]>
Commit | Line | Data |
---|---|---|
991d9fa0 | 1 | /* |
e49e5829 | 2 | * Copyright (C) 2011-2012 Red Hat UK. |
991d9fa0 JT |
3 | * |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-thin-metadata.h" | |
1f4e0ff0 | 8 | #include "dm.h" |
991d9fa0 JT |
9 | |
10 | #include <linux/device-mapper.h> | |
11 | #include <linux/dm-io.h> | |
12 | #include <linux/dm-kcopyd.h> | |
13 | #include <linux/list.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/slab.h> | |
17 | ||
18 | #define DM_MSG_PREFIX "thin" | |
19 | ||
20 | /* | |
21 | * Tunable constants | |
22 | */ | |
7768ed33 | 23 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
24 | #define DEFERRED_SET_SIZE 64 |
25 | #define MAPPING_POOL_SIZE 1024 | |
26 | #define PRISON_CELLS 1024 | |
905e51b3 | 27 | #define COMMIT_PERIOD HZ |
991d9fa0 JT |
28 | |
29 | /* | |
30 | * The block size of the device holding pool data must be | |
31 | * between 64KB and 1GB. | |
32 | */ | |
33 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
34 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
35 | ||
991d9fa0 JT |
36 | /* |
37 | * Device id is restricted to 24 bits. | |
38 | */ | |
39 | #define MAX_DEV_ID ((1 << 24) - 1) | |
40 | ||
41 | /* | |
42 | * How do we handle breaking sharing of data blocks? | |
43 | * ================================================= | |
44 | * | |
45 | * We use a standard copy-on-write btree to store the mappings for the | |
46 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
47 | * the data). When you take an internal snapshot you clone the root node | |
48 | * of the origin btree. After this there is no concept of an origin or a | |
49 | * snapshot. They are just two device trees that happen to point to the | |
50 | * same data blocks. | |
51 | * | |
52 | * When we get a write in we decide if it's to a shared data block using | |
53 | * some timestamp magic. If it is, we have to break sharing. | |
54 | * | |
55 | * Let's say we write to a shared block in what was the origin. The | |
56 | * steps are: | |
57 | * | |
58 | * i) plug io further to this physical block. (see bio_prison code). | |
59 | * | |
60 | * ii) quiesce any read io to that shared data block. Obviously | |
61 | * including all devices that share this block. (see deferred_set code) | |
62 | * | |
63 | * iii) copy the data block to a newly allocate block. This step can be | |
64 | * missed out if the io covers the block. (schedule_copy). | |
65 | * | |
66 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 67 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
68 | * sharing of btree nodes between the two devices. Breaking sharing only |
69 | * effects the btree of that specific device. Btrees for the other | |
70 | * devices that share the block never change. The btree for the origin | |
71 | * device as it was after the last commit is untouched, ie. we're using | |
72 | * persistent data structures in the functional programming sense. | |
73 | * | |
74 | * v) unplug io to this physical block, including the io that triggered | |
75 | * the breaking of sharing. | |
76 | * | |
77 | * Steps (ii) and (iii) occur in parallel. | |
78 | * | |
79 | * The metadata _doesn't_ need to be committed before the io continues. We | |
80 | * get away with this because the io is always written to a _new_ block. | |
81 | * If there's a crash, then: | |
82 | * | |
83 | * - The origin mapping will point to the old origin block (the shared | |
84 | * one). This will contain the data as it was before the io that triggered | |
85 | * the breaking of sharing came in. | |
86 | * | |
87 | * - The snap mapping still points to the old block. As it would after | |
88 | * the commit. | |
89 | * | |
90 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
91 | * will continue to think that data block in the snapshot device is shared | |
92 | * even after the write to the origin has broken sharing. I suspect data | |
93 | * blocks will typically be shared by many different devices, so we're | |
94 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
95 | * devices that reference this data block. At the moment I think the | |
96 | * benefits far, far outweigh the disadvantages. | |
97 | */ | |
98 | ||
99 | /*----------------------------------------------------------------*/ | |
100 | ||
101 | /* | |
102 | * Sometimes we can't deal with a bio straight away. We put them in prison | |
103 | * where they can't cause any mischief. Bios are put in a cell identified | |
104 | * by a key, multiple bios can be in the same cell. When the cell is | |
105 | * subsequently unlocked the bios become available. | |
106 | */ | |
107 | struct bio_prison; | |
108 | ||
109 | struct cell_key { | |
110 | int virtual; | |
111 | dm_thin_id dev; | |
112 | dm_block_t block; | |
113 | }; | |
114 | ||
a24c2569 | 115 | struct dm_bio_prison_cell { |
991d9fa0 JT |
116 | struct hlist_node list; |
117 | struct bio_prison *prison; | |
118 | struct cell_key key; | |
6f94a4c4 | 119 | struct bio *holder; |
991d9fa0 JT |
120 | struct bio_list bios; |
121 | }; | |
122 | ||
123 | struct bio_prison { | |
124 | spinlock_t lock; | |
125 | mempool_t *cell_pool; | |
126 | ||
127 | unsigned nr_buckets; | |
128 | unsigned hash_mask; | |
129 | struct hlist_head *cells; | |
130 | }; | |
131 | ||
132 | static uint32_t calc_nr_buckets(unsigned nr_cells) | |
133 | { | |
134 | uint32_t n = 128; | |
135 | ||
136 | nr_cells /= 4; | |
137 | nr_cells = min(nr_cells, 8192u); | |
138 | ||
139 | while (n < nr_cells) | |
140 | n <<= 1; | |
141 | ||
142 | return n; | |
143 | } | |
144 | ||
a24c2569 MS |
145 | static struct kmem_cache *_cell_cache; |
146 | ||
991d9fa0 JT |
147 | /* |
148 | * @nr_cells should be the number of cells you want in use _concurrently_. | |
149 | * Don't confuse it with the number of distinct keys. | |
150 | */ | |
151 | static struct bio_prison *prison_create(unsigned nr_cells) | |
152 | { | |
153 | unsigned i; | |
154 | uint32_t nr_buckets = calc_nr_buckets(nr_cells); | |
155 | size_t len = sizeof(struct bio_prison) + | |
156 | (sizeof(struct hlist_head) * nr_buckets); | |
157 | struct bio_prison *prison = kmalloc(len, GFP_KERNEL); | |
158 | ||
159 | if (!prison) | |
160 | return NULL; | |
161 | ||
162 | spin_lock_init(&prison->lock); | |
a24c2569 | 163 | prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache); |
991d9fa0 JT |
164 | if (!prison->cell_pool) { |
165 | kfree(prison); | |
166 | return NULL; | |
167 | } | |
168 | ||
169 | prison->nr_buckets = nr_buckets; | |
170 | prison->hash_mask = nr_buckets - 1; | |
171 | prison->cells = (struct hlist_head *) (prison + 1); | |
172 | for (i = 0; i < nr_buckets; i++) | |
173 | INIT_HLIST_HEAD(prison->cells + i); | |
174 | ||
175 | return prison; | |
176 | } | |
177 | ||
178 | static void prison_destroy(struct bio_prison *prison) | |
179 | { | |
180 | mempool_destroy(prison->cell_pool); | |
181 | kfree(prison); | |
182 | } | |
183 | ||
184 | static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key) | |
185 | { | |
186 | const unsigned long BIG_PRIME = 4294967291UL; | |
187 | uint64_t hash = key->block * BIG_PRIME; | |
188 | ||
189 | return (uint32_t) (hash & prison->hash_mask); | |
190 | } | |
191 | ||
192 | static int keys_equal(struct cell_key *lhs, struct cell_key *rhs) | |
193 | { | |
194 | return (lhs->virtual == rhs->virtual) && | |
195 | (lhs->dev == rhs->dev) && | |
196 | (lhs->block == rhs->block); | |
197 | } | |
198 | ||
a24c2569 MS |
199 | static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket, |
200 | struct cell_key *key) | |
991d9fa0 | 201 | { |
a24c2569 | 202 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
203 | struct hlist_node *tmp; |
204 | ||
205 | hlist_for_each_entry(cell, tmp, bucket, list) | |
206 | if (keys_equal(&cell->key, key)) | |
207 | return cell; | |
208 | ||
209 | return NULL; | |
210 | } | |
211 | ||
212 | /* | |
213 | * This may block if a new cell needs allocating. You must ensure that | |
214 | * cells will be unlocked even if the calling thread is blocked. | |
215 | * | |
6f94a4c4 | 216 | * Returns 1 if the cell was already held, 0 if @inmate is the new holder. |
991d9fa0 JT |
217 | */ |
218 | static int bio_detain(struct bio_prison *prison, struct cell_key *key, | |
a24c2569 | 219 | struct bio *inmate, struct dm_bio_prison_cell **ref) |
991d9fa0 | 220 | { |
6f94a4c4 | 221 | int r = 1; |
991d9fa0 JT |
222 | unsigned long flags; |
223 | uint32_t hash = hash_key(prison, key); | |
a24c2569 | 224 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
225 | |
226 | BUG_ON(hash > prison->nr_buckets); | |
227 | ||
228 | spin_lock_irqsave(&prison->lock, flags); | |
991d9fa0 | 229 | |
6f94a4c4 JT |
230 | cell = __search_bucket(prison->cells + hash, key); |
231 | if (cell) { | |
232 | bio_list_add(&cell->bios, inmate); | |
233 | goto out; | |
991d9fa0 JT |
234 | } |
235 | ||
6f94a4c4 JT |
236 | /* |
237 | * Allocate a new cell | |
238 | */ | |
991d9fa0 | 239 | spin_unlock_irqrestore(&prison->lock, flags); |
6f94a4c4 JT |
240 | cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO); |
241 | spin_lock_irqsave(&prison->lock, flags); | |
991d9fa0 | 242 | |
6f94a4c4 JT |
243 | /* |
244 | * We've been unlocked, so we have to double check that | |
245 | * nobody else has inserted this cell in the meantime. | |
246 | */ | |
247 | cell = __search_bucket(prison->cells + hash, key); | |
248 | if (cell) { | |
991d9fa0 | 249 | mempool_free(cell2, prison->cell_pool); |
6f94a4c4 JT |
250 | bio_list_add(&cell->bios, inmate); |
251 | goto out; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Use new cell. | |
256 | */ | |
257 | cell = cell2; | |
258 | ||
259 | cell->prison = prison; | |
260 | memcpy(&cell->key, key, sizeof(cell->key)); | |
261 | cell->holder = inmate; | |
262 | bio_list_init(&cell->bios); | |
263 | hlist_add_head(&cell->list, prison->cells + hash); | |
264 | ||
265 | r = 0; | |
266 | ||
267 | out: | |
268 | spin_unlock_irqrestore(&prison->lock, flags); | |
991d9fa0 JT |
269 | |
270 | *ref = cell; | |
271 | ||
272 | return r; | |
273 | } | |
274 | ||
275 | /* | |
276 | * @inmates must have been initialised prior to this call | |
277 | */ | |
a24c2569 | 278 | static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates) |
991d9fa0 JT |
279 | { |
280 | struct bio_prison *prison = cell->prison; | |
281 | ||
282 | hlist_del(&cell->list); | |
283 | ||
03aaae7c MS |
284 | if (inmates) { |
285 | bio_list_add(inmates, cell->holder); | |
286 | bio_list_merge(inmates, &cell->bios); | |
287 | } | |
991d9fa0 JT |
288 | |
289 | mempool_free(cell, prison->cell_pool); | |
290 | } | |
291 | ||
a24c2569 | 292 | static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios) |
991d9fa0 JT |
293 | { |
294 | unsigned long flags; | |
295 | struct bio_prison *prison = cell->prison; | |
296 | ||
297 | spin_lock_irqsave(&prison->lock, flags); | |
298 | __cell_release(cell, bios); | |
299 | spin_unlock_irqrestore(&prison->lock, flags); | |
300 | } | |
301 | ||
302 | /* | |
303 | * There are a couple of places where we put a bio into a cell briefly | |
304 | * before taking it out again. In these situations we know that no other | |
305 | * bio may be in the cell. This function releases the cell, and also does | |
306 | * a sanity check. | |
307 | */ | |
a24c2569 | 308 | static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
6f94a4c4 | 309 | { |
6f94a4c4 JT |
310 | BUG_ON(cell->holder != bio); |
311 | BUG_ON(!bio_list_empty(&cell->bios)); | |
03aaae7c MS |
312 | |
313 | __cell_release(cell, NULL); | |
6f94a4c4 JT |
314 | } |
315 | ||
a24c2569 | 316 | static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 | 317 | { |
991d9fa0 | 318 | unsigned long flags; |
6f94a4c4 | 319 | struct bio_prison *prison = cell->prison; |
991d9fa0 JT |
320 | |
321 | spin_lock_irqsave(&prison->lock, flags); | |
6f94a4c4 | 322 | __cell_release_singleton(cell, bio); |
991d9fa0 | 323 | spin_unlock_irqrestore(&prison->lock, flags); |
6f94a4c4 JT |
324 | } |
325 | ||
326 | /* | |
327 | * Sometimes we don't want the holder, just the additional bios. | |
328 | */ | |
a24c2569 MS |
329 | static void __cell_release_no_holder(struct dm_bio_prison_cell *cell, |
330 | struct bio_list *inmates) | |
6f94a4c4 JT |
331 | { |
332 | struct bio_prison *prison = cell->prison; | |
333 | ||
334 | hlist_del(&cell->list); | |
335 | bio_list_merge(inmates, &cell->bios); | |
336 | ||
337 | mempool_free(cell, prison->cell_pool); | |
338 | } | |
339 | ||
a24c2569 MS |
340 | static void cell_release_no_holder(struct dm_bio_prison_cell *cell, |
341 | struct bio_list *inmates) | |
6f94a4c4 JT |
342 | { |
343 | unsigned long flags; | |
344 | struct bio_prison *prison = cell->prison; | |
991d9fa0 | 345 | |
6f94a4c4 JT |
346 | spin_lock_irqsave(&prison->lock, flags); |
347 | __cell_release_no_holder(cell, inmates); | |
348 | spin_unlock_irqrestore(&prison->lock, flags); | |
991d9fa0 JT |
349 | } |
350 | ||
a24c2569 | 351 | static void cell_error(struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
352 | { |
353 | struct bio_prison *prison = cell->prison; | |
354 | struct bio_list bios; | |
355 | struct bio *bio; | |
356 | unsigned long flags; | |
357 | ||
358 | bio_list_init(&bios); | |
359 | ||
360 | spin_lock_irqsave(&prison->lock, flags); | |
361 | __cell_release(cell, &bios); | |
362 | spin_unlock_irqrestore(&prison->lock, flags); | |
363 | ||
364 | while ((bio = bio_list_pop(&bios))) | |
365 | bio_io_error(bio); | |
366 | } | |
367 | ||
368 | /*----------------------------------------------------------------*/ | |
369 | ||
370 | /* | |
371 | * We use the deferred set to keep track of pending reads to shared blocks. | |
372 | * We do this to ensure the new mapping caused by a write isn't performed | |
373 | * until these prior reads have completed. Otherwise the insertion of the | |
374 | * new mapping could free the old block that the read bios are mapped to. | |
375 | */ | |
376 | ||
377 | struct deferred_set; | |
378 | struct deferred_entry { | |
379 | struct deferred_set *ds; | |
380 | unsigned count; | |
381 | struct list_head work_items; | |
382 | }; | |
383 | ||
384 | struct deferred_set { | |
385 | spinlock_t lock; | |
386 | unsigned current_entry; | |
387 | unsigned sweeper; | |
388 | struct deferred_entry entries[DEFERRED_SET_SIZE]; | |
389 | }; | |
390 | ||
391 | static void ds_init(struct deferred_set *ds) | |
392 | { | |
393 | int i; | |
394 | ||
395 | spin_lock_init(&ds->lock); | |
396 | ds->current_entry = 0; | |
397 | ds->sweeper = 0; | |
398 | for (i = 0; i < DEFERRED_SET_SIZE; i++) { | |
399 | ds->entries[i].ds = ds; | |
400 | ds->entries[i].count = 0; | |
401 | INIT_LIST_HEAD(&ds->entries[i].work_items); | |
402 | } | |
403 | } | |
404 | ||
405 | static struct deferred_entry *ds_inc(struct deferred_set *ds) | |
406 | { | |
407 | unsigned long flags; | |
408 | struct deferred_entry *entry; | |
409 | ||
410 | spin_lock_irqsave(&ds->lock, flags); | |
411 | entry = ds->entries + ds->current_entry; | |
412 | entry->count++; | |
413 | spin_unlock_irqrestore(&ds->lock, flags); | |
414 | ||
415 | return entry; | |
416 | } | |
417 | ||
418 | static unsigned ds_next(unsigned index) | |
419 | { | |
420 | return (index + 1) % DEFERRED_SET_SIZE; | |
421 | } | |
422 | ||
423 | static void __sweep(struct deferred_set *ds, struct list_head *head) | |
424 | { | |
425 | while ((ds->sweeper != ds->current_entry) && | |
426 | !ds->entries[ds->sweeper].count) { | |
427 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); | |
428 | ds->sweeper = ds_next(ds->sweeper); | |
429 | } | |
430 | ||
431 | if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count) | |
432 | list_splice_init(&ds->entries[ds->sweeper].work_items, head); | |
433 | } | |
434 | ||
435 | static void ds_dec(struct deferred_entry *entry, struct list_head *head) | |
436 | { | |
437 | unsigned long flags; | |
438 | ||
439 | spin_lock_irqsave(&entry->ds->lock, flags); | |
440 | BUG_ON(!entry->count); | |
441 | --entry->count; | |
442 | __sweep(entry->ds, head); | |
443 | spin_unlock_irqrestore(&entry->ds->lock, flags); | |
444 | } | |
445 | ||
446 | /* | |
447 | * Returns 1 if deferred or 0 if no pending items to delay job. | |
448 | */ | |
449 | static int ds_add_work(struct deferred_set *ds, struct list_head *work) | |
450 | { | |
451 | int r = 1; | |
452 | unsigned long flags; | |
453 | unsigned next_entry; | |
454 | ||
455 | spin_lock_irqsave(&ds->lock, flags); | |
456 | if ((ds->sweeper == ds->current_entry) && | |
457 | !ds->entries[ds->current_entry].count) | |
458 | r = 0; | |
459 | else { | |
460 | list_add(work, &ds->entries[ds->current_entry].work_items); | |
461 | next_entry = ds_next(ds->current_entry); | |
462 | if (!ds->entries[next_entry].count) | |
463 | ds->current_entry = next_entry; | |
464 | } | |
465 | spin_unlock_irqrestore(&ds->lock, flags); | |
466 | ||
467 | return r; | |
468 | } | |
469 | ||
470 | /*----------------------------------------------------------------*/ | |
471 | ||
472 | /* | |
473 | * Key building. | |
474 | */ | |
475 | static void build_data_key(struct dm_thin_device *td, | |
476 | dm_block_t b, struct cell_key *key) | |
477 | { | |
478 | key->virtual = 0; | |
479 | key->dev = dm_thin_dev_id(td); | |
480 | key->block = b; | |
481 | } | |
482 | ||
483 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
484 | struct cell_key *key) | |
485 | { | |
486 | key->virtual = 1; | |
487 | key->dev = dm_thin_dev_id(td); | |
488 | key->block = b; | |
489 | } | |
490 | ||
491 | /*----------------------------------------------------------------*/ | |
492 | ||
493 | /* | |
494 | * A pool device ties together a metadata device and a data device. It | |
495 | * also provides the interface for creating and destroying internal | |
496 | * devices. | |
497 | */ | |
a24c2569 | 498 | struct dm_thin_new_mapping; |
67e2e2b2 | 499 | |
e49e5829 JT |
500 | /* |
501 | * The pool runs in 3 modes. Ordered in degraded order for comparisons. | |
502 | */ | |
503 | enum pool_mode { | |
504 | PM_WRITE, /* metadata may be changed */ | |
505 | PM_READ_ONLY, /* metadata may not be changed */ | |
506 | PM_FAIL, /* all I/O fails */ | |
507 | }; | |
508 | ||
67e2e2b2 | 509 | struct pool_features { |
e49e5829 JT |
510 | enum pool_mode mode; |
511 | ||
67e2e2b2 JT |
512 | unsigned zero_new_blocks:1; |
513 | unsigned discard_enabled:1; | |
514 | unsigned discard_passdown:1; | |
515 | }; | |
516 | ||
e49e5829 JT |
517 | struct thin_c; |
518 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
519 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
520 | ||
991d9fa0 JT |
521 | struct pool { |
522 | struct list_head list; | |
523 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
524 | ||
525 | struct mapped_device *pool_md; | |
526 | struct block_device *md_dev; | |
527 | struct dm_pool_metadata *pmd; | |
528 | ||
991d9fa0 | 529 | dm_block_t low_water_blocks; |
55f2b8bd | 530 | uint32_t sectors_per_block; |
f9a8e0cd | 531 | int sectors_per_block_shift; |
991d9fa0 | 532 | |
67e2e2b2 | 533 | struct pool_features pf; |
991d9fa0 JT |
534 | unsigned low_water_triggered:1; /* A dm event has been sent */ |
535 | unsigned no_free_space:1; /* A -ENOSPC warning has been issued */ | |
536 | ||
537 | struct bio_prison *prison; | |
538 | struct dm_kcopyd_client *copier; | |
539 | ||
540 | struct workqueue_struct *wq; | |
541 | struct work_struct worker; | |
905e51b3 | 542 | struct delayed_work waker; |
991d9fa0 | 543 | |
905e51b3 | 544 | unsigned long last_commit_jiffies; |
55f2b8bd | 545 | unsigned ref_count; |
991d9fa0 JT |
546 | |
547 | spinlock_t lock; | |
548 | struct bio_list deferred_bios; | |
549 | struct bio_list deferred_flush_bios; | |
550 | struct list_head prepared_mappings; | |
104655fd | 551 | struct list_head prepared_discards; |
991d9fa0 JT |
552 | |
553 | struct bio_list retry_on_resume_list; | |
554 | ||
eb2aa48d | 555 | struct deferred_set shared_read_ds; |
104655fd | 556 | struct deferred_set all_io_ds; |
991d9fa0 | 557 | |
a24c2569 | 558 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 JT |
559 | mempool_t *mapping_pool; |
560 | mempool_t *endio_hook_pool; | |
e49e5829 JT |
561 | |
562 | process_bio_fn process_bio; | |
563 | process_bio_fn process_discard; | |
564 | ||
565 | process_mapping_fn process_prepared_mapping; | |
566 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
567 | }; |
568 | ||
e49e5829 JT |
569 | static enum pool_mode get_pool_mode(struct pool *pool); |
570 | static void set_pool_mode(struct pool *pool, enum pool_mode mode); | |
571 | ||
991d9fa0 JT |
572 | /* |
573 | * Target context for a pool. | |
574 | */ | |
575 | struct pool_c { | |
576 | struct dm_target *ti; | |
577 | struct pool *pool; | |
578 | struct dm_dev *data_dev; | |
579 | struct dm_dev *metadata_dev; | |
580 | struct dm_target_callbacks callbacks; | |
581 | ||
582 | dm_block_t low_water_blocks; | |
67e2e2b2 | 583 | struct pool_features pf; |
991d9fa0 JT |
584 | }; |
585 | ||
586 | /* | |
587 | * Target context for a thin. | |
588 | */ | |
589 | struct thin_c { | |
590 | struct dm_dev *pool_dev; | |
2dd9c257 | 591 | struct dm_dev *origin_dev; |
991d9fa0 JT |
592 | dm_thin_id dev_id; |
593 | ||
594 | struct pool *pool; | |
595 | struct dm_thin_device *td; | |
596 | }; | |
597 | ||
598 | /*----------------------------------------------------------------*/ | |
599 | ||
600 | /* | |
601 | * A global list of pools that uses a struct mapped_device as a key. | |
602 | */ | |
603 | static struct dm_thin_pool_table { | |
604 | struct mutex mutex; | |
605 | struct list_head pools; | |
606 | } dm_thin_pool_table; | |
607 | ||
608 | static void pool_table_init(void) | |
609 | { | |
610 | mutex_init(&dm_thin_pool_table.mutex); | |
611 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
612 | } | |
613 | ||
614 | static void __pool_table_insert(struct pool *pool) | |
615 | { | |
616 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
617 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
618 | } | |
619 | ||
620 | static void __pool_table_remove(struct pool *pool) | |
621 | { | |
622 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
623 | list_del(&pool->list); | |
624 | } | |
625 | ||
626 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
627 | { | |
628 | struct pool *pool = NULL, *tmp; | |
629 | ||
630 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
631 | ||
632 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
633 | if (tmp->pool_md == md) { | |
634 | pool = tmp; | |
635 | break; | |
636 | } | |
637 | } | |
638 | ||
639 | return pool; | |
640 | } | |
641 | ||
642 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
643 | { | |
644 | struct pool *pool = NULL, *tmp; | |
645 | ||
646 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
647 | ||
648 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
649 | if (tmp->md_dev == md_dev) { | |
650 | pool = tmp; | |
651 | break; | |
652 | } | |
653 | } | |
654 | ||
655 | return pool; | |
656 | } | |
657 | ||
658 | /*----------------------------------------------------------------*/ | |
659 | ||
a24c2569 | 660 | struct dm_thin_endio_hook { |
eb2aa48d JT |
661 | struct thin_c *tc; |
662 | struct deferred_entry *shared_read_entry; | |
104655fd | 663 | struct deferred_entry *all_io_entry; |
a24c2569 | 664 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
665 | }; |
666 | ||
991d9fa0 JT |
667 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
668 | { | |
669 | struct bio *bio; | |
670 | struct bio_list bios; | |
671 | ||
672 | bio_list_init(&bios); | |
673 | bio_list_merge(&bios, master); | |
674 | bio_list_init(master); | |
675 | ||
676 | while ((bio = bio_list_pop(&bios))) { | |
a24c2569 MS |
677 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
678 | ||
eb2aa48d | 679 | if (h->tc == tc) |
991d9fa0 JT |
680 | bio_endio(bio, DM_ENDIO_REQUEUE); |
681 | else | |
682 | bio_list_add(master, bio); | |
683 | } | |
684 | } | |
685 | ||
686 | static void requeue_io(struct thin_c *tc) | |
687 | { | |
688 | struct pool *pool = tc->pool; | |
689 | unsigned long flags; | |
690 | ||
691 | spin_lock_irqsave(&pool->lock, flags); | |
692 | __requeue_bio_list(tc, &pool->deferred_bios); | |
693 | __requeue_bio_list(tc, &pool->retry_on_resume_list); | |
694 | spin_unlock_irqrestore(&pool->lock, flags); | |
695 | } | |
696 | ||
697 | /* | |
698 | * This section of code contains the logic for processing a thin device's IO. | |
699 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
700 | * but most is exclusively called from the thin target rather than the thin-pool | |
701 | * target. | |
702 | */ | |
703 | ||
704 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
705 | { | |
55f2b8bd MS |
706 | sector_t block_nr = bio->bi_sector; |
707 | ||
f9a8e0cd MP |
708 | if (tc->pool->sectors_per_block_shift < 0) |
709 | (void) sector_div(block_nr, tc->pool->sectors_per_block); | |
710 | else | |
711 | block_nr >>= tc->pool->sectors_per_block_shift; | |
55f2b8bd MS |
712 | |
713 | return block_nr; | |
991d9fa0 JT |
714 | } |
715 | ||
716 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
717 | { | |
718 | struct pool *pool = tc->pool; | |
55f2b8bd | 719 | sector_t bi_sector = bio->bi_sector; |
991d9fa0 JT |
720 | |
721 | bio->bi_bdev = tc->pool_dev->bdev; | |
f9a8e0cd MP |
722 | if (tc->pool->sectors_per_block_shift < 0) |
723 | bio->bi_sector = (block * pool->sectors_per_block) + | |
724 | sector_div(bi_sector, pool->sectors_per_block); | |
725 | else | |
726 | bio->bi_sector = (block << pool->sectors_per_block_shift) | | |
727 | (bi_sector & (pool->sectors_per_block - 1)); | |
991d9fa0 JT |
728 | } |
729 | ||
2dd9c257 JT |
730 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
731 | { | |
732 | bio->bi_bdev = tc->origin_dev->bdev; | |
733 | } | |
734 | ||
4afdd680 JT |
735 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
736 | { | |
737 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
738 | dm_thin_changed_this_transaction(tc->td); | |
739 | } | |
740 | ||
2dd9c257 | 741 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
742 | { |
743 | struct pool *pool = tc->pool; | |
744 | unsigned long flags; | |
745 | ||
e49e5829 JT |
746 | if (!bio_triggers_commit(tc, bio)) { |
747 | generic_make_request(bio); | |
748 | return; | |
749 | } | |
750 | ||
991d9fa0 | 751 | /* |
e49e5829 JT |
752 | * Complete bio with an error if earlier I/O caused changes to |
753 | * the metadata that can't be committed e.g, due to I/O errors | |
754 | * on the metadata device. | |
991d9fa0 | 755 | */ |
e49e5829 JT |
756 | if (dm_thin_aborted_changes(tc->td)) { |
757 | bio_io_error(bio); | |
758 | return; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Batch together any bios that trigger commits and then issue a | |
763 | * single commit for them in process_deferred_bios(). | |
764 | */ | |
765 | spin_lock_irqsave(&pool->lock, flags); | |
766 | bio_list_add(&pool->deferred_flush_bios, bio); | |
767 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
768 | } |
769 | ||
2dd9c257 JT |
770 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
771 | { | |
772 | remap_to_origin(tc, bio); | |
773 | issue(tc, bio); | |
774 | } | |
775 | ||
776 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
777 | dm_block_t block) | |
778 | { | |
779 | remap(tc, bio, block); | |
780 | issue(tc, bio); | |
781 | } | |
782 | ||
991d9fa0 JT |
783 | /* |
784 | * wake_worker() is used when new work is queued and when pool_resume is | |
785 | * ready to continue deferred IO processing. | |
786 | */ | |
787 | static void wake_worker(struct pool *pool) | |
788 | { | |
789 | queue_work(pool->wq, &pool->worker); | |
790 | } | |
791 | ||
792 | /*----------------------------------------------------------------*/ | |
793 | ||
794 | /* | |
795 | * Bio endio functions. | |
796 | */ | |
a24c2569 | 797 | struct dm_thin_new_mapping { |
991d9fa0 JT |
798 | struct list_head list; |
799 | ||
eb2aa48d JT |
800 | unsigned quiesced:1; |
801 | unsigned prepared:1; | |
104655fd | 802 | unsigned pass_discard:1; |
991d9fa0 JT |
803 | |
804 | struct thin_c *tc; | |
805 | dm_block_t virt_block; | |
806 | dm_block_t data_block; | |
a24c2569 | 807 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
808 | int err; |
809 | ||
810 | /* | |
811 | * If the bio covers the whole area of a block then we can avoid | |
812 | * zeroing or copying. Instead this bio is hooked. The bio will | |
813 | * still be in the cell, so care has to be taken to avoid issuing | |
814 | * the bio twice. | |
815 | */ | |
816 | struct bio *bio; | |
817 | bio_end_io_t *saved_bi_end_io; | |
818 | }; | |
819 | ||
a24c2569 | 820 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
821 | { |
822 | struct pool *pool = m->tc->pool; | |
823 | ||
eb2aa48d | 824 | if (m->quiesced && m->prepared) { |
991d9fa0 JT |
825 | list_add(&m->list, &pool->prepared_mappings); |
826 | wake_worker(pool); | |
827 | } | |
828 | } | |
829 | ||
830 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
831 | { | |
832 | unsigned long flags; | |
a24c2569 | 833 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
834 | struct pool *pool = m->tc->pool; |
835 | ||
836 | m->err = read_err || write_err ? -EIO : 0; | |
837 | ||
838 | spin_lock_irqsave(&pool->lock, flags); | |
839 | m->prepared = 1; | |
840 | __maybe_add_mapping(m); | |
841 | spin_unlock_irqrestore(&pool->lock, flags); | |
842 | } | |
843 | ||
844 | static void overwrite_endio(struct bio *bio, int err) | |
845 | { | |
846 | unsigned long flags; | |
a24c2569 MS |
847 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
848 | struct dm_thin_new_mapping *m = h->overwrite_mapping; | |
991d9fa0 JT |
849 | struct pool *pool = m->tc->pool; |
850 | ||
851 | m->err = err; | |
852 | ||
853 | spin_lock_irqsave(&pool->lock, flags); | |
854 | m->prepared = 1; | |
855 | __maybe_add_mapping(m); | |
856 | spin_unlock_irqrestore(&pool->lock, flags); | |
857 | } | |
858 | ||
991d9fa0 JT |
859 | /*----------------------------------------------------------------*/ |
860 | ||
861 | /* | |
862 | * Workqueue. | |
863 | */ | |
864 | ||
865 | /* | |
866 | * Prepared mapping jobs. | |
867 | */ | |
868 | ||
869 | /* | |
870 | * This sends the bios in the cell back to the deferred_bios list. | |
871 | */ | |
a24c2569 | 872 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
873 | dm_block_t data_block) |
874 | { | |
875 | struct pool *pool = tc->pool; | |
876 | unsigned long flags; | |
877 | ||
878 | spin_lock_irqsave(&pool->lock, flags); | |
879 | cell_release(cell, &pool->deferred_bios); | |
880 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
881 | ||
882 | wake_worker(pool); | |
883 | } | |
884 | ||
885 | /* | |
886 | * Same as cell_defer above, except it omits one particular detainee, | |
887 | * a write bio that covers the block and has already been processed. | |
888 | */ | |
a24c2569 | 889 | static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
890 | { |
891 | struct bio_list bios; | |
991d9fa0 JT |
892 | struct pool *pool = tc->pool; |
893 | unsigned long flags; | |
894 | ||
895 | bio_list_init(&bios); | |
991d9fa0 JT |
896 | |
897 | spin_lock_irqsave(&pool->lock, flags); | |
6f94a4c4 | 898 | cell_release_no_holder(cell, &pool->deferred_bios); |
991d9fa0 JT |
899 | spin_unlock_irqrestore(&pool->lock, flags); |
900 | ||
901 | wake_worker(pool); | |
902 | } | |
903 | ||
e49e5829 JT |
904 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
905 | { | |
906 | if (m->bio) | |
907 | m->bio->bi_end_io = m->saved_bi_end_io; | |
908 | cell_error(m->cell); | |
909 | list_del(&m->list); | |
910 | mempool_free(m, m->tc->pool->mapping_pool); | |
911 | } | |
a24c2569 | 912 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
913 | { |
914 | struct thin_c *tc = m->tc; | |
915 | struct bio *bio; | |
916 | int r; | |
917 | ||
918 | bio = m->bio; | |
919 | if (bio) | |
920 | bio->bi_end_io = m->saved_bi_end_io; | |
921 | ||
922 | if (m->err) { | |
923 | cell_error(m->cell); | |
905386f8 | 924 | goto out; |
991d9fa0 JT |
925 | } |
926 | ||
927 | /* | |
928 | * Commit the prepared block into the mapping btree. | |
929 | * Any I/O for this block arriving after this point will get | |
930 | * remapped to it directly. | |
931 | */ | |
932 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
933 | if (r) { | |
934 | DMERR("dm_thin_insert_block() failed"); | |
935 | cell_error(m->cell); | |
905386f8 | 936 | goto out; |
991d9fa0 JT |
937 | } |
938 | ||
939 | /* | |
940 | * Release any bios held while the block was being provisioned. | |
941 | * If we are processing a write bio that completely covers the block, | |
942 | * we already processed it so can ignore it now when processing | |
943 | * the bios in the cell. | |
944 | */ | |
945 | if (bio) { | |
6f94a4c4 | 946 | cell_defer_except(tc, m->cell); |
991d9fa0 JT |
947 | bio_endio(bio, 0); |
948 | } else | |
949 | cell_defer(tc, m->cell, m->data_block); | |
950 | ||
905386f8 | 951 | out: |
991d9fa0 JT |
952 | list_del(&m->list); |
953 | mempool_free(m, tc->pool->mapping_pool); | |
954 | } | |
955 | ||
e49e5829 | 956 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 957 | { |
104655fd JT |
958 | struct thin_c *tc = m->tc; |
959 | ||
e49e5829 JT |
960 | bio_io_error(m->bio); |
961 | cell_defer_except(tc, m->cell); | |
962 | cell_defer_except(tc, m->cell2); | |
963 | mempool_free(m, tc->pool->mapping_pool); | |
964 | } | |
965 | ||
966 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
967 | { | |
968 | struct thin_c *tc = m->tc; | |
104655fd | 969 | |
104655fd JT |
970 | if (m->pass_discard) |
971 | remap_and_issue(tc, m->bio, m->data_block); | |
972 | else | |
973 | bio_endio(m->bio, 0); | |
974 | ||
975 | cell_defer_except(tc, m->cell); | |
976 | cell_defer_except(tc, m->cell2); | |
977 | mempool_free(m, tc->pool->mapping_pool); | |
978 | } | |
979 | ||
e49e5829 JT |
980 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
981 | { | |
982 | int r; | |
983 | struct thin_c *tc = m->tc; | |
984 | ||
985 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
986 | if (r) | |
987 | DMERR("dm_thin_remove_block() failed"); | |
988 | ||
989 | process_prepared_discard_passdown(m); | |
990 | } | |
991 | ||
104655fd | 992 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 993 | process_mapping_fn *fn) |
991d9fa0 JT |
994 | { |
995 | unsigned long flags; | |
996 | struct list_head maps; | |
a24c2569 | 997 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
998 | |
999 | INIT_LIST_HEAD(&maps); | |
1000 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 1001 | list_splice_init(head, &maps); |
991d9fa0 JT |
1002 | spin_unlock_irqrestore(&pool->lock, flags); |
1003 | ||
1004 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1005 | (*fn)(m); |
991d9fa0 JT |
1006 | } |
1007 | ||
1008 | /* | |
1009 | * Deferred bio jobs. | |
1010 | */ | |
104655fd | 1011 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1012 | { |
f9a8e0cd | 1013 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
1014 | } |
1015 | ||
1016 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1017 | { | |
1018 | return (bio_data_dir(bio) == WRITE) && | |
1019 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1020 | } |
1021 | ||
1022 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1023 | bio_end_io_t *fn) | |
1024 | { | |
1025 | *save = bio->bi_end_io; | |
1026 | bio->bi_end_io = fn; | |
1027 | } | |
1028 | ||
1029 | static int ensure_next_mapping(struct pool *pool) | |
1030 | { | |
1031 | if (pool->next_mapping) | |
1032 | return 0; | |
1033 | ||
1034 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
1035 | ||
1036 | return pool->next_mapping ? 0 : -ENOMEM; | |
1037 | } | |
1038 | ||
a24c2569 | 1039 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1040 | { |
a24c2569 | 1041 | struct dm_thin_new_mapping *r = pool->next_mapping; |
991d9fa0 JT |
1042 | |
1043 | BUG_ON(!pool->next_mapping); | |
1044 | ||
1045 | pool->next_mapping = NULL; | |
1046 | ||
1047 | return r; | |
1048 | } | |
1049 | ||
1050 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
1051 | struct dm_dev *origin, dm_block_t data_origin, |
1052 | dm_block_t data_dest, | |
a24c2569 | 1053 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
1054 | { |
1055 | int r; | |
1056 | struct pool *pool = tc->pool; | |
a24c2569 | 1057 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 JT |
1058 | |
1059 | INIT_LIST_HEAD(&m->list); | |
eb2aa48d | 1060 | m->quiesced = 0; |
991d9fa0 JT |
1061 | m->prepared = 0; |
1062 | m->tc = tc; | |
1063 | m->virt_block = virt_block; | |
1064 | m->data_block = data_dest; | |
1065 | m->cell = cell; | |
1066 | m->err = 0; | |
1067 | m->bio = NULL; | |
1068 | ||
eb2aa48d JT |
1069 | if (!ds_add_work(&pool->shared_read_ds, &m->list)) |
1070 | m->quiesced = 1; | |
991d9fa0 JT |
1071 | |
1072 | /* | |
1073 | * IO to pool_dev remaps to the pool target's data_dev. | |
1074 | * | |
1075 | * If the whole block of data is being overwritten, we can issue the | |
1076 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1077 | */ | |
1078 | if (io_overwrites_block(pool, bio)) { | |
a24c2569 MS |
1079 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
1080 | ||
eb2aa48d | 1081 | h->overwrite_mapping = m; |
991d9fa0 JT |
1082 | m->bio = bio; |
1083 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
991d9fa0 JT |
1084 | remap_and_issue(tc, bio, data_dest); |
1085 | } else { | |
1086 | struct dm_io_region from, to; | |
1087 | ||
2dd9c257 | 1088 | from.bdev = origin->bdev; |
991d9fa0 JT |
1089 | from.sector = data_origin * pool->sectors_per_block; |
1090 | from.count = pool->sectors_per_block; | |
1091 | ||
1092 | to.bdev = tc->pool_dev->bdev; | |
1093 | to.sector = data_dest * pool->sectors_per_block; | |
1094 | to.count = pool->sectors_per_block; | |
1095 | ||
1096 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
1097 | 0, copy_complete, m); | |
1098 | if (r < 0) { | |
1099 | mempool_free(m, pool->mapping_pool); | |
1100 | DMERR("dm_kcopyd_copy() failed"); | |
1101 | cell_error(cell); | |
1102 | } | |
1103 | } | |
1104 | } | |
1105 | ||
2dd9c257 JT |
1106 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1107 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1108 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1109 | { |
1110 | schedule_copy(tc, virt_block, tc->pool_dev, | |
1111 | data_origin, data_dest, cell, bio); | |
1112 | } | |
1113 | ||
1114 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
1115 | dm_block_t data_dest, | |
a24c2569 | 1116 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1117 | { |
1118 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1119 | virt_block, data_dest, cell, bio); | |
1120 | } | |
1121 | ||
991d9fa0 | 1122 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1123 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1124 | struct bio *bio) |
1125 | { | |
1126 | struct pool *pool = tc->pool; | |
a24c2569 | 1127 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 JT |
1128 | |
1129 | INIT_LIST_HEAD(&m->list); | |
eb2aa48d | 1130 | m->quiesced = 1; |
991d9fa0 JT |
1131 | m->prepared = 0; |
1132 | m->tc = tc; | |
1133 | m->virt_block = virt_block; | |
1134 | m->data_block = data_block; | |
1135 | m->cell = cell; | |
1136 | m->err = 0; | |
1137 | m->bio = NULL; | |
1138 | ||
1139 | /* | |
1140 | * If the whole block of data is being overwritten or we are not | |
1141 | * zeroing pre-existing data, we can issue the bio immediately. | |
1142 | * Otherwise we use kcopyd to zero the data first. | |
1143 | */ | |
67e2e2b2 | 1144 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
1145 | process_prepared_mapping(m); |
1146 | ||
1147 | else if (io_overwrites_block(pool, bio)) { | |
a24c2569 MS |
1148 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
1149 | ||
eb2aa48d | 1150 | h->overwrite_mapping = m; |
991d9fa0 JT |
1151 | m->bio = bio; |
1152 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
991d9fa0 | 1153 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
1154 | } else { |
1155 | int r; | |
1156 | struct dm_io_region to; | |
1157 | ||
1158 | to.bdev = tc->pool_dev->bdev; | |
1159 | to.sector = data_block * pool->sectors_per_block; | |
1160 | to.count = pool->sectors_per_block; | |
1161 | ||
1162 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
1163 | if (r < 0) { | |
1164 | mempool_free(m, pool->mapping_pool); | |
1165 | DMERR("dm_kcopyd_zero() failed"); | |
1166 | cell_error(cell); | |
1167 | } | |
1168 | } | |
1169 | } | |
1170 | ||
e49e5829 JT |
1171 | static int commit(struct pool *pool) |
1172 | { | |
1173 | int r; | |
1174 | ||
1175 | r = dm_pool_commit_metadata(pool->pmd); | |
1176 | if (r) | |
1177 | DMERR("commit failed, error = %d", r); | |
1178 | ||
1179 | return r; | |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * A non-zero return indicates read_only or fail_io mode. | |
1184 | * Many callers don't care about the return value. | |
1185 | */ | |
1186 | static int commit_or_fallback(struct pool *pool) | |
1187 | { | |
1188 | int r; | |
1189 | ||
1190 | if (get_pool_mode(pool) != PM_WRITE) | |
1191 | return -EINVAL; | |
1192 | ||
1193 | r = commit(pool); | |
1194 | if (r) | |
1195 | set_pool_mode(pool, PM_READ_ONLY); | |
1196 | ||
1197 | return r; | |
1198 | } | |
1199 | ||
991d9fa0 JT |
1200 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1201 | { | |
1202 | int r; | |
1203 | dm_block_t free_blocks; | |
1204 | unsigned long flags; | |
1205 | struct pool *pool = tc->pool; | |
1206 | ||
1207 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); | |
1208 | if (r) | |
1209 | return r; | |
1210 | ||
1211 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1212 | DMWARN("%s: reached low water mark, sending event.", | |
1213 | dm_device_name(pool->pool_md)); | |
1214 | spin_lock_irqsave(&pool->lock, flags); | |
1215 | pool->low_water_triggered = 1; | |
1216 | spin_unlock_irqrestore(&pool->lock, flags); | |
1217 | dm_table_event(pool->ti->table); | |
1218 | } | |
1219 | ||
1220 | if (!free_blocks) { | |
1221 | if (pool->no_free_space) | |
1222 | return -ENOSPC; | |
1223 | else { | |
1224 | /* | |
1225 | * Try to commit to see if that will free up some | |
1226 | * more space. | |
1227 | */ | |
e49e5829 | 1228 | (void) commit_or_fallback(pool); |
991d9fa0 JT |
1229 | |
1230 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); | |
1231 | if (r) | |
1232 | return r; | |
1233 | ||
1234 | /* | |
1235 | * If we still have no space we set a flag to avoid | |
1236 | * doing all this checking and return -ENOSPC. | |
1237 | */ | |
1238 | if (!free_blocks) { | |
1239 | DMWARN("%s: no free space available.", | |
1240 | dm_device_name(pool->pool_md)); | |
1241 | spin_lock_irqsave(&pool->lock, flags); | |
1242 | pool->no_free_space = 1; | |
1243 | spin_unlock_irqrestore(&pool->lock, flags); | |
1244 | return -ENOSPC; | |
1245 | } | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
1250 | if (r) | |
1251 | return r; | |
1252 | ||
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | /* | |
1257 | * If we have run out of space, queue bios until the device is | |
1258 | * resumed, presumably after having been reloaded with more space. | |
1259 | */ | |
1260 | static void retry_on_resume(struct bio *bio) | |
1261 | { | |
a24c2569 | 1262 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
eb2aa48d | 1263 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1264 | struct pool *pool = tc->pool; |
1265 | unsigned long flags; | |
1266 | ||
1267 | spin_lock_irqsave(&pool->lock, flags); | |
1268 | bio_list_add(&pool->retry_on_resume_list, bio); | |
1269 | spin_unlock_irqrestore(&pool->lock, flags); | |
1270 | } | |
1271 | ||
a24c2569 | 1272 | static void no_space(struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1273 | { |
1274 | struct bio *bio; | |
1275 | struct bio_list bios; | |
1276 | ||
1277 | bio_list_init(&bios); | |
1278 | cell_release(cell, &bios); | |
1279 | ||
1280 | while ((bio = bio_list_pop(&bios))) | |
1281 | retry_on_resume(bio); | |
1282 | } | |
1283 | ||
104655fd JT |
1284 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1285 | { | |
1286 | int r; | |
c3a0ce2e | 1287 | unsigned long flags; |
104655fd | 1288 | struct pool *pool = tc->pool; |
a24c2569 | 1289 | struct dm_bio_prison_cell *cell, *cell2; |
104655fd JT |
1290 | struct cell_key key, key2; |
1291 | dm_block_t block = get_bio_block(tc, bio); | |
1292 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1293 | struct dm_thin_new_mapping *m; |
104655fd JT |
1294 | |
1295 | build_virtual_key(tc->td, block, &key); | |
1296 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) | |
1297 | return; | |
1298 | ||
1299 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1300 | switch (r) { | |
1301 | case 0: | |
1302 | /* | |
1303 | * Check nobody is fiddling with this pool block. This can | |
1304 | * happen if someone's in the process of breaking sharing | |
1305 | * on this block. | |
1306 | */ | |
1307 | build_data_key(tc->td, lookup_result.block, &key2); | |
1308 | if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) { | |
1309 | cell_release_singleton(cell, bio); | |
1310 | break; | |
1311 | } | |
1312 | ||
1313 | if (io_overlaps_block(pool, bio)) { | |
1314 | /* | |
1315 | * IO may still be going to the destination block. We must | |
1316 | * quiesce before we can do the removal. | |
1317 | */ | |
1318 | m = get_next_mapping(pool); | |
1319 | m->tc = tc; | |
17b7d63f | 1320 | m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown; |
104655fd JT |
1321 | m->virt_block = block; |
1322 | m->data_block = lookup_result.block; | |
1323 | m->cell = cell; | |
1324 | m->cell2 = cell2; | |
1325 | m->err = 0; | |
1326 | m->bio = bio; | |
1327 | ||
1328 | if (!ds_add_work(&pool->all_io_ds, &m->list)) { | |
c3a0ce2e | 1329 | spin_lock_irqsave(&pool->lock, flags); |
104655fd | 1330 | list_add(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1331 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1332 | wake_worker(pool); |
1333 | } | |
1334 | } else { | |
1335 | /* | |
49296309 MP |
1336 | * The DM core makes sure that the discard doesn't span |
1337 | * a block boundary. So we submit the discard of a | |
1338 | * partial block appropriately. | |
104655fd | 1339 | */ |
104655fd JT |
1340 | cell_release_singleton(cell, bio); |
1341 | cell_release_singleton(cell2, bio); | |
650d2a06 MP |
1342 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1343 | remap_and_issue(tc, bio, lookup_result.block); | |
1344 | else | |
1345 | bio_endio(bio, 0); | |
104655fd JT |
1346 | } |
1347 | break; | |
1348 | ||
1349 | case -ENODATA: | |
1350 | /* | |
1351 | * It isn't provisioned, just forget it. | |
1352 | */ | |
1353 | cell_release_singleton(cell, bio); | |
1354 | bio_endio(bio, 0); | |
1355 | break; | |
1356 | ||
1357 | default: | |
1358 | DMERR("discard: find block unexpectedly returned %d", r); | |
1359 | cell_release_singleton(cell, bio); | |
1360 | bio_io_error(bio); | |
1361 | break; | |
1362 | } | |
1363 | } | |
1364 | ||
991d9fa0 JT |
1365 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
1366 | struct cell_key *key, | |
1367 | struct dm_thin_lookup_result *lookup_result, | |
a24c2569 | 1368 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1369 | { |
1370 | int r; | |
1371 | dm_block_t data_block; | |
1372 | ||
1373 | r = alloc_data_block(tc, &data_block); | |
1374 | switch (r) { | |
1375 | case 0: | |
2dd9c257 JT |
1376 | schedule_internal_copy(tc, block, lookup_result->block, |
1377 | data_block, cell, bio); | |
991d9fa0 JT |
1378 | break; |
1379 | ||
1380 | case -ENOSPC: | |
1381 | no_space(cell); | |
1382 | break; | |
1383 | ||
1384 | default: | |
1385 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); | |
1386 | cell_error(cell); | |
1387 | break; | |
1388 | } | |
1389 | } | |
1390 | ||
1391 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1392 | dm_block_t block, | |
1393 | struct dm_thin_lookup_result *lookup_result) | |
1394 | { | |
a24c2569 | 1395 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
1396 | struct pool *pool = tc->pool; |
1397 | struct cell_key key; | |
1398 | ||
1399 | /* | |
1400 | * If cell is already occupied, then sharing is already in the process | |
1401 | * of being broken so we have nothing further to do here. | |
1402 | */ | |
1403 | build_data_key(tc->td, lookup_result->block, &key); | |
1404 | if (bio_detain(pool->prison, &key, bio, &cell)) | |
1405 | return; | |
1406 | ||
60049701 | 1407 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1408 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1409 | else { | |
a24c2569 | 1410 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
991d9fa0 | 1411 | |
eb2aa48d | 1412 | h->shared_read_entry = ds_inc(&pool->shared_read_ds); |
991d9fa0 JT |
1413 | |
1414 | cell_release_singleton(cell, bio); | |
1415 | remap_and_issue(tc, bio, lookup_result->block); | |
1416 | } | |
1417 | } | |
1418 | ||
1419 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1420 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1421 | { |
1422 | int r; | |
1423 | dm_block_t data_block; | |
1424 | ||
1425 | /* | |
1426 | * Remap empty bios (flushes) immediately, without provisioning. | |
1427 | */ | |
1428 | if (!bio->bi_size) { | |
1429 | cell_release_singleton(cell, bio); | |
1430 | remap_and_issue(tc, bio, 0); | |
1431 | return; | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Fill read bios with zeroes and complete them immediately. | |
1436 | */ | |
1437 | if (bio_data_dir(bio) == READ) { | |
1438 | zero_fill_bio(bio); | |
1439 | cell_release_singleton(cell, bio); | |
1440 | bio_endio(bio, 0); | |
1441 | return; | |
1442 | } | |
1443 | ||
1444 | r = alloc_data_block(tc, &data_block); | |
1445 | switch (r) { | |
1446 | case 0: | |
2dd9c257 JT |
1447 | if (tc->origin_dev) |
1448 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1449 | else | |
1450 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1451 | break; |
1452 | ||
1453 | case -ENOSPC: | |
1454 | no_space(cell); | |
1455 | break; | |
1456 | ||
1457 | default: | |
1458 | DMERR("%s: alloc_data_block() failed, error = %d", __func__, r); | |
e49e5829 | 1459 | set_pool_mode(tc->pool, PM_READ_ONLY); |
991d9fa0 JT |
1460 | cell_error(cell); |
1461 | break; | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1466 | { | |
1467 | int r; | |
1468 | dm_block_t block = get_bio_block(tc, bio); | |
a24c2569 | 1469 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
1470 | struct cell_key key; |
1471 | struct dm_thin_lookup_result lookup_result; | |
1472 | ||
1473 | /* | |
1474 | * If cell is already occupied, then the block is already | |
1475 | * being provisioned so we have nothing further to do here. | |
1476 | */ | |
1477 | build_virtual_key(tc->td, block, &key); | |
1478 | if (bio_detain(tc->pool->prison, &key, bio, &cell)) | |
1479 | return; | |
1480 | ||
1481 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1482 | switch (r) { | |
1483 | case 0: | |
1484 | /* | |
1485 | * We can release this cell now. This thread is the only | |
1486 | * one that puts bios into a cell, and we know there were | |
1487 | * no preceding bios. | |
1488 | */ | |
1489 | /* | |
1490 | * TODO: this will probably have to change when discard goes | |
1491 | * back in. | |
1492 | */ | |
1493 | cell_release_singleton(cell, bio); | |
1494 | ||
1495 | if (lookup_result.shared) | |
1496 | process_shared_bio(tc, bio, block, &lookup_result); | |
1497 | else | |
1498 | remap_and_issue(tc, bio, lookup_result.block); | |
1499 | break; | |
1500 | ||
1501 | case -ENODATA: | |
2dd9c257 JT |
1502 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
1503 | cell_release_singleton(cell, bio); | |
1504 | remap_to_origin_and_issue(tc, bio); | |
1505 | } else | |
1506 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1507 | break; |
1508 | ||
1509 | default: | |
1510 | DMERR("dm_thin_find_block() failed, error = %d", r); | |
104655fd | 1511 | cell_release_singleton(cell, bio); |
991d9fa0 JT |
1512 | bio_io_error(bio); |
1513 | break; | |
1514 | } | |
1515 | } | |
1516 | ||
e49e5829 JT |
1517 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1518 | { | |
1519 | int r; | |
1520 | int rw = bio_data_dir(bio); | |
1521 | dm_block_t block = get_bio_block(tc, bio); | |
1522 | struct dm_thin_lookup_result lookup_result; | |
1523 | ||
1524 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1525 | switch (r) { | |
1526 | case 0: | |
1527 | if (lookup_result.shared && (rw == WRITE) && bio->bi_size) | |
1528 | bio_io_error(bio); | |
1529 | else | |
1530 | remap_and_issue(tc, bio, lookup_result.block); | |
1531 | break; | |
1532 | ||
1533 | case -ENODATA: | |
1534 | if (rw != READ) { | |
1535 | bio_io_error(bio); | |
1536 | break; | |
1537 | } | |
1538 | ||
1539 | if (tc->origin_dev) { | |
1540 | remap_to_origin_and_issue(tc, bio); | |
1541 | break; | |
1542 | } | |
1543 | ||
1544 | zero_fill_bio(bio); | |
1545 | bio_endio(bio, 0); | |
1546 | break; | |
1547 | ||
1548 | default: | |
1549 | DMERR("dm_thin_find_block() failed, error = %d", r); | |
1550 | bio_io_error(bio); | |
1551 | break; | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) | |
1556 | { | |
1557 | bio_io_error(bio); | |
1558 | } | |
1559 | ||
905e51b3 JT |
1560 | static int need_commit_due_to_time(struct pool *pool) |
1561 | { | |
1562 | return jiffies < pool->last_commit_jiffies || | |
1563 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1564 | } | |
1565 | ||
991d9fa0 JT |
1566 | static void process_deferred_bios(struct pool *pool) |
1567 | { | |
1568 | unsigned long flags; | |
1569 | struct bio *bio; | |
1570 | struct bio_list bios; | |
991d9fa0 JT |
1571 | |
1572 | bio_list_init(&bios); | |
1573 | ||
1574 | spin_lock_irqsave(&pool->lock, flags); | |
1575 | bio_list_merge(&bios, &pool->deferred_bios); | |
1576 | bio_list_init(&pool->deferred_bios); | |
1577 | spin_unlock_irqrestore(&pool->lock, flags); | |
1578 | ||
1579 | while ((bio = bio_list_pop(&bios))) { | |
a24c2569 | 1580 | struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr; |
eb2aa48d JT |
1581 | struct thin_c *tc = h->tc; |
1582 | ||
991d9fa0 JT |
1583 | /* |
1584 | * If we've got no free new_mapping structs, and processing | |
1585 | * this bio might require one, we pause until there are some | |
1586 | * prepared mappings to process. | |
1587 | */ | |
1588 | if (ensure_next_mapping(pool)) { | |
1589 | spin_lock_irqsave(&pool->lock, flags); | |
1590 | bio_list_merge(&pool->deferred_bios, &bios); | |
1591 | spin_unlock_irqrestore(&pool->lock, flags); | |
1592 | ||
1593 | break; | |
1594 | } | |
104655fd JT |
1595 | |
1596 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1597 | pool->process_discard(tc, bio); |
104655fd | 1598 | else |
e49e5829 | 1599 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1600 | } |
1601 | ||
1602 | /* | |
1603 | * If there are any deferred flush bios, we must commit | |
1604 | * the metadata before issuing them. | |
1605 | */ | |
1606 | bio_list_init(&bios); | |
1607 | spin_lock_irqsave(&pool->lock, flags); | |
1608 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1609 | bio_list_init(&pool->deferred_flush_bios); | |
1610 | spin_unlock_irqrestore(&pool->lock, flags); | |
1611 | ||
905e51b3 | 1612 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1613 | return; |
1614 | ||
e49e5829 | 1615 | if (commit_or_fallback(pool)) { |
991d9fa0 JT |
1616 | while ((bio = bio_list_pop(&bios))) |
1617 | bio_io_error(bio); | |
1618 | return; | |
1619 | } | |
905e51b3 | 1620 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1621 | |
1622 | while ((bio = bio_list_pop(&bios))) | |
1623 | generic_make_request(bio); | |
1624 | } | |
1625 | ||
1626 | static void do_worker(struct work_struct *ws) | |
1627 | { | |
1628 | struct pool *pool = container_of(ws, struct pool, worker); | |
1629 | ||
e49e5829 JT |
1630 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1631 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1632 | process_deferred_bios(pool); |
1633 | } | |
1634 | ||
905e51b3 JT |
1635 | /* |
1636 | * We want to commit periodically so that not too much | |
1637 | * unwritten data builds up. | |
1638 | */ | |
1639 | static void do_waker(struct work_struct *ws) | |
1640 | { | |
1641 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1642 | wake_worker(pool); | |
1643 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1644 | } | |
1645 | ||
991d9fa0 JT |
1646 | /*----------------------------------------------------------------*/ |
1647 | ||
e49e5829 JT |
1648 | static enum pool_mode get_pool_mode(struct pool *pool) |
1649 | { | |
1650 | return pool->pf.mode; | |
1651 | } | |
1652 | ||
1653 | static void set_pool_mode(struct pool *pool, enum pool_mode mode) | |
1654 | { | |
1655 | int r; | |
1656 | ||
1657 | pool->pf.mode = mode; | |
1658 | ||
1659 | switch (mode) { | |
1660 | case PM_FAIL: | |
1661 | DMERR("switching pool to failure mode"); | |
1662 | pool->process_bio = process_bio_fail; | |
1663 | pool->process_discard = process_bio_fail; | |
1664 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1665 | pool->process_prepared_discard = process_prepared_discard_fail; | |
1666 | break; | |
1667 | ||
1668 | case PM_READ_ONLY: | |
1669 | DMERR("switching pool to read-only mode"); | |
1670 | r = dm_pool_abort_metadata(pool->pmd); | |
1671 | if (r) { | |
1672 | DMERR("aborting transaction failed"); | |
1673 | set_pool_mode(pool, PM_FAIL); | |
1674 | } else { | |
1675 | dm_pool_metadata_read_only(pool->pmd); | |
1676 | pool->process_bio = process_bio_read_only; | |
1677 | pool->process_discard = process_discard; | |
1678 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1679 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1680 | } | |
1681 | break; | |
1682 | ||
1683 | case PM_WRITE: | |
1684 | pool->process_bio = process_bio; | |
1685 | pool->process_discard = process_discard; | |
1686 | pool->process_prepared_mapping = process_prepared_mapping; | |
1687 | pool->process_prepared_discard = process_prepared_discard; | |
1688 | break; | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | /*----------------------------------------------------------------*/ | |
1693 | ||
991d9fa0 JT |
1694 | /* |
1695 | * Mapping functions. | |
1696 | */ | |
1697 | ||
1698 | /* | |
1699 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1700 | */ | |
1701 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1702 | { | |
1703 | unsigned long flags; | |
1704 | struct pool *pool = tc->pool; | |
1705 | ||
1706 | spin_lock_irqsave(&pool->lock, flags); | |
1707 | bio_list_add(&pool->deferred_bios, bio); | |
1708 | spin_unlock_irqrestore(&pool->lock, flags); | |
1709 | ||
1710 | wake_worker(pool); | |
1711 | } | |
1712 | ||
a24c2569 | 1713 | static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d JT |
1714 | { |
1715 | struct pool *pool = tc->pool; | |
a24c2569 | 1716 | struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO); |
eb2aa48d JT |
1717 | |
1718 | h->tc = tc; | |
1719 | h->shared_read_entry = NULL; | |
104655fd | 1720 | h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds); |
eb2aa48d JT |
1721 | h->overwrite_mapping = NULL; |
1722 | ||
1723 | return h; | |
1724 | } | |
1725 | ||
991d9fa0 JT |
1726 | /* |
1727 | * Non-blocking function called from the thin target's map function. | |
1728 | */ | |
1729 | static int thin_bio_map(struct dm_target *ti, struct bio *bio, | |
1730 | union map_info *map_context) | |
1731 | { | |
1732 | int r; | |
1733 | struct thin_c *tc = ti->private; | |
1734 | dm_block_t block = get_bio_block(tc, bio); | |
1735 | struct dm_thin_device *td = tc->td; | |
1736 | struct dm_thin_lookup_result result; | |
1737 | ||
eb2aa48d | 1738 | map_context->ptr = thin_hook_bio(tc, bio); |
e49e5829 JT |
1739 | |
1740 | if (get_pool_mode(tc->pool) == PM_FAIL) { | |
1741 | bio_io_error(bio); | |
1742 | return DM_MAPIO_SUBMITTED; | |
1743 | } | |
1744 | ||
104655fd | 1745 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1746 | thin_defer_bio(tc, bio); |
1747 | return DM_MAPIO_SUBMITTED; | |
1748 | } | |
1749 | ||
1750 | r = dm_thin_find_block(td, block, 0, &result); | |
1751 | ||
1752 | /* | |
1753 | * Note that we defer readahead too. | |
1754 | */ | |
1755 | switch (r) { | |
1756 | case 0: | |
1757 | if (unlikely(result.shared)) { | |
1758 | /* | |
1759 | * We have a race condition here between the | |
1760 | * result.shared value returned by the lookup and | |
1761 | * snapshot creation, which may cause new | |
1762 | * sharing. | |
1763 | * | |
1764 | * To avoid this always quiesce the origin before | |
1765 | * taking the snap. You want to do this anyway to | |
1766 | * ensure a consistent application view | |
1767 | * (i.e. lockfs). | |
1768 | * | |
1769 | * More distant ancestors are irrelevant. The | |
1770 | * shared flag will be set in their case. | |
1771 | */ | |
1772 | thin_defer_bio(tc, bio); | |
1773 | r = DM_MAPIO_SUBMITTED; | |
1774 | } else { | |
1775 | remap(tc, bio, result.block); | |
1776 | r = DM_MAPIO_REMAPPED; | |
1777 | } | |
1778 | break; | |
1779 | ||
1780 | case -ENODATA: | |
e49e5829 JT |
1781 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1782 | /* | |
1783 | * This block isn't provisioned, and we have no way | |
1784 | * of doing so. Just error it. | |
1785 | */ | |
1786 | bio_io_error(bio); | |
1787 | r = DM_MAPIO_SUBMITTED; | |
1788 | break; | |
1789 | } | |
1790 | /* fall through */ | |
1791 | ||
1792 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1793 | /* |
1794 | * In future, the failed dm_thin_find_block above could | |
1795 | * provide the hint to load the metadata into cache. | |
1796 | */ | |
991d9fa0 JT |
1797 | thin_defer_bio(tc, bio); |
1798 | r = DM_MAPIO_SUBMITTED; | |
1799 | break; | |
e49e5829 JT |
1800 | |
1801 | default: | |
1802 | /* | |
1803 | * Must always call bio_io_error on failure. | |
1804 | * dm_thin_find_block can fail with -EINVAL if the | |
1805 | * pool is switched to fail-io mode. | |
1806 | */ | |
1807 | bio_io_error(bio); | |
1808 | r = DM_MAPIO_SUBMITTED; | |
1809 | break; | |
991d9fa0 JT |
1810 | } |
1811 | ||
1812 | return r; | |
1813 | } | |
1814 | ||
1815 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1816 | { | |
1817 | int r; | |
1818 | unsigned long flags; | |
1819 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1820 | ||
1821 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1822 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1823 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1824 | ||
1825 | if (!r) { | |
1826 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1827 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1828 | } | |
1829 | ||
1830 | return r; | |
1831 | } | |
1832 | ||
1833 | static void __requeue_bios(struct pool *pool) | |
1834 | { | |
1835 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1836 | bio_list_init(&pool->retry_on_resume_list); | |
1837 | } | |
1838 | ||
1839 | /*---------------------------------------------------------------- | |
1840 | * Binding of control targets to a pool object | |
1841 | *--------------------------------------------------------------*/ | |
1842 | static int bind_control_target(struct pool *pool, struct dm_target *ti) | |
1843 | { | |
1844 | struct pool_c *pt = ti->private; | |
1845 | ||
e49e5829 JT |
1846 | /* |
1847 | * We want to make sure that degraded pools are never upgraded. | |
1848 | */ | |
1849 | enum pool_mode old_mode = pool->pf.mode; | |
1850 | enum pool_mode new_mode = pt->pf.mode; | |
1851 | ||
1852 | if (old_mode > new_mode) | |
1853 | new_mode = old_mode; | |
1854 | ||
991d9fa0 JT |
1855 | pool->ti = ti; |
1856 | pool->low_water_blocks = pt->low_water_blocks; | |
67e2e2b2 | 1857 | pool->pf = pt->pf; |
e49e5829 | 1858 | set_pool_mode(pool, new_mode); |
991d9fa0 | 1859 | |
f402693d MS |
1860 | /* |
1861 | * If discard_passdown was enabled verify that the data device | |
1862 | * supports discards. Disable discard_passdown if not; otherwise | |
1863 | * -EOPNOTSUPP will be returned. | |
1864 | */ | |
e49e5829 | 1865 | /* FIXME: pull this out into a sep fn. */ |
f402693d MS |
1866 | if (pt->pf.discard_passdown) { |
1867 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1868 | if (!q || !blk_queue_discard(q)) { | |
1869 | char buf[BDEVNAME_SIZE]; | |
1870 | DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.", | |
1871 | bdevname(pt->data_dev->bdev, buf)); | |
1872 | pool->pf.discard_passdown = 0; | |
1873 | } | |
1874 | } | |
1875 | ||
991d9fa0 JT |
1876 | return 0; |
1877 | } | |
1878 | ||
1879 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1880 | { | |
1881 | if (pool->ti == ti) | |
1882 | pool->ti = NULL; | |
1883 | } | |
1884 | ||
1885 | /*---------------------------------------------------------------- | |
1886 | * Pool creation | |
1887 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1888 | /* Initialize pool features. */ |
1889 | static void pool_features_init(struct pool_features *pf) | |
1890 | { | |
e49e5829 | 1891 | pf->mode = PM_WRITE; |
67e2e2b2 JT |
1892 | pf->zero_new_blocks = 1; |
1893 | pf->discard_enabled = 1; | |
1894 | pf->discard_passdown = 1; | |
1895 | } | |
1896 | ||
991d9fa0 JT |
1897 | static void __pool_destroy(struct pool *pool) |
1898 | { | |
1899 | __pool_table_remove(pool); | |
1900 | ||
1901 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1902 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1903 | ||
1904 | prison_destroy(pool->prison); | |
1905 | dm_kcopyd_client_destroy(pool->copier); | |
1906 | ||
1907 | if (pool->wq) | |
1908 | destroy_workqueue(pool->wq); | |
1909 | ||
1910 | if (pool->next_mapping) | |
1911 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1912 | mempool_destroy(pool->mapping_pool); | |
1913 | mempool_destroy(pool->endio_hook_pool); | |
1914 | kfree(pool); | |
1915 | } | |
1916 | ||
a24c2569 MS |
1917 | static struct kmem_cache *_new_mapping_cache; |
1918 | static struct kmem_cache *_endio_hook_cache; | |
1919 | ||
991d9fa0 JT |
1920 | static struct pool *pool_create(struct mapped_device *pool_md, |
1921 | struct block_device *metadata_dev, | |
e49e5829 JT |
1922 | unsigned long block_size, |
1923 | int read_only, char **error) | |
991d9fa0 JT |
1924 | { |
1925 | int r; | |
1926 | void *err_p; | |
1927 | struct pool *pool; | |
1928 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1929 | bool format_device = read_only ? false : true; |
991d9fa0 | 1930 | |
e49e5829 | 1931 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1932 | if (IS_ERR(pmd)) { |
1933 | *error = "Error creating metadata object"; | |
1934 | return (struct pool *)pmd; | |
1935 | } | |
1936 | ||
1937 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1938 | if (!pool) { | |
1939 | *error = "Error allocating memory for pool"; | |
1940 | err_p = ERR_PTR(-ENOMEM); | |
1941 | goto bad_pool; | |
1942 | } | |
1943 | ||
1944 | pool->pmd = pmd; | |
1945 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1946 | if (block_size & (block_size - 1)) |
1947 | pool->sectors_per_block_shift = -1; | |
1948 | else | |
1949 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1950 | pool->low_water_blocks = 0; |
67e2e2b2 | 1951 | pool_features_init(&pool->pf); |
991d9fa0 JT |
1952 | pool->prison = prison_create(PRISON_CELLS); |
1953 | if (!pool->prison) { | |
1954 | *error = "Error creating pool's bio prison"; | |
1955 | err_p = ERR_PTR(-ENOMEM); | |
1956 | goto bad_prison; | |
1957 | } | |
1958 | ||
1959 | pool->copier = dm_kcopyd_client_create(); | |
1960 | if (IS_ERR(pool->copier)) { | |
1961 | r = PTR_ERR(pool->copier); | |
1962 | *error = "Error creating pool's kcopyd client"; | |
1963 | err_p = ERR_PTR(r); | |
1964 | goto bad_kcopyd_client; | |
1965 | } | |
1966 | ||
1967 | /* | |
1968 | * Create singlethreaded workqueue that will service all devices | |
1969 | * that use this metadata. | |
1970 | */ | |
1971 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1972 | if (!pool->wq) { | |
1973 | *error = "Error creating pool's workqueue"; | |
1974 | err_p = ERR_PTR(-ENOMEM); | |
1975 | goto bad_wq; | |
1976 | } | |
1977 | ||
1978 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1979 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1980 | spin_lock_init(&pool->lock); |
1981 | bio_list_init(&pool->deferred_bios); | |
1982 | bio_list_init(&pool->deferred_flush_bios); | |
1983 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1984 | INIT_LIST_HEAD(&pool->prepared_discards); |
991d9fa0 JT |
1985 | pool->low_water_triggered = 0; |
1986 | pool->no_free_space = 0; | |
1987 | bio_list_init(&pool->retry_on_resume_list); | |
eb2aa48d | 1988 | ds_init(&pool->shared_read_ds); |
104655fd | 1989 | ds_init(&pool->all_io_ds); |
991d9fa0 JT |
1990 | |
1991 | pool->next_mapping = NULL; | |
a24c2569 MS |
1992 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1993 | _new_mapping_cache); | |
991d9fa0 JT |
1994 | if (!pool->mapping_pool) { |
1995 | *error = "Error creating pool's mapping mempool"; | |
1996 | err_p = ERR_PTR(-ENOMEM); | |
1997 | goto bad_mapping_pool; | |
1998 | } | |
1999 | ||
a24c2569 MS |
2000 | pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE, |
2001 | _endio_hook_cache); | |
991d9fa0 JT |
2002 | if (!pool->endio_hook_pool) { |
2003 | *error = "Error creating pool's endio_hook mempool"; | |
2004 | err_p = ERR_PTR(-ENOMEM); | |
2005 | goto bad_endio_hook_pool; | |
2006 | } | |
2007 | pool->ref_count = 1; | |
905e51b3 | 2008 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2009 | pool->pool_md = pool_md; |
2010 | pool->md_dev = metadata_dev; | |
2011 | __pool_table_insert(pool); | |
2012 | ||
2013 | return pool; | |
2014 | ||
2015 | bad_endio_hook_pool: | |
2016 | mempool_destroy(pool->mapping_pool); | |
2017 | bad_mapping_pool: | |
2018 | destroy_workqueue(pool->wq); | |
2019 | bad_wq: | |
2020 | dm_kcopyd_client_destroy(pool->copier); | |
2021 | bad_kcopyd_client: | |
2022 | prison_destroy(pool->prison); | |
2023 | bad_prison: | |
2024 | kfree(pool); | |
2025 | bad_pool: | |
2026 | if (dm_pool_metadata_close(pmd)) | |
2027 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2028 | ||
2029 | return err_p; | |
2030 | } | |
2031 | ||
2032 | static void __pool_inc(struct pool *pool) | |
2033 | { | |
2034 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2035 | pool->ref_count++; | |
2036 | } | |
2037 | ||
2038 | static void __pool_dec(struct pool *pool) | |
2039 | { | |
2040 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2041 | BUG_ON(!pool->ref_count); | |
2042 | if (!--pool->ref_count) | |
2043 | __pool_destroy(pool); | |
2044 | } | |
2045 | ||
2046 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2047 | struct block_device *metadata_dev, | |
e49e5829 JT |
2048 | unsigned long block_size, int read_only, |
2049 | char **error, int *created) | |
991d9fa0 JT |
2050 | { |
2051 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2052 | ||
2053 | if (pool) { | |
f09996c9 MS |
2054 | if (pool->pool_md != pool_md) { |
2055 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2056 | return ERR_PTR(-EBUSY); |
f09996c9 | 2057 | } |
991d9fa0 JT |
2058 | __pool_inc(pool); |
2059 | ||
2060 | } else { | |
2061 | pool = __pool_table_lookup(pool_md); | |
2062 | if (pool) { | |
f09996c9 MS |
2063 | if (pool->md_dev != metadata_dev) { |
2064 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2065 | return ERR_PTR(-EINVAL); |
f09996c9 | 2066 | } |
991d9fa0 JT |
2067 | __pool_inc(pool); |
2068 | ||
67e2e2b2 | 2069 | } else { |
e49e5829 | 2070 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2071 | *created = 1; |
2072 | } | |
991d9fa0 JT |
2073 | } |
2074 | ||
2075 | return pool; | |
2076 | } | |
2077 | ||
2078 | /*---------------------------------------------------------------- | |
2079 | * Pool target methods | |
2080 | *--------------------------------------------------------------*/ | |
2081 | static void pool_dtr(struct dm_target *ti) | |
2082 | { | |
2083 | struct pool_c *pt = ti->private; | |
2084 | ||
2085 | mutex_lock(&dm_thin_pool_table.mutex); | |
2086 | ||
2087 | unbind_control_target(pt->pool, ti); | |
2088 | __pool_dec(pt->pool); | |
2089 | dm_put_device(ti, pt->metadata_dev); | |
2090 | dm_put_device(ti, pt->data_dev); | |
2091 | kfree(pt); | |
2092 | ||
2093 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2094 | } | |
2095 | ||
991d9fa0 JT |
2096 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2097 | struct dm_target *ti) | |
2098 | { | |
2099 | int r; | |
2100 | unsigned argc; | |
2101 | const char *arg_name; | |
2102 | ||
2103 | static struct dm_arg _args[] = { | |
67e2e2b2 | 2104 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2105 | }; |
2106 | ||
2107 | /* | |
2108 | * No feature arguments supplied. | |
2109 | */ | |
2110 | if (!as->argc) | |
2111 | return 0; | |
2112 | ||
2113 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2114 | if (r) | |
2115 | return -EINVAL; | |
2116 | ||
2117 | while (argc && !r) { | |
2118 | arg_name = dm_shift_arg(as); | |
2119 | argc--; | |
2120 | ||
e49e5829 | 2121 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
991d9fa0 | 2122 | pf->zero_new_blocks = 0; |
e49e5829 JT |
2123 | |
2124 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
67e2e2b2 | 2125 | pf->discard_enabled = 0; |
e49e5829 JT |
2126 | |
2127 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
67e2e2b2 | 2128 | pf->discard_passdown = 0; |
991d9fa0 | 2129 | |
e49e5829 JT |
2130 | else if (!strcasecmp(arg_name, "read_only")) |
2131 | pf->mode = PM_READ_ONLY; | |
2132 | ||
2133 | else { | |
2134 | ti->error = "Unrecognised pool feature requested"; | |
2135 | r = -EINVAL; | |
2136 | break; | |
2137 | } | |
991d9fa0 JT |
2138 | } |
2139 | ||
2140 | return r; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * thin-pool <metadata dev> <data dev> | |
2145 | * <data block size (sectors)> | |
2146 | * <low water mark (blocks)> | |
2147 | * [<#feature args> [<arg>]*] | |
2148 | * | |
2149 | * Optional feature arguments are: | |
2150 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2151 | * ignore_discard: disable discard |
2152 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
2153 | */ |
2154 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2155 | { | |
67e2e2b2 | 2156 | int r, pool_created = 0; |
991d9fa0 JT |
2157 | struct pool_c *pt; |
2158 | struct pool *pool; | |
2159 | struct pool_features pf; | |
2160 | struct dm_arg_set as; | |
2161 | struct dm_dev *data_dev; | |
2162 | unsigned long block_size; | |
2163 | dm_block_t low_water_blocks; | |
2164 | struct dm_dev *metadata_dev; | |
2165 | sector_t metadata_dev_size; | |
c4a69ecd | 2166 | char b[BDEVNAME_SIZE]; |
991d9fa0 JT |
2167 | |
2168 | /* | |
2169 | * FIXME Remove validation from scope of lock. | |
2170 | */ | |
2171 | mutex_lock(&dm_thin_pool_table.mutex); | |
2172 | ||
2173 | if (argc < 4) { | |
2174 | ti->error = "Invalid argument count"; | |
2175 | r = -EINVAL; | |
2176 | goto out_unlock; | |
2177 | } | |
2178 | as.argc = argc; | |
2179 | as.argv = argv; | |
2180 | ||
2181 | r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev); | |
2182 | if (r) { | |
2183 | ti->error = "Error opening metadata block device"; | |
2184 | goto out_unlock; | |
2185 | } | |
2186 | ||
2187 | metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
c4a69ecd MS |
2188 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
2189 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
2190 | bdevname(metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS); | |
991d9fa0 JT |
2191 | |
2192 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2193 | if (r) { | |
2194 | ti->error = "Error getting data device"; | |
2195 | goto out_metadata; | |
2196 | } | |
2197 | ||
2198 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2199 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2200 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2201 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2202 | ti->error = "Invalid block size"; |
2203 | r = -EINVAL; | |
2204 | goto out; | |
2205 | } | |
2206 | ||
2207 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2208 | ti->error = "Invalid low water mark"; | |
2209 | r = -EINVAL; | |
2210 | goto out; | |
2211 | } | |
2212 | ||
2213 | /* | |
2214 | * Set default pool features. | |
2215 | */ | |
67e2e2b2 | 2216 | pool_features_init(&pf); |
991d9fa0 JT |
2217 | |
2218 | dm_consume_args(&as, 4); | |
2219 | r = parse_pool_features(&as, &pf, ti); | |
2220 | if (r) | |
2221 | goto out; | |
2222 | ||
2223 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); | |
2224 | if (!pt) { | |
2225 | r = -ENOMEM; | |
2226 | goto out; | |
2227 | } | |
2228 | ||
2229 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2230 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2231 | if (IS_ERR(pool)) { |
2232 | r = PTR_ERR(pool); | |
2233 | goto out_free_pt; | |
2234 | } | |
2235 | ||
67e2e2b2 JT |
2236 | /* |
2237 | * 'pool_created' reflects whether this is the first table load. | |
2238 | * Top level discard support is not allowed to be changed after | |
2239 | * initial load. This would require a pool reload to trigger thin | |
2240 | * device changes. | |
2241 | */ | |
2242 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2243 | ti->error = "Discard support cannot be disabled once enabled"; | |
2244 | r = -EINVAL; | |
2245 | goto out_flags_changed; | |
2246 | } | |
2247 | ||
55f2b8bd MS |
2248 | /* |
2249 | * The block layer requires discard_granularity to be a power of 2. | |
2250 | */ | |
2251 | if (pf.discard_enabled && !is_power_of_2(block_size)) { | |
2252 | ti->error = "Discard support must be disabled when the block size is not a power of 2"; | |
2253 | r = -EINVAL; | |
2254 | goto out_flags_changed; | |
2255 | } | |
2256 | ||
991d9fa0 JT |
2257 | pt->pool = pool; |
2258 | pt->ti = ti; | |
2259 | pt->metadata_dev = metadata_dev; | |
2260 | pt->data_dev = data_dev; | |
2261 | pt->low_water_blocks = low_water_blocks; | |
67e2e2b2 | 2262 | pt->pf = pf; |
991d9fa0 | 2263 | ti->num_flush_requests = 1; |
67e2e2b2 JT |
2264 | /* |
2265 | * Only need to enable discards if the pool should pass | |
2266 | * them down to the data device. The thin device's discard | |
2267 | * processing will cause mappings to be removed from the btree. | |
2268 | */ | |
2269 | if (pf.discard_enabled && pf.discard_passdown) { | |
2270 | ti->num_discard_requests = 1; | |
2271 | /* | |
2272 | * Setting 'discards_supported' circumvents the normal | |
2273 | * stacking of discard limits (this keeps the pool and | |
2274 | * thin devices' discard limits consistent). | |
2275 | */ | |
0ac55489 | 2276 | ti->discards_supported = true; |
67e2e2b2 | 2277 | } |
991d9fa0 JT |
2278 | ti->private = pt; |
2279 | ||
2280 | pt->callbacks.congested_fn = pool_is_congested; | |
2281 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2282 | ||
2283 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2284 | ||
2285 | return 0; | |
2286 | ||
67e2e2b2 JT |
2287 | out_flags_changed: |
2288 | __pool_dec(pool); | |
991d9fa0 JT |
2289 | out_free_pt: |
2290 | kfree(pt); | |
2291 | out: | |
2292 | dm_put_device(ti, data_dev); | |
2293 | out_metadata: | |
2294 | dm_put_device(ti, metadata_dev); | |
2295 | out_unlock: | |
2296 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2297 | ||
2298 | return r; | |
2299 | } | |
2300 | ||
2301 | static int pool_map(struct dm_target *ti, struct bio *bio, | |
2302 | union map_info *map_context) | |
2303 | { | |
2304 | int r; | |
2305 | struct pool_c *pt = ti->private; | |
2306 | struct pool *pool = pt->pool; | |
2307 | unsigned long flags; | |
2308 | ||
2309 | /* | |
2310 | * As this is a singleton target, ti->begin is always zero. | |
2311 | */ | |
2312 | spin_lock_irqsave(&pool->lock, flags); | |
2313 | bio->bi_bdev = pt->data_dev->bdev; | |
2314 | r = DM_MAPIO_REMAPPED; | |
2315 | spin_unlock_irqrestore(&pool->lock, flags); | |
2316 | ||
2317 | return r; | |
2318 | } | |
2319 | ||
2320 | /* | |
2321 | * Retrieves the number of blocks of the data device from | |
2322 | * the superblock and compares it to the actual device size, | |
2323 | * thus resizing the data device in case it has grown. | |
2324 | * | |
2325 | * This both copes with opening preallocated data devices in the ctr | |
2326 | * being followed by a resume | |
2327 | * -and- | |
2328 | * calling the resume method individually after userspace has | |
2329 | * grown the data device in reaction to a table event. | |
2330 | */ | |
2331 | static int pool_preresume(struct dm_target *ti) | |
2332 | { | |
2333 | int r; | |
2334 | struct pool_c *pt = ti->private; | |
2335 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2336 | sector_t data_size = ti->len; |
2337 | dm_block_t sb_data_size; | |
991d9fa0 JT |
2338 | |
2339 | /* | |
2340 | * Take control of the pool object. | |
2341 | */ | |
2342 | r = bind_control_target(pool, ti); | |
2343 | if (r) | |
2344 | return r; | |
2345 | ||
55f2b8bd MS |
2346 | (void) sector_div(data_size, pool->sectors_per_block); |
2347 | ||
991d9fa0 JT |
2348 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2349 | if (r) { | |
2350 | DMERR("failed to retrieve data device size"); | |
2351 | return r; | |
2352 | } | |
2353 | ||
2354 | if (data_size < sb_data_size) { | |
2355 | DMERR("pool target too small, is %llu blocks (expected %llu)", | |
55f2b8bd | 2356 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2357 | return -EINVAL; |
2358 | ||
2359 | } else if (data_size > sb_data_size) { | |
2360 | r = dm_pool_resize_data_dev(pool->pmd, data_size); | |
2361 | if (r) { | |
2362 | DMERR("failed to resize data device"); | |
e49e5829 JT |
2363 | /* FIXME Stricter than necessary: Rollback transaction instead here */ |
2364 | set_pool_mode(pool, PM_READ_ONLY); | |
991d9fa0 JT |
2365 | return r; |
2366 | } | |
2367 | ||
e49e5829 | 2368 | (void) commit_or_fallback(pool); |
991d9fa0 JT |
2369 | } |
2370 | ||
2371 | return 0; | |
2372 | } | |
2373 | ||
2374 | static void pool_resume(struct dm_target *ti) | |
2375 | { | |
2376 | struct pool_c *pt = ti->private; | |
2377 | struct pool *pool = pt->pool; | |
2378 | unsigned long flags; | |
2379 | ||
2380 | spin_lock_irqsave(&pool->lock, flags); | |
2381 | pool->low_water_triggered = 0; | |
2382 | pool->no_free_space = 0; | |
2383 | __requeue_bios(pool); | |
2384 | spin_unlock_irqrestore(&pool->lock, flags); | |
2385 | ||
905e51b3 | 2386 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2387 | } |
2388 | ||
2389 | static void pool_postsuspend(struct dm_target *ti) | |
2390 | { | |
991d9fa0 JT |
2391 | struct pool_c *pt = ti->private; |
2392 | struct pool *pool = pt->pool; | |
2393 | ||
905e51b3 | 2394 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2395 | flush_workqueue(pool->wq); |
e49e5829 | 2396 | (void) commit_or_fallback(pool); |
991d9fa0 JT |
2397 | } |
2398 | ||
2399 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2400 | { | |
2401 | if (argc != args_required) { | |
2402 | DMWARN("Message received with %u arguments instead of %u.", | |
2403 | argc, args_required); | |
2404 | return -EINVAL; | |
2405 | } | |
2406 | ||
2407 | return 0; | |
2408 | } | |
2409 | ||
2410 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2411 | { | |
2412 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2413 | *dev_id <= MAX_DEV_ID) | |
2414 | return 0; | |
2415 | ||
2416 | if (warning) | |
2417 | DMWARN("Message received with invalid device id: %s", arg); | |
2418 | ||
2419 | return -EINVAL; | |
2420 | } | |
2421 | ||
2422 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2423 | { | |
2424 | dm_thin_id dev_id; | |
2425 | int r; | |
2426 | ||
2427 | r = check_arg_count(argc, 2); | |
2428 | if (r) | |
2429 | return r; | |
2430 | ||
2431 | r = read_dev_id(argv[1], &dev_id, 1); | |
2432 | if (r) | |
2433 | return r; | |
2434 | ||
2435 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2436 | if (r) { | |
2437 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2438 | argv[1]); | |
2439 | return r; | |
2440 | } | |
2441 | ||
2442 | return 0; | |
2443 | } | |
2444 | ||
2445 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2446 | { | |
2447 | dm_thin_id dev_id; | |
2448 | dm_thin_id origin_dev_id; | |
2449 | int r; | |
2450 | ||
2451 | r = check_arg_count(argc, 3); | |
2452 | if (r) | |
2453 | return r; | |
2454 | ||
2455 | r = read_dev_id(argv[1], &dev_id, 1); | |
2456 | if (r) | |
2457 | return r; | |
2458 | ||
2459 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2460 | if (r) | |
2461 | return r; | |
2462 | ||
2463 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2464 | if (r) { | |
2465 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2466 | argv[1], argv[2]); | |
2467 | return r; | |
2468 | } | |
2469 | ||
2470 | return 0; | |
2471 | } | |
2472 | ||
2473 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2474 | { | |
2475 | dm_thin_id dev_id; | |
2476 | int r; | |
2477 | ||
2478 | r = check_arg_count(argc, 2); | |
2479 | if (r) | |
2480 | return r; | |
2481 | ||
2482 | r = read_dev_id(argv[1], &dev_id, 1); | |
2483 | if (r) | |
2484 | return r; | |
2485 | ||
2486 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2487 | if (r) | |
2488 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2489 | ||
2490 | return r; | |
2491 | } | |
2492 | ||
2493 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2494 | { | |
2495 | dm_thin_id old_id, new_id; | |
2496 | int r; | |
2497 | ||
2498 | r = check_arg_count(argc, 3); | |
2499 | if (r) | |
2500 | return r; | |
2501 | ||
2502 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2503 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2504 | return -EINVAL; | |
2505 | } | |
2506 | ||
2507 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2508 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2509 | return -EINVAL; | |
2510 | } | |
2511 | ||
2512 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2513 | if (r) { | |
2514 | DMWARN("Failed to change transaction id from %s to %s.", | |
2515 | argv[1], argv[2]); | |
2516 | return r; | |
2517 | } | |
2518 | ||
2519 | return 0; | |
2520 | } | |
2521 | ||
cc8394d8 JT |
2522 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2523 | { | |
2524 | int r; | |
2525 | ||
2526 | r = check_arg_count(argc, 1); | |
2527 | if (r) | |
2528 | return r; | |
2529 | ||
e49e5829 | 2530 | (void) commit_or_fallback(pool); |
0d200aef | 2531 | |
cc8394d8 JT |
2532 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2533 | if (r) | |
2534 | DMWARN("reserve_metadata_snap message failed."); | |
2535 | ||
2536 | return r; | |
2537 | } | |
2538 | ||
2539 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2540 | { | |
2541 | int r; | |
2542 | ||
2543 | r = check_arg_count(argc, 1); | |
2544 | if (r) | |
2545 | return r; | |
2546 | ||
2547 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2548 | if (r) | |
2549 | DMWARN("release_metadata_snap message failed."); | |
2550 | ||
2551 | return r; | |
2552 | } | |
2553 | ||
991d9fa0 JT |
2554 | /* |
2555 | * Messages supported: | |
2556 | * create_thin <dev_id> | |
2557 | * create_snap <dev_id> <origin_id> | |
2558 | * delete <dev_id> | |
2559 | * trim <dev_id> <new_size_in_sectors> | |
2560 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2561 | * reserve_metadata_snap |
2562 | * release_metadata_snap | |
991d9fa0 JT |
2563 | */ |
2564 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2565 | { | |
2566 | int r = -EINVAL; | |
2567 | struct pool_c *pt = ti->private; | |
2568 | struct pool *pool = pt->pool; | |
2569 | ||
2570 | if (!strcasecmp(argv[0], "create_thin")) | |
2571 | r = process_create_thin_mesg(argc, argv, pool); | |
2572 | ||
2573 | else if (!strcasecmp(argv[0], "create_snap")) | |
2574 | r = process_create_snap_mesg(argc, argv, pool); | |
2575 | ||
2576 | else if (!strcasecmp(argv[0], "delete")) | |
2577 | r = process_delete_mesg(argc, argv, pool); | |
2578 | ||
2579 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2580 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2581 | ||
cc8394d8 JT |
2582 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2583 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2584 | ||
2585 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2586 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2587 | ||
991d9fa0 JT |
2588 | else |
2589 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2590 | ||
e49e5829 JT |
2591 | if (!r) |
2592 | (void) commit_or_fallback(pool); | |
991d9fa0 JT |
2593 | |
2594 | return r; | |
2595 | } | |
2596 | ||
e49e5829 JT |
2597 | static void emit_flags(struct pool_features *pf, char *result, |
2598 | unsigned sz, unsigned maxlen) | |
2599 | { | |
2600 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
2601 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY); | |
2602 | DMEMIT("%u ", count); | |
2603 | ||
2604 | if (!pf->zero_new_blocks) | |
2605 | DMEMIT("skip_block_zeroing "); | |
2606 | ||
2607 | if (!pf->discard_enabled) | |
2608 | DMEMIT("ignore_discard "); | |
2609 | ||
2610 | if (!pf->discard_passdown) | |
2611 | DMEMIT("no_discard_passdown "); | |
2612 | ||
2613 | if (pf->mode == PM_READ_ONLY) | |
2614 | DMEMIT("read_only "); | |
2615 | } | |
2616 | ||
991d9fa0 JT |
2617 | /* |
2618 | * Status line is: | |
2619 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2620 | * <used data sectors>/<total data sectors> <held metadata root> | |
2621 | */ | |
2622 | static int pool_status(struct dm_target *ti, status_type_t type, | |
1f4e0ff0 | 2623 | unsigned status_flags, char *result, unsigned maxlen) |
991d9fa0 | 2624 | { |
e49e5829 | 2625 | int r; |
991d9fa0 JT |
2626 | unsigned sz = 0; |
2627 | uint64_t transaction_id; | |
2628 | dm_block_t nr_free_blocks_data; | |
2629 | dm_block_t nr_free_blocks_metadata; | |
2630 | dm_block_t nr_blocks_data; | |
2631 | dm_block_t nr_blocks_metadata; | |
2632 | dm_block_t held_root; | |
2633 | char buf[BDEVNAME_SIZE]; | |
2634 | char buf2[BDEVNAME_SIZE]; | |
2635 | struct pool_c *pt = ti->private; | |
2636 | struct pool *pool = pt->pool; | |
2637 | ||
2638 | switch (type) { | |
2639 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2640 | if (get_pool_mode(pool) == PM_FAIL) { |
2641 | DMEMIT("Fail"); | |
2642 | break; | |
2643 | } | |
2644 | ||
1f4e0ff0 AK |
2645 | /* Commit to ensure statistics aren't out-of-date */ |
2646 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
2647 | (void) commit_or_fallback(pool); | |
2648 | ||
991d9fa0 JT |
2649 | r = dm_pool_get_metadata_transaction_id(pool->pmd, |
2650 | &transaction_id); | |
2651 | if (r) | |
2652 | return r; | |
2653 | ||
2654 | r = dm_pool_get_free_metadata_block_count(pool->pmd, | |
2655 | &nr_free_blocks_metadata); | |
2656 | if (r) | |
2657 | return r; | |
2658 | ||
2659 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
2660 | if (r) | |
2661 | return r; | |
2662 | ||
2663 | r = dm_pool_get_free_block_count(pool->pmd, | |
2664 | &nr_free_blocks_data); | |
2665 | if (r) | |
2666 | return r; | |
2667 | ||
2668 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
2669 | if (r) | |
2670 | return r; | |
2671 | ||
cc8394d8 | 2672 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
991d9fa0 JT |
2673 | if (r) |
2674 | return r; | |
2675 | ||
2676 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2677 | (unsigned long long)transaction_id, | |
2678 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2679 | (unsigned long long)nr_blocks_metadata, | |
2680 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2681 | (unsigned long long)nr_blocks_data); | |
2682 | ||
2683 | if (held_root) | |
e49e5829 JT |
2684 | DMEMIT("%llu ", held_root); |
2685 | else | |
2686 | DMEMIT("- "); | |
2687 | ||
2688 | if (pool->pf.mode == PM_READ_ONLY) | |
2689 | DMEMIT("ro "); | |
991d9fa0 | 2690 | else |
e49e5829 JT |
2691 | DMEMIT("rw "); |
2692 | ||
2693 | if (pool->pf.discard_enabled && pool->pf.discard_passdown) | |
2694 | DMEMIT("discard_passdown"); | |
2695 | else | |
2696 | DMEMIT("no_discard_passdown"); | |
991d9fa0 JT |
2697 | |
2698 | break; | |
2699 | ||
2700 | case STATUSTYPE_TABLE: | |
2701 | DMEMIT("%s %s %lu %llu ", | |
2702 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2703 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2704 | (unsigned long)pool->sectors_per_block, | |
2705 | (unsigned long long)pt->low_water_blocks); | |
e49e5829 | 2706 | emit_flags(&pt->pf, result, sz, maxlen); |
991d9fa0 JT |
2707 | break; |
2708 | } | |
2709 | ||
2710 | return 0; | |
2711 | } | |
2712 | ||
2713 | static int pool_iterate_devices(struct dm_target *ti, | |
2714 | iterate_devices_callout_fn fn, void *data) | |
2715 | { | |
2716 | struct pool_c *pt = ti->private; | |
2717 | ||
2718 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2719 | } | |
2720 | ||
2721 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2722 | struct bio_vec *biovec, int max_size) | |
2723 | { | |
2724 | struct pool_c *pt = ti->private; | |
2725 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2726 | ||
2727 | if (!q->merge_bvec_fn) | |
2728 | return max_size; | |
2729 | ||
2730 | bvm->bi_bdev = pt->data_dev->bdev; | |
2731 | ||
2732 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2733 | } | |
2734 | ||
104655fd JT |
2735 | static void set_discard_limits(struct pool *pool, struct queue_limits *limits) |
2736 | { | |
67e2e2b2 JT |
2737 | /* |
2738 | * FIXME: these limits may be incompatible with the pool's data device | |
2739 | */ | |
104655fd JT |
2740 | limits->max_discard_sectors = pool->sectors_per_block; |
2741 | ||
2742 | /* | |
2743 | * This is just a hint, and not enforced. We have to cope with | |
49296309 MP |
2744 | * bios that cover a block partially. A discard that spans a block |
2745 | * boundary is not sent to this target. | |
104655fd JT |
2746 | */ |
2747 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
67e2e2b2 | 2748 | limits->discard_zeroes_data = pool->pf.zero_new_blocks; |
104655fd JT |
2749 | } |
2750 | ||
991d9fa0 JT |
2751 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2752 | { | |
2753 | struct pool_c *pt = ti->private; | |
2754 | struct pool *pool = pt->pool; | |
2755 | ||
2756 | blk_limits_io_min(limits, 0); | |
2757 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
67e2e2b2 JT |
2758 | if (pool->pf.discard_enabled) |
2759 | set_discard_limits(pool, limits); | |
991d9fa0 JT |
2760 | } |
2761 | ||
2762 | static struct target_type pool_target = { | |
2763 | .name = "thin-pool", | |
2764 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2765 | DM_TARGET_IMMUTABLE, | |
e49e5829 | 2766 | .version = {1, 3, 0}, |
991d9fa0 JT |
2767 | .module = THIS_MODULE, |
2768 | .ctr = pool_ctr, | |
2769 | .dtr = pool_dtr, | |
2770 | .map = pool_map, | |
2771 | .postsuspend = pool_postsuspend, | |
2772 | .preresume = pool_preresume, | |
2773 | .resume = pool_resume, | |
2774 | .message = pool_message, | |
2775 | .status = pool_status, | |
2776 | .merge = pool_merge, | |
2777 | .iterate_devices = pool_iterate_devices, | |
2778 | .io_hints = pool_io_hints, | |
2779 | }; | |
2780 | ||
2781 | /*---------------------------------------------------------------- | |
2782 | * Thin target methods | |
2783 | *--------------------------------------------------------------*/ | |
2784 | static void thin_dtr(struct dm_target *ti) | |
2785 | { | |
2786 | struct thin_c *tc = ti->private; | |
2787 | ||
2788 | mutex_lock(&dm_thin_pool_table.mutex); | |
2789 | ||
2790 | __pool_dec(tc->pool); | |
2791 | dm_pool_close_thin_device(tc->td); | |
2792 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2793 | if (tc->origin_dev) |
2794 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2795 | kfree(tc); |
2796 | ||
2797 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2798 | } | |
2799 | ||
2800 | /* | |
2801 | * Thin target parameters: | |
2802 | * | |
2dd9c257 | 2803 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2804 | * |
2805 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2806 | * dev_id: the internal device identifier | |
2dd9c257 | 2807 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2808 | * |
2809 | * If the pool device has discards disabled, they get disabled for the thin | |
2810 | * device as well. | |
991d9fa0 JT |
2811 | */ |
2812 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2813 | { | |
2814 | int r; | |
2815 | struct thin_c *tc; | |
2dd9c257 | 2816 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2817 | struct mapped_device *pool_md; |
2818 | ||
2819 | mutex_lock(&dm_thin_pool_table.mutex); | |
2820 | ||
2dd9c257 | 2821 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2822 | ti->error = "Invalid argument count"; |
2823 | r = -EINVAL; | |
2824 | goto out_unlock; | |
2825 | } | |
2826 | ||
2827 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2828 | if (!tc) { | |
2829 | ti->error = "Out of memory"; | |
2830 | r = -ENOMEM; | |
2831 | goto out_unlock; | |
2832 | } | |
2833 | ||
2dd9c257 JT |
2834 | if (argc == 3) { |
2835 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2836 | if (r) { | |
2837 | ti->error = "Error opening origin device"; | |
2838 | goto bad_origin_dev; | |
2839 | } | |
2840 | tc->origin_dev = origin_dev; | |
2841 | } | |
2842 | ||
991d9fa0 JT |
2843 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2844 | if (r) { | |
2845 | ti->error = "Error opening pool device"; | |
2846 | goto bad_pool_dev; | |
2847 | } | |
2848 | tc->pool_dev = pool_dev; | |
2849 | ||
2850 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2851 | ti->error = "Invalid device id"; | |
2852 | r = -EINVAL; | |
2853 | goto bad_common; | |
2854 | } | |
2855 | ||
2856 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2857 | if (!pool_md) { | |
2858 | ti->error = "Couldn't get pool mapped device"; | |
2859 | r = -EINVAL; | |
2860 | goto bad_common; | |
2861 | } | |
2862 | ||
2863 | tc->pool = __pool_table_lookup(pool_md); | |
2864 | if (!tc->pool) { | |
2865 | ti->error = "Couldn't find pool object"; | |
2866 | r = -EINVAL; | |
2867 | goto bad_pool_lookup; | |
2868 | } | |
2869 | __pool_inc(tc->pool); | |
2870 | ||
e49e5829 JT |
2871 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2872 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
2873 | goto bad_thin_open; | |
2874 | } | |
2875 | ||
991d9fa0 JT |
2876 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
2877 | if (r) { | |
2878 | ti->error = "Couldn't open thin internal device"; | |
2879 | goto bad_thin_open; | |
2880 | } | |
2881 | ||
542f9038 MS |
2882 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2883 | if (r) | |
2884 | goto bad_thin_open; | |
2885 | ||
991d9fa0 | 2886 | ti->num_flush_requests = 1; |
16ad3d10 | 2887 | ti->flush_supported = true; |
67e2e2b2 JT |
2888 | |
2889 | /* In case the pool supports discards, pass them on. */ | |
2890 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 2891 | ti->discards_supported = true; |
67e2e2b2 | 2892 | ti->num_discard_requests = 1; |
0ac55489 | 2893 | ti->discard_zeroes_data_unsupported = true; |
49296309 | 2894 | /* Discard requests must be split on a block boundary */ |
0ac55489 | 2895 | ti->split_discard_requests = true; |
67e2e2b2 | 2896 | } |
991d9fa0 JT |
2897 | |
2898 | dm_put(pool_md); | |
2899 | ||
2900 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2901 | ||
2902 | return 0; | |
2903 | ||
2904 | bad_thin_open: | |
2905 | __pool_dec(tc->pool); | |
2906 | bad_pool_lookup: | |
2907 | dm_put(pool_md); | |
2908 | bad_common: | |
2909 | dm_put_device(ti, tc->pool_dev); | |
2910 | bad_pool_dev: | |
2dd9c257 JT |
2911 | if (tc->origin_dev) |
2912 | dm_put_device(ti, tc->origin_dev); | |
2913 | bad_origin_dev: | |
991d9fa0 JT |
2914 | kfree(tc); |
2915 | out_unlock: | |
2916 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2917 | ||
2918 | return r; | |
2919 | } | |
2920 | ||
2921 | static int thin_map(struct dm_target *ti, struct bio *bio, | |
2922 | union map_info *map_context) | |
2923 | { | |
6efd6e83 | 2924 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 JT |
2925 | |
2926 | return thin_bio_map(ti, bio, map_context); | |
2927 | } | |
2928 | ||
eb2aa48d JT |
2929 | static int thin_endio(struct dm_target *ti, |
2930 | struct bio *bio, int err, | |
2931 | union map_info *map_context) | |
2932 | { | |
2933 | unsigned long flags; | |
a24c2569 | 2934 | struct dm_thin_endio_hook *h = map_context->ptr; |
eb2aa48d | 2935 | struct list_head work; |
a24c2569 | 2936 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2937 | struct pool *pool = h->tc->pool; |
2938 | ||
2939 | if (h->shared_read_entry) { | |
2940 | INIT_LIST_HEAD(&work); | |
2941 | ds_dec(h->shared_read_entry, &work); | |
2942 | ||
2943 | spin_lock_irqsave(&pool->lock, flags); | |
2944 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2945 | list_del(&m->list); | |
2946 | m->quiesced = 1; | |
2947 | __maybe_add_mapping(m); | |
2948 | } | |
2949 | spin_unlock_irqrestore(&pool->lock, flags); | |
2950 | } | |
2951 | ||
104655fd JT |
2952 | if (h->all_io_entry) { |
2953 | INIT_LIST_HEAD(&work); | |
2954 | ds_dec(h->all_io_entry, &work); | |
c3a0ce2e | 2955 | spin_lock_irqsave(&pool->lock, flags); |
104655fd JT |
2956 | list_for_each_entry_safe(m, tmp, &work, list) |
2957 | list_add(&m->list, &pool->prepared_discards); | |
c3a0ce2e | 2958 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
2959 | } |
2960 | ||
eb2aa48d JT |
2961 | mempool_free(h, pool->endio_hook_pool); |
2962 | ||
2963 | return 0; | |
2964 | } | |
2965 | ||
991d9fa0 JT |
2966 | static void thin_postsuspend(struct dm_target *ti) |
2967 | { | |
2968 | if (dm_noflush_suspending(ti)) | |
2969 | requeue_io((struct thin_c *)ti->private); | |
2970 | } | |
2971 | ||
2972 | /* | |
2973 | * <nr mapped sectors> <highest mapped sector> | |
2974 | */ | |
2975 | static int thin_status(struct dm_target *ti, status_type_t type, | |
1f4e0ff0 | 2976 | unsigned status_flags, char *result, unsigned maxlen) |
991d9fa0 JT |
2977 | { |
2978 | int r; | |
2979 | ssize_t sz = 0; | |
2980 | dm_block_t mapped, highest; | |
2981 | char buf[BDEVNAME_SIZE]; | |
2982 | struct thin_c *tc = ti->private; | |
2983 | ||
e49e5829 JT |
2984 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2985 | DMEMIT("Fail"); | |
2986 | return 0; | |
2987 | } | |
2988 | ||
991d9fa0 JT |
2989 | if (!tc->td) |
2990 | DMEMIT("-"); | |
2991 | else { | |
2992 | switch (type) { | |
2993 | case STATUSTYPE_INFO: | |
2994 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
2995 | if (r) | |
2996 | return r; | |
2997 | ||
2998 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
2999 | if (r < 0) | |
3000 | return r; | |
3001 | ||
3002 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3003 | if (r) | |
3004 | DMEMIT("%llu", ((highest + 1) * | |
3005 | tc->pool->sectors_per_block) - 1); | |
3006 | else | |
3007 | DMEMIT("-"); | |
3008 | break; | |
3009 | ||
3010 | case STATUSTYPE_TABLE: | |
3011 | DMEMIT("%s %lu", | |
3012 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3013 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3014 | if (tc->origin_dev) |
3015 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3016 | break; |
3017 | } | |
3018 | } | |
3019 | ||
3020 | return 0; | |
3021 | } | |
3022 | ||
3023 | static int thin_iterate_devices(struct dm_target *ti, | |
3024 | iterate_devices_callout_fn fn, void *data) | |
3025 | { | |
55f2b8bd | 3026 | sector_t blocks; |
991d9fa0 | 3027 | struct thin_c *tc = ti->private; |
55f2b8bd | 3028 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3029 | |
3030 | /* | |
3031 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3032 | * we follow a more convoluted path through to the pool's target. | |
3033 | */ | |
55f2b8bd | 3034 | if (!pool->ti) |
991d9fa0 JT |
3035 | return 0; /* nothing is bound */ |
3036 | ||
55f2b8bd MS |
3037 | blocks = pool->ti->len; |
3038 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3039 | if (blocks) |
55f2b8bd | 3040 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3041 | |
3042 | return 0; | |
3043 | } | |
3044 | ||
3045 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
3046 | { | |
3047 | struct thin_c *tc = ti->private; | |
104655fd | 3048 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3049 | |
3050 | blk_limits_io_min(limits, 0); | |
104655fd JT |
3051 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
3052 | set_discard_limits(pool, limits); | |
991d9fa0 JT |
3053 | } |
3054 | ||
3055 | static struct target_type thin_target = { | |
3056 | .name = "thin", | |
e49e5829 | 3057 | .version = {1, 3, 0}, |
991d9fa0 JT |
3058 | .module = THIS_MODULE, |
3059 | .ctr = thin_ctr, | |
3060 | .dtr = thin_dtr, | |
3061 | .map = thin_map, | |
eb2aa48d | 3062 | .end_io = thin_endio, |
991d9fa0 JT |
3063 | .postsuspend = thin_postsuspend, |
3064 | .status = thin_status, | |
3065 | .iterate_devices = thin_iterate_devices, | |
3066 | .io_hints = thin_io_hints, | |
3067 | }; | |
3068 | ||
3069 | /*----------------------------------------------------------------*/ | |
3070 | ||
3071 | static int __init dm_thin_init(void) | |
3072 | { | |
3073 | int r; | |
3074 | ||
3075 | pool_table_init(); | |
3076 | ||
3077 | r = dm_register_target(&thin_target); | |
3078 | if (r) | |
3079 | return r; | |
3080 | ||
3081 | r = dm_register_target(&pool_target); | |
3082 | if (r) | |
a24c2569 MS |
3083 | goto bad_pool_target; |
3084 | ||
3085 | r = -ENOMEM; | |
3086 | ||
3087 | _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0); | |
3088 | if (!_cell_cache) | |
3089 | goto bad_cell_cache; | |
3090 | ||
3091 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); | |
3092 | if (!_new_mapping_cache) | |
3093 | goto bad_new_mapping_cache; | |
3094 | ||
3095 | _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0); | |
3096 | if (!_endio_hook_cache) | |
3097 | goto bad_endio_hook_cache; | |
3098 | ||
3099 | return 0; | |
3100 | ||
3101 | bad_endio_hook_cache: | |
3102 | kmem_cache_destroy(_new_mapping_cache); | |
3103 | bad_new_mapping_cache: | |
3104 | kmem_cache_destroy(_cell_cache); | |
3105 | bad_cell_cache: | |
3106 | dm_unregister_target(&pool_target); | |
3107 | bad_pool_target: | |
3108 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3109 | |
3110 | return r; | |
3111 | } | |
3112 | ||
3113 | static void dm_thin_exit(void) | |
3114 | { | |
3115 | dm_unregister_target(&thin_target); | |
3116 | dm_unregister_target(&pool_target); | |
a24c2569 MS |
3117 | |
3118 | kmem_cache_destroy(_cell_cache); | |
3119 | kmem_cache_destroy(_new_mapping_cache); | |
3120 | kmem_cache_destroy(_endio_hook_cache); | |
991d9fa0 JT |
3121 | } |
3122 | ||
3123 | module_init(dm_thin_init); | |
3124 | module_exit(dm_thin_exit); | |
3125 | ||
7cab8bf1 | 3126 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3127 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3128 | MODULE_LICENSE("GPL"); |