]>
Commit | Line | Data |
---|---|---|
991d9fa0 JT |
1 | /* |
2 | * Copyright (C) 2011 Red Hat UK. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-thin-metadata.h" | |
8 | ||
9 | #include <linux/device-mapper.h> | |
10 | #include <linux/dm-io.h> | |
11 | #include <linux/dm-kcopyd.h> | |
12 | #include <linux/list.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/slab.h> | |
16 | ||
17 | #define DM_MSG_PREFIX "thin" | |
18 | ||
19 | /* | |
20 | * Tunable constants | |
21 | */ | |
7768ed33 | 22 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
23 | #define DEFERRED_SET_SIZE 64 |
24 | #define MAPPING_POOL_SIZE 1024 | |
25 | #define PRISON_CELLS 1024 | |
905e51b3 | 26 | #define COMMIT_PERIOD HZ |
991d9fa0 JT |
27 | |
28 | /* | |
29 | * The block size of the device holding pool data must be | |
30 | * between 64KB and 1GB. | |
31 | */ | |
32 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
33 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
34 | ||
991d9fa0 JT |
35 | /* |
36 | * Device id is restricted to 24 bits. | |
37 | */ | |
38 | #define MAX_DEV_ID ((1 << 24) - 1) | |
39 | ||
40 | /* | |
41 | * How do we handle breaking sharing of data blocks? | |
42 | * ================================================= | |
43 | * | |
44 | * We use a standard copy-on-write btree to store the mappings for the | |
45 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
46 | * the data). When you take an internal snapshot you clone the root node | |
47 | * of the origin btree. After this there is no concept of an origin or a | |
48 | * snapshot. They are just two device trees that happen to point to the | |
49 | * same data blocks. | |
50 | * | |
51 | * When we get a write in we decide if it's to a shared data block using | |
52 | * some timestamp magic. If it is, we have to break sharing. | |
53 | * | |
54 | * Let's say we write to a shared block in what was the origin. The | |
55 | * steps are: | |
56 | * | |
57 | * i) plug io further to this physical block. (see bio_prison code). | |
58 | * | |
59 | * ii) quiesce any read io to that shared data block. Obviously | |
60 | * including all devices that share this block. (see deferred_set code) | |
61 | * | |
62 | * iii) copy the data block to a newly allocate block. This step can be | |
63 | * missed out if the io covers the block. (schedule_copy). | |
64 | * | |
65 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 66 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
67 | * sharing of btree nodes between the two devices. Breaking sharing only |
68 | * effects the btree of that specific device. Btrees for the other | |
69 | * devices that share the block never change. The btree for the origin | |
70 | * device as it was after the last commit is untouched, ie. we're using | |
71 | * persistent data structures in the functional programming sense. | |
72 | * | |
73 | * v) unplug io to this physical block, including the io that triggered | |
74 | * the breaking of sharing. | |
75 | * | |
76 | * Steps (ii) and (iii) occur in parallel. | |
77 | * | |
78 | * The metadata _doesn't_ need to be committed before the io continues. We | |
79 | * get away with this because the io is always written to a _new_ block. | |
80 | * If there's a crash, then: | |
81 | * | |
82 | * - The origin mapping will point to the old origin block (the shared | |
83 | * one). This will contain the data as it was before the io that triggered | |
84 | * the breaking of sharing came in. | |
85 | * | |
86 | * - The snap mapping still points to the old block. As it would after | |
87 | * the commit. | |
88 | * | |
89 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
90 | * will continue to think that data block in the snapshot device is shared | |
91 | * even after the write to the origin has broken sharing. I suspect data | |
92 | * blocks will typically be shared by many different devices, so we're | |
93 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
94 | * devices that reference this data block. At the moment I think the | |
95 | * benefits far, far outweigh the disadvantages. | |
96 | */ | |
97 | ||
98 | /*----------------------------------------------------------------*/ | |
99 | ||
100 | /* | |
101 | * Sometimes we can't deal with a bio straight away. We put them in prison | |
102 | * where they can't cause any mischief. Bios are put in a cell identified | |
103 | * by a key, multiple bios can be in the same cell. When the cell is | |
104 | * subsequently unlocked the bios become available. | |
105 | */ | |
106 | struct bio_prison; | |
107 | ||
108 | struct cell_key { | |
109 | int virtual; | |
110 | dm_thin_id dev; | |
111 | dm_block_t block; | |
112 | }; | |
113 | ||
a24c2569 | 114 | struct dm_bio_prison_cell { |
991d9fa0 JT |
115 | struct hlist_node list; |
116 | struct bio_prison *prison; | |
117 | struct cell_key key; | |
6f94a4c4 | 118 | struct bio *holder; |
991d9fa0 JT |
119 | struct bio_list bios; |
120 | }; | |
121 | ||
122 | struct bio_prison { | |
123 | spinlock_t lock; | |
124 | mempool_t *cell_pool; | |
125 | ||
126 | unsigned nr_buckets; | |
127 | unsigned hash_mask; | |
128 | struct hlist_head *cells; | |
129 | }; | |
130 | ||
131 | static uint32_t calc_nr_buckets(unsigned nr_cells) | |
132 | { | |
133 | uint32_t n = 128; | |
134 | ||
135 | nr_cells /= 4; | |
136 | nr_cells = min(nr_cells, 8192u); | |
137 | ||
138 | while (n < nr_cells) | |
139 | n <<= 1; | |
140 | ||
141 | return n; | |
142 | } | |
143 | ||
a24c2569 MS |
144 | static struct kmem_cache *_cell_cache; |
145 | ||
991d9fa0 JT |
146 | /* |
147 | * @nr_cells should be the number of cells you want in use _concurrently_. | |
148 | * Don't confuse it with the number of distinct keys. | |
149 | */ | |
150 | static struct bio_prison *prison_create(unsigned nr_cells) | |
151 | { | |
152 | unsigned i; | |
153 | uint32_t nr_buckets = calc_nr_buckets(nr_cells); | |
154 | size_t len = sizeof(struct bio_prison) + | |
155 | (sizeof(struct hlist_head) * nr_buckets); | |
156 | struct bio_prison *prison = kmalloc(len, GFP_KERNEL); | |
157 | ||
158 | if (!prison) | |
159 | return NULL; | |
160 | ||
161 | spin_lock_init(&prison->lock); | |
a24c2569 | 162 | prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache); |
991d9fa0 JT |
163 | if (!prison->cell_pool) { |
164 | kfree(prison); | |
165 | return NULL; | |
166 | } | |
167 | ||
168 | prison->nr_buckets = nr_buckets; | |
169 | prison->hash_mask = nr_buckets - 1; | |
170 | prison->cells = (struct hlist_head *) (prison + 1); | |
171 | for (i = 0; i < nr_buckets; i++) | |
172 | INIT_HLIST_HEAD(prison->cells + i); | |
173 | ||
174 | return prison; | |
175 | } | |
176 | ||
177 | static void prison_destroy(struct bio_prison *prison) | |
178 | { | |
179 | mempool_destroy(prison->cell_pool); | |
180 | kfree(prison); | |
181 | } | |
182 | ||
183 | static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key) | |
184 | { | |
185 | const unsigned long BIG_PRIME = 4294967291UL; | |
186 | uint64_t hash = key->block * BIG_PRIME; | |
187 | ||
188 | return (uint32_t) (hash & prison->hash_mask); | |
189 | } | |
190 | ||
191 | static int keys_equal(struct cell_key *lhs, struct cell_key *rhs) | |
192 | { | |
193 | return (lhs->virtual == rhs->virtual) && | |
194 | (lhs->dev == rhs->dev) && | |
195 | (lhs->block == rhs->block); | |
196 | } | |
197 | ||
a24c2569 MS |
198 | static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket, |
199 | struct cell_key *key) | |
991d9fa0 | 200 | { |
a24c2569 | 201 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
202 | struct hlist_node *tmp; |
203 | ||
204 | hlist_for_each_entry(cell, tmp, bucket, list) | |
205 | if (keys_equal(&cell->key, key)) | |
206 | return cell; | |
207 | ||
208 | return NULL; | |
209 | } | |
210 | ||
211 | /* | |
212 | * This may block if a new cell needs allocating. You must ensure that | |
213 | * cells will be unlocked even if the calling thread is blocked. | |
214 | * | |
6f94a4c4 | 215 | * Returns 1 if the cell was already held, 0 if @inmate is the new holder. |
991d9fa0 JT |
216 | */ |
217 | static int bio_detain(struct bio_prison *prison, struct cell_key *key, | |
a24c2569 | 218 | struct bio *inmate, struct dm_bio_prison_cell **ref) |
991d9fa0 | 219 | { |
6f94a4c4 | 220 | int r = 1; |
991d9fa0 JT |
221 | unsigned long flags; |
222 | uint32_t hash = hash_key(prison, key); | |
a24c2569 | 223 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
224 | |
225 | BUG_ON(hash > prison->nr_buckets); | |
226 | ||
227 | spin_lock_irqsave(&prison->lock, flags); | |
991d9fa0 | 228 | |
6f94a4c4 JT |
229 | cell = __search_bucket(prison->cells + hash, key); |
230 | if (cell) { | |
231 | bio_list_add(&cell->bios, inmate); | |
232 | goto out; | |
991d9fa0 JT |
233 | } |
234 | ||
6f94a4c4 JT |
235 | /* |
236 | * Allocate a new cell | |
237 | */ | |
991d9fa0 | 238 | spin_unlock_irqrestore(&prison->lock, flags); |
6f94a4c4 JT |
239 | cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO); |
240 | spin_lock_irqsave(&prison->lock, flags); | |
991d9fa0 | 241 | |
6f94a4c4 JT |
242 | /* |
243 | * We've been unlocked, so we have to double check that | |
244 | * nobody else has inserted this cell in the meantime. | |
245 | */ | |
246 | cell = __search_bucket(prison->cells + hash, key); | |
247 | if (cell) { | |
991d9fa0 | 248 | mempool_free(cell2, prison->cell_pool); |
6f94a4c4 JT |
249 | bio_list_add(&cell->bios, inmate); |
250 | goto out; | |
251 | } | |
252 | ||
253 | /* | |
254 | * Use new cell. | |
255 | */ | |
256 | cell = cell2; | |
257 | ||
258 | cell->prison = prison; | |
259 | memcpy(&cell->key, key, sizeof(cell->key)); | |
260 | cell->holder = inmate; | |
261 | bio_list_init(&cell->bios); | |
262 | hlist_add_head(&cell->list, prison->cells + hash); | |
263 | ||
264 | r = 0; | |
265 | ||
266 | out: | |
267 | spin_unlock_irqrestore(&prison->lock, flags); | |
991d9fa0 JT |
268 | |
269 | *ref = cell; | |
270 | ||
271 | return r; | |
272 | } | |
273 | ||
274 | /* | |
275 | * @inmates must have been initialised prior to this call | |
276 | */ | |
a24c2569 | 277 | static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates) |
991d9fa0 JT |
278 | { |
279 | struct bio_prison *prison = cell->prison; | |
280 | ||
281 | hlist_del(&cell->list); | |
282 | ||
03aaae7c MS |
283 | if (inmates) { |
284 | bio_list_add(inmates, cell->holder); | |
285 | bio_list_merge(inmates, &cell->bios); | |
286 | } | |
991d9fa0 JT |
287 | |
288 | mempool_free(cell, prison->cell_pool); | |
289 | } | |
290 | ||
a24c2569 | 291 | static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios) |
991d9fa0 JT |
292 | { |
293 | unsigned long flags; | |
294 | struct bio_prison *prison = cell->prison; | |
295 | ||
296 | spin_lock_irqsave(&prison->lock, flags); | |
297 | __cell_release(cell, bios); | |
298 | spin_unlock_irqrestore(&prison->lock, flags); | |
299 | } | |
300 | ||
301 | /* | |
302 | * There are a couple of places where we put a bio into a cell briefly | |
303 | * before taking it out again. In these situations we know that no other | |
304 | * bio may be in the cell. This function releases the cell, and also does | |
305 | * a sanity check. | |
306 | */ | |
a24c2569 | 307 | static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
6f94a4c4 | 308 | { |
6f94a4c4 JT |
309 | BUG_ON(cell->holder != bio); |
310 | BUG_ON(!bio_list_empty(&cell->bios)); | |
03aaae7c MS |
311 | |
312 | __cell_release(cell, NULL); | |
6f94a4c4 JT |
313 | } |
314 | ||
a24c2569 | 315 | static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 | 316 | { |
991d9fa0 | 317 | unsigned long flags; |
6f94a4c4 | 318 | struct bio_prison *prison = cell->prison; |
991d9fa0 JT |
319 | |
320 | spin_lock_irqsave(&prison->lock, flags); | |
6f94a4c4 | 321 | __cell_release_singleton(cell, bio); |
991d9fa0 | 322 | spin_unlock_irqrestore(&prison->lock, flags); |
6f94a4c4 JT |
323 | } |
324 | ||
325 | /* | |
326 | * Sometimes we don't want the holder, just the additional bios. | |
327 | */ | |
a24c2569 MS |
328 | static void __cell_release_no_holder(struct dm_bio_prison_cell *cell, |
329 | struct bio_list *inmates) | |
6f94a4c4 JT |
330 | { |
331 | struct bio_prison *prison = cell->prison; | |
332 | ||
333 | hlist_del(&cell->list); | |
334 | bio_list_merge(inmates, &cell->bios); | |
335 | ||
336 | mempool_free(cell, prison->cell_pool); | |
337 | } | |
338 | ||
a24c2569 MS |
339 | static void cell_release_no_holder(struct dm_bio_prison_cell *cell, |
340 | struct bio_list *inmates) | |
6f94a4c4 JT |
341 | { |
342 | unsigned long flags; | |
343 | struct bio_prison *prison = cell->prison; | |
991d9fa0 | 344 | |
6f94a4c4 JT |
345 | spin_lock_irqsave(&prison->lock, flags); |
346 | __cell_release_no_holder(cell, inmates); | |
347 | spin_unlock_irqrestore(&prison->lock, flags); | |
991d9fa0 JT |
348 | } |
349 | ||
a24c2569 | 350 | static void cell_error(struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
351 | { |
352 | struct bio_prison *prison = cell->prison; | |
353 | struct bio_list bios; | |
354 | struct bio *bio; | |
355 | unsigned long flags; | |
356 | ||
357 | bio_list_init(&bios); | |
358 | ||
359 | spin_lock_irqsave(&prison->lock, flags); | |
360 | __cell_release(cell, &bios); | |
361 | spin_unlock_irqrestore(&prison->lock, flags); | |
362 | ||
363 | while ((bio = bio_list_pop(&bios))) | |
364 | bio_io_error(bio); | |
365 | } | |
366 | ||
367 | /*----------------------------------------------------------------*/ | |
368 | ||
369 | /* | |
370 | * We use the deferred set to keep track of pending reads to shared blocks. | |
371 | * We do this to ensure the new mapping caused by a write isn't performed | |
372 | * until these prior reads have completed. Otherwise the insertion of the | |
373 | * new mapping could free the old block that the read bios are mapped to. | |
374 | */ | |
375 | ||
376 | struct deferred_set; | |
377 | struct deferred_entry { | |
378 | struct deferred_set *ds; | |
379 | unsigned count; | |
380 | struct list_head work_items; | |
381 | }; | |
382 | ||
383 | struct deferred_set { | |
384 | spinlock_t lock; | |
385 | unsigned current_entry; | |
386 | unsigned sweeper; | |
387 | struct deferred_entry entries[DEFERRED_SET_SIZE]; | |
388 | }; | |
389 | ||
390 | static void ds_init(struct deferred_set *ds) | |
391 | { | |
392 | int i; | |
393 | ||
394 | spin_lock_init(&ds->lock); | |
395 | ds->current_entry = 0; | |
396 | ds->sweeper = 0; | |
397 | for (i = 0; i < DEFERRED_SET_SIZE; i++) { | |
398 | ds->entries[i].ds = ds; | |
399 | ds->entries[i].count = 0; | |
400 | INIT_LIST_HEAD(&ds->entries[i].work_items); | |
401 | } | |
402 | } | |
403 | ||
404 | static struct deferred_entry *ds_inc(struct deferred_set *ds) | |
405 | { | |
406 | unsigned long flags; | |
407 | struct deferred_entry *entry; | |
408 | ||
409 | spin_lock_irqsave(&ds->lock, flags); | |
410 | entry = ds->entries + ds->current_entry; | |
411 | entry->count++; | |
412 | spin_unlock_irqrestore(&ds->lock, flags); | |
413 | ||
414 | return entry; | |
415 | } | |
416 | ||
417 | static unsigned ds_next(unsigned index) | |
418 | { | |
419 | return (index + 1) % DEFERRED_SET_SIZE; | |
420 | } | |
421 | ||
422 | static void __sweep(struct deferred_set *ds, struct list_head *head) | |
423 | { | |
424 | while ((ds->sweeper != ds->current_entry) && | |
425 | !ds->entries[ds->sweeper].count) { | |
426 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); | |
427 | ds->sweeper = ds_next(ds->sweeper); | |
428 | } | |
429 | ||
430 | if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count) | |
431 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); | |
432 | } | |
433 | ||
434 | static void ds_dec(struct deferred_entry *entry, struct list_head *head) | |
435 | { | |
436 | unsigned long flags; | |
437 | ||
438 | spin_lock_irqsave(&entry->ds->lock, flags); | |
439 | BUG_ON(!entry->count); | |
440 | --entry->count; | |
441 | __sweep(entry->ds, head); | |
442 | spin_unlock_irqrestore(&entry->ds->lock, flags); | |
443 | } | |
444 | ||
445 | /* | |
446 | * Returns 1 if deferred or 0 if no pending items to delay job. | |
447 | */ | |
448 | static int ds_add_work(struct deferred_set *ds, struct list_head *work) | |
449 | { | |
450 | int r = 1; | |
451 | unsigned long flags; | |
452 | unsigned next_entry; | |
453 | ||
454 | spin_lock_irqsave(&ds->lock, flags); | |
455 | if ((ds->sweeper == ds->current_entry) && | |
456 | !ds->entries[ds->current_entry].count) | |
457 | r = 0; | |
458 | else { | |
459 | list_add(work, &ds->entries[ds->current_entry].work_items); | |
460 | next_entry = ds_next(ds->current_entry); | |
461 | if (!ds->entries[next_entry].count) | |
462 | ds->current_entry = next_entry; | |
463 | } | |
464 | spin_unlock_irqrestore(&ds->lock, flags); | |
465 | ||
466 | return r; | |
467 | } | |
468 | ||
469 | /*----------------------------------------------------------------*/ | |
470 | ||
471 | /* | |
472 | * Key building. | |
473 | */ | |
474 | static void build_data_key(struct dm_thin_device *td, | |
475 | dm_block_t b, struct cell_key *key) | |
476 | { | |
477 | key->virtual = 0; | |
478 | key->dev = dm_thin_dev_id(td); | |
479 | key->block = b; | |
480 | } | |
481 | ||
482 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
483 | struct cell_key *key) | |
484 | { | |
485 | key->virtual = 1; | |
486 | key->dev = dm_thin_dev_id(td); | |
487 | key->block = b; | |
488 | } | |
489 | ||
490 | /*----------------------------------------------------------------*/ | |
491 | ||
492 | /* | |
493 | * A pool device ties together a metadata device and a data device. It | |
494 | * also provides the interface for creating and destroying internal | |
495 | * devices. | |
496 | */ | |
a24c2569 | 497 | struct dm_thin_new_mapping; |
67e2e2b2 JT |
498 | |
499 | struct pool_features { | |
500 | unsigned zero_new_blocks:1; | |
501 | unsigned discard_enabled:1; | |
502 | unsigned discard_passdown:1; | |
503 | }; | |
504 | ||
991d9fa0 JT |
505 | struct pool { |
506 | struct list_head list; | |
507 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
508 | ||
509 | struct mapped_device *pool_md; | |
510 | struct block_device *md_dev; | |
511 | struct dm_pool_metadata *pmd; | |
512 | ||
991d9fa0 | 513 | dm_block_t low_water_blocks; |
55f2b8bd | 514 | uint32_t sectors_per_block; |
f9a8e0cd | 515 | int sectors_per_block_shift; |
991d9fa0 | 516 | |
67e2e2b2 | 517 | struct pool_features pf; |
991d9fa0 JT |
518 | unsigned low_water_triggered:1; /* A dm event has been sent */ |
519 | unsigned no_free_space:1; /* A -ENOSPC warning has been issued */ | |
520 | ||
521 | struct bio_prison *prison; | |
522 | struct dm_kcopyd_client *copier; | |
523 | ||
524 | struct workqueue_struct *wq; | |
525 | struct work_struct worker; | |
905e51b3 | 526 | struct delayed_work waker; |
991d9fa0 | 527 | |
905e51b3 | 528 | unsigned long last_commit_jiffies; |
55f2b8bd | 529 | unsigned ref_count; |
991d9fa0 JT |
530 | |
531 | spinlock_t lock; | |
532 | struct bio_list deferred_bios; | |
533 | struct bio_list deferred_flush_bios; | |
534 | struct list_head prepared_mappings; | |
104655fd | 535 | struct list_head prepared_discards; |
991d9fa0 JT |
536 | |
537 | struct bio_list retry_on_resume_list; | |
538 | ||
eb2aa48d | 539 | struct deferred_set shared_read_ds; |
104655fd | 540 | struct deferred_set all_io_ds; |
991d9fa0 | 541 | |
a24c2569 | 542 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 JT |
543 | mempool_t *mapping_pool; |
544 | mempool_t *endio_hook_pool; | |
545 | }; | |
546 | ||
547 | /* | |
548 | * Target context for a pool. | |
549 | */ | |
550 | struct pool_c { | |
551 | struct dm_target *ti; | |
552 | struct pool *pool; | |
553 | struct dm_dev *data_dev; | |
554 | struct dm_dev *metadata_dev; | |
555 | struct dm_target_callbacks callbacks; | |
556 | ||
557 | dm_block_t low_water_blocks; | |
67e2e2b2 | 558 | struct pool_features pf; |
991d9fa0 JT |
559 | }; |
560 | ||
561 | /* | |
562 | * Target context for a thin. | |
563 | */ | |
564 | struct thin_c { | |
565 | struct dm_dev *pool_dev; | |
2dd9c257 | 566 | struct dm_dev *origin_dev; |
991d9fa0 JT |
567 | dm_thin_id dev_id; |
568 | ||
569 | struct pool *pool; | |
570 | struct dm_thin_device *td; | |
571 | }; | |
572 | ||
573 | /*----------------------------------------------------------------*/ | |
574 | ||
575 | /* | |
576 | * A global list of pools that uses a struct mapped_device as a key. | |
577 | */ | |
578 | static struct dm_thin_pool_table { | |
579 | struct mutex mutex; | |
580 | struct list_head pools; | |
581 | } dm_thin_pool_table; | |
582 | ||
583 | static void pool_table_init(void) | |
584 | { | |
585 | mutex_init(&dm_thin_pool_table.mutex); | |
586 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
587 | } | |
588 | ||
589 | static void __pool_table_insert(struct pool *pool) | |
590 | { | |
591 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
592 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
593 | } | |
594 | ||
595 | static void __pool_table_remove(struct pool *pool) | |
596 | { | |
597 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
598 | list_del(&pool->list); | |
599 | } | |
600 | ||
601 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
602 | { | |
603 | struct pool *pool = NULL, *tmp; | |
604 | ||
605 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
606 | ||
607 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
608 | if (tmp->pool_md == md) { | |
609 | pool = tmp; | |
610 | break; | |
611 | } | |
612 | } | |
613 | ||
614 | return pool; | |
615 | } | |
616 | ||
617 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
618 | { | |
619 | struct pool *pool = NULL, *tmp; | |
620 | ||
621 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
622 | ||
623 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
624 | if (tmp->md_dev == md_dev) { | |
625 | pool = tmp; | |
626 | break; | |
627 | } | |
628 | } | |
629 | ||
630 | return pool; | |
631 | } | |
632 | ||
633 | /*----------------------------------------------------------------*/ | |
634 | ||
a24c2569 | 635 | struct dm_thin_endio_hook { |
eb2aa48d JT |
636 | struct thin_c *tc; |
637 | struct deferred_entry *shared_read_entry; | |
104655fd | 638 | struct deferred_entry *all_io_entry; |
a24c2569 | 639 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
640 | }; |
641 | ||
991d9fa0 JT |
642 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
643 | { | |
644 | struct bio *bio; | |
645 | struct bio_list bios; | |
646 | ||
647 | bio_list_init(&bios); | |
648 | bio_list_merge(&bios, master); | |
649 | bio_list_init(master); | |
650 | ||
651 | while ((bio = bio_list_pop(&bios))) { | |
a24c2569 MS |
652 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
653 | ||
eb2aa48d | 654 | if (h->tc == tc) |
991d9fa0 JT |
655 | bio_endio(bio, DM_ENDIO_REQUEUE); |
656 | else | |
657 | bio_list_add(master, bio); | |
658 | } | |
659 | } | |
660 | ||
661 | static void requeue_io(struct thin_c *tc) | |
662 | { | |
663 | struct pool *pool = tc->pool; | |
664 | unsigned long flags; | |
665 | ||
666 | spin_lock_irqsave(&pool->lock, flags); | |
667 | __requeue_bio_list(tc, &pool->deferred_bios); | |
668 | __requeue_bio_list(tc, &pool->retry_on_resume_list); | |
669 | spin_unlock_irqrestore(&pool->lock, flags); | |
670 | } | |
671 | ||
672 | /* | |
673 | * This section of code contains the logic for processing a thin device's IO. | |
674 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
675 | * but most is exclusively called from the thin target rather than the thin-pool | |
676 | * target. | |
677 | */ | |
678 | ||
679 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
680 | { | |
55f2b8bd MS |
681 | sector_t block_nr = bio->bi_sector; |
682 | ||
f9a8e0cd MP |
683 | if (tc->pool->sectors_per_block_shift < 0) |
684 | (void) sector_div(block_nr, tc->pool->sectors_per_block); | |
685 | else | |
686 | block_nr >>= tc->pool->sectors_per_block_shift; | |
55f2b8bd MS |
687 | |
688 | return block_nr; | |
991d9fa0 JT |
689 | } |
690 | ||
691 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
692 | { | |
693 | struct pool *pool = tc->pool; | |
55f2b8bd | 694 | sector_t bi_sector = bio->bi_sector; |
991d9fa0 JT |
695 | |
696 | bio->bi_bdev = tc->pool_dev->bdev; | |
f9a8e0cd MP |
697 | if (tc->pool->sectors_per_block_shift < 0) |
698 | bio->bi_sector = (block * pool->sectors_per_block) + | |
699 | sector_div(bi_sector, pool->sectors_per_block); | |
700 | else | |
701 | bio->bi_sector = (block << pool->sectors_per_block_shift) | | |
702 | (bi_sector & (pool->sectors_per_block - 1)); | |
991d9fa0 JT |
703 | } |
704 | ||
2dd9c257 JT |
705 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
706 | { | |
707 | bio->bi_bdev = tc->origin_dev->bdev; | |
708 | } | |
709 | ||
710 | static void issue(struct thin_c *tc, struct bio *bio) | |
991d9fa0 JT |
711 | { |
712 | struct pool *pool = tc->pool; | |
713 | unsigned long flags; | |
714 | ||
991d9fa0 JT |
715 | /* |
716 | * Batch together any FUA/FLUSH bios we find and then issue | |
717 | * a single commit for them in process_deferred_bios(). | |
718 | */ | |
719 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { | |
720 | spin_lock_irqsave(&pool->lock, flags); | |
721 | bio_list_add(&pool->deferred_flush_bios, bio); | |
722 | spin_unlock_irqrestore(&pool->lock, flags); | |
723 | } else | |
724 | generic_make_request(bio); | |
725 | } | |
726 | ||
2dd9c257 JT |
727 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
728 | { | |
729 | remap_to_origin(tc, bio); | |
730 | issue(tc, bio); | |
731 | } | |
732 | ||
733 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
734 | dm_block_t block) | |
735 | { | |
736 | remap(tc, bio, block); | |
737 | issue(tc, bio); | |
738 | } | |
739 | ||
991d9fa0 JT |
740 | /* |
741 | * wake_worker() is used when new work is queued and when pool_resume is | |
742 | * ready to continue deferred IO processing. | |
743 | */ | |
744 | static void wake_worker(struct pool *pool) | |
745 | { | |
746 | queue_work(pool->wq, &pool->worker); | |
747 | } | |
748 | ||
749 | /*----------------------------------------------------------------*/ | |
750 | ||
751 | /* | |
752 | * Bio endio functions. | |
753 | */ | |
a24c2569 | 754 | struct dm_thin_new_mapping { |
991d9fa0 JT |
755 | struct list_head list; |
756 | ||
eb2aa48d JT |
757 | unsigned quiesced:1; |
758 | unsigned prepared:1; | |
104655fd | 759 | unsigned pass_discard:1; |
991d9fa0 JT |
760 | |
761 | struct thin_c *tc; | |
762 | dm_block_t virt_block; | |
763 | dm_block_t data_block; | |
a24c2569 | 764 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
765 | int err; |
766 | ||
767 | /* | |
768 | * If the bio covers the whole area of a block then we can avoid | |
769 | * zeroing or copying. Instead this bio is hooked. The bio will | |
770 | * still be in the cell, so care has to be taken to avoid issuing | |
771 | * the bio twice. | |
772 | */ | |
773 | struct bio *bio; | |
774 | bio_end_io_t *saved_bi_end_io; | |
775 | }; | |
776 | ||
a24c2569 | 777 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
778 | { |
779 | struct pool *pool = m->tc->pool; | |
780 | ||
eb2aa48d | 781 | if (m->quiesced && m->prepared) { |
991d9fa0 JT |
782 | list_add(&m->list, &pool->prepared_mappings); |
783 | wake_worker(pool); | |
784 | } | |
785 | } | |
786 | ||
787 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
788 | { | |
789 | unsigned long flags; | |
a24c2569 | 790 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
791 | struct pool *pool = m->tc->pool; |
792 | ||
793 | m->err = read_err || write_err ? -EIO : 0; | |
794 | ||
795 | spin_lock_irqsave(&pool->lock, flags); | |
796 | m->prepared = 1; | |
797 | __maybe_add_mapping(m); | |
798 | spin_unlock_irqrestore(&pool->lock, flags); | |
799 | } | |
800 | ||
801 | static void overwrite_endio(struct bio *bio, int err) | |
802 | { | |
803 | unsigned long flags; | |
a24c2569 MS |
804 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
805 | struct dm_thin_new_mapping *m = h->overwrite_mapping; | |
991d9fa0 JT |
806 | struct pool *pool = m->tc->pool; |
807 | ||
808 | m->err = err; | |
809 | ||
810 | spin_lock_irqsave(&pool->lock, flags); | |
811 | m->prepared = 1; | |
812 | __maybe_add_mapping(m); | |
813 | spin_unlock_irqrestore(&pool->lock, flags); | |
814 | } | |
815 | ||
991d9fa0 JT |
816 | /*----------------------------------------------------------------*/ |
817 | ||
818 | /* | |
819 | * Workqueue. | |
820 | */ | |
821 | ||
822 | /* | |
823 | * Prepared mapping jobs. | |
824 | */ | |
825 | ||
826 | /* | |
827 | * This sends the bios in the cell back to the deferred_bios list. | |
828 | */ | |
a24c2569 | 829 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
830 | dm_block_t data_block) |
831 | { | |
832 | struct pool *pool = tc->pool; | |
833 | unsigned long flags; | |
834 | ||
835 | spin_lock_irqsave(&pool->lock, flags); | |
836 | cell_release(cell, &pool->deferred_bios); | |
837 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
838 | ||
839 | wake_worker(pool); | |
840 | } | |
841 | ||
842 | /* | |
843 | * Same as cell_defer above, except it omits one particular detainee, | |
844 | * a write bio that covers the block and has already been processed. | |
845 | */ | |
a24c2569 | 846 | static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
847 | { |
848 | struct bio_list bios; | |
991d9fa0 JT |
849 | struct pool *pool = tc->pool; |
850 | unsigned long flags; | |
851 | ||
852 | bio_list_init(&bios); | |
991d9fa0 JT |
853 | |
854 | spin_lock_irqsave(&pool->lock, flags); | |
6f94a4c4 | 855 | cell_release_no_holder(cell, &pool->deferred_bios); |
991d9fa0 JT |
856 | spin_unlock_irqrestore(&pool->lock, flags); |
857 | ||
858 | wake_worker(pool); | |
859 | } | |
860 | ||
a24c2569 | 861 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
862 | { |
863 | struct thin_c *tc = m->tc; | |
864 | struct bio *bio; | |
865 | int r; | |
866 | ||
867 | bio = m->bio; | |
868 | if (bio) | |
869 | bio->bi_end_io = m->saved_bi_end_io; | |
870 | ||
871 | if (m->err) { | |
872 | cell_error(m->cell); | |
905386f8 | 873 | goto out; |
991d9fa0 JT |
874 | } |
875 | ||
876 | /* | |
877 | * Commit the prepared block into the mapping btree. | |
878 | * Any I/O for this block arriving after this point will get | |
879 | * remapped to it directly. | |
880 | */ | |
881 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
882 | if (r) { | |
883 | DMERR("dm_thin_insert_block() failed"); | |
884 | cell_error(m->cell); | |
905386f8 | 885 | goto out; |
991d9fa0 JT |
886 | } |
887 | ||
888 | /* | |
889 | * Release any bios held while the block was being provisioned. | |
890 | * If we are processing a write bio that completely covers the block, | |
891 | * we already processed it so can ignore it now when processing | |
892 | * the bios in the cell. | |
893 | */ | |
894 | if (bio) { | |
6f94a4c4 | 895 | cell_defer_except(tc, m->cell); |
991d9fa0 JT |
896 | bio_endio(bio, 0); |
897 | } else | |
898 | cell_defer(tc, m->cell, m->data_block); | |
899 | ||
905386f8 | 900 | out: |
991d9fa0 JT |
901 | list_del(&m->list); |
902 | mempool_free(m, tc->pool->mapping_pool); | |
903 | } | |
904 | ||
a24c2569 | 905 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
104655fd JT |
906 | { |
907 | int r; | |
908 | struct thin_c *tc = m->tc; | |
909 | ||
910 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
911 | if (r) | |
912 | DMERR("dm_thin_remove_block() failed"); | |
913 | ||
914 | /* | |
915 | * Pass the discard down to the underlying device? | |
916 | */ | |
917 | if (m->pass_discard) | |
918 | remap_and_issue(tc, m->bio, m->data_block); | |
919 | else | |
920 | bio_endio(m->bio, 0); | |
921 | ||
922 | cell_defer_except(tc, m->cell); | |
923 | cell_defer_except(tc, m->cell2); | |
924 | mempool_free(m, tc->pool->mapping_pool); | |
925 | } | |
926 | ||
927 | static void process_prepared(struct pool *pool, struct list_head *head, | |
a24c2569 | 928 | void (*fn)(struct dm_thin_new_mapping *)) |
991d9fa0 JT |
929 | { |
930 | unsigned long flags; | |
931 | struct list_head maps; | |
a24c2569 | 932 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
933 | |
934 | INIT_LIST_HEAD(&maps); | |
935 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 936 | list_splice_init(head, &maps); |
991d9fa0 JT |
937 | spin_unlock_irqrestore(&pool->lock, flags); |
938 | ||
939 | list_for_each_entry_safe(m, tmp, &maps, list) | |
104655fd | 940 | fn(m); |
991d9fa0 JT |
941 | } |
942 | ||
943 | /* | |
944 | * Deferred bio jobs. | |
945 | */ | |
104655fd | 946 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 947 | { |
f9a8e0cd | 948 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
949 | } |
950 | ||
951 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
952 | { | |
953 | return (bio_data_dir(bio) == WRITE) && | |
954 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
955 | } |
956 | ||
957 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
958 | bio_end_io_t *fn) | |
959 | { | |
960 | *save = bio->bi_end_io; | |
961 | bio->bi_end_io = fn; | |
962 | } | |
963 | ||
964 | static int ensure_next_mapping(struct pool *pool) | |
965 | { | |
966 | if (pool->next_mapping) | |
967 | return 0; | |
968 | ||
969 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
970 | ||
971 | return pool->next_mapping ? 0 : -ENOMEM; | |
972 | } | |
973 | ||
a24c2569 | 974 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 975 | { |
a24c2569 | 976 | struct dm_thin_new_mapping *r = pool->next_mapping; |
991d9fa0 JT |
977 | |
978 | BUG_ON(!pool->next_mapping); | |
979 | ||
980 | pool->next_mapping = NULL; | |
981 | ||
982 | return r; | |
983 | } | |
984 | ||
985 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
986 | struct dm_dev *origin, dm_block_t data_origin, |
987 | dm_block_t data_dest, | |
a24c2569 | 988 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
989 | { |
990 | int r; | |
991 | struct pool *pool = tc->pool; | |
a24c2569 | 992 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 JT |
993 | |
994 | INIT_LIST_HEAD(&m->list); | |
eb2aa48d | 995 | m->quiesced = 0; |
991d9fa0 JT |
996 | m->prepared = 0; |
997 | m->tc = tc; | |
998 | m->virt_block = virt_block; | |
999 | m->data_block = data_dest; | |
1000 | m->cell = cell; | |
1001 | m->err = 0; | |
1002 | m->bio = NULL; | |
1003 | ||
eb2aa48d JT |
1004 | if (!ds_add_work(&pool->shared_read_ds, &m->list)) |
1005 | m->quiesced = 1; | |
991d9fa0 JT |
1006 | |
1007 | /* | |
1008 | * IO to pool_dev remaps to the pool target's data_dev. | |
1009 | * | |
1010 | * If the whole block of data is being overwritten, we can issue the | |
1011 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1012 | */ | |
1013 | if (io_overwrites_block(pool, bio)) { | |
a24c2569 MS |
1014 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
1015 | ||
eb2aa48d | 1016 | h->overwrite_mapping = m; |
991d9fa0 JT |
1017 | m->bio = bio; |
1018 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
991d9fa0 JT |
1019 | remap_and_issue(tc, bio, data_dest); |
1020 | } else { | |
1021 | struct dm_io_region from, to; | |
1022 | ||
2dd9c257 | 1023 | from.bdev = origin->bdev; |
991d9fa0 JT |
1024 | from.sector = data_origin * pool->sectors_per_block; |
1025 | from.count = pool->sectors_per_block; | |
1026 | ||
1027 | to.bdev = tc->pool_dev->bdev; | |
1028 | to.sector = data_dest * pool->sectors_per_block; | |
1029 | to.count = pool->sectors_per_block; | |
1030 | ||
1031 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1032 | 0, copy_complete, m); | |
1033 | if (r < 0) { | |
1034 | mempool_free(m, pool->mapping_pool); | |
1035 | DMERR("dm_kcopyd_copy() failed"); | |
1036 | cell_error(cell); | |
1037 | } | |
1038 | } | |
1039 | } | |
1040 | ||
2dd9c257 JT |
1041 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1042 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1043 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1044 | { |
1045 | schedule_copy(tc, virt_block, tc->pool_dev, | |
1046 | data_origin, data_dest, cell, bio); | |
1047 | } | |
1048 | ||
1049 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
1050 | dm_block_t data_dest, | |
a24c2569 | 1051 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1052 | { |
1053 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1054 | virt_block, data_dest, cell, bio); | |
1055 | } | |
1056 | ||
991d9fa0 | 1057 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1058 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1059 | struct bio *bio) |
1060 | { | |
1061 | struct pool *pool = tc->pool; | |
a24c2569 | 1062 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 JT |
1063 | |
1064 | INIT_LIST_HEAD(&m->list); | |
eb2aa48d | 1065 | m->quiesced = 1; |
991d9fa0 JT |
1066 | m->prepared = 0; |
1067 | m->tc = tc; | |
1068 | m->virt_block = virt_block; | |
1069 | m->data_block = data_block; | |
1070 | m->cell = cell; | |
1071 | m->err = 0; | |
1072 | m->bio = NULL; | |
1073 | ||
1074 | /* | |
1075 | * If the whole block of data is being overwritten or we are not | |
1076 | * zeroing pre-existing data, we can issue the bio immediately. | |
1077 | * Otherwise we use kcopyd to zero the data first. | |
1078 | */ | |
67e2e2b2 | 1079 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
1080 | process_prepared_mapping(m); |
1081 | ||
1082 | else if (io_overwrites_block(pool, bio)) { | |
a24c2569 MS |
1083 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
1084 | ||
eb2aa48d | 1085 | h->overwrite_mapping = m; |
991d9fa0 JT |
1086 | m->bio = bio; |
1087 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
991d9fa0 | 1088 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
1089 | } else { |
1090 | int r; | |
1091 | struct dm_io_region to; | |
1092 | ||
1093 | to.bdev = tc->pool_dev->bdev; | |
1094 | to.sector = data_block * pool->sectors_per_block; | |
1095 | to.count = pool->sectors_per_block; | |
1096 | ||
1097 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
1098 | if (r < 0) { | |
1099 | mempool_free(m, pool->mapping_pool); | |
1100 | DMERR("dm_kcopyd_zero() failed"); | |
1101 | cell_error(cell); | |
1102 | } | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) | |
1107 | { | |
1108 | int r; | |
1109 | dm_block_t free_blocks; | |
1110 | unsigned long flags; | |
1111 | struct pool *pool = tc->pool; | |
1112 | ||
1113 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); | |
1114 | if (r) | |
1115 | return r; | |
1116 | ||
1117 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1118 | DMWARN("%s: reached low water mark, sending event.", | |
1119 | dm_device_name(pool->pool_md)); | |
1120 | spin_lock_irqsave(&pool->lock, flags); | |
1121 | pool->low_water_triggered = 1; | |
1122 | spin_unlock_irqrestore(&pool->lock, flags); | |
1123 | dm_table_event(pool->ti->table); | |
1124 | } | |
1125 | ||
1126 | if (!free_blocks) { | |
1127 | if (pool->no_free_space) | |
1128 | return -ENOSPC; | |
1129 | else { | |
1130 | /* | |
1131 | * Try to commit to see if that will free up some | |
1132 | * more space. | |
1133 | */ | |
1134 | r = dm_pool_commit_metadata(pool->pmd); | |
1135 | if (r) { | |
1136 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", | |
1137 | __func__, r); | |
1138 | return r; | |
1139 | } | |
1140 | ||
1141 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); | |
1142 | if (r) | |
1143 | return r; | |
1144 | ||
1145 | /* | |
1146 | * If we still have no space we set a flag to avoid | |
1147 | * doing all this checking and return -ENOSPC. | |
1148 | */ | |
1149 | if (!free_blocks) { | |
1150 | DMWARN("%s: no free space available.", | |
1151 | dm_device_name(pool->pool_md)); | |
1152 | spin_lock_irqsave(&pool->lock, flags); | |
1153 | pool->no_free_space = 1; | |
1154 | spin_unlock_irqrestore(&pool->lock, flags); | |
1155 | return -ENOSPC; | |
1156 | } | |
1157 | } | |
1158 | } | |
1159 | ||
1160 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
1161 | if (r) | |
1162 | return r; | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * If we have run out of space, queue bios until the device is | |
1169 | * resumed, presumably after having been reloaded with more space. | |
1170 | */ | |
1171 | static void retry_on_resume(struct bio *bio) | |
1172 | { | |
a24c2569 | 1173 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
eb2aa48d | 1174 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1175 | struct pool *pool = tc->pool; |
1176 | unsigned long flags; | |
1177 | ||
1178 | spin_lock_irqsave(&pool->lock, flags); | |
1179 | bio_list_add(&pool->retry_on_resume_list, bio); | |
1180 | spin_unlock_irqrestore(&pool->lock, flags); | |
1181 | } | |
1182 | ||
a24c2569 | 1183 | static void no_space(struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1184 | { |
1185 | struct bio *bio; | |
1186 | struct bio_list bios; | |
1187 | ||
1188 | bio_list_init(&bios); | |
1189 | cell_release(cell, &bios); | |
1190 | ||
1191 | while ((bio = bio_list_pop(&bios))) | |
1192 | retry_on_resume(bio); | |
1193 | } | |
1194 | ||
104655fd JT |
1195 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1196 | { | |
1197 | int r; | |
c3a0ce2e | 1198 | unsigned long flags; |
104655fd | 1199 | struct pool *pool = tc->pool; |
a24c2569 | 1200 | struct dm_bio_prison_cell *cell, *cell2; |
104655fd JT |
1201 | struct cell_key key, key2; |
1202 | dm_block_t block = get_bio_block(tc, bio); | |
1203 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1204 | struct dm_thin_new_mapping *m; |
104655fd JT |
1205 | |
1206 | build_virtual_key(tc->td, block, &key); | |
1207 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) | |
1208 | return; | |
1209 | ||
1210 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1211 | switch (r) { | |
1212 | case 0: | |
1213 | /* | |
1214 | * Check nobody is fiddling with this pool block. This can | |
1215 | * happen if someone's in the process of breaking sharing | |
1216 | * on this block. | |
1217 | */ | |
1218 | build_data_key(tc->td, lookup_result.block, &key2); | |
1219 | if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) { | |
1220 | cell_release_singleton(cell, bio); | |
1221 | break; | |
1222 | } | |
1223 | ||
1224 | if (io_overlaps_block(pool, bio)) { | |
1225 | /* | |
1226 | * IO may still be going to the destination block. We must | |
1227 | * quiesce before we can do the removal. | |
1228 | */ | |
1229 | m = get_next_mapping(pool); | |
1230 | m->tc = tc; | |
17b7d63f | 1231 | m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown; |
104655fd JT |
1232 | m->virt_block = block; |
1233 | m->data_block = lookup_result.block; | |
1234 | m->cell = cell; | |
1235 | m->cell2 = cell2; | |
1236 | m->err = 0; | |
1237 | m->bio = bio; | |
1238 | ||
1239 | if (!ds_add_work(&pool->all_io_ds, &m->list)) { | |
c3a0ce2e | 1240 | spin_lock_irqsave(&pool->lock, flags); |
104655fd | 1241 | list_add(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1242 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1243 | wake_worker(pool); |
1244 | } | |
1245 | } else { | |
1246 | /* | |
49296309 MP |
1247 | * The DM core makes sure that the discard doesn't span |
1248 | * a block boundary. So we submit the discard of a | |
1249 | * partial block appropriately. | |
104655fd | 1250 | */ |
104655fd JT |
1251 | cell_release_singleton(cell, bio); |
1252 | cell_release_singleton(cell2, bio); | |
650d2a06 MP |
1253 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1254 | remap_and_issue(tc, bio, lookup_result.block); | |
1255 | else | |
1256 | bio_endio(bio, 0); | |
104655fd JT |
1257 | } |
1258 | break; | |
1259 | ||
1260 | case -ENODATA: | |
1261 | /* | |
1262 | * It isn't provisioned, just forget it. | |
1263 | */ | |
1264 | cell_release_singleton(cell, bio); | |
1265 | bio_endio(bio, 0); | |
1266 | break; | |
1267 | ||
1268 | default: | |
1269 | DMERR("discard: find block unexpectedly returned %d", r); | |
1270 | cell_release_singleton(cell, bio); | |
1271 | bio_io_error(bio); | |
1272 | break; | |
1273 | } | |
1274 | } | |
1275 | ||
991d9fa0 JT |
1276 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
1277 | struct cell_key *key, | |
1278 | struct dm_thin_lookup_result *lookup_result, | |
a24c2569 | 1279 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1280 | { |
1281 | int r; | |
1282 | dm_block_t data_block; | |
1283 | ||
1284 | r = alloc_data_block(tc, &data_block); | |
1285 | switch (r) { | |
1286 | case 0: | |
2dd9c257 JT |
1287 | schedule_internal_copy(tc, block, lookup_result->block, |
1288 | data_block, cell, bio); | |
991d9fa0 JT |
1289 | break; |
1290 | ||
1291 | case -ENOSPC: | |
1292 | no_space(cell); | |
1293 | break; | |
1294 | ||
1295 | default: | |
1296 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); | |
1297 | cell_error(cell); | |
1298 | break; | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1303 | dm_block_t block, | |
1304 | struct dm_thin_lookup_result *lookup_result) | |
1305 | { | |
a24c2569 | 1306 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
1307 | struct pool *pool = tc->pool; |
1308 | struct cell_key key; | |
1309 | ||
1310 | /* | |
1311 | * If cell is already occupied, then sharing is already in the process | |
1312 | * of being broken so we have nothing further to do here. | |
1313 | */ | |
1314 | build_data_key(tc->td, lookup_result->block, &key); | |
1315 | if (bio_detain(pool->prison, &key, bio, &cell)) | |
1316 | return; | |
1317 | ||
60049701 | 1318 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1319 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1320 | else { | |
a24c2569 | 1321 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
991d9fa0 | 1322 | |
eb2aa48d | 1323 | h->shared_read_entry = ds_inc(&pool->shared_read_ds); |
991d9fa0 JT |
1324 | |
1325 | cell_release_singleton(cell, bio); | |
1326 | remap_and_issue(tc, bio, lookup_result->block); | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1331 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1332 | { |
1333 | int r; | |
1334 | dm_block_t data_block; | |
1335 | ||
1336 | /* | |
1337 | * Remap empty bios (flushes) immediately, without provisioning. | |
1338 | */ | |
1339 | if (!bio->bi_size) { | |
1340 | cell_release_singleton(cell, bio); | |
1341 | remap_and_issue(tc, bio, 0); | |
1342 | return; | |
1343 | } | |
1344 | ||
1345 | /* | |
1346 | * Fill read bios with zeroes and complete them immediately. | |
1347 | */ | |
1348 | if (bio_data_dir(bio) == READ) { | |
1349 | zero_fill_bio(bio); | |
1350 | cell_release_singleton(cell, bio); | |
1351 | bio_endio(bio, 0); | |
1352 | return; | |
1353 | } | |
1354 | ||
1355 | r = alloc_data_block(tc, &data_block); | |
1356 | switch (r) { | |
1357 | case 0: | |
2dd9c257 JT |
1358 | if (tc->origin_dev) |
1359 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1360 | else | |
1361 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1362 | break; |
1363 | ||
1364 | case -ENOSPC: | |
1365 | no_space(cell); | |
1366 | break; | |
1367 | ||
1368 | default: | |
1369 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); | |
1370 | cell_error(cell); | |
1371 | break; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1376 | { | |
1377 | int r; | |
1378 | dm_block_t block = get_bio_block(tc, bio); | |
a24c2569 | 1379 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
1380 | struct cell_key key; |
1381 | struct dm_thin_lookup_result lookup_result; | |
1382 | ||
1383 | /* | |
1384 | * If cell is already occupied, then the block is already | |
1385 | * being provisioned so we have nothing further to do here. | |
1386 | */ | |
1387 | build_virtual_key(tc->td, block, &key); | |
1388 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) | |
1389 | return; | |
1390 | ||
1391 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1392 | switch (r) { | |
1393 | case 0: | |
1394 | /* | |
1395 | * We can release this cell now. This thread is the only | |
1396 | * one that puts bios into a cell, and we know there were | |
1397 | * no preceding bios. | |
1398 | */ | |
1399 | /* | |
1400 | * TODO: this will probably have to change when discard goes | |
1401 | * back in. | |
1402 | */ | |
1403 | cell_release_singleton(cell, bio); | |
1404 | ||
1405 | if (lookup_result.shared) | |
1406 | process_shared_bio(tc, bio, block, &lookup_result); | |
1407 | else | |
1408 | remap_and_issue(tc, bio, lookup_result.block); | |
1409 | break; | |
1410 | ||
1411 | case -ENODATA: | |
2dd9c257 JT |
1412 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
1413 | cell_release_singleton(cell, bio); | |
1414 | remap_to_origin_and_issue(tc, bio); | |
1415 | } else | |
1416 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1417 | break; |
1418 | ||
1419 | default: | |
1420 | DMERR("dm_thin_find_block() failed, error = %d", r); | |
104655fd | 1421 | cell_release_singleton(cell, bio); |
991d9fa0 JT |
1422 | bio_io_error(bio); |
1423 | break; | |
1424 | } | |
1425 | } | |
1426 | ||
905e51b3 JT |
1427 | static int need_commit_due_to_time(struct pool *pool) |
1428 | { | |
1429 | return jiffies < pool->last_commit_jiffies || | |
1430 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1431 | } | |
1432 | ||
991d9fa0 JT |
1433 | static void process_deferred_bios(struct pool *pool) |
1434 | { | |
1435 | unsigned long flags; | |
1436 | struct bio *bio; | |
1437 | struct bio_list bios; | |
1438 | int r; | |
1439 | ||
1440 | bio_list_init(&bios); | |
1441 | ||
1442 | spin_lock_irqsave(&pool->lock, flags); | |
1443 | bio_list_merge(&bios, &pool->deferred_bios); | |
1444 | bio_list_init(&pool->deferred_bios); | |
1445 | spin_unlock_irqrestore(&pool->lock, flags); | |
1446 | ||
1447 | while ((bio = bio_list_pop(&bios))) { | |
a24c2569 | 1448 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
eb2aa48d JT |
1449 | struct thin_c *tc = h->tc; |
1450 | ||
991d9fa0 JT |
1451 | /* |
1452 | * If we've got no free new_mapping structs, and processing | |
1453 | * this bio might require one, we pause until there are some | |
1454 | * prepared mappings to process. | |
1455 | */ | |
1456 | if (ensure_next_mapping(pool)) { | |
1457 | spin_lock_irqsave(&pool->lock, flags); | |
1458 | bio_list_merge(&pool->deferred_bios, &bios); | |
1459 | spin_unlock_irqrestore(&pool->lock, flags); | |
1460 | ||
1461 | break; | |
1462 | } | |
104655fd JT |
1463 | |
1464 | if (bio->bi_rw & REQ_DISCARD) | |
1465 | process_discard(tc, bio); | |
1466 | else | |
1467 | process_bio(tc, bio); | |
991d9fa0 JT |
1468 | } |
1469 | ||
1470 | /* | |
1471 | * If there are any deferred flush bios, we must commit | |
1472 | * the metadata before issuing them. | |
1473 | */ | |
1474 | bio_list_init(&bios); | |
1475 | spin_lock_irqsave(&pool->lock, flags); | |
1476 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1477 | bio_list_init(&pool->deferred_flush_bios); | |
1478 | spin_unlock_irqrestore(&pool->lock, flags); | |
1479 | ||
905e51b3 | 1480 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1481 | return; |
1482 | ||
1483 | r = dm_pool_commit_metadata(pool->pmd); | |
1484 | if (r) { | |
1485 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", | |
1486 | __func__, r); | |
1487 | while ((bio = bio_list_pop(&bios))) | |
1488 | bio_io_error(bio); | |
1489 | return; | |
1490 | } | |
905e51b3 | 1491 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1492 | |
1493 | while ((bio = bio_list_pop(&bios))) | |
1494 | generic_make_request(bio); | |
1495 | } | |
1496 | ||
1497 | static void do_worker(struct work_struct *ws) | |
1498 | { | |
1499 | struct pool *pool = container_of(ws, struct pool, worker); | |
1500 | ||
104655fd JT |
1501 | process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping); |
1502 | process_prepared(pool, &pool->prepared_discards, process_prepared_discard); | |
991d9fa0 JT |
1503 | process_deferred_bios(pool); |
1504 | } | |
1505 | ||
905e51b3 JT |
1506 | /* |
1507 | * We want to commit periodically so that not too much | |
1508 | * unwritten data builds up. | |
1509 | */ | |
1510 | static void do_waker(struct work_struct *ws) | |
1511 | { | |
1512 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1513 | wake_worker(pool); | |
1514 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1515 | } | |
1516 | ||
991d9fa0 JT |
1517 | /*----------------------------------------------------------------*/ |
1518 | ||
1519 | /* | |
1520 | * Mapping functions. | |
1521 | */ | |
1522 | ||
1523 | /* | |
1524 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1525 | */ | |
1526 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1527 | { | |
1528 | unsigned long flags; | |
1529 | struct pool *pool = tc->pool; | |
1530 | ||
1531 | spin_lock_irqsave(&pool->lock, flags); | |
1532 | bio_list_add(&pool->deferred_bios, bio); | |
1533 | spin_unlock_irqrestore(&pool->lock, flags); | |
1534 | ||
1535 | wake_worker(pool); | |
1536 | } | |
1537 | ||
a24c2569 | 1538 | static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d JT |
1539 | { |
1540 | struct pool *pool = tc->pool; | |
a24c2569 | 1541 | struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO); |
eb2aa48d JT |
1542 | |
1543 | h->tc = tc; | |
1544 | h->shared_read_entry = NULL; | |
104655fd | 1545 | h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds); |
eb2aa48d JT |
1546 | h->overwrite_mapping = NULL; |
1547 | ||
1548 | return h; | |
1549 | } | |
1550 | ||
991d9fa0 JT |
1551 | /* |
1552 | * Non-blocking function called from the thin target's map function. | |
1553 | */ | |
1554 | static int thin_bio_map(struct dm_target *ti, struct bio *bio, | |
1555 | union map_info *map_context) | |
1556 | { | |
1557 | int r; | |
1558 | struct thin_c *tc = ti->private; | |
1559 | dm_block_t block = get_bio_block(tc, bio); | |
1560 | struct dm_thin_device *td = tc->td; | |
1561 | struct dm_thin_lookup_result result; | |
1562 | ||
eb2aa48d | 1563 | map_context->ptr = thin_hook_bio(tc, bio); |
104655fd | 1564 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1565 | thin_defer_bio(tc, bio); |
1566 | return DM_MAPIO_SUBMITTED; | |
1567 | } | |
1568 | ||
1569 | r = dm_thin_find_block(td, block, 0, &result); | |
1570 | ||
1571 | /* | |
1572 | * Note that we defer readahead too. | |
1573 | */ | |
1574 | switch (r) { | |
1575 | case 0: | |
1576 | if (unlikely(result.shared)) { | |
1577 | /* | |
1578 | * We have a race condition here between the | |
1579 | * result.shared value returned by the lookup and | |
1580 | * snapshot creation, which may cause new | |
1581 | * sharing. | |
1582 | * | |
1583 | * To avoid this always quiesce the origin before | |
1584 | * taking the snap. You want to do this anyway to | |
1585 | * ensure a consistent application view | |
1586 | * (i.e. lockfs). | |
1587 | * | |
1588 | * More distant ancestors are irrelevant. The | |
1589 | * shared flag will be set in their case. | |
1590 | */ | |
1591 | thin_defer_bio(tc, bio); | |
1592 | r = DM_MAPIO_SUBMITTED; | |
1593 | } else { | |
1594 | remap(tc, bio, result.block); | |
1595 | r = DM_MAPIO_REMAPPED; | |
1596 | } | |
1597 | break; | |
1598 | ||
1599 | case -ENODATA: | |
1600 | /* | |
1601 | * In future, the failed dm_thin_find_block above could | |
1602 | * provide the hint to load the metadata into cache. | |
1603 | */ | |
1604 | case -EWOULDBLOCK: | |
1605 | thin_defer_bio(tc, bio); | |
1606 | r = DM_MAPIO_SUBMITTED; | |
1607 | break; | |
1608 | } | |
1609 | ||
1610 | return r; | |
1611 | } | |
1612 | ||
1613 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1614 | { | |
1615 | int r; | |
1616 | unsigned long flags; | |
1617 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1618 | ||
1619 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1620 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1621 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1622 | ||
1623 | if (!r) { | |
1624 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1625 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1626 | } | |
1627 | ||
1628 | return r; | |
1629 | } | |
1630 | ||
1631 | static void __requeue_bios(struct pool *pool) | |
1632 | { | |
1633 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1634 | bio_list_init(&pool->retry_on_resume_list); | |
1635 | } | |
1636 | ||
1637 | /*---------------------------------------------------------------- | |
1638 | * Binding of control targets to a pool object | |
1639 | *--------------------------------------------------------------*/ | |
1640 | static int bind_control_target(struct pool *pool, struct dm_target *ti) | |
1641 | { | |
1642 | struct pool_c *pt = ti->private; | |
1643 | ||
1644 | pool->ti = ti; | |
1645 | pool->low_water_blocks = pt->low_water_blocks; | |
67e2e2b2 | 1646 | pool->pf = pt->pf; |
991d9fa0 | 1647 | |
f402693d MS |
1648 | /* |
1649 | * If discard_passdown was enabled verify that the data device | |
1650 | * supports discards. Disable discard_passdown if not; otherwise | |
1651 | * -EOPNOTSUPP will be returned. | |
1652 | */ | |
1653 | if (pt->pf.discard_passdown) { | |
1654 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1655 | if (!q || !blk_queue_discard(q)) { | |
1656 | char buf[BDEVNAME_SIZE]; | |
1657 | DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.", | |
1658 | bdevname(pt->data_dev->bdev, buf)); | |
1659 | pool->pf.discard_passdown = 0; | |
1660 | } | |
1661 | } | |
1662 | ||
991d9fa0 JT |
1663 | return 0; |
1664 | } | |
1665 | ||
1666 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1667 | { | |
1668 | if (pool->ti == ti) | |
1669 | pool->ti = NULL; | |
1670 | } | |
1671 | ||
1672 | /*---------------------------------------------------------------- | |
1673 | * Pool creation | |
1674 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1675 | /* Initialize pool features. */ |
1676 | static void pool_features_init(struct pool_features *pf) | |
1677 | { | |
1678 | pf->zero_new_blocks = 1; | |
1679 | pf->discard_enabled = 1; | |
1680 | pf->discard_passdown = 1; | |
1681 | } | |
1682 | ||
991d9fa0 JT |
1683 | static void __pool_destroy(struct pool *pool) |
1684 | { | |
1685 | __pool_table_remove(pool); | |
1686 | ||
1687 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1688 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1689 | ||
1690 | prison_destroy(pool->prison); | |
1691 | dm_kcopyd_client_destroy(pool->copier); | |
1692 | ||
1693 | if (pool->wq) | |
1694 | destroy_workqueue(pool->wq); | |
1695 | ||
1696 | if (pool->next_mapping) | |
1697 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1698 | mempool_destroy(pool->mapping_pool); | |
1699 | mempool_destroy(pool->endio_hook_pool); | |
1700 | kfree(pool); | |
1701 | } | |
1702 | ||
a24c2569 MS |
1703 | static struct kmem_cache *_new_mapping_cache; |
1704 | static struct kmem_cache *_endio_hook_cache; | |
1705 | ||
991d9fa0 JT |
1706 | static struct pool *pool_create(struct mapped_device *pool_md, |
1707 | struct block_device *metadata_dev, | |
1708 | unsigned long block_size, char **error) | |
1709 | { | |
1710 | int r; | |
1711 | void *err_p; | |
1712 | struct pool *pool; | |
1713 | struct dm_pool_metadata *pmd; | |
1714 | ||
66b1edc0 | 1715 | pmd = dm_pool_metadata_open(metadata_dev, block_size, true); |
991d9fa0 JT |
1716 | if (IS_ERR(pmd)) { |
1717 | *error = "Error creating metadata object"; | |
1718 | return (struct pool *)pmd; | |
1719 | } | |
1720 | ||
1721 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1722 | if (!pool) { | |
1723 | *error = "Error allocating memory for pool"; | |
1724 | err_p = ERR_PTR(-ENOMEM); | |
1725 | goto bad_pool; | |
1726 | } | |
1727 | ||
1728 | pool->pmd = pmd; | |
1729 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1730 | if (block_size & (block_size - 1)) |
1731 | pool->sectors_per_block_shift = -1; | |
1732 | else | |
1733 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1734 | pool->low_water_blocks = 0; |
67e2e2b2 | 1735 | pool_features_init(&pool->pf); |
991d9fa0 JT |
1736 | pool->prison = prison_create(PRISON_CELLS); |
1737 | if (!pool->prison) { | |
1738 | *error = "Error creating pool's bio prison"; | |
1739 | err_p = ERR_PTR(-ENOMEM); | |
1740 | goto bad_prison; | |
1741 | } | |
1742 | ||
1743 | pool->copier = dm_kcopyd_client_create(); | |
1744 | if (IS_ERR(pool->copier)) { | |
1745 | r = PTR_ERR(pool->copier); | |
1746 | *error = "Error creating pool's kcopyd client"; | |
1747 | err_p = ERR_PTR(r); | |
1748 | goto bad_kcopyd_client; | |
1749 | } | |
1750 | ||
1751 | /* | |
1752 | * Create singlethreaded workqueue that will service all devices | |
1753 | * that use this metadata. | |
1754 | */ | |
1755 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1756 | if (!pool->wq) { | |
1757 | *error = "Error creating pool's workqueue"; | |
1758 | err_p = ERR_PTR(-ENOMEM); | |
1759 | goto bad_wq; | |
1760 | } | |
1761 | ||
1762 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1763 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1764 | spin_lock_init(&pool->lock); |
1765 | bio_list_init(&pool->deferred_bios); | |
1766 | bio_list_init(&pool->deferred_flush_bios); | |
1767 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1768 | INIT_LIST_HEAD(&pool->prepared_discards); |
991d9fa0 JT |
1769 | pool->low_water_triggered = 0; |
1770 | pool->no_free_space = 0; | |
1771 | bio_list_init(&pool->retry_on_resume_list); | |
eb2aa48d | 1772 | ds_init(&pool->shared_read_ds); |
104655fd | 1773 | ds_init(&pool->all_io_ds); |
991d9fa0 JT |
1774 | |
1775 | pool->next_mapping = NULL; | |
a24c2569 MS |
1776 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1777 | _new_mapping_cache); | |
991d9fa0 JT |
1778 | if (!pool->mapping_pool) { |
1779 | *error = "Error creating pool's mapping mempool"; | |
1780 | err_p = ERR_PTR(-ENOMEM); | |
1781 | goto bad_mapping_pool; | |
1782 | } | |
1783 | ||
a24c2569 MS |
1784 | pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE, |
1785 | _endio_hook_cache); | |
991d9fa0 JT |
1786 | if (!pool->endio_hook_pool) { |
1787 | *error = "Error creating pool's endio_hook mempool"; | |
1788 | err_p = ERR_PTR(-ENOMEM); | |
1789 | goto bad_endio_hook_pool; | |
1790 | } | |
1791 | pool->ref_count = 1; | |
905e51b3 | 1792 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1793 | pool->pool_md = pool_md; |
1794 | pool->md_dev = metadata_dev; | |
1795 | __pool_table_insert(pool); | |
1796 | ||
1797 | return pool; | |
1798 | ||
1799 | bad_endio_hook_pool: | |
1800 | mempool_destroy(pool->mapping_pool); | |
1801 | bad_mapping_pool: | |
1802 | destroy_workqueue(pool->wq); | |
1803 | bad_wq: | |
1804 | dm_kcopyd_client_destroy(pool->copier); | |
1805 | bad_kcopyd_client: | |
1806 | prison_destroy(pool->prison); | |
1807 | bad_prison: | |
1808 | kfree(pool); | |
1809 | bad_pool: | |
1810 | if (dm_pool_metadata_close(pmd)) | |
1811 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1812 | ||
1813 | return err_p; | |
1814 | } | |
1815 | ||
1816 | static void __pool_inc(struct pool *pool) | |
1817 | { | |
1818 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1819 | pool->ref_count++; | |
1820 | } | |
1821 | ||
1822 | static void __pool_dec(struct pool *pool) | |
1823 | { | |
1824 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1825 | BUG_ON(!pool->ref_count); | |
1826 | if (!--pool->ref_count) | |
1827 | __pool_destroy(pool); | |
1828 | } | |
1829 | ||
1830 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
1831 | struct block_device *metadata_dev, | |
67e2e2b2 JT |
1832 | unsigned long block_size, char **error, |
1833 | int *created) | |
991d9fa0 JT |
1834 | { |
1835 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
1836 | ||
1837 | if (pool) { | |
f09996c9 MS |
1838 | if (pool->pool_md != pool_md) { |
1839 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 1840 | return ERR_PTR(-EBUSY); |
f09996c9 | 1841 | } |
991d9fa0 JT |
1842 | __pool_inc(pool); |
1843 | ||
1844 | } else { | |
1845 | pool = __pool_table_lookup(pool_md); | |
1846 | if (pool) { | |
f09996c9 MS |
1847 | if (pool->md_dev != metadata_dev) { |
1848 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 1849 | return ERR_PTR(-EINVAL); |
f09996c9 | 1850 | } |
991d9fa0 JT |
1851 | __pool_inc(pool); |
1852 | ||
67e2e2b2 | 1853 | } else { |
991d9fa0 | 1854 | pool = pool_create(pool_md, metadata_dev, block_size, error); |
67e2e2b2 JT |
1855 | *created = 1; |
1856 | } | |
991d9fa0 JT |
1857 | } |
1858 | ||
1859 | return pool; | |
1860 | } | |
1861 | ||
1862 | /*---------------------------------------------------------------- | |
1863 | * Pool target methods | |
1864 | *--------------------------------------------------------------*/ | |
1865 | static void pool_dtr(struct dm_target *ti) | |
1866 | { | |
1867 | struct pool_c *pt = ti->private; | |
1868 | ||
1869 | mutex_lock(&dm_thin_pool_table.mutex); | |
1870 | ||
1871 | unbind_control_target(pt->pool, ti); | |
1872 | __pool_dec(pt->pool); | |
1873 | dm_put_device(ti, pt->metadata_dev); | |
1874 | dm_put_device(ti, pt->data_dev); | |
1875 | kfree(pt); | |
1876 | ||
1877 | mutex_unlock(&dm_thin_pool_table.mutex); | |
1878 | } | |
1879 | ||
991d9fa0 JT |
1880 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
1881 | struct dm_target *ti) | |
1882 | { | |
1883 | int r; | |
1884 | unsigned argc; | |
1885 | const char *arg_name; | |
1886 | ||
1887 | static struct dm_arg _args[] = { | |
67e2e2b2 | 1888 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
1889 | }; |
1890 | ||
1891 | /* | |
1892 | * No feature arguments supplied. | |
1893 | */ | |
1894 | if (!as->argc) | |
1895 | return 0; | |
1896 | ||
1897 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1898 | if (r) | |
1899 | return -EINVAL; | |
1900 | ||
1901 | while (argc && !r) { | |
1902 | arg_name = dm_shift_arg(as); | |
1903 | argc--; | |
1904 | ||
1905 | if (!strcasecmp(arg_name, "skip_block_zeroing")) { | |
1906 | pf->zero_new_blocks = 0; | |
1907 | continue; | |
67e2e2b2 JT |
1908 | } else if (!strcasecmp(arg_name, "ignore_discard")) { |
1909 | pf->discard_enabled = 0; | |
1910 | continue; | |
1911 | } else if (!strcasecmp(arg_name, "no_discard_passdown")) { | |
1912 | pf->discard_passdown = 0; | |
1913 | continue; | |
991d9fa0 JT |
1914 | } |
1915 | ||
1916 | ti->error = "Unrecognised pool feature requested"; | |
1917 | r = -EINVAL; | |
1918 | } | |
1919 | ||
1920 | return r; | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * thin-pool <metadata dev> <data dev> | |
1925 | * <data block size (sectors)> | |
1926 | * <low water mark (blocks)> | |
1927 | * [<#feature args> [<arg>]*] | |
1928 | * | |
1929 | * Optional feature arguments are: | |
1930 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
1931 | * ignore_discard: disable discard |
1932 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
1933 | */ |
1934 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
1935 | { | |
67e2e2b2 | 1936 | int r, pool_created = 0; |
991d9fa0 JT |
1937 | struct pool_c *pt; |
1938 | struct pool *pool; | |
1939 | struct pool_features pf; | |
1940 | struct dm_arg_set as; | |
1941 | struct dm_dev *data_dev; | |
1942 | unsigned long block_size; | |
1943 | dm_block_t low_water_blocks; | |
1944 | struct dm_dev *metadata_dev; | |
1945 | sector_t metadata_dev_size; | |
c4a69ecd | 1946 | char b[BDEVNAME_SIZE]; |
991d9fa0 JT |
1947 | |
1948 | /* | |
1949 | * FIXME Remove validation from scope of lock. | |
1950 | */ | |
1951 | mutex_lock(&dm_thin_pool_table.mutex); | |
1952 | ||
1953 | if (argc < 4) { | |
1954 | ti->error = "Invalid argument count"; | |
1955 | r = -EINVAL; | |
1956 | goto out_unlock; | |
1957 | } | |
1958 | as.argc = argc; | |
1959 | as.argv = argv; | |
1960 | ||
1961 | r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev); | |
1962 | if (r) { | |
1963 | ti->error = "Error opening metadata block device"; | |
1964 | goto out_unlock; | |
1965 | } | |
1966 | ||
1967 | metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
c4a69ecd MS |
1968 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
1969 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
1970 | bdevname(metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
991d9fa0 JT |
1971 | |
1972 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
1973 | if (r) { | |
1974 | ti->error = "Error getting data device"; | |
1975 | goto out_metadata; | |
1976 | } | |
1977 | ||
1978 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
1979 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
1980 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 1981 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
1982 | ti->error = "Invalid block size"; |
1983 | r = -EINVAL; | |
1984 | goto out; | |
1985 | } | |
1986 | ||
1987 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
1988 | ti->error = "Invalid low water mark"; | |
1989 | r = -EINVAL; | |
1990 | goto out; | |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * Set default pool features. | |
1995 | */ | |
67e2e2b2 | 1996 | pool_features_init(&pf); |
991d9fa0 JT |
1997 | |
1998 | dm_consume_args(&as, 4); | |
1999 | r = parse_pool_features(&as, &pf, ti); | |
2000 | if (r) | |
2001 | goto out; | |
2002 | ||
2003 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); | |
2004 | if (!pt) { | |
2005 | r = -ENOMEM; | |
2006 | goto out; | |
2007 | } | |
2008 | ||
2009 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
67e2e2b2 | 2010 | block_size, &ti->error, &pool_created); |
991d9fa0 JT |
2011 | if (IS_ERR(pool)) { |
2012 | r = PTR_ERR(pool); | |
2013 | goto out_free_pt; | |
2014 | } | |
2015 | ||
67e2e2b2 JT |
2016 | /* |
2017 | * 'pool_created' reflects whether this is the first table load. | |
2018 | * Top level discard support is not allowed to be changed after | |
2019 | * initial load. This would require a pool reload to trigger thin | |
2020 | * device changes. | |
2021 | */ | |
2022 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2023 | ti->error = "Discard support cannot be disabled once enabled"; | |
2024 | r = -EINVAL; | |
2025 | goto out_flags_changed; | |
2026 | } | |
2027 | ||
55f2b8bd MS |
2028 | /* |
2029 | * The block layer requires discard_granularity to be a power of 2. | |
2030 | */ | |
2031 | if (pf.discard_enabled && !is_power_of_2(block_size)) { | |
2032 | ti->error = "Discard support must be disabled when the block size is not a power of 2"; | |
2033 | r = -EINVAL; | |
2034 | goto out_flags_changed; | |
2035 | } | |
2036 | ||
991d9fa0 JT |
2037 | pt->pool = pool; |
2038 | pt->ti = ti; | |
2039 | pt->metadata_dev = metadata_dev; | |
2040 | pt->data_dev = data_dev; | |
2041 | pt->low_water_blocks = low_water_blocks; | |
67e2e2b2 | 2042 | pt->pf = pf; |
991d9fa0 | 2043 | ti->num_flush_requests = 1; |
67e2e2b2 JT |
2044 | /* |
2045 | * Only need to enable discards if the pool should pass | |
2046 | * them down to the data device. The thin device's discard | |
2047 | * processing will cause mappings to be removed from the btree. | |
2048 | */ | |
2049 | if (pf.discard_enabled && pf.discard_passdown) { | |
2050 | ti->num_discard_requests = 1; | |
2051 | /* | |
2052 | * Setting 'discards_supported' circumvents the normal | |
2053 | * stacking of discard limits (this keeps the pool and | |
2054 | * thin devices' discard limits consistent). | |
2055 | */ | |
0ac55489 | 2056 | ti->discards_supported = true; |
67e2e2b2 | 2057 | } |
991d9fa0 JT |
2058 | ti->private = pt; |
2059 | ||
2060 | pt->callbacks.congested_fn = pool_is_congested; | |
2061 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2062 | ||
2063 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2064 | ||
2065 | return 0; | |
2066 | ||
67e2e2b2 JT |
2067 | out_flags_changed: |
2068 | __pool_dec(pool); | |
991d9fa0 JT |
2069 | out_free_pt: |
2070 | kfree(pt); | |
2071 | out: | |
2072 | dm_put_device(ti, data_dev); | |
2073 | out_metadata: | |
2074 | dm_put_device(ti, metadata_dev); | |
2075 | out_unlock: | |
2076 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2077 | ||
2078 | return r; | |
2079 | } | |
2080 | ||
2081 | static int pool_map(struct dm_target *ti, struct bio *bio, | |
2082 | union map_info *map_context) | |
2083 | { | |
2084 | int r; | |
2085 | struct pool_c *pt = ti->private; | |
2086 | struct pool *pool = pt->pool; | |
2087 | unsigned long flags; | |
2088 | ||
2089 | /* | |
2090 | * As this is a singleton target, ti->begin is always zero. | |
2091 | */ | |
2092 | spin_lock_irqsave(&pool->lock, flags); | |
2093 | bio->bi_bdev = pt->data_dev->bdev; | |
2094 | r = DM_MAPIO_REMAPPED; | |
2095 | spin_unlock_irqrestore(&pool->lock, flags); | |
2096 | ||
2097 | return r; | |
2098 | } | |
2099 | ||
2100 | /* | |
2101 | * Retrieves the number of blocks of the data device from | |
2102 | * the superblock and compares it to the actual device size, | |
2103 | * thus resizing the data device in case it has grown. | |
2104 | * | |
2105 | * This both copes with opening preallocated data devices in the ctr | |
2106 | * being followed by a resume | |
2107 | * -and- | |
2108 | * calling the resume method individually after userspace has | |
2109 | * grown the data device in reaction to a table event. | |
2110 | */ | |
2111 | static int pool_preresume(struct dm_target *ti) | |
2112 | { | |
2113 | int r; | |
2114 | struct pool_c *pt = ti->private; | |
2115 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2116 | sector_t data_size = ti->len; |
2117 | dm_block_t sb_data_size; | |
991d9fa0 JT |
2118 | |
2119 | /* | |
2120 | * Take control of the pool object. | |
2121 | */ | |
2122 | r = bind_control_target(pool, ti); | |
2123 | if (r) | |
2124 | return r; | |
2125 | ||
55f2b8bd MS |
2126 | (void) sector_div(data_size, pool->sectors_per_block); |
2127 | ||
991d9fa0 JT |
2128 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2129 | if (r) { | |
2130 | DMERR("failed to retrieve data device size"); | |
2131 | return r; | |
2132 | } | |
2133 | ||
2134 | if (data_size < sb_data_size) { | |
2135 | DMERR("pool target too small, is %llu blocks (expected %llu)", | |
55f2b8bd | 2136 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2137 | return -EINVAL; |
2138 | ||
2139 | } else if (data_size > sb_data_size) { | |
2140 | r = dm_pool_resize_data_dev(pool->pmd, data_size); | |
2141 | if (r) { | |
2142 | DMERR("failed to resize data device"); | |
2143 | return r; | |
2144 | } | |
2145 | ||
2146 | r = dm_pool_commit_metadata(pool->pmd); | |
2147 | if (r) { | |
2148 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", | |
2149 | __func__, r); | |
2150 | return r; | |
2151 | } | |
2152 | } | |
2153 | ||
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | static void pool_resume(struct dm_target *ti) | |
2158 | { | |
2159 | struct pool_c *pt = ti->private; | |
2160 | struct pool *pool = pt->pool; | |
2161 | unsigned long flags; | |
2162 | ||
2163 | spin_lock_irqsave(&pool->lock, flags); | |
2164 | pool->low_water_triggered = 0; | |
2165 | pool->no_free_space = 0; | |
2166 | __requeue_bios(pool); | |
2167 | spin_unlock_irqrestore(&pool->lock, flags); | |
2168 | ||
905e51b3 | 2169 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2170 | } |
2171 | ||
2172 | static void pool_postsuspend(struct dm_target *ti) | |
2173 | { | |
2174 | int r; | |
2175 | struct pool_c *pt = ti->private; | |
2176 | struct pool *pool = pt->pool; | |
2177 | ||
905e51b3 | 2178 | cancel_delayed_work(&pool->waker); |
991d9fa0 JT |
2179 | flush_workqueue(pool->wq); |
2180 | ||
2181 | r = dm_pool_commit_metadata(pool->pmd); | |
2182 | if (r < 0) { | |
2183 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", | |
2184 | __func__, r); | |
2185 | /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/ | |
2186 | } | |
2187 | } | |
2188 | ||
2189 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2190 | { | |
2191 | if (argc != args_required) { | |
2192 | DMWARN("Message received with %u arguments instead of %u.", | |
2193 | argc, args_required); | |
2194 | return -EINVAL; | |
2195 | } | |
2196 | ||
2197 | return 0; | |
2198 | } | |
2199 | ||
2200 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2201 | { | |
2202 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2203 | *dev_id <= MAX_DEV_ID) | |
2204 | return 0; | |
2205 | ||
2206 | if (warning) | |
2207 | DMWARN("Message received with invalid device id: %s", arg); | |
2208 | ||
2209 | return -EINVAL; | |
2210 | } | |
2211 | ||
2212 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2213 | { | |
2214 | dm_thin_id dev_id; | |
2215 | int r; | |
2216 | ||
2217 | r = check_arg_count(argc, 2); | |
2218 | if (r) | |
2219 | return r; | |
2220 | ||
2221 | r = read_dev_id(argv[1], &dev_id, 1); | |
2222 | if (r) | |
2223 | return r; | |
2224 | ||
2225 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2226 | if (r) { | |
2227 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2228 | argv[1]); | |
2229 | return r; | |
2230 | } | |
2231 | ||
2232 | return 0; | |
2233 | } | |
2234 | ||
2235 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2236 | { | |
2237 | dm_thin_id dev_id; | |
2238 | dm_thin_id origin_dev_id; | |
2239 | int r; | |
2240 | ||
2241 | r = check_arg_count(argc, 3); | |
2242 | if (r) | |
2243 | return r; | |
2244 | ||
2245 | r = read_dev_id(argv[1], &dev_id, 1); | |
2246 | if (r) | |
2247 | return r; | |
2248 | ||
2249 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2250 | if (r) | |
2251 | return r; | |
2252 | ||
2253 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2254 | if (r) { | |
2255 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2256 | argv[1], argv[2]); | |
2257 | return r; | |
2258 | } | |
2259 | ||
2260 | return 0; | |
2261 | } | |
2262 | ||
2263 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2264 | { | |
2265 | dm_thin_id dev_id; | |
2266 | int r; | |
2267 | ||
2268 | r = check_arg_count(argc, 2); | |
2269 | if (r) | |
2270 | return r; | |
2271 | ||
2272 | r = read_dev_id(argv[1], &dev_id, 1); | |
2273 | if (r) | |
2274 | return r; | |
2275 | ||
2276 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2277 | if (r) | |
2278 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2279 | ||
2280 | return r; | |
2281 | } | |
2282 | ||
2283 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2284 | { | |
2285 | dm_thin_id old_id, new_id; | |
2286 | int r; | |
2287 | ||
2288 | r = check_arg_count(argc, 3); | |
2289 | if (r) | |
2290 | return r; | |
2291 | ||
2292 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2293 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2294 | return -EINVAL; | |
2295 | } | |
2296 | ||
2297 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2298 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2299 | return -EINVAL; | |
2300 | } | |
2301 | ||
2302 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2303 | if (r) { | |
2304 | DMWARN("Failed to change transaction id from %s to %s.", | |
2305 | argv[1], argv[2]); | |
2306 | return r; | |
2307 | } | |
2308 | ||
2309 | return 0; | |
2310 | } | |
2311 | ||
cc8394d8 JT |
2312 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2313 | { | |
2314 | int r; | |
2315 | ||
2316 | r = check_arg_count(argc, 1); | |
2317 | if (r) | |
2318 | return r; | |
2319 | ||
0d200aef JT |
2320 | r = dm_pool_commit_metadata(pool->pmd); |
2321 | if (r) { | |
2322 | DMERR("%s: dm_pool_commit_metadata() failed, error = %d", | |
2323 | __func__, r); | |
2324 | return r; | |
2325 | } | |
2326 | ||
cc8394d8 JT |
2327 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2328 | if (r) | |
2329 | DMWARN("reserve_metadata_snap message failed."); | |
2330 | ||
2331 | return r; | |
2332 | } | |
2333 | ||
2334 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2335 | { | |
2336 | int r; | |
2337 | ||
2338 | r = check_arg_count(argc, 1); | |
2339 | if (r) | |
2340 | return r; | |
2341 | ||
2342 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2343 | if (r) | |
2344 | DMWARN("release_metadata_snap message failed."); | |
2345 | ||
2346 | return r; | |
2347 | } | |
2348 | ||
991d9fa0 JT |
2349 | /* |
2350 | * Messages supported: | |
2351 | * create_thin <dev_id> | |
2352 | * create_snap <dev_id> <origin_id> | |
2353 | * delete <dev_id> | |
2354 | * trim <dev_id> <new_size_in_sectors> | |
2355 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2356 | * reserve_metadata_snap |
2357 | * release_metadata_snap | |
991d9fa0 JT |
2358 | */ |
2359 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2360 | { | |
2361 | int r = -EINVAL; | |
2362 | struct pool_c *pt = ti->private; | |
2363 | struct pool *pool = pt->pool; | |
2364 | ||
2365 | if (!strcasecmp(argv[0], "create_thin")) | |
2366 | r = process_create_thin_mesg(argc, argv, pool); | |
2367 | ||
2368 | else if (!strcasecmp(argv[0], "create_snap")) | |
2369 | r = process_create_snap_mesg(argc, argv, pool); | |
2370 | ||
2371 | else if (!strcasecmp(argv[0], "delete")) | |
2372 | r = process_delete_mesg(argc, argv, pool); | |
2373 | ||
2374 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2375 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2376 | ||
cc8394d8 JT |
2377 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2378 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2379 | ||
2380 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2381 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2382 | ||
991d9fa0 JT |
2383 | else |
2384 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2385 | ||
2386 | if (!r) { | |
2387 | r = dm_pool_commit_metadata(pool->pmd); | |
2388 | if (r) | |
2389 | DMERR("%s message: dm_pool_commit_metadata() failed, error = %d", | |
2390 | argv[0], r); | |
2391 | } | |
2392 | ||
2393 | return r; | |
2394 | } | |
2395 | ||
2396 | /* | |
2397 | * Status line is: | |
2398 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2399 | * <used data sectors>/<total data sectors> <held metadata root> | |
2400 | */ | |
2401 | static int pool_status(struct dm_target *ti, status_type_t type, | |
2402 | char *result, unsigned maxlen) | |
2403 | { | |
67e2e2b2 | 2404 | int r, count; |
991d9fa0 JT |
2405 | unsigned sz = 0; |
2406 | uint64_t transaction_id; | |
2407 | dm_block_t nr_free_blocks_data; | |
2408 | dm_block_t nr_free_blocks_metadata; | |
2409 | dm_block_t nr_blocks_data; | |
2410 | dm_block_t nr_blocks_metadata; | |
2411 | dm_block_t held_root; | |
2412 | char buf[BDEVNAME_SIZE]; | |
2413 | char buf2[BDEVNAME_SIZE]; | |
2414 | struct pool_c *pt = ti->private; | |
2415 | struct pool *pool = pt->pool; | |
2416 | ||
2417 | switch (type) { | |
2418 | case STATUSTYPE_INFO: | |
2419 | r = dm_pool_get_metadata_transaction_id(pool->pmd, | |
2420 | &transaction_id); | |
2421 | if (r) | |
2422 | return r; | |
2423 | ||
2424 | r = dm_pool_get_free_metadata_block_count(pool->pmd, | |
2425 | &nr_free_blocks_metadata); | |
2426 | if (r) | |
2427 | return r; | |
2428 | ||
2429 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
2430 | if (r) | |
2431 | return r; | |
2432 | ||
2433 | r = dm_pool_get_free_block_count(pool->pmd, | |
2434 | &nr_free_blocks_data); | |
2435 | if (r) | |
2436 | return r; | |
2437 | ||
2438 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
2439 | if (r) | |
2440 | return r; | |
2441 | ||
cc8394d8 | 2442 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
991d9fa0 JT |
2443 | if (r) |
2444 | return r; | |
2445 | ||
2446 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2447 | (unsigned long long)transaction_id, | |
2448 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2449 | (unsigned long long)nr_blocks_metadata, | |
2450 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2451 | (unsigned long long)nr_blocks_data); | |
2452 | ||
2453 | if (held_root) | |
2454 | DMEMIT("%llu", held_root); | |
2455 | else | |
2456 | DMEMIT("-"); | |
2457 | ||
2458 | break; | |
2459 | ||
2460 | case STATUSTYPE_TABLE: | |
2461 | DMEMIT("%s %s %lu %llu ", | |
2462 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2463 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2464 | (unsigned long)pool->sectors_per_block, | |
2465 | (unsigned long long)pt->low_water_blocks); | |
2466 | ||
67e2e2b2 | 2467 | count = !pool->pf.zero_new_blocks + !pool->pf.discard_enabled + |
f402693d | 2468 | !pt->pf.discard_passdown; |
67e2e2b2 | 2469 | DMEMIT("%u ", count); |
991d9fa0 | 2470 | |
67e2e2b2 | 2471 | if (!pool->pf.zero_new_blocks) |
991d9fa0 | 2472 | DMEMIT("skip_block_zeroing "); |
67e2e2b2 JT |
2473 | |
2474 | if (!pool->pf.discard_enabled) | |
2475 | DMEMIT("ignore_discard "); | |
2476 | ||
f402693d | 2477 | if (!pt->pf.discard_passdown) |
67e2e2b2 JT |
2478 | DMEMIT("no_discard_passdown "); |
2479 | ||
991d9fa0 JT |
2480 | break; |
2481 | } | |
2482 | ||
2483 | return 0; | |
2484 | } | |
2485 | ||
2486 | static int pool_iterate_devices(struct dm_target *ti, | |
2487 | iterate_devices_callout_fn fn, void *data) | |
2488 | { | |
2489 | struct pool_c *pt = ti->private; | |
2490 | ||
2491 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2492 | } | |
2493 | ||
2494 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2495 | struct bio_vec *biovec, int max_size) | |
2496 | { | |
2497 | struct pool_c *pt = ti->private; | |
2498 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2499 | ||
2500 | if (!q->merge_bvec_fn) | |
2501 | return max_size; | |
2502 | ||
2503 | bvm->bi_bdev = pt->data_dev->bdev; | |
2504 | ||
2505 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2506 | } | |
2507 | ||
104655fd JT |
2508 | static void set_discard_limits(struct pool *pool, struct queue_limits *limits) |
2509 | { | |
67e2e2b2 JT |
2510 | /* |
2511 | * FIXME: these limits may be incompatible with the pool's data device | |
2512 | */ | |
104655fd JT |
2513 | limits->max_discard_sectors = pool->sectors_per_block; |
2514 | ||
2515 | /* | |
2516 | * This is just a hint, and not enforced. We have to cope with | |
49296309 MP |
2517 | * bios that cover a block partially. A discard that spans a block |
2518 | * boundary is not sent to this target. | |
104655fd JT |
2519 | */ |
2520 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
67e2e2b2 | 2521 | limits->discard_zeroes_data = pool->pf.zero_new_blocks; |
104655fd JT |
2522 | } |
2523 | ||
991d9fa0 JT |
2524 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2525 | { | |
2526 | struct pool_c *pt = ti->private; | |
2527 | struct pool *pool = pt->pool; | |
2528 | ||
2529 | blk_limits_io_min(limits, 0); | |
2530 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
67e2e2b2 JT |
2531 | if (pool->pf.discard_enabled) |
2532 | set_discard_limits(pool, limits); | |
991d9fa0 JT |
2533 | } |
2534 | ||
2535 | static struct target_type pool_target = { | |
2536 | .name = "thin-pool", | |
2537 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2538 | DM_TARGET_IMMUTABLE, | |
cc8394d8 | 2539 | .version = {1, 2, 0}, |
991d9fa0 JT |
2540 | .module = THIS_MODULE, |
2541 | .ctr = pool_ctr, | |
2542 | .dtr = pool_dtr, | |
2543 | .map = pool_map, | |
2544 | .postsuspend = pool_postsuspend, | |
2545 | .preresume = pool_preresume, | |
2546 | .resume = pool_resume, | |
2547 | .message = pool_message, | |
2548 | .status = pool_status, | |
2549 | .merge = pool_merge, | |
2550 | .iterate_devices = pool_iterate_devices, | |
2551 | .io_hints = pool_io_hints, | |
2552 | }; | |
2553 | ||
2554 | /*---------------------------------------------------------------- | |
2555 | * Thin target methods | |
2556 | *--------------------------------------------------------------*/ | |
2557 | static void thin_dtr(struct dm_target *ti) | |
2558 | { | |
2559 | struct thin_c *tc = ti->private; | |
2560 | ||
2561 | mutex_lock(&dm_thin_pool_table.mutex); | |
2562 | ||
2563 | __pool_dec(tc->pool); | |
2564 | dm_pool_close_thin_device(tc->td); | |
2565 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2566 | if (tc->origin_dev) |
2567 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2568 | kfree(tc); |
2569 | ||
2570 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * Thin target parameters: | |
2575 | * | |
2dd9c257 | 2576 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2577 | * |
2578 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2579 | * dev_id: the internal device identifier | |
2dd9c257 | 2580 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2581 | * |
2582 | * If the pool device has discards disabled, they get disabled for the thin | |
2583 | * device as well. | |
991d9fa0 JT |
2584 | */ |
2585 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2586 | { | |
2587 | int r; | |
2588 | struct thin_c *tc; | |
2dd9c257 | 2589 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2590 | struct mapped_device *pool_md; |
2591 | ||
2592 | mutex_lock(&dm_thin_pool_table.mutex); | |
2593 | ||
2dd9c257 | 2594 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2595 | ti->error = "Invalid argument count"; |
2596 | r = -EINVAL; | |
2597 | goto out_unlock; | |
2598 | } | |
2599 | ||
2600 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2601 | if (!tc) { | |
2602 | ti->error = "Out of memory"; | |
2603 | r = -ENOMEM; | |
2604 | goto out_unlock; | |
2605 | } | |
2606 | ||
2dd9c257 JT |
2607 | if (argc == 3) { |
2608 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2609 | if (r) { | |
2610 | ti->error = "Error opening origin device"; | |
2611 | goto bad_origin_dev; | |
2612 | } | |
2613 | tc->origin_dev = origin_dev; | |
2614 | } | |
2615 | ||
991d9fa0 JT |
2616 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2617 | if (r) { | |
2618 | ti->error = "Error opening pool device"; | |
2619 | goto bad_pool_dev; | |
2620 | } | |
2621 | tc->pool_dev = pool_dev; | |
2622 | ||
2623 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2624 | ti->error = "Invalid device id"; | |
2625 | r = -EINVAL; | |
2626 | goto bad_common; | |
2627 | } | |
2628 | ||
2629 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2630 | if (!pool_md) { | |
2631 | ti->error = "Couldn't get pool mapped device"; | |
2632 | r = -EINVAL; | |
2633 | goto bad_common; | |
2634 | } | |
2635 | ||
2636 | tc->pool = __pool_table_lookup(pool_md); | |
2637 | if (!tc->pool) { | |
2638 | ti->error = "Couldn't find pool object"; | |
2639 | r = -EINVAL; | |
2640 | goto bad_pool_lookup; | |
2641 | } | |
2642 | __pool_inc(tc->pool); | |
2643 | ||
2644 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); | |
2645 | if (r) { | |
2646 | ti->error = "Couldn't open thin internal device"; | |
2647 | goto bad_thin_open; | |
2648 | } | |
2649 | ||
542f9038 MS |
2650 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2651 | if (r) | |
2652 | goto bad_thin_open; | |
2653 | ||
991d9fa0 | 2654 | ti->num_flush_requests = 1; |
16ad3d10 | 2655 | ti->flush_supported = true; |
67e2e2b2 JT |
2656 | |
2657 | /* In case the pool supports discards, pass them on. */ | |
2658 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 2659 | ti->discards_supported = true; |
67e2e2b2 | 2660 | ti->num_discard_requests = 1; |
0ac55489 | 2661 | ti->discard_zeroes_data_unsupported = true; |
49296309 | 2662 | /* Discard requests must be split on a block boundary */ |
0ac55489 | 2663 | ti->split_discard_requests = true; |
67e2e2b2 | 2664 | } |
991d9fa0 JT |
2665 | |
2666 | dm_put(pool_md); | |
2667 | ||
2668 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2669 | ||
2670 | return 0; | |
2671 | ||
2672 | bad_thin_open: | |
2673 | __pool_dec(tc->pool); | |
2674 | bad_pool_lookup: | |
2675 | dm_put(pool_md); | |
2676 | bad_common: | |
2677 | dm_put_device(ti, tc->pool_dev); | |
2678 | bad_pool_dev: | |
2dd9c257 JT |
2679 | if (tc->origin_dev) |
2680 | dm_put_device(ti, tc->origin_dev); | |
2681 | bad_origin_dev: | |
991d9fa0 JT |
2682 | kfree(tc); |
2683 | out_unlock: | |
2684 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2685 | ||
2686 | return r; | |
2687 | } | |
2688 | ||
2689 | static int thin_map(struct dm_target *ti, struct bio *bio, | |
2690 | union map_info *map_context) | |
2691 | { | |
6efd6e83 | 2692 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 JT |
2693 | |
2694 | return thin_bio_map(ti, bio, map_context); | |
2695 | } | |
2696 | ||
eb2aa48d JT |
2697 | static int thin_endio(struct dm_target *ti, |
2698 | struct bio *bio, int err, | |
2699 | union map_info *map_context) | |
2700 | { | |
2701 | unsigned long flags; | |
a24c2569 | 2702 | struct dm_thin_endio_hook *h = map_context->ptr; |
eb2aa48d | 2703 | struct list_head work; |
a24c2569 | 2704 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2705 | struct pool *pool = h->tc->pool; |
2706 | ||
2707 | if (h->shared_read_entry) { | |
2708 | INIT_LIST_HEAD(&work); | |
2709 | ds_dec(h->shared_read_entry, &work); | |
2710 | ||
2711 | spin_lock_irqsave(&pool->lock, flags); | |
2712 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2713 | list_del(&m->list); | |
2714 | m->quiesced = 1; | |
2715 | __maybe_add_mapping(m); | |
2716 | } | |
2717 | spin_unlock_irqrestore(&pool->lock, flags); | |
2718 | } | |
2719 | ||
104655fd JT |
2720 | if (h->all_io_entry) { |
2721 | INIT_LIST_HEAD(&work); | |
2722 | ds_dec(h->all_io_entry, &work); | |
c3a0ce2e | 2723 | spin_lock_irqsave(&pool->lock, flags); |
104655fd JT |
2724 | list_for_each_entry_safe(m, tmp, &work, list) |
2725 | list_add(&m->list, &pool->prepared_discards); | |
c3a0ce2e | 2726 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
2727 | } |
2728 | ||
eb2aa48d JT |
2729 | mempool_free(h, pool->endio_hook_pool); |
2730 | ||
2731 | return 0; | |
2732 | } | |
2733 | ||
991d9fa0 JT |
2734 | static void thin_postsuspend(struct dm_target *ti) |
2735 | { | |
2736 | if (dm_noflush_suspending(ti)) | |
2737 | requeue_io((struct thin_c *)ti->private); | |
2738 | } | |
2739 | ||
2740 | /* | |
2741 | * <nr mapped sectors> <highest mapped sector> | |
2742 | */ | |
2743 | static int thin_status(struct dm_target *ti, status_type_t type, | |
2744 | char *result, unsigned maxlen) | |
2745 | { | |
2746 | int r; | |
2747 | ssize_t sz = 0; | |
2748 | dm_block_t mapped, highest; | |
2749 | char buf[BDEVNAME_SIZE]; | |
2750 | struct thin_c *tc = ti->private; | |
2751 | ||
2752 | if (!tc->td) | |
2753 | DMEMIT("-"); | |
2754 | else { | |
2755 | switch (type) { | |
2756 | case STATUSTYPE_INFO: | |
2757 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
2758 | if (r) | |
2759 | return r; | |
2760 | ||
2761 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
2762 | if (r < 0) | |
2763 | return r; | |
2764 | ||
2765 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
2766 | if (r) | |
2767 | DMEMIT("%llu", ((highest + 1) * | |
2768 | tc->pool->sectors_per_block) - 1); | |
2769 | else | |
2770 | DMEMIT("-"); | |
2771 | break; | |
2772 | ||
2773 | case STATUSTYPE_TABLE: | |
2774 | DMEMIT("%s %lu", | |
2775 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
2776 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
2777 | if (tc->origin_dev) |
2778 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
2779 | break; |
2780 | } | |
2781 | } | |
2782 | ||
2783 | return 0; | |
2784 | } | |
2785 | ||
2786 | static int thin_iterate_devices(struct dm_target *ti, | |
2787 | iterate_devices_callout_fn fn, void *data) | |
2788 | { | |
55f2b8bd | 2789 | sector_t blocks; |
991d9fa0 | 2790 | struct thin_c *tc = ti->private; |
55f2b8bd | 2791 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2792 | |
2793 | /* | |
2794 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
2795 | * we follow a more convoluted path through to the pool's target. | |
2796 | */ | |
55f2b8bd | 2797 | if (!pool->ti) |
991d9fa0 JT |
2798 | return 0; /* nothing is bound */ |
2799 | ||
55f2b8bd MS |
2800 | blocks = pool->ti->len; |
2801 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 2802 | if (blocks) |
55f2b8bd | 2803 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
2804 | |
2805 | return 0; | |
2806 | } | |
2807 | ||
2808 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
2809 | { | |
2810 | struct thin_c *tc = ti->private; | |
104655fd | 2811 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2812 | |
2813 | blk_limits_io_min(limits, 0); | |
104655fd JT |
2814 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
2815 | set_discard_limits(pool, limits); | |
991d9fa0 JT |
2816 | } |
2817 | ||
2818 | static struct target_type thin_target = { | |
2819 | .name = "thin", | |
55f2b8bd | 2820 | .version = {1, 2, 0}, |
991d9fa0 JT |
2821 | .module = THIS_MODULE, |
2822 | .ctr = thin_ctr, | |
2823 | .dtr = thin_dtr, | |
2824 | .map = thin_map, | |
eb2aa48d | 2825 | .end_io = thin_endio, |
991d9fa0 JT |
2826 | .postsuspend = thin_postsuspend, |
2827 | .status = thin_status, | |
2828 | .iterate_devices = thin_iterate_devices, | |
2829 | .io_hints = thin_io_hints, | |
2830 | }; | |
2831 | ||
2832 | /*----------------------------------------------------------------*/ | |
2833 | ||
2834 | static int __init dm_thin_init(void) | |
2835 | { | |
2836 | int r; | |
2837 | ||
2838 | pool_table_init(); | |
2839 | ||
2840 | r = dm_register_target(&thin_target); | |
2841 | if (r) | |
2842 | return r; | |
2843 | ||
2844 | r = dm_register_target(&pool_target); | |
2845 | if (r) | |
a24c2569 MS |
2846 | goto bad_pool_target; |
2847 | ||
2848 | r = -ENOMEM; | |
2849 | ||
2850 | _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0); | |
2851 | if (!_cell_cache) | |
2852 | goto bad_cell_cache; | |
2853 | ||
2854 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); | |
2855 | if (!_new_mapping_cache) | |
2856 | goto bad_new_mapping_cache; | |
2857 | ||
2858 | _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0); | |
2859 | if (!_endio_hook_cache) | |
2860 | goto bad_endio_hook_cache; | |
2861 | ||
2862 | return 0; | |
2863 | ||
2864 | bad_endio_hook_cache: | |
2865 | kmem_cache_destroy(_new_mapping_cache); | |
2866 | bad_new_mapping_cache: | |
2867 | kmem_cache_destroy(_cell_cache); | |
2868 | bad_cell_cache: | |
2869 | dm_unregister_target(&pool_target); | |
2870 | bad_pool_target: | |
2871 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
2872 | |
2873 | return r; | |
2874 | } | |
2875 | ||
2876 | static void dm_thin_exit(void) | |
2877 | { | |
2878 | dm_unregister_target(&thin_target); | |
2879 | dm_unregister_target(&pool_target); | |
a24c2569 MS |
2880 | |
2881 | kmem_cache_destroy(_cell_cache); | |
2882 | kmem_cache_destroy(_new_mapping_cache); | |
2883 | kmem_cache_destroy(_endio_hook_cache); | |
991d9fa0 JT |
2884 | } |
2885 | ||
2886 | module_init(dm_thin_init); | |
2887 | module_exit(dm_thin_exit); | |
2888 | ||
7cab8bf1 | 2889 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
2890 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
2891 | MODULE_LICENSE("GPL"); |