pool) | |
2379 | { | |
2380 | struct pool_c *pt = pool->ti->private; | |
2381 | ||
2382 | if (passdown_enabled(pt)) { | |
2383 | pool->process_discard_cell = process_discard_cell_passdown; | |
2384 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
2385 | } else { | |
2386 | pool->process_discard_cell = process_discard_cell_no_passdown; | |
2387 | pool->process_prepared_discard = process_prepared_discard_no_passdown; | |
2388 | } | |
2389 | } | |
2390 | ||
8b64e881 | 2391 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2392 | { |
cdc2b415 | 2393 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2394 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2395 | enum pool_mode old_mode = get_pool_mode(pool); | |
80c57893 | 2396 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2397 | |
2398 | /* | |
2399 | * Never allow the pool to transition to PM_WRITE mode if user | |
2400 | * intervention is required to verify metadata and data consistency. | |
2401 | */ | |
2402 | if (new_mode == PM_WRITE && needs_check) { | |
2403 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2404 | dm_device_name(pool->pool_md)); | |
2405 | if (old_mode != new_mode) | |
2406 | new_mode = old_mode; | |
2407 | else | |
2408 | new_mode = PM_READ_ONLY; | |
2409 | } | |
2410 | /* | |
2411 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2412 | * not going to recover without a thin_repair. So we never let the | |
2413 | * pool move out of the old mode. | |
2414 | */ | |
2415 | if (old_mode == PM_FAIL) | |
2416 | new_mode = old_mode; | |
e49e5829 | 2417 | |
8b64e881 | 2418 | switch (new_mode) { |
e49e5829 | 2419 | case PM_FAIL: |
8b64e881 | 2420 | if (old_mode != new_mode) |
3e1a0699 | 2421 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 2422 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2423 | pool->process_bio = process_bio_fail; |
2424 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2425 | pool->process_cell = process_cell_fail; |
2426 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2427 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2428 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2429 | |
2430 | error_retry_list(pool); | |
e49e5829 JT |
2431 | break; |
2432 | ||
2433 | case PM_READ_ONLY: | |
8b64e881 | 2434 | if (old_mode != new_mode) |
3e1a0699 JT |
2435 | notify_of_pool_mode_change(pool, "read-only"); |
2436 | dm_pool_metadata_read_only(pool->pmd); | |
2437 | pool->process_bio = process_bio_read_only; | |
2438 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2439 | pool->process_cell = process_cell_read_only; |
2440 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 | 2441 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
34fbcf62 | 2442 | pool->process_prepared_discard = process_prepared_discard_success; |
3e1a0699 JT |
2443 | |
2444 | error_retry_list(pool); | |
2445 | break; | |
2446 | ||
2447 | case PM_OUT_OF_DATA_SPACE: | |
2448 | /* | |
2449 | * Ideally we'd never hit this state; the low water mark | |
2450 | * would trigger userland to extend the pool before we | |
2451 | * completely run out of data space. However, many small | |
2452 | * IOs to unprovisioned space can consume data space at an | |
2453 | * alarming rate. Adjust your low water mark if you're | |
2454 | * frequently seeing this mode. | |
2455 | */ | |
2456 | if (old_mode != new_mode) | |
2457 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
2458 | pool->process_bio = process_bio_read_only; | |
a374bb21 JT |
2459 | pool->process_discard = process_discard_bio; |
2460 | pool->process_cell = process_cell_read_only; | |
3e1a0699 | 2461 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2462 | set_discard_callbacks(pool); |
85ad643b | 2463 | |
80c57893 MS |
2464 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2465 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2466 | break; |
2467 | ||
2468 | case PM_WRITE: | |
8b64e881 | 2469 | if (old_mode != new_mode) |
3e1a0699 | 2470 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 2471 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2472 | pool->process_bio = process_bio; |
a374bb21 JT |
2473 | pool->process_discard = process_discard_bio; |
2474 | pool->process_cell = process_cell; | |
e49e5829 | 2475 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2476 | set_discard_callbacks(pool); |
e49e5829 JT |
2477 | break; |
2478 | } | |
8b64e881 MS |
2479 | |
2480 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2481 | /* |
2482 | * The pool mode may have changed, sync it so bind_control_target() | |
2483 | * doesn't cause an unexpected mode transition on resume. | |
2484 | */ | |
2485 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
2486 | } |
2487 | ||
07f2b6e0 | 2488 | static void abort_transaction(struct pool *pool) |
b5330655 | 2489 | { |
07f2b6e0 MS |
2490 | const char *dev_name = dm_device_name(pool->pool_md); |
2491 | ||
2492 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2493 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2494 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2495 | set_pool_mode(pool, PM_FAIL); | |
2496 | } | |
2497 | ||
2498 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2499 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2500 | set_pool_mode(pool, PM_FAIL); | |
2501 | } | |
2502 | } | |
399caddf | 2503 | |
07f2b6e0 MS |
2504 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2505 | { | |
b5330655 JT |
2506 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2507 | dm_device_name(pool->pool_md), op, r); | |
2508 | ||
07f2b6e0 | 2509 | abort_transaction(pool); |
b5330655 JT |
2510 | set_pool_mode(pool, PM_READ_ONLY); |
2511 | } | |
2512 | ||
e49e5829 JT |
2513 | /*----------------------------------------------------------------*/ |
2514 | ||
991d9fa0 JT |
2515 | /* |
2516 | * Mapping functions. | |
2517 | */ | |
2518 | ||
2519 | /* | |
2520 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2521 | */ | |
2522 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2523 | { | |
2524 | unsigned long flags; | |
2525 | struct pool *pool = tc->pool; | |
2526 | ||
c140e1c4 MS |
2527 | spin_lock_irqsave(&tc->lock, flags); |
2528 | bio_list_add(&tc->deferred_bio_list, bio); | |
2529 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2530 | |
2531 | wake_worker(pool); | |
2532 | } | |
2533 | ||
7d327fe0 JT |
2534 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2535 | { | |
2536 | struct pool *pool = tc->pool; | |
2537 | ||
2538 | throttle_lock(&pool->throttle); | |
2539 | thin_defer_bio(tc, bio); | |
2540 | throttle_unlock(&pool->throttle); | |
2541 | } | |
2542 | ||
a374bb21 JT |
2543 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2544 | { | |
2545 | unsigned long flags; | |
2546 | struct pool *pool = tc->pool; | |
2547 | ||
2548 | throttle_lock(&pool->throttle); | |
2549 | spin_lock_irqsave(&tc->lock, flags); | |
2550 | list_add_tail(&cell->user_list, &tc->deferred_cells); | |
2551 | spin_unlock_irqrestore(&tc->lock, flags); | |
2552 | throttle_unlock(&pool->throttle); | |
2553 | ||
2554 | wake_worker(pool); | |
2555 | } | |
2556 | ||
59c3d2c6 | 2557 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2558 | { |
59c3d2c6 | 2559 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2560 | |
2561 | h->tc = tc; | |
2562 | h->shared_read_entry = NULL; | |
e8088073 | 2563 | h->all_io_entry = NULL; |
eb2aa48d | 2564 | h->overwrite_mapping = NULL; |
34fbcf62 | 2565 | h->cell = NULL; |
eb2aa48d JT |
2566 | } |
2567 | ||
991d9fa0 JT |
2568 | /* |
2569 | * Non-blocking function called from the thin target's map function. | |
2570 | */ | |
7de3ee57 | 2571 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2572 | { |
2573 | int r; | |
2574 | struct thin_c *tc = ti->private; | |
2575 | dm_block_t block = get_bio_block(tc, bio); | |
2576 | struct dm_thin_device *td = tc->td; | |
2577 | struct dm_thin_lookup_result result; | |
a374bb21 | 2578 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2579 | struct dm_cell_key key; |
991d9fa0 | 2580 | |
59c3d2c6 | 2581 | thin_hook_bio(tc, bio); |
e49e5829 | 2582 | |
738211f7 JT |
2583 | if (tc->requeue_mode) { |
2584 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
2585 | return DM_MAPIO_SUBMITTED; | |
2586 | } | |
2587 | ||
e49e5829 JT |
2588 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2589 | bio_io_error(bio); | |
2590 | return DM_MAPIO_SUBMITTED; | |
2591 | } | |
2592 | ||
104655fd | 2593 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
7d327fe0 | 2594 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2595 | return DM_MAPIO_SUBMITTED; |
2596 | } | |
2597 | ||
c822ed96 JT |
2598 | /* |
2599 | * We must hold the virtual cell before doing the lookup, otherwise | |
2600 | * there's a race with discard. | |
2601 | */ | |
2602 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2603 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2604 | return DM_MAPIO_SUBMITTED; |
2605 | ||
991d9fa0 JT |
2606 | r = dm_thin_find_block(td, block, 0, &result); |
2607 | ||
2608 | /* | |
2609 | * Note that we defer readahead too. | |
2610 | */ | |
2611 | switch (r) { | |
2612 | case 0: | |
2613 | if (unlikely(result.shared)) { | |
2614 | /* | |
2615 | * We have a race condition here between the | |
2616 | * result.shared value returned by the lookup and | |
2617 | * snapshot creation, which may cause new | |
2618 | * sharing. | |
2619 | * | |
2620 | * To avoid this always quiesce the origin before | |
2621 | * taking the snap. You want to do this anyway to | |
2622 | * ensure a consistent application view | |
2623 | * (i.e. lockfs). | |
2624 | * | |
2625 | * More distant ancestors are irrelevant. The | |
2626 | * shared flag will be set in their case. | |
2627 | */ | |
a374bb21 | 2628 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2629 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2630 | } |
e8088073 | 2631 | |
e8088073 | 2632 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2633 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2634 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2635 | return DM_MAPIO_SUBMITTED; |
2636 | } | |
2637 | ||
2638 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2639 | cell_defer_no_holder(tc, data_cell); |
2640 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2641 | |
2642 | remap(tc, bio, result.block); | |
2643 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2644 | |
2645 | case -ENODATA: | |
e49e5829 | 2646 | case -EWOULDBLOCK: |
a374bb21 | 2647 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2648 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2649 | |
2650 | default: | |
2651 | /* | |
2652 | * Must always call bio_io_error on failure. | |
2653 | * dm_thin_find_block can fail with -EINVAL if the | |
2654 | * pool is switched to fail-io mode. | |
2655 | */ | |
2656 | bio_io_error(bio); | |
a374bb21 | 2657 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2658 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2659 | } |
991d9fa0 JT |
2660 | } |
2661 | ||
2662 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2663 | { | |
991d9fa0 | 2664 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 2665 | struct request_queue *q; |
991d9fa0 | 2666 | |
760fe67e MS |
2667 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
2668 | return 1; | |
991d9fa0 | 2669 | |
760fe67e MS |
2670 | q = bdev_get_queue(pt->data_dev->bdev); |
2671 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
2672 | } |
2673 | ||
c140e1c4 | 2674 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2675 | { |
c140e1c4 MS |
2676 | unsigned long flags; |
2677 | struct thin_c *tc; | |
2678 | ||
2679 | rcu_read_lock(); | |
2680 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
2681 | spin_lock_irqsave(&tc->lock, flags); | |
2682 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
2683 | bio_list_init(&tc->retry_on_resume_list); | |
2684 | spin_unlock_irqrestore(&tc->lock, flags); | |
2685 | } | |
2686 | rcu_read_unlock(); | |
991d9fa0 JT |
2687 | } |
2688 | ||
2689 | /*---------------------------------------------------------------- | |
2690 | * Binding of control targets to a pool object | |
2691 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2692 | static bool data_dev_supports_discard(struct pool_c *pt) |
2693 | { | |
2694 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2695 | ||
2696 | return q && blk_queue_discard(q); | |
2697 | } | |
2698 | ||
58051b94 JT |
2699 | static bool is_factor(sector_t block_size, uint32_t n) |
2700 | { | |
2701 | return !sector_div(block_size, n); | |
2702 | } | |
2703 | ||
9bc142dd MS |
2704 | /* |
2705 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2706 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2707 | */ |
0424caa1 | 2708 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2709 | { |
0424caa1 MS |
2710 | struct pool *pool = pt->pool; |
2711 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2712 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
0424caa1 | 2713 | const char *reason = NULL; |
9bc142dd MS |
2714 | char buf[BDEVNAME_SIZE]; |
2715 | ||
0424caa1 | 2716 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2717 | return; |
2718 | ||
0424caa1 MS |
2719 | if (!data_dev_supports_discard(pt)) |
2720 | reason = "discard unsupported"; | |
2721 | ||
2722 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2723 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2724 | |
0424caa1 MS |
2725 | if (reason) { |
2726 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2727 | pt->adjusted_pf.discard_passdown = false; | |
2728 | } | |
9bc142dd MS |
2729 | } |
2730 | ||
991d9fa0 JT |
2731 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2732 | { | |
2733 | struct pool_c *pt = ti->private; | |
2734 | ||
e49e5829 | 2735 | /* |
9b7aaa64 | 2736 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2737 | */ |
07f2b6e0 | 2738 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2739 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2740 | |
8b64e881 MS |
2741 | /* |
2742 | * Don't change the pool's mode until set_pool_mode() below. | |
2743 | * Otherwise the pool's process_* function pointers may | |
2744 | * not match the desired pool mode. | |
2745 | */ | |
2746 | pt->adjusted_pf.mode = old_mode; | |
2747 | ||
2748 | pool->ti = ti; | |
2749 | pool->pf = pt->adjusted_pf; | |
2750 | pool->low_water_blocks = pt->low_water_blocks; | |
2751 | ||
9bc142dd | 2752 | set_pool_mode(pool, new_mode); |
f402693d | 2753 | |
991d9fa0 JT |
2754 | return 0; |
2755 | } | |
2756 | ||
2757 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2758 | { | |
2759 | if (pool->ti == ti) | |
2760 | pool->ti = NULL; | |
2761 | } | |
2762 | ||
2763 | /*---------------------------------------------------------------- | |
2764 | * Pool creation | |
2765 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2766 | /* Initialize pool features. */ |
2767 | static void pool_features_init(struct pool_features *pf) | |
2768 | { | |
e49e5829 | 2769 | pf->mode = PM_WRITE; |
9bc142dd MS |
2770 | pf->zero_new_blocks = true; |
2771 | pf->discard_enabled = true; | |
2772 | pf->discard_passdown = true; | |
787a996c | 2773 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2774 | } |
2775 | ||
991d9fa0 JT |
2776 | static void __pool_destroy(struct pool *pool) |
2777 | { | |
2778 | __pool_table_remove(pool); | |
2779 | ||
2780 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2781 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2782 | ||
44feb387 | 2783 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2784 | dm_kcopyd_client_destroy(pool->copier); |
2785 | ||
2786 | if (pool->wq) | |
2787 | destroy_workqueue(pool->wq); | |
2788 | ||
2789 | if (pool->next_mapping) | |
2790 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2791 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2792 | dm_deferred_set_destroy(pool->shared_read_ds); |
2793 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2794 | kfree(pool); |
2795 | } | |
2796 | ||
a24c2569 | 2797 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2798 | |
991d9fa0 JT |
2799 | static struct pool *pool_create(struct mapped_device *pool_md, |
2800 | struct block_device *metadata_dev, | |
e49e5829 JT |
2801 | unsigned long block_size, |
2802 | int read_only, char **error) | |
991d9fa0 JT |
2803 | { |
2804 | int r; | |
2805 | void *err_p; | |
2806 | struct pool *pool; | |
2807 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2808 | bool format_device = read_only ? false : true; |
991d9fa0 | 2809 | |
e49e5829 | 2810 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2811 | if (IS_ERR(pmd)) { |
2812 | *error = "Error creating metadata object"; | |
2813 | return (struct pool *)pmd; | |
2814 | } | |
2815 | ||
2816 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2817 | if (!pool) { | |
2818 | *error = "Error allocating memory for pool"; | |
2819 | err_p = ERR_PTR(-ENOMEM); | |
2820 | goto bad_pool; | |
2821 | } | |
2822 | ||
2823 | pool->pmd = pmd; | |
2824 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2825 | if (block_size & (block_size - 1)) |
2826 | pool->sectors_per_block_shift = -1; | |
2827 | else | |
2828 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2829 | pool->low_water_blocks = 0; |
67e2e2b2 | 2830 | pool_features_init(&pool->pf); |
a195db2d | 2831 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2832 | if (!pool->prison) { |
2833 | *error = "Error creating pool's bio prison"; | |
2834 | err_p = ERR_PTR(-ENOMEM); | |
2835 | goto bad_prison; | |
2836 | } | |
2837 | ||
df5d2e90 | 2838 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2839 | if (IS_ERR(pool->copier)) { |
2840 | r = PTR_ERR(pool->copier); | |
2841 | *error = "Error creating pool's kcopyd client"; | |
2842 | err_p = ERR_PTR(r); | |
2843 | goto bad_kcopyd_client; | |
2844 | } | |
2845 | ||
2846 | /* | |
2847 | * Create singlethreaded workqueue that will service all devices | |
2848 | * that use this metadata. | |
2849 | */ | |
2850 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2851 | if (!pool->wq) { | |
2852 | *error = "Error creating pool's workqueue"; | |
2853 | err_p = ERR_PTR(-ENOMEM); | |
2854 | goto bad_wq; | |
2855 | } | |
2856 | ||
7d327fe0 | 2857 | throttle_init(&pool->throttle); |
991d9fa0 | 2858 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 2859 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2860 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2861 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2862 | bio_list_init(&pool->deferred_flush_bios); |
2863 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2864 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2865 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2866 | pool->low_water_triggered = false; |
80e96c54 | 2867 | pool->suspended = true; |
44feb387 MS |
2868 | |
2869 | pool->shared_read_ds = dm_deferred_set_create(); | |
2870 | if (!pool->shared_read_ds) { | |
2871 | *error = "Error creating pool's shared read deferred set"; | |
2872 | err_p = ERR_PTR(-ENOMEM); | |
2873 | goto bad_shared_read_ds; | |
2874 | } | |
2875 | ||
2876 | pool->all_io_ds = dm_deferred_set_create(); | |
2877 | if (!pool->all_io_ds) { | |
2878 | *error = "Error creating pool's all io deferred set"; | |
2879 | err_p = ERR_PTR(-ENOMEM); | |
2880 | goto bad_all_io_ds; | |
2881 | } | |
991d9fa0 JT |
2882 | |
2883 | pool->next_mapping = NULL; | |
a24c2569 MS |
2884 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2885 | _new_mapping_cache); | |
991d9fa0 JT |
2886 | if (!pool->mapping_pool) { |
2887 | *error = "Error creating pool's mapping mempool"; | |
2888 | err_p = ERR_PTR(-ENOMEM); | |
2889 | goto bad_mapping_pool; | |
2890 | } | |
2891 | ||
991d9fa0 | 2892 | pool->ref_count = 1; |
905e51b3 | 2893 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2894 | pool->pool_md = pool_md; |
2895 | pool->md_dev = metadata_dev; | |
2896 | __pool_table_insert(pool); | |
2897 | ||
2898 | return pool; | |
2899 | ||
991d9fa0 | 2900 | bad_mapping_pool: |
44feb387 MS |
2901 | dm_deferred_set_destroy(pool->all_io_ds); |
2902 | bad_all_io_ds: | |
2903 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2904 | bad_shared_read_ds: | |
991d9fa0 JT |
2905 | destroy_workqueue(pool->wq); |
2906 | bad_wq: | |
2907 | dm_kcopyd_client_destroy(pool->copier); | |
2908 | bad_kcopyd_client: | |
44feb387 | 2909 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2910 | bad_prison: |
2911 | kfree(pool); | |
2912 | bad_pool: | |
2913 | if (dm_pool_metadata_close(pmd)) | |
2914 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2915 | ||
2916 | return err_p; | |
2917 | } | |
2918 | ||
2919 | static void __pool_inc(struct pool *pool) | |
2920 | { | |
2921 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2922 | pool->ref_count++; | |
2923 | } | |
2924 | ||
2925 | static void __pool_dec(struct pool *pool) | |
2926 | { | |
2927 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2928 | BUG_ON(!pool->ref_count); | |
2929 | if (!--pool->ref_count) | |
2930 | __pool_destroy(pool); | |
2931 | } | |
2932 | ||
2933 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2934 | struct block_device *metadata_dev, | |
e49e5829 JT |
2935 | unsigned long block_size, int read_only, |
2936 | char **error, int *created) | |
991d9fa0 JT |
2937 | { |
2938 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2939 | ||
2940 | if (pool) { | |
f09996c9 MS |
2941 | if (pool->pool_md != pool_md) { |
2942 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2943 | return ERR_PTR(-EBUSY); |
f09996c9 | 2944 | } |
991d9fa0 JT |
2945 | __pool_inc(pool); |
2946 | ||
2947 | } else { | |
2948 | pool = __pool_table_lookup(pool_md); | |
2949 | if (pool) { | |
f09996c9 MS |
2950 | if (pool->md_dev != metadata_dev) { |
2951 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2952 | return ERR_PTR(-EINVAL); |
f09996c9 | 2953 | } |
991d9fa0 JT |
2954 | __pool_inc(pool); |
2955 | ||
67e2e2b2 | 2956 | } else { |
e49e5829 | 2957 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2958 | *created = 1; |
2959 | } | |
991d9fa0 JT |
2960 | } |
2961 | ||
2962 | return pool; | |
2963 | } | |
2964 | ||
2965 | /*---------------------------------------------------------------- | |
2966 | * Pool target methods | |
2967 | *--------------------------------------------------------------*/ | |
2968 | static void pool_dtr(struct dm_target *ti) | |
2969 | { | |
2970 | struct pool_c *pt = ti->private; | |
2971 | ||
2972 | mutex_lock(&dm_thin_pool_table.mutex); | |
2973 | ||
2974 | unbind_control_target(pt->pool, ti); | |
2975 | __pool_dec(pt->pool); | |
2976 | dm_put_device(ti, pt->metadata_dev); | |
2977 | dm_put_device(ti, pt->data_dev); | |
2978 | kfree(pt); | |
2979 | ||
2980 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2981 | } | |
2982 | ||
991d9fa0 JT |
2983 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2984 | struct dm_target *ti) | |
2985 | { | |
2986 | int r; | |
2987 | unsigned argc; | |
2988 | const char *arg_name; | |
2989 | ||
2990 | static struct dm_arg _args[] = { | |
74aa45c3 | 2991 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2992 | }; |
2993 | ||
2994 | /* | |
2995 | * No feature arguments supplied. | |
2996 | */ | |
2997 | if (!as->argc) | |
2998 | return 0; | |
2999 | ||
3000 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
3001 | if (r) | |
3002 | return -EINVAL; | |
3003 | ||
3004 | while (argc && !r) { | |
3005 | arg_name = dm_shift_arg(as); | |
3006 | argc--; | |
3007 | ||
e49e5829 | 3008 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 3009 | pf->zero_new_blocks = false; |
e49e5829 JT |
3010 | |
3011 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 3012 | pf->discard_enabled = false; |
e49e5829 JT |
3013 | |
3014 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 3015 | pf->discard_passdown = false; |
991d9fa0 | 3016 | |
e49e5829 JT |
3017 | else if (!strcasecmp(arg_name, "read_only")) |
3018 | pf->mode = PM_READ_ONLY; | |
3019 | ||
787a996c MS |
3020 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
3021 | pf->error_if_no_space = true; | |
3022 | ||
e49e5829 JT |
3023 | else { |
3024 | ti->error = "Unrecognised pool feature requested"; | |
3025 | r = -EINVAL; | |
3026 | break; | |
3027 | } | |
991d9fa0 JT |
3028 | } |
3029 | ||
3030 | return r; | |
3031 | } | |
3032 | ||
ac8c3f3d JT |
3033 | static void metadata_low_callback(void *context) |
3034 | { | |
3035 | struct pool *pool = context; | |
3036 | ||
3037 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
3038 | dm_device_name(pool->pool_md)); | |
3039 | ||
3040 | dm_table_event(pool->ti->table); | |
3041 | } | |
3042 | ||
7d48935e MS |
3043 | static sector_t get_dev_size(struct block_device *bdev) |
3044 | { | |
3045 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
3046 | } | |
3047 | ||
3048 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 3049 | { |
7d48935e | 3050 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
3051 | char buffer[BDEVNAME_SIZE]; |
3052 | ||
7d48935e | 3053 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
3054 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
3055 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
3056 | } |
3057 | ||
3058 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
3059 | { | |
3060 | sector_t metadata_dev_size = get_dev_size(bdev); | |
3061 | ||
3062 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
3063 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
3064 | |
3065 | return metadata_dev_size; | |
3066 | } | |
3067 | ||
24347e95 JT |
3068 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
3069 | { | |
3070 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
3071 | ||
7d48935e | 3072 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
3073 | |
3074 | return metadata_dev_size; | |
3075 | } | |
3076 | ||
ac8c3f3d JT |
3077 | /* |
3078 | * When a metadata threshold is crossed a dm event is triggered, and | |
3079 | * userland should respond by growing the metadata device. We could let | |
3080 | * userland set the threshold, like we do with the data threshold, but I'm | |
3081 | * not sure they know enough to do this well. | |
3082 | */ | |
3083 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
3084 | { | |
3085 | /* | |
3086 | * 4M is ample for all ops with the possible exception of thin | |
3087 | * device deletion which is harmless if it fails (just retry the | |
3088 | * delete after you've grown the device). | |
3089 | */ | |
3090 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
3091 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
3092 | } | |
3093 | ||
991d9fa0 JT |
3094 | /* |
3095 | * thin-pool <metadata dev> <data dev> | |
3096 | * <data block size (sectors)> | |
3097 | * <low water mark (blocks)> | |
3098 | * [<#feature args> [<arg>]*] | |
3099 | * | |
3100 | * Optional feature arguments are: | |
3101 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
3102 | * ignore_discard: disable discard |
3103 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
3104 | * read_only: Don't allow any changes to be made to the pool metadata. |
3105 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
3106 | */ |
3107 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3108 | { | |
67e2e2b2 | 3109 | int r, pool_created = 0; |
991d9fa0 JT |
3110 | struct pool_c *pt; |
3111 | struct pool *pool; | |
3112 | struct pool_features pf; | |
3113 | struct dm_arg_set as; | |
3114 | struct dm_dev *data_dev; | |
3115 | unsigned long block_size; | |
3116 | dm_block_t low_water_blocks; | |
3117 | struct dm_dev *metadata_dev; | |
5d0db96d | 3118 | fmode_t metadata_mode; |
991d9fa0 JT |
3119 | |
3120 | /* | |
3121 | * FIXME Remove validation from scope of lock. | |
3122 | */ | |
3123 | mutex_lock(&dm_thin_pool_table.mutex); | |
3124 | ||
3125 | if (argc < 4) { | |
3126 | ti->error = "Invalid argument count"; | |
3127 | r = -EINVAL; | |
3128 | goto out_unlock; | |
3129 | } | |
5d0db96d | 3130 | |
991d9fa0 JT |
3131 | as.argc = argc; |
3132 | as.argv = argv; | |
3133 | ||
5d0db96d JT |
3134 | /* |
3135 | * Set default pool features. | |
3136 | */ | |
3137 | pool_features_init(&pf); | |
3138 | ||
3139 | dm_consume_args(&as, 4); | |
3140 | r = parse_pool_features(&as, &pf, ti); | |
3141 | if (r) | |
3142 | goto out_unlock; | |
3143 | ||
3144 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
3145 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
3146 | if (r) { |
3147 | ti->error = "Error opening metadata block device"; | |
3148 | goto out_unlock; | |
3149 | } | |
7d48935e | 3150 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
3151 | |
3152 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
3153 | if (r) { | |
3154 | ti->error = "Error getting data device"; | |
3155 | goto out_metadata; | |
3156 | } | |
3157 | ||
3158 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
3159 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
3160 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 3161 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
3162 | ti->error = "Invalid block size"; |
3163 | r = -EINVAL; | |
3164 | goto out; | |
3165 | } | |
3166 | ||
3167 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
3168 | ti->error = "Invalid low water mark"; | |
3169 | r = -EINVAL; | |
3170 | goto out; | |
3171 | } | |
3172 | ||
991d9fa0 JT |
3173 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
3174 | if (!pt) { | |
3175 | r = -ENOMEM; | |
3176 | goto out; | |
3177 | } | |
3178 | ||
3179 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 3180 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
3181 | if (IS_ERR(pool)) { |
3182 | r = PTR_ERR(pool); | |
3183 | goto out_free_pt; | |
3184 | } | |
3185 | ||
67e2e2b2 JT |
3186 | /* |
3187 | * 'pool_created' reflects whether this is the first table load. | |
3188 | * Top level discard support is not allowed to be changed after | |
3189 | * initial load. This would require a pool reload to trigger thin | |
3190 | * device changes. | |
3191 | */ | |
3192 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
3193 | ti->error = "Discard support cannot be disabled once enabled"; | |
3194 | r = -EINVAL; | |
3195 | goto out_flags_changed; | |
3196 | } | |
3197 | ||
991d9fa0 JT |
3198 | pt->pool = pool; |
3199 | pt->ti = ti; | |
3200 | pt->metadata_dev = metadata_dev; | |
3201 | pt->data_dev = data_dev; | |
3202 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 3203 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 3204 | ti->num_flush_bios = 1; |
9bc142dd | 3205 | |
67e2e2b2 JT |
3206 | /* |
3207 | * Only need to enable discards if the pool should pass | |
3208 | * them down to the data device. The thin device's discard | |
3209 | * processing will cause mappings to be removed from the btree. | |
3210 | */ | |
b60ab990 | 3211 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3212 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 3213 | ti->num_discard_bios = 1; |
9bc142dd | 3214 | |
67e2e2b2 JT |
3215 | /* |
3216 | * Setting 'discards_supported' circumvents the normal | |
3217 | * stacking of discard limits (this keeps the pool and | |
3218 | * thin devices' discard limits consistent). | |
3219 | */ | |
0ac55489 | 3220 | ti->discards_supported = true; |
67e2e2b2 | 3221 | } |
991d9fa0 JT |
3222 | ti->private = pt; |
3223 | ||
ac8c3f3d JT |
3224 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
3225 | calc_metadata_threshold(pt), | |
3226 | metadata_low_callback, | |
3227 | pool); | |
3228 | if (r) | |
3229 | goto out_free_pt; | |
3230 | ||
991d9fa0 JT |
3231 | pt->callbacks.congested_fn = pool_is_congested; |
3232 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
3233 | ||
3234 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3235 | ||
3236 | return 0; | |
3237 | ||
67e2e2b2 JT |
3238 | out_flags_changed: |
3239 | __pool_dec(pool); | |
991d9fa0 JT |
3240 | out_free_pt: |
3241 | kfree(pt); | |
3242 | out: | |
3243 | dm_put_device(ti, data_dev); | |
3244 | out_metadata: | |
3245 | dm_put_device(ti, metadata_dev); | |
3246 | out_unlock: | |
3247 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3248 | ||
3249 | return r; | |
3250 | } | |
3251 | ||
7de3ee57 | 3252 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
3253 | { |
3254 | int r; | |
3255 | struct pool_c *pt = ti->private; | |
3256 | struct pool *pool = pt->pool; | |
3257 | unsigned long flags; | |
3258 | ||
3259 | /* | |
3260 | * As this is a singleton target, ti->begin is always zero. | |
3261 | */ | |
3262 | spin_lock_irqsave(&pool->lock, flags); | |
3263 | bio->bi_bdev = pt->data_dev->bdev; | |
3264 | r = DM_MAPIO_REMAPPED; | |
3265 | spin_unlock_irqrestore(&pool->lock, flags); | |
3266 | ||
3267 | return r; | |
3268 | } | |
3269 | ||
b17446df | 3270 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
3271 | { |
3272 | int r; | |
3273 | struct pool_c *pt = ti->private; | |
3274 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
3275 | sector_t data_size = ti->len; |
3276 | dm_block_t sb_data_size; | |
991d9fa0 | 3277 | |
b17446df | 3278 | *need_commit = false; |
991d9fa0 | 3279 | |
55f2b8bd MS |
3280 | (void) sector_div(data_size, pool->sectors_per_block); |
3281 | ||
991d9fa0 JT |
3282 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
3283 | if (r) { | |
4fa5971a MS |
3284 | DMERR("%s: failed to retrieve data device size", |
3285 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
3286 | return r; |
3287 | } | |
3288 | ||
3289 | if (data_size < sb_data_size) { | |
4fa5971a MS |
3290 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
3291 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3292 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3293 | return -EINVAL; |
3294 | ||
3295 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3296 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3297 | DMERR("%s: unable to grow the data device until repaired.", | |
3298 | dm_device_name(pool->pool_md)); | |
3299 | return 0; | |
3300 | } | |
3301 | ||
6f7f51d4 MS |
3302 | if (sb_data_size) |
3303 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3304 | dm_device_name(pool->pool_md), | |
3305 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3306 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3307 | if (r) { | |
b5330655 | 3308 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3309 | return r; |
3310 | } | |
3311 | ||
b17446df | 3312 | *need_commit = true; |
991d9fa0 JT |
3313 | } |
3314 | ||
3315 | return 0; | |
3316 | } | |
3317 | ||
24347e95 JT |
3318 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3319 | { | |
3320 | int r; | |
3321 | struct pool_c *pt = ti->private; | |
3322 | struct pool *pool = pt->pool; | |
3323 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3324 | ||
3325 | *need_commit = false; | |
3326 | ||
610bba8b | 3327 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3328 | |
3329 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3330 | if (r) { | |
4fa5971a MS |
3331 | DMERR("%s: failed to retrieve metadata device size", |
3332 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3333 | return r; |
3334 | } | |
3335 | ||
3336 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3337 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3338 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3339 | metadata_dev_size, sb_metadata_dev_size); |
3340 | return -EINVAL; | |
3341 | ||
3342 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3343 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3344 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3345 | dm_device_name(pool->pool_md)); | |
3346 | return 0; | |
3347 | } | |
3348 | ||
7d48935e | 3349 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3350 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3351 | dm_device_name(pool->pool_md), | |
3352 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
3353 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3354 | if (r) { | |
b5330655 | 3355 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3356 | return r; |
3357 | } | |
3358 | ||
3359 | *need_commit = true; | |
3360 | } | |
3361 | ||
3362 | return 0; | |
3363 | } | |
3364 | ||
b17446df JT |
3365 | /* |
3366 | * Retrieves the number of blocks of the data device from | |
3367 | * the superblock and compares it to the actual device size, | |
3368 | * thus resizing the data device in case it has grown. | |
3369 | * | |
3370 | * This both copes with opening preallocated data devices in the ctr | |
3371 | * being followed by a resume | |
3372 | * -and- | |
3373 | * calling the resume method individually after userspace has | |
3374 | * grown the data device in reaction to a table event. | |
3375 | */ | |
3376 | static int pool_preresume(struct dm_target *ti) | |
3377 | { | |
3378 | int r; | |
24347e95 | 3379 | bool need_commit1, need_commit2; |
b17446df JT |
3380 | struct pool_c *pt = ti->private; |
3381 | struct pool *pool = pt->pool; | |
3382 | ||
3383 | /* | |
3384 | * Take control of the pool object. | |
3385 | */ | |
3386 | r = bind_control_target(pool, ti); | |
3387 | if (r) | |
3388 | return r; | |
3389 | ||
3390 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3391 | if (r) | |
3392 | return r; | |
3393 | ||
24347e95 JT |
3394 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3395 | if (r) | |
3396 | return r; | |
3397 | ||
3398 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3399 | (void) commit(pool); |
b17446df JT |
3400 | |
3401 | return 0; | |
3402 | } | |
3403 | ||
583024d2 MS |
3404 | static void pool_suspend_active_thins(struct pool *pool) |
3405 | { | |
3406 | struct thin_c *tc; | |
3407 | ||
3408 | /* Suspend all active thin devices */ | |
3409 | tc = get_first_thin(pool); | |
3410 | while (tc) { | |
3411 | dm_internal_suspend_noflush(tc->thin_md); | |
3412 | tc = get_next_thin(pool, tc); | |
3413 | } | |
3414 | } | |
3415 | ||
3416 | static void pool_resume_active_thins(struct pool *pool) | |
3417 | { | |
3418 | struct thin_c *tc; | |
3419 | ||
3420 | /* Resume all active thin devices */ | |
3421 | tc = get_first_thin(pool); | |
3422 | while (tc) { | |
3423 | dm_internal_resume(tc->thin_md); | |
3424 | tc = get_next_thin(pool, tc); | |
3425 | } | |
3426 | } | |
3427 | ||
991d9fa0 JT |
3428 | static void pool_resume(struct dm_target *ti) |
3429 | { | |
3430 | struct pool_c *pt = ti->private; | |
3431 | struct pool *pool = pt->pool; | |
3432 | unsigned long flags; | |
3433 | ||
583024d2 MS |
3434 | /* |
3435 | * Must requeue active_thins' bios and then resume | |
3436 | * active_thins _before_ clearing 'suspend' flag. | |
3437 | */ | |
3438 | requeue_bios(pool); | |
3439 | pool_resume_active_thins(pool); | |
3440 | ||
991d9fa0 | 3441 | spin_lock_irqsave(&pool->lock, flags); |
88a6621b | 3442 | pool->low_water_triggered = false; |
80e96c54 | 3443 | pool->suspended = false; |
991d9fa0 | 3444 | spin_unlock_irqrestore(&pool->lock, flags); |
80e96c54 | 3445 | |
905e51b3 | 3446 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3447 | } |
3448 | ||
80e96c54 MS |
3449 | static void pool_presuspend(struct dm_target *ti) |
3450 | { | |
3451 | struct pool_c *pt = ti->private; | |
3452 | struct pool *pool = pt->pool; | |
3453 | unsigned long flags; | |
3454 | ||
3455 | spin_lock_irqsave(&pool->lock, flags); | |
3456 | pool->suspended = true; | |
3457 | spin_unlock_irqrestore(&pool->lock, flags); | |
583024d2 MS |
3458 | |
3459 | pool_suspend_active_thins(pool); | |
80e96c54 MS |
3460 | } |
3461 | ||
3462 | static void pool_presuspend_undo(struct dm_target *ti) | |
3463 | { | |
3464 | struct pool_c *pt = ti->private; | |
3465 | struct pool *pool = pt->pool; | |
3466 | unsigned long flags; | |
3467 | ||
583024d2 MS |
3468 | pool_resume_active_thins(pool); |
3469 | ||
80e96c54 MS |
3470 | spin_lock_irqsave(&pool->lock, flags); |
3471 | pool->suspended = false; | |
3472 | spin_unlock_irqrestore(&pool->lock, flags); | |
3473 | } | |
3474 | ||
991d9fa0 JT |
3475 | static void pool_postsuspend(struct dm_target *ti) |
3476 | { | |
991d9fa0 JT |
3477 | struct pool_c *pt = ti->private; |
3478 | struct pool *pool = pt->pool; | |
3479 | ||
905e51b3 | 3480 | cancel_delayed_work(&pool->waker); |
85ad643b | 3481 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 3482 | flush_workqueue(pool->wq); |
020cc3b5 | 3483 | (void) commit(pool); |
991d9fa0 JT |
3484 | } |
3485 | ||
3486 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3487 | { | |
3488 | if (argc != args_required) { | |
3489 | DMWARN("Message received with %u arguments instead of %u.", | |
3490 | argc, args_required); | |
3491 | return -EINVAL; | |
3492 | } | |
3493 | ||
3494 | return 0; | |
3495 | } | |
3496 | ||
3497 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3498 | { | |
3499 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3500 | *dev_id <= MAX_DEV_ID) | |
3501 | return 0; | |
3502 | ||
3503 | if (warning) | |
3504 | DMWARN("Message received with invalid device id: %s", arg); | |
3505 | ||
3506 | return -EINVAL; | |
3507 | } | |
3508 | ||
3509 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3510 | { | |
3511 | dm_thin_id dev_id; | |
3512 | int r; | |
3513 | ||
3514 | r = check_arg_count(argc, 2); | |
3515 | if (r) | |
3516 | return r; | |
3517 | ||
3518 | r = read_dev_id(argv[1], &dev_id, 1); | |
3519 | if (r) | |
3520 | return r; | |
3521 | ||
3522 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3523 | if (r) { | |
3524 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3525 | argv[1]); | |
3526 | return r; | |
3527 | } | |
3528 | ||
3529 | return 0; | |
3530 | } | |
3531 | ||
3532 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3533 | { | |
3534 | dm_thin_id dev_id; | |
3535 | dm_thin_id origin_dev_id; | |
3536 | int r; | |
3537 | ||
3538 | r = check_arg_count(argc, 3); | |
3539 | if (r) | |
3540 | return r; | |
3541 | ||
3542 | r = read_dev_id(argv[1], &dev_id, 1); | |
3543 | if (r) | |
3544 | return r; | |
3545 | ||
3546 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3547 | if (r) | |
3548 | return r; | |
3549 | ||
3550 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3551 | if (r) { | |
3552 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3553 | argv[1], argv[2]); | |
3554 | return r; | |
3555 | } | |
3556 | ||
3557 | return 0; | |
3558 | } | |
3559 | ||
3560 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3561 | { | |
3562 | dm_thin_id dev_id; | |
3563 | int r; | |
3564 | ||
3565 | r = check_arg_count(argc, 2); | |
3566 | if (r) | |
3567 | return r; | |
3568 | ||
3569 | r = read_dev_id(argv[1], &dev_id, 1); | |
3570 | if (r) | |
3571 | return r; | |
3572 | ||
3573 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3574 | if (r) | |
3575 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3576 | ||
3577 | return r; | |
3578 | } | |
3579 | ||
3580 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3581 | { | |
3582 | dm_thin_id old_id, new_id; | |
3583 | int r; | |
3584 | ||
3585 | r = check_arg_count(argc, 3); | |
3586 | if (r) | |
3587 | return r; | |
3588 | ||
3589 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3590 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3591 | return -EINVAL; | |
3592 | } | |
3593 | ||
3594 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3595 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3596 | return -EINVAL; | |
3597 | } | |
3598 | ||
3599 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3600 | if (r) { | |
3601 | DMWARN("Failed to change transaction id from %s to %s.", | |
3602 | argv[1], argv[2]); | |
3603 | return r; | |
3604 | } | |
3605 | ||
3606 | return 0; | |
3607 | } | |
3608 | ||
cc8394d8 JT |
3609 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3610 | { | |
3611 | int r; | |
3612 | ||
3613 | r = check_arg_count(argc, 1); | |
3614 | if (r) | |
3615 | return r; | |
3616 | ||
020cc3b5 | 3617 | (void) commit(pool); |
0d200aef | 3618 | |
cc8394d8 JT |
3619 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3620 | if (r) | |
3621 | DMWARN("reserve_metadata_snap message failed."); | |
3622 | ||
3623 | return r; | |
3624 | } | |
3625 | ||
3626 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3627 | { | |
3628 | int r; | |
3629 | ||
3630 | r = check_arg_count(argc, 1); | |
3631 | if (r) | |
3632 | return r; | |
3633 | ||
3634 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3635 | if (r) | |
3636 | DMWARN("release_metadata_snap message failed."); | |
3637 | ||
3638 | return r; | |
3639 | } | |
3640 | ||
991d9fa0 JT |
3641 | /* |
3642 | * Messages supported: | |
3643 | * create_thin <dev_id> | |
3644 | * create_snap <dev_id> <origin_id> | |
3645 | * delete <dev_id> | |
991d9fa0 | 3646 | * set_transaction_id <current_trans_id> <new_trans_id> |
cc8394d8 JT |
3647 | * reserve_metadata_snap |
3648 | * release_metadata_snap | |
991d9fa0 JT |
3649 | */ |
3650 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
3651 | { | |
3652 | int r = -EINVAL; | |
3653 | struct pool_c *pt = ti->private; | |
3654 | struct pool *pool = pt->pool; | |
3655 | ||
2a7eaea0 JT |
3656 | if (get_pool_mode(pool) >= PM_READ_ONLY) { |
3657 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", | |
3658 | dm_device_name(pool->pool_md)); | |
fd467696 | 3659 | return -EOPNOTSUPP; |
2a7eaea0 JT |
3660 | } |
3661 | ||
991d9fa0 JT |
3662 | if (!strcasecmp(argv[0], "create_thin")) |
3663 | r = process_create_thin_mesg(argc, argv, pool); | |
3664 | ||
3665 | else if (!strcasecmp(argv[0], "create_snap")) | |
3666 | r = process_create_snap_mesg(argc, argv, pool); | |
3667 | ||
3668 | else if (!strcasecmp(argv[0], "delete")) | |
3669 | r = process_delete_mesg(argc, argv, pool); | |
3670 | ||
3671 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3672 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3673 | ||
cc8394d8 JT |
3674 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3675 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3676 | ||
3677 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3678 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3679 | ||
991d9fa0 JT |
3680 | else |
3681 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3682 | ||
e49e5829 | 3683 | if (!r) |
020cc3b5 | 3684 | (void) commit(pool); |
991d9fa0 JT |
3685 | |
3686 | return r; | |
3687 | } | |
3688 | ||
e49e5829 JT |
3689 | static void emit_flags(struct pool_features *pf, char *result, |
3690 | unsigned sz, unsigned maxlen) | |
3691 | { | |
3692 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3693 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3694 | pf->error_if_no_space; | |
e49e5829 JT |
3695 | DMEMIT("%u ", count); |
3696 | ||
3697 | if (!pf->zero_new_blocks) | |
3698 | DMEMIT("skip_block_zeroing "); | |
3699 | ||
3700 | if (!pf->discard_enabled) | |
3701 | DMEMIT("ignore_discard "); | |
3702 | ||
3703 | if (!pf->discard_passdown) | |
3704 | DMEMIT("no_discard_passdown "); | |
3705 | ||
3706 | if (pf->mode == PM_READ_ONLY) | |
3707 | DMEMIT("read_only "); | |
787a996c MS |
3708 | |
3709 | if (pf->error_if_no_space) | |
3710 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3711 | } |
3712 | ||
991d9fa0 JT |
3713 | /* |
3714 | * Status line is: | |
3715 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3716 | * <used data sectors>/<total data sectors> <held metadata root> | |
3717 | */ | |
fd7c092e MP |
3718 | static void pool_status(struct dm_target *ti, status_type_t type, |
3719 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3720 | { |
e49e5829 | 3721 | int r; |
991d9fa0 JT |
3722 | unsigned sz = 0; |
3723 | uint64_t transaction_id; | |
3724 | dm_block_t nr_free_blocks_data; | |
3725 | dm_block_t nr_free_blocks_metadata; | |
3726 | dm_block_t nr_blocks_data; | |
3727 | dm_block_t nr_blocks_metadata; | |
3728 | dm_block_t held_root; | |
3729 | char buf[BDEVNAME_SIZE]; | |
3730 | char buf2[BDEVNAME_SIZE]; | |
3731 | struct pool_c *pt = ti->private; | |
3732 | struct pool *pool = pt->pool; | |
3733 | ||
3734 | switch (type) { | |
3735 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3736 | if (get_pool_mode(pool) == PM_FAIL) { |
3737 | DMEMIT("Fail"); | |
3738 | break; | |
3739 | } | |
3740 | ||
1f4e0ff0 AK |
3741 | /* Commit to ensure statistics aren't out-of-date */ |
3742 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3743 | (void) commit(pool); |
1f4e0ff0 | 3744 | |
fd7c092e MP |
3745 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3746 | if (r) { | |
4fa5971a MS |
3747 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3748 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3749 | goto err; |
3750 | } | |
991d9fa0 | 3751 | |
fd7c092e MP |
3752 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3753 | if (r) { | |
4fa5971a MS |
3754 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3755 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3756 | goto err; |
3757 | } | |
991d9fa0 JT |
3758 | |
3759 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3760 | if (r) { |
4fa5971a MS |
3761 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3762 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3763 | goto err; |
3764 | } | |
991d9fa0 | 3765 | |
fd7c092e MP |
3766 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3767 | if (r) { | |
4fa5971a MS |
3768 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3769 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3770 | goto err; |
3771 | } | |
991d9fa0 JT |
3772 | |
3773 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3774 | if (r) { |
4fa5971a MS |
3775 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3776 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3777 | goto err; |
3778 | } | |
991d9fa0 | 3779 | |
cc8394d8 | 3780 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3781 | if (r) { |
4fa5971a MS |
3782 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3783 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3784 | goto err; |
3785 | } | |
991d9fa0 JT |
3786 | |
3787 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3788 | (unsigned long long)transaction_id, | |
3789 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3790 | (unsigned long long)nr_blocks_metadata, | |
3791 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3792 | (unsigned long long)nr_blocks_data); | |
3793 | ||
3794 | if (held_root) | |
e49e5829 JT |
3795 | DMEMIT("%llu ", held_root); |
3796 | else | |
3797 | DMEMIT("- "); | |
3798 | ||
3e1a0699 JT |
3799 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3800 | DMEMIT("out_of_data_space "); | |
3801 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3802 | DMEMIT("ro "); |
991d9fa0 | 3803 | else |
e49e5829 JT |
3804 | DMEMIT("rw "); |
3805 | ||
018debea | 3806 | if (!pool->pf.discard_enabled) |
787a996c | 3807 | DMEMIT("ignore_discard "); |
018debea | 3808 | else if (pool->pf.discard_passdown) |
787a996c MS |
3809 | DMEMIT("discard_passdown "); |
3810 | else | |
3811 | DMEMIT("no_discard_passdown "); | |
3812 | ||
3813 | if (pool->pf.error_if_no_space) | |
3814 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3815 | else |
787a996c | 3816 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3817 | |
3818 | break; | |
3819 | ||
3820 | case STATUSTYPE_TABLE: | |
3821 | DMEMIT("%s %s %lu %llu ", | |
3822 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3823 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3824 | (unsigned long)pool->sectors_per_block, | |
3825 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3826 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3827 | break; |
3828 | } | |
fd7c092e | 3829 | return; |
991d9fa0 | 3830 | |
fd7c092e MP |
3831 | err: |
3832 | DMEMIT("Error"); | |
991d9fa0 JT |
3833 | } |
3834 | ||
3835 | static int pool_iterate_devices(struct dm_target *ti, | |
3836 | iterate_devices_callout_fn fn, void *data) | |
3837 | { | |
3838 | struct pool_c *pt = ti->private; | |
3839 | ||
3840 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3841 | } | |
3842 | ||
3843 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3844 | struct bio_vec *biovec, int max_size) | |
3845 | { | |
3846 | struct pool_c *pt = ti->private; | |
3847 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3848 | ||
3849 | if (!q->merge_bvec_fn) | |
3850 | return max_size; | |
3851 | ||
3852 | bvm->bi_bdev = pt->data_dev->bdev; | |
3853 | ||
3854 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3855 | } | |
3856 | ||
3857 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3858 | { | |
3859 | struct pool_c *pt = ti->private; | |
3860 | struct pool *pool = pt->pool; | |
604ea906 MS |
3861 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
3862 | ||
3863 | /* | |
d200c30e MS |
3864 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
3865 | * to the highest possible power-of-2 factor of pool->sectors_per_block. | |
3866 | * This is especially beneficial when the pool's data device is a RAID | |
3867 | * device that has a full stripe width that matches pool->sectors_per_block | |
3868 | * -- because even though partial RAID stripe-sized IOs will be issued to a | |
3869 | * single RAID stripe; when aggregated they will end on a full RAID stripe | |
3870 | * boundary.. which avoids additional partial RAID stripe writes cascading | |
604ea906 | 3871 | */ |
604ea906 MS |
3872 | if (limits->max_sectors < pool->sectors_per_block) { |
3873 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
3874 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
3875 | limits->max_sectors--; | |
3876 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
3877 | } | |
604ea906 | 3878 | } |
991d9fa0 | 3879 | |
0cc67cd9 MS |
3880 | /* |
3881 | * If the system-determined stacked limits are compatible with the | |
3882 | * pool's blocksize (io_opt is a factor) do not override them. | |
3883 | */ | |
3884 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
3885 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
3886 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
3887 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
3888 | else | |
3889 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
3890 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
3891 | } | |
0424caa1 MS |
3892 | |
3893 | /* | |
3894 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3895 | * They get transferred to the live pool in bind_control_target() | |
3896 | * called from pool_preresume(). | |
3897 | */ | |
b60ab990 MS |
3898 | if (!pt->adjusted_pf.discard_enabled) { |
3899 | /* | |
3900 | * Must explicitly disallow stacking discard limits otherwise the | |
3901 | * block layer will stack them if pool's data device has support. | |
3902 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3903 | * user to see that, so make sure to set all discard limits to 0. | |
3904 | */ | |
3905 | limits->discard_granularity = 0; | |
0424caa1 | 3906 | return; |
b60ab990 | 3907 | } |
0424caa1 MS |
3908 | |
3909 | disable_passdown_if_not_supported(pt); | |
3910 | ||
34fbcf62 JT |
3911 | /* |
3912 | * The pool uses the same discard limits as the underlying data | |
3913 | * device. DM core has already set this up. | |
3914 | */ | |
991d9fa0 JT |
3915 | } |
3916 | ||
3917 | static struct target_type pool_target = { | |
3918 | .name = "thin-pool", | |
3919 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3920 | DM_TARGET_IMMUTABLE, | |
34fbcf62 | 3921 | .version = {1, 15, 0}, |
991d9fa0 JT |
3922 | .module = THIS_MODULE, |
3923 | .ctr = pool_ctr, | |
3924 | .dtr = pool_dtr, | |
3925 | .map = pool_map, | |
80e96c54 MS |
3926 | .presuspend = pool_presuspend, |
3927 | .presuspend_undo = pool_presuspend_undo, | |
991d9fa0 JT |
3928 | .postsuspend = pool_postsuspend, |
3929 | .preresume = pool_preresume, | |
3930 | .resume = pool_resume, | |
3931 | .message = pool_message, | |
3932 | .status = pool_status, | |
3933 | .merge = pool_merge, | |
3934 | .iterate_devices = pool_iterate_devices, | |
3935 | .io_hints = pool_io_hints, | |
3936 | }; | |
3937 | ||
3938 | /*---------------------------------------------------------------- | |
3939 | * Thin target methods | |
3940 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3941 | static void thin_get(struct thin_c *tc) |
3942 | { | |
3943 | atomic_inc(&tc->refcount); | |
3944 | } | |
3945 | ||
3946 | static void thin_put(struct thin_c *tc) | |
3947 | { | |
3948 | if (atomic_dec_and_test(&tc->refcount)) | |
3949 | complete(&tc->can_destroy); | |
3950 | } | |
3951 | ||
991d9fa0 JT |
3952 | static void thin_dtr(struct dm_target *ti) |
3953 | { | |
3954 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3955 | unsigned long flags; |
3956 | ||
3957 | spin_lock_irqsave(&tc->pool->lock, flags); | |
3958 | list_del_rcu(&tc->list); | |
3959 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3960 | synchronize_rcu(); | |
991d9fa0 | 3961 | |
17181fb7 MP |
3962 | thin_put(tc); |
3963 | wait_for_completion(&tc->can_destroy); | |
3964 | ||
991d9fa0 JT |
3965 | mutex_lock(&dm_thin_pool_table.mutex); |
3966 | ||
3967 | __pool_dec(tc->pool); | |
3968 | dm_pool_close_thin_device(tc->td); | |
3969 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3970 | if (tc->origin_dev) |
3971 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3972 | kfree(tc); |
3973 | ||
3974 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3975 | } | |
3976 | ||
3977 | /* | |
3978 | * Thin target parameters: | |
3979 | * | |
2dd9c257 | 3980 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3981 | * |
3982 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3983 | * dev_id: the internal device identifier | |
2dd9c257 | 3984 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3985 | * |
3986 | * If the pool device has discards disabled, they get disabled for the thin | |
3987 | * device as well. | |
991d9fa0 JT |
3988 | */ |
3989 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3990 | { | |
3991 | int r; | |
3992 | struct thin_c *tc; | |
2dd9c257 | 3993 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3994 | struct mapped_device *pool_md; |
5e3283e2 | 3995 | unsigned long flags; |
991d9fa0 JT |
3996 | |
3997 | mutex_lock(&dm_thin_pool_table.mutex); | |
3998 | ||
2dd9c257 | 3999 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
4000 | ti->error = "Invalid argument count"; |
4001 | r = -EINVAL; | |
4002 | goto out_unlock; | |
4003 | } | |
4004 | ||
4005 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
4006 | if (!tc) { | |
4007 | ti->error = "Out of memory"; | |
4008 | r = -ENOMEM; | |
4009 | goto out_unlock; | |
4010 | } | |
583024d2 | 4011 | tc->thin_md = dm_table_get_md(ti->table); |
c140e1c4 | 4012 | spin_lock_init(&tc->lock); |
a374bb21 | 4013 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
4014 | bio_list_init(&tc->deferred_bio_list); |
4015 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 4016 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 4017 | |
2dd9c257 JT |
4018 | if (argc == 3) { |
4019 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
4020 | if (r) { | |
4021 | ti->error = "Error opening origin device"; | |
4022 | goto bad_origin_dev; | |
4023 | } | |
4024 | tc->origin_dev = origin_dev; | |
4025 | } | |
4026 | ||
991d9fa0 JT |
4027 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
4028 | if (r) { | |
4029 | ti->error = "Error opening pool device"; | |
4030 | goto bad_pool_dev; | |
4031 | } | |
4032 | tc->pool_dev = pool_dev; | |
4033 | ||
4034 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
4035 | ti->error = "Invalid device id"; | |
4036 | r = -EINVAL; | |
4037 | goto bad_common; | |
4038 | } | |
4039 | ||
4040 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
4041 | if (!pool_md) { | |
4042 | ti->error = "Couldn't get pool mapped device"; | |
4043 | r = -EINVAL; | |
4044 | goto bad_common; | |
4045 | } | |
4046 | ||
4047 | tc->pool = __pool_table_lookup(pool_md); | |
4048 | if (!tc->pool) { | |
4049 | ti->error = "Couldn't find pool object"; | |
4050 | r = -EINVAL; | |
4051 | goto bad_pool_lookup; | |
4052 | } | |
4053 | __pool_inc(tc->pool); | |
4054 | ||
e49e5829 JT |
4055 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4056 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 4057 | r = -EINVAL; |
80e96c54 | 4058 | goto bad_pool; |
e49e5829 JT |
4059 | } |
4060 | ||
991d9fa0 JT |
4061 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
4062 | if (r) { | |
4063 | ti->error = "Couldn't open thin internal device"; | |
80e96c54 | 4064 | goto bad_pool; |
991d9fa0 JT |
4065 | } |
4066 | ||
542f9038 MS |
4067 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
4068 | if (r) | |
80e96c54 | 4069 | goto bad; |
542f9038 | 4070 | |
55a62eef | 4071 | ti->num_flush_bios = 1; |
16ad3d10 | 4072 | ti->flush_supported = true; |
59c3d2c6 | 4073 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
4074 | |
4075 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 4076 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 4077 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 4078 | ti->discards_supported = true; |
55a62eef | 4079 | ti->num_discard_bios = 1; |
34fbcf62 | 4080 | ti->split_discard_bios = false; |
67e2e2b2 | 4081 | } |
991d9fa0 | 4082 | |
991d9fa0 JT |
4083 | mutex_unlock(&dm_thin_pool_table.mutex); |
4084 | ||
5e3283e2 | 4085 | spin_lock_irqsave(&tc->pool->lock, flags); |
80e96c54 MS |
4086 | if (tc->pool->suspended) { |
4087 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
4088 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ | |
4089 | ti->error = "Unable to activate thin device while pool is suspended"; | |
4090 | r = -EINVAL; | |
4091 | goto bad; | |
4092 | } | |
2b94e896 MD |
4093 | atomic_set(&tc->refcount, 1); |
4094 | init_completion(&tc->can_destroy); | |
c140e1c4 | 4095 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 4096 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
4097 | /* |
4098 | * This synchronize_rcu() call is needed here otherwise we risk a | |
4099 | * wake_worker() call finding no bios to process (because the newly | |
4100 | * added tc isn't yet visible). So this reduces latency since we | |
4101 | * aren't then dependent on the periodic commit to wake_worker(). | |
4102 | */ | |
4103 | synchronize_rcu(); | |
4104 | ||
80e96c54 MS |
4105 | dm_put(pool_md); |
4106 | ||
991d9fa0 JT |
4107 | return 0; |
4108 | ||
80e96c54 | 4109 | bad: |
1acacc07 | 4110 | dm_pool_close_thin_device(tc->td); |
80e96c54 | 4111 | bad_pool: |
991d9fa0 JT |
4112 | __pool_dec(tc->pool); |
4113 | bad_pool_lookup: | |
4114 | dm_put(pool_md); | |
4115 | bad_common: | |
4116 | dm_put_device(ti, tc->pool_dev); | |
4117 | bad_pool_dev: | |
2dd9c257 JT |
4118 | if (tc->origin_dev) |
4119 | dm_put_device(ti, tc->origin_dev); | |
4120 | bad_origin_dev: | |
991d9fa0 JT |
4121 | kfree(tc); |
4122 | out_unlock: | |
4123 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4124 | ||
4125 | return r; | |
4126 | } | |
4127 | ||
7de3ee57 | 4128 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 4129 | { |
4f024f37 | 4130 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 4131 | |
7de3ee57 | 4132 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
4133 | } |
4134 | ||
7de3ee57 | 4135 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
4136 | { |
4137 | unsigned long flags; | |
59c3d2c6 | 4138 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 4139 | struct list_head work; |
a24c2569 | 4140 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
4141 | struct pool *pool = h->tc->pool; |
4142 | ||
4143 | if (h->shared_read_entry) { | |
4144 | INIT_LIST_HEAD(&work); | |
44feb387 | 4145 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
4146 | |
4147 | spin_lock_irqsave(&pool->lock, flags); | |
4148 | list_for_each_entry_safe(m, tmp, &work, list) { | |
4149 | list_del(&m->list); | |
50f3c3ef | 4150 | __complete_mapping_preparation(m); |
eb2aa48d JT |
4151 | } |
4152 | spin_unlock_irqrestore(&pool->lock, flags); | |
4153 | } | |
4154 | ||
104655fd JT |
4155 | if (h->all_io_entry) { |
4156 | INIT_LIST_HEAD(&work); | |
44feb387 | 4157 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
4158 | if (!list_empty(&work)) { |
4159 | spin_lock_irqsave(&pool->lock, flags); | |
4160 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 4161 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
4162 | spin_unlock_irqrestore(&pool->lock, flags); |
4163 | wake_worker(pool); | |
4164 | } | |
104655fd JT |
4165 | } |
4166 | ||
34fbcf62 JT |
4167 | if (h->cell) |
4168 | cell_defer_no_holder(h->tc, h->cell); | |
4169 | ||
eb2aa48d JT |
4170 | return 0; |
4171 | } | |
4172 | ||
738211f7 | 4173 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 4174 | { |
738211f7 JT |
4175 | struct thin_c *tc = ti->private; |
4176 | ||
991d9fa0 | 4177 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
4178 | noflush_work(tc, do_noflush_start); |
4179 | } | |
4180 | ||
4181 | static void thin_postsuspend(struct dm_target *ti) | |
4182 | { | |
4183 | struct thin_c *tc = ti->private; | |
4184 | ||
4185 | /* | |
4186 | * The dm_noflush_suspending flag has been cleared by now, so | |
4187 | * unfortunately we must always run this. | |
4188 | */ | |
4189 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
4190 | } |
4191 | ||
e5aea7b4 JT |
4192 | static int thin_preresume(struct dm_target *ti) |
4193 | { | |
4194 | struct thin_c *tc = ti->private; | |
4195 | ||
4196 | if (tc->origin_dev) | |
4197 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
4198 | ||
4199 | return 0; | |
4200 | } | |
4201 | ||
991d9fa0 JT |
4202 | /* |
4203 | * <nr mapped sectors> <highest mapped sector> | |
4204 | */ | |
fd7c092e MP |
4205 | static void thin_status(struct dm_target *ti, status_type_t type, |
4206 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
4207 | { |
4208 | int r; | |
4209 | ssize_t sz = 0; | |
4210 | dm_block_t mapped, highest; | |
4211 | char buf[BDEVNAME_SIZE]; | |
4212 | struct thin_c *tc = ti->private; | |
4213 | ||
e49e5829 JT |
4214 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4215 | DMEMIT("Fail"); | |
fd7c092e | 4216 | return; |
e49e5829 JT |
4217 | } |
4218 | ||
991d9fa0 JT |
4219 | if (!tc->td) |
4220 | DMEMIT("-"); | |
4221 | else { | |
4222 | switch (type) { | |
4223 | case STATUSTYPE_INFO: | |
4224 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
4225 | if (r) { |
4226 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
4227 | goto err; | |
4228 | } | |
991d9fa0 JT |
4229 | |
4230 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
4231 | if (r < 0) { |
4232 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
4233 | goto err; | |
4234 | } | |
991d9fa0 JT |
4235 | |
4236 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
4237 | if (r) | |
4238 | DMEMIT("%llu", ((highest + 1) * | |
4239 | tc->pool->sectors_per_block) - 1); | |
4240 | else | |
4241 | DMEMIT("-"); | |
4242 | break; | |
4243 | ||
4244 | case STATUSTYPE_TABLE: | |
4245 | DMEMIT("%s %lu", | |
4246 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
4247 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
4248 | if (tc->origin_dev) |
4249 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
4250 | break; |
4251 | } | |
4252 | } | |
4253 | ||
fd7c092e MP |
4254 | return; |
4255 | ||
4256 | err: | |
4257 | DMEMIT("Error"); | |
991d9fa0 JT |
4258 | } |
4259 | ||
36f12aeb MS |
4260 | static int thin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
4261 | struct bio_vec *biovec, int max_size) | |
4262 | { | |
4263 | struct thin_c *tc = ti->private; | |
4264 | struct request_queue *q = bdev_get_queue(tc->pool_dev->bdev); | |
4265 | ||
4266 | if (!q->merge_bvec_fn) | |
4267 | return max_size; | |
4268 | ||
4269 | bvm->bi_bdev = tc->pool_dev->bdev; | |
4270 | bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector); | |
4271 | ||
4272 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
4273 | } | |
4274 | ||
991d9fa0 JT |
4275 | static int thin_iterate_devices(struct dm_target *ti, |
4276 | iterate_devices_callout_fn fn, void *data) | |
4277 | { | |
55f2b8bd | 4278 | sector_t blocks; |
991d9fa0 | 4279 | struct thin_c *tc = ti->private; |
55f2b8bd | 4280 | struct pool *pool = tc->pool; |
991d9fa0 JT |
4281 | |
4282 | /* | |
4283 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
4284 | * we follow a more convoluted path through to the pool's target. | |
4285 | */ | |
55f2b8bd | 4286 | if (!pool->ti) |
991d9fa0 JT |
4287 | return 0; /* nothing is bound */ |
4288 | ||
55f2b8bd MS |
4289 | blocks = pool->ti->len; |
4290 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 4291 | if (blocks) |
55f2b8bd | 4292 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
4293 | |
4294 | return 0; | |
4295 | } | |
4296 | ||
34fbcf62 JT |
4297 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4298 | { | |
4299 | struct thin_c *tc = ti->private; | |
4300 | struct pool *pool = tc->pool; | |
4301 | ||
4302 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
4303 | limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ | |
4304 | } | |
4305 | ||
991d9fa0 JT |
4306 | static struct target_type thin_target = { |
4307 | .name = "thin", | |
34fbcf62 | 4308 | .version = {1, 15, 0}, |
991d9fa0 JT |
4309 | .module = THIS_MODULE, |
4310 | .ctr = thin_ctr, | |
4311 | .dtr = thin_dtr, | |
4312 | .map = thin_map, | |
eb2aa48d | 4313 | .end_io = thin_endio, |
e5aea7b4 | 4314 | .preresume = thin_preresume, |
738211f7 | 4315 | .presuspend = thin_presuspend, |
991d9fa0 JT |
4316 | .postsuspend = thin_postsuspend, |
4317 | .status = thin_status, | |
36f12aeb | 4318 | .merge = thin_merge, |
991d9fa0 | 4319 | .iterate_devices = thin_iterate_devices, |
34fbcf62 | 4320 | .io_hints = thin_io_hints, |
991d9fa0 JT |
4321 | }; |
4322 | ||
4323 | /*----------------------------------------------------------------*/ | |
4324 | ||
4325 | static int __init dm_thin_init(void) | |
4326 | { | |
4327 | int r; | |
4328 | ||
4329 | pool_table_init(); | |
4330 | ||
4331 | r = dm_register_target(&thin_target); | |
4332 | if (r) | |
4333 | return r; | |
4334 | ||
4335 | r = dm_register_target(&pool_target); | |
4336 | if (r) | |
a24c2569 MS |
4337 | goto bad_pool_target; |
4338 | ||
4339 | r = -ENOMEM; | |
4340 | ||
a24c2569 MS |
4341 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
4342 | if (!_new_mapping_cache) | |
4343 | goto bad_new_mapping_cache; | |
4344 | ||
a24c2569 MS |
4345 | return 0; |
4346 | ||
a24c2569 | 4347 | bad_new_mapping_cache: |
a24c2569 MS |
4348 | dm_unregister_target(&pool_target); |
4349 | bad_pool_target: | |
4350 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
4351 | |
4352 | return r; | |
4353 | } | |
4354 | ||
4355 | static void dm_thin_exit(void) | |
4356 | { | |
4357 | dm_unregister_target(&thin_target); | |
4358 | dm_unregister_target(&pool_target); | |
a24c2569 | 4359 | |
a24c2569 | 4360 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
4361 | } |
4362 | ||
4363 | module_init(dm_thin_init); | |
4364 | module_exit(dm_thin_exit); | |
4365 | ||
80c57893 MS |
4366 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
4367 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
4368 | ||
7cab8bf1 | 4369 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4370 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4371 | MODULE_LICENSE("GPL"); |