Commit | Line | Data |
---|---|---|
991d9fa0 | 1 | /* |
e49e5829 | 2 | * Copyright (C) 2011-2012 Red Hat UK. |
991d9fa0 JT |
3 | * |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-thin-metadata.h" | |
4f81a417 | 8 | #include "dm-bio-prison.h" |
1f4e0ff0 | 9 | #include "dm.h" |
991d9fa0 JT |
10 | |
11 | #include <linux/device-mapper.h> | |
12 | #include <linux/dm-io.h> | |
13 | #include <linux/dm-kcopyd.h> | |
14 | #include <linux/list.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/slab.h> | |
18 | ||
19 | #define DM_MSG_PREFIX "thin" | |
20 | ||
21 | /* | |
22 | * Tunable constants | |
23 | */ | |
7768ed33 | 24 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
25 | #define MAPPING_POOL_SIZE 1024 |
26 | #define PRISON_CELLS 1024 | |
905e51b3 | 27 | #define COMMIT_PERIOD HZ |
991d9fa0 | 28 | |
df5d2e90 MP |
29 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
30 | "A percentage of time allocated for copy on write"); | |
31 | ||
991d9fa0 JT |
32 | /* |
33 | * The block size of the device holding pool data must be | |
34 | * between 64KB and 1GB. | |
35 | */ | |
36 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
37 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
38 | ||
991d9fa0 JT |
39 | /* |
40 | * Device id is restricted to 24 bits. | |
41 | */ | |
42 | #define MAX_DEV_ID ((1 << 24) - 1) | |
43 | ||
44 | /* | |
45 | * How do we handle breaking sharing of data blocks? | |
46 | * ================================================= | |
47 | * | |
48 | * We use a standard copy-on-write btree to store the mappings for the | |
49 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
50 | * the data). When you take an internal snapshot you clone the root node | |
51 | * of the origin btree. After this there is no concept of an origin or a | |
52 | * snapshot. They are just two device trees that happen to point to the | |
53 | * same data blocks. | |
54 | * | |
55 | * When we get a write in we decide if it's to a shared data block using | |
56 | * some timestamp magic. If it is, we have to break sharing. | |
57 | * | |
58 | * Let's say we write to a shared block in what was the origin. The | |
59 | * steps are: | |
60 | * | |
61 | * i) plug io further to this physical block. (see bio_prison code). | |
62 | * | |
63 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 64 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
65 | * |
66 | * iii) copy the data block to a newly allocate block. This step can be | |
67 | * missed out if the io covers the block. (schedule_copy). | |
68 | * | |
69 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 70 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
71 | * sharing of btree nodes between the two devices. Breaking sharing only |
72 | * effects the btree of that specific device. Btrees for the other | |
73 | * devices that share the block never change. The btree for the origin | |
74 | * device as it was after the last commit is untouched, ie. we're using | |
75 | * persistent data structures in the functional programming sense. | |
76 | * | |
77 | * v) unplug io to this physical block, including the io that triggered | |
78 | * the breaking of sharing. | |
79 | * | |
80 | * Steps (ii) and (iii) occur in parallel. | |
81 | * | |
82 | * The metadata _doesn't_ need to be committed before the io continues. We | |
83 | * get away with this because the io is always written to a _new_ block. | |
84 | * If there's a crash, then: | |
85 | * | |
86 | * - The origin mapping will point to the old origin block (the shared | |
87 | * one). This will contain the data as it was before the io that triggered | |
88 | * the breaking of sharing came in. | |
89 | * | |
90 | * - The snap mapping still points to the old block. As it would after | |
91 | * the commit. | |
92 | * | |
93 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
94 | * will continue to think that data block in the snapshot device is shared | |
95 | * even after the write to the origin has broken sharing. I suspect data | |
96 | * blocks will typically be shared by many different devices, so we're | |
97 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
98 | * devices that reference this data block. At the moment I think the | |
99 | * benefits far, far outweigh the disadvantages. | |
100 | */ | |
101 | ||
102 | /*----------------------------------------------------------------*/ | |
103 | ||
991d9fa0 JT |
104 | /* |
105 | * Key building. | |
106 | */ | |
107 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 108 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
109 | { |
110 | key->virtual = 0; | |
111 | key->dev = dm_thin_dev_id(td); | |
112 | key->block = b; | |
113 | } | |
114 | ||
115 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 116 | struct dm_cell_key *key) |
991d9fa0 JT |
117 | { |
118 | key->virtual = 1; | |
119 | key->dev = dm_thin_dev_id(td); | |
120 | key->block = b; | |
121 | } | |
122 | ||
123 | /*----------------------------------------------------------------*/ | |
124 | ||
125 | /* | |
126 | * A pool device ties together a metadata device and a data device. It | |
127 | * also provides the interface for creating and destroying internal | |
128 | * devices. | |
129 | */ | |
a24c2569 | 130 | struct dm_thin_new_mapping; |
67e2e2b2 | 131 | |
e49e5829 JT |
132 | /* |
133 | * The pool runs in 3 modes. Ordered in degraded order for comparisons. | |
134 | */ | |
135 | enum pool_mode { | |
136 | PM_WRITE, /* metadata may be changed */ | |
137 | PM_READ_ONLY, /* metadata may not be changed */ | |
138 | PM_FAIL, /* all I/O fails */ | |
139 | }; | |
140 | ||
67e2e2b2 | 141 | struct pool_features { |
e49e5829 JT |
142 | enum pool_mode mode; |
143 | ||
9bc142dd MS |
144 | bool zero_new_blocks:1; |
145 | bool discard_enabled:1; | |
146 | bool discard_passdown:1; | |
67e2e2b2 JT |
147 | }; |
148 | ||
e49e5829 JT |
149 | struct thin_c; |
150 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
151 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
152 | ||
991d9fa0 JT |
153 | struct pool { |
154 | struct list_head list; | |
155 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
156 | ||
157 | struct mapped_device *pool_md; | |
158 | struct block_device *md_dev; | |
159 | struct dm_pool_metadata *pmd; | |
160 | ||
991d9fa0 | 161 | dm_block_t low_water_blocks; |
55f2b8bd | 162 | uint32_t sectors_per_block; |
f9a8e0cd | 163 | int sectors_per_block_shift; |
991d9fa0 | 164 | |
67e2e2b2 | 165 | struct pool_features pf; |
88a6621b JT |
166 | bool low_water_triggered:1; /* A dm event has been sent */ |
167 | bool no_free_space:1; /* A -ENOSPC warning has been issued */ | |
991d9fa0 | 168 | |
44feb387 | 169 | struct dm_bio_prison *prison; |
991d9fa0 JT |
170 | struct dm_kcopyd_client *copier; |
171 | ||
172 | struct workqueue_struct *wq; | |
173 | struct work_struct worker; | |
905e51b3 | 174 | struct delayed_work waker; |
991d9fa0 | 175 | |
905e51b3 | 176 | unsigned long last_commit_jiffies; |
55f2b8bd | 177 | unsigned ref_count; |
991d9fa0 JT |
178 | |
179 | spinlock_t lock; | |
180 | struct bio_list deferred_bios; | |
181 | struct bio_list deferred_flush_bios; | |
182 | struct list_head prepared_mappings; | |
104655fd | 183 | struct list_head prepared_discards; |
991d9fa0 JT |
184 | |
185 | struct bio_list retry_on_resume_list; | |
186 | ||
44feb387 MS |
187 | struct dm_deferred_set *shared_read_ds; |
188 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 189 | |
a24c2569 | 190 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 191 | mempool_t *mapping_pool; |
e49e5829 JT |
192 | |
193 | process_bio_fn process_bio; | |
194 | process_bio_fn process_discard; | |
195 | ||
196 | process_mapping_fn process_prepared_mapping; | |
197 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
198 | }; |
199 | ||
e49e5829 JT |
200 | static enum pool_mode get_pool_mode(struct pool *pool); |
201 | static void set_pool_mode(struct pool *pool, enum pool_mode mode); | |
202 | ||
991d9fa0 JT |
203 | /* |
204 | * Target context for a pool. | |
205 | */ | |
206 | struct pool_c { | |
207 | struct dm_target *ti; | |
208 | struct pool *pool; | |
209 | struct dm_dev *data_dev; | |
210 | struct dm_dev *metadata_dev; | |
211 | struct dm_target_callbacks callbacks; | |
212 | ||
213 | dm_block_t low_water_blocks; | |
0424caa1 MS |
214 | struct pool_features requested_pf; /* Features requested during table load */ |
215 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
216 | }; |
217 | ||
218 | /* | |
219 | * Target context for a thin. | |
220 | */ | |
221 | struct thin_c { | |
222 | struct dm_dev *pool_dev; | |
2dd9c257 | 223 | struct dm_dev *origin_dev; |
991d9fa0 JT |
224 | dm_thin_id dev_id; |
225 | ||
226 | struct pool *pool; | |
227 | struct dm_thin_device *td; | |
228 | }; | |
229 | ||
230 | /*----------------------------------------------------------------*/ | |
231 | ||
025b9685 JT |
232 | /* |
233 | * wake_worker() is used when new work is queued and when pool_resume is | |
234 | * ready to continue deferred IO processing. | |
235 | */ | |
236 | static void wake_worker(struct pool *pool) | |
237 | { | |
238 | queue_work(pool->wq, &pool->worker); | |
239 | } | |
240 | ||
241 | /*----------------------------------------------------------------*/ | |
242 | ||
6beca5eb JT |
243 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
244 | struct dm_bio_prison_cell **cell_result) | |
245 | { | |
246 | int r; | |
247 | struct dm_bio_prison_cell *cell_prealloc; | |
248 | ||
249 | /* | |
250 | * Allocate a cell from the prison's mempool. | |
251 | * This might block but it can't fail. | |
252 | */ | |
253 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
254 | ||
255 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
256 | if (r) | |
257 | /* | |
258 | * We reused an old cell; we can get rid of | |
259 | * the new one. | |
260 | */ | |
261 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
262 | ||
263 | return r; | |
264 | } | |
265 | ||
266 | static void cell_release(struct pool *pool, | |
267 | struct dm_bio_prison_cell *cell, | |
268 | struct bio_list *bios) | |
269 | { | |
270 | dm_cell_release(pool->prison, cell, bios); | |
271 | dm_bio_prison_free_cell(pool->prison, cell); | |
272 | } | |
273 | ||
274 | static void cell_release_no_holder(struct pool *pool, | |
275 | struct dm_bio_prison_cell *cell, | |
276 | struct bio_list *bios) | |
277 | { | |
278 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
279 | dm_bio_prison_free_cell(pool->prison, cell); | |
280 | } | |
281 | ||
025b9685 JT |
282 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
283 | struct dm_bio_prison_cell *cell) | |
284 | { | |
285 | struct pool *pool = tc->pool; | |
286 | unsigned long flags; | |
287 | ||
288 | spin_lock_irqsave(&pool->lock, flags); | |
289 | dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios); | |
290 | spin_unlock_irqrestore(&pool->lock, flags); | |
291 | ||
292 | wake_worker(pool); | |
293 | } | |
294 | ||
6beca5eb JT |
295 | static void cell_error(struct pool *pool, |
296 | struct dm_bio_prison_cell *cell) | |
297 | { | |
298 | dm_cell_error(pool->prison, cell); | |
299 | dm_bio_prison_free_cell(pool->prison, cell); | |
300 | } | |
301 | ||
302 | /*----------------------------------------------------------------*/ | |
303 | ||
991d9fa0 JT |
304 | /* |
305 | * A global list of pools that uses a struct mapped_device as a key. | |
306 | */ | |
307 | static struct dm_thin_pool_table { | |
308 | struct mutex mutex; | |
309 | struct list_head pools; | |
310 | } dm_thin_pool_table; | |
311 | ||
312 | static void pool_table_init(void) | |
313 | { | |
314 | mutex_init(&dm_thin_pool_table.mutex); | |
315 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
316 | } | |
317 | ||
318 | static void __pool_table_insert(struct pool *pool) | |
319 | { | |
320 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
321 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
322 | } | |
323 | ||
324 | static void __pool_table_remove(struct pool *pool) | |
325 | { | |
326 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
327 | list_del(&pool->list); | |
328 | } | |
329 | ||
330 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
331 | { | |
332 | struct pool *pool = NULL, *tmp; | |
333 | ||
334 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
335 | ||
336 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
337 | if (tmp->pool_md == md) { | |
338 | pool = tmp; | |
339 | break; | |
340 | } | |
341 | } | |
342 | ||
343 | return pool; | |
344 | } | |
345 | ||
346 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
347 | { | |
348 | struct pool *pool = NULL, *tmp; | |
349 | ||
350 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
351 | ||
352 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
353 | if (tmp->md_dev == md_dev) { | |
354 | pool = tmp; | |
355 | break; | |
356 | } | |
357 | } | |
358 | ||
359 | return pool; | |
360 | } | |
361 | ||
362 | /*----------------------------------------------------------------*/ | |
363 | ||
a24c2569 | 364 | struct dm_thin_endio_hook { |
eb2aa48d | 365 | struct thin_c *tc; |
44feb387 MS |
366 | struct dm_deferred_entry *shared_read_entry; |
367 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 368 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
369 | }; |
370 | ||
991d9fa0 JT |
371 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
372 | { | |
373 | struct bio *bio; | |
374 | struct bio_list bios; | |
375 | ||
376 | bio_list_init(&bios); | |
377 | bio_list_merge(&bios, master); | |
378 | bio_list_init(master); | |
379 | ||
380 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 381 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 382 | |
eb2aa48d | 383 | if (h->tc == tc) |
991d9fa0 JT |
384 | bio_endio(bio, DM_ENDIO_REQUEUE); |
385 | else | |
386 | bio_list_add(master, bio); | |
387 | } | |
388 | } | |
389 | ||
390 | static void requeue_io(struct thin_c *tc) | |
391 | { | |
392 | struct pool *pool = tc->pool; | |
393 | unsigned long flags; | |
394 | ||
395 | spin_lock_irqsave(&pool->lock, flags); | |
396 | __requeue_bio_list(tc, &pool->deferred_bios); | |
397 | __requeue_bio_list(tc, &pool->retry_on_resume_list); | |
398 | spin_unlock_irqrestore(&pool->lock, flags); | |
399 | } | |
400 | ||
401 | /* | |
402 | * This section of code contains the logic for processing a thin device's IO. | |
403 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
404 | * but most is exclusively called from the thin target rather than the thin-pool | |
405 | * target. | |
406 | */ | |
407 | ||
58f77a21 MS |
408 | static bool block_size_is_power_of_two(struct pool *pool) |
409 | { | |
410 | return pool->sectors_per_block_shift >= 0; | |
411 | } | |
412 | ||
991d9fa0 JT |
413 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
414 | { | |
58f77a21 | 415 | struct pool *pool = tc->pool; |
55f2b8bd MS |
416 | sector_t block_nr = bio->bi_sector; |
417 | ||
58f77a21 MS |
418 | if (block_size_is_power_of_two(pool)) |
419 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 420 | else |
58f77a21 | 421 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
422 | |
423 | return block_nr; | |
991d9fa0 JT |
424 | } |
425 | ||
426 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
427 | { | |
428 | struct pool *pool = tc->pool; | |
55f2b8bd | 429 | sector_t bi_sector = bio->bi_sector; |
991d9fa0 JT |
430 | |
431 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 432 | if (block_size_is_power_of_two(pool)) |
f9a8e0cd MP |
433 | bio->bi_sector = (block << pool->sectors_per_block_shift) | |
434 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 MS |
435 | else |
436 | bio->bi_sector = (block * pool->sectors_per_block) + | |
437 | sector_div(bi_sector, pool->sectors_per_block); | |
991d9fa0 JT |
438 | } |
439 | ||
2dd9c257 JT |
440 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
441 | { | |
442 | bio->bi_bdev = tc->origin_dev->bdev; | |
443 | } | |
444 | ||
4afdd680 JT |
445 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
446 | { | |
447 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
448 | dm_thin_changed_this_transaction(tc->td); | |
449 | } | |
450 | ||
e8088073 JT |
451 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
452 | { | |
453 | struct dm_thin_endio_hook *h; | |
454 | ||
455 | if (bio->bi_rw & REQ_DISCARD) | |
456 | return; | |
457 | ||
59c3d2c6 | 458 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
459 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
460 | } | |
461 | ||
2dd9c257 | 462 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
463 | { |
464 | struct pool *pool = tc->pool; | |
465 | unsigned long flags; | |
466 | ||
e49e5829 JT |
467 | if (!bio_triggers_commit(tc, bio)) { |
468 | generic_make_request(bio); | |
469 | return; | |
470 | } | |
471 | ||
991d9fa0 | 472 | /* |
e49e5829 JT |
473 | * Complete bio with an error if earlier I/O caused changes to |
474 | * the metadata that can't be committed e.g, due to I/O errors | |
475 | * on the metadata device. | |
991d9fa0 | 476 | */ |
e49e5829 JT |
477 | if (dm_thin_aborted_changes(tc->td)) { |
478 | bio_io_error(bio); | |
479 | return; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Batch together any bios that trigger commits and then issue a | |
484 | * single commit for them in process_deferred_bios(). | |
485 | */ | |
486 | spin_lock_irqsave(&pool->lock, flags); | |
487 | bio_list_add(&pool->deferred_flush_bios, bio); | |
488 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
489 | } |
490 | ||
2dd9c257 JT |
491 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
492 | { | |
493 | remap_to_origin(tc, bio); | |
494 | issue(tc, bio); | |
495 | } | |
496 | ||
497 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
498 | dm_block_t block) | |
499 | { | |
500 | remap(tc, bio, block); | |
501 | issue(tc, bio); | |
502 | } | |
503 | ||
991d9fa0 JT |
504 | /*----------------------------------------------------------------*/ |
505 | ||
506 | /* | |
507 | * Bio endio functions. | |
508 | */ | |
a24c2569 | 509 | struct dm_thin_new_mapping { |
991d9fa0 JT |
510 | struct list_head list; |
511 | ||
7f214665 MS |
512 | bool quiesced:1; |
513 | bool prepared:1; | |
514 | bool pass_discard:1; | |
515 | bool definitely_not_shared:1; | |
991d9fa0 | 516 | |
7f214665 | 517 | int err; |
991d9fa0 JT |
518 | struct thin_c *tc; |
519 | dm_block_t virt_block; | |
520 | dm_block_t data_block; | |
a24c2569 | 521 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
522 | |
523 | /* | |
524 | * If the bio covers the whole area of a block then we can avoid | |
525 | * zeroing or copying. Instead this bio is hooked. The bio will | |
526 | * still be in the cell, so care has to be taken to avoid issuing | |
527 | * the bio twice. | |
528 | */ | |
529 | struct bio *bio; | |
530 | bio_end_io_t *saved_bi_end_io; | |
531 | }; | |
532 | ||
a24c2569 | 533 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
534 | { |
535 | struct pool *pool = m->tc->pool; | |
536 | ||
eb2aa48d | 537 | if (m->quiesced && m->prepared) { |
daec338b | 538 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
539 | wake_worker(pool); |
540 | } | |
541 | } | |
542 | ||
543 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
544 | { | |
545 | unsigned long flags; | |
a24c2569 | 546 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
547 | struct pool *pool = m->tc->pool; |
548 | ||
549 | m->err = read_err || write_err ? -EIO : 0; | |
550 | ||
551 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 552 | m->prepared = true; |
991d9fa0 JT |
553 | __maybe_add_mapping(m); |
554 | spin_unlock_irqrestore(&pool->lock, flags); | |
555 | } | |
556 | ||
557 | static void overwrite_endio(struct bio *bio, int err) | |
558 | { | |
559 | unsigned long flags; | |
59c3d2c6 | 560 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 561 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
562 | struct pool *pool = m->tc->pool; |
563 | ||
564 | m->err = err; | |
565 | ||
566 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 567 | m->prepared = true; |
991d9fa0 JT |
568 | __maybe_add_mapping(m); |
569 | spin_unlock_irqrestore(&pool->lock, flags); | |
570 | } | |
571 | ||
991d9fa0 JT |
572 | /*----------------------------------------------------------------*/ |
573 | ||
574 | /* | |
575 | * Workqueue. | |
576 | */ | |
577 | ||
578 | /* | |
579 | * Prepared mapping jobs. | |
580 | */ | |
581 | ||
582 | /* | |
583 | * This sends the bios in the cell back to the deferred_bios list. | |
584 | */ | |
2aab3850 | 585 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
586 | { |
587 | struct pool *pool = tc->pool; | |
588 | unsigned long flags; | |
589 | ||
590 | spin_lock_irqsave(&pool->lock, flags); | |
6beca5eb | 591 | cell_release(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
592 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
593 | ||
594 | wake_worker(pool); | |
595 | } | |
596 | ||
597 | /* | |
6beca5eb | 598 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 599 | */ |
f286ba0e | 600 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 601 | { |
991d9fa0 JT |
602 | struct pool *pool = tc->pool; |
603 | unsigned long flags; | |
604 | ||
991d9fa0 | 605 | spin_lock_irqsave(&pool->lock, flags); |
6beca5eb | 606 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
607 | spin_unlock_irqrestore(&pool->lock, flags); |
608 | ||
609 | wake_worker(pool); | |
610 | } | |
611 | ||
e49e5829 JT |
612 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
613 | { | |
614 | if (m->bio) | |
615 | m->bio->bi_end_io = m->saved_bi_end_io; | |
6beca5eb | 616 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
617 | list_del(&m->list); |
618 | mempool_free(m, m->tc->pool->mapping_pool); | |
619 | } | |
025b9685 | 620 | |
a24c2569 | 621 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
622 | { |
623 | struct thin_c *tc = m->tc; | |
6beca5eb | 624 | struct pool *pool = tc->pool; |
991d9fa0 JT |
625 | struct bio *bio; |
626 | int r; | |
627 | ||
628 | bio = m->bio; | |
629 | if (bio) | |
630 | bio->bi_end_io = m->saved_bi_end_io; | |
631 | ||
632 | if (m->err) { | |
6beca5eb | 633 | cell_error(pool, m->cell); |
905386f8 | 634 | goto out; |
991d9fa0 JT |
635 | } |
636 | ||
637 | /* | |
638 | * Commit the prepared block into the mapping btree. | |
639 | * Any I/O for this block arriving after this point will get | |
640 | * remapped to it directly. | |
641 | */ | |
642 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
643 | if (r) { | |
fafc7a81 JT |
644 | DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d", |
645 | dm_device_name(pool->pool_md), r); | |
646 | set_pool_mode(pool, PM_READ_ONLY); | |
6beca5eb | 647 | cell_error(pool, m->cell); |
905386f8 | 648 | goto out; |
991d9fa0 JT |
649 | } |
650 | ||
651 | /* | |
652 | * Release any bios held while the block was being provisioned. | |
653 | * If we are processing a write bio that completely covers the block, | |
654 | * we already processed it so can ignore it now when processing | |
655 | * the bios in the cell. | |
656 | */ | |
657 | if (bio) { | |
f286ba0e | 658 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
659 | bio_endio(bio, 0); |
660 | } else | |
2aab3850 | 661 | cell_defer(tc, m->cell); |
991d9fa0 | 662 | |
905386f8 | 663 | out: |
991d9fa0 | 664 | list_del(&m->list); |
6beca5eb | 665 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
666 | } |
667 | ||
e49e5829 | 668 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 669 | { |
104655fd JT |
670 | struct thin_c *tc = m->tc; |
671 | ||
e49e5829 | 672 | bio_io_error(m->bio); |
f286ba0e JT |
673 | cell_defer_no_holder(tc, m->cell); |
674 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
675 | mempool_free(m, tc->pool->mapping_pool); |
676 | } | |
677 | ||
678 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
679 | { | |
680 | struct thin_c *tc = m->tc; | |
104655fd | 681 | |
e8088073 | 682 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
683 | cell_defer_no_holder(tc, m->cell); |
684 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 685 | |
104655fd | 686 | if (m->pass_discard) |
19fa1a67 JT |
687 | if (m->definitely_not_shared) |
688 | remap_and_issue(tc, m->bio, m->data_block); | |
689 | else { | |
690 | bool used = false; | |
691 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
692 | bio_endio(m->bio, 0); | |
693 | else | |
694 | remap_and_issue(tc, m->bio, m->data_block); | |
695 | } | |
104655fd JT |
696 | else |
697 | bio_endio(m->bio, 0); | |
698 | ||
104655fd JT |
699 | mempool_free(m, tc->pool->mapping_pool); |
700 | } | |
701 | ||
e49e5829 JT |
702 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
703 | { | |
704 | int r; | |
705 | struct thin_c *tc = m->tc; | |
706 | ||
707 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
708 | if (r) | |
c397741c | 709 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
710 | |
711 | process_prepared_discard_passdown(m); | |
712 | } | |
713 | ||
104655fd | 714 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 715 | process_mapping_fn *fn) |
991d9fa0 JT |
716 | { |
717 | unsigned long flags; | |
718 | struct list_head maps; | |
a24c2569 | 719 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
720 | |
721 | INIT_LIST_HEAD(&maps); | |
722 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 723 | list_splice_init(head, &maps); |
991d9fa0 JT |
724 | spin_unlock_irqrestore(&pool->lock, flags); |
725 | ||
726 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 727 | (*fn)(m); |
991d9fa0 JT |
728 | } |
729 | ||
730 | /* | |
731 | * Deferred bio jobs. | |
732 | */ | |
104655fd | 733 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 734 | { |
f9a8e0cd | 735 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
736 | } |
737 | ||
738 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
739 | { | |
740 | return (bio_data_dir(bio) == WRITE) && | |
741 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
742 | } |
743 | ||
744 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
745 | bio_end_io_t *fn) | |
746 | { | |
747 | *save = bio->bi_end_io; | |
748 | bio->bi_end_io = fn; | |
749 | } | |
750 | ||
751 | static int ensure_next_mapping(struct pool *pool) | |
752 | { | |
753 | if (pool->next_mapping) | |
754 | return 0; | |
755 | ||
756 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
757 | ||
758 | return pool->next_mapping ? 0 : -ENOMEM; | |
759 | } | |
760 | ||
a24c2569 | 761 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 762 | { |
16961b04 | 763 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
764 | |
765 | BUG_ON(!pool->next_mapping); | |
766 | ||
16961b04 MS |
767 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
768 | INIT_LIST_HEAD(&m->list); | |
769 | m->bio = NULL; | |
770 | ||
991d9fa0 JT |
771 | pool->next_mapping = NULL; |
772 | ||
16961b04 | 773 | return m; |
991d9fa0 JT |
774 | } |
775 | ||
776 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
777 | struct dm_dev *origin, dm_block_t data_origin, |
778 | dm_block_t data_dest, | |
a24c2569 | 779 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
780 | { |
781 | int r; | |
782 | struct pool *pool = tc->pool; | |
a24c2569 | 783 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 784 | |
991d9fa0 JT |
785 | m->tc = tc; |
786 | m->virt_block = virt_block; | |
787 | m->data_block = data_dest; | |
788 | m->cell = cell; | |
991d9fa0 | 789 | |
44feb387 | 790 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 791 | m->quiesced = true; |
991d9fa0 JT |
792 | |
793 | /* | |
794 | * IO to pool_dev remaps to the pool target's data_dev. | |
795 | * | |
796 | * If the whole block of data is being overwritten, we can issue the | |
797 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
798 | */ | |
799 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 800 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 801 | |
eb2aa48d | 802 | h->overwrite_mapping = m; |
991d9fa0 JT |
803 | m->bio = bio; |
804 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 805 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
806 | remap_and_issue(tc, bio, data_dest); |
807 | } else { | |
808 | struct dm_io_region from, to; | |
809 | ||
2dd9c257 | 810 | from.bdev = origin->bdev; |
991d9fa0 JT |
811 | from.sector = data_origin * pool->sectors_per_block; |
812 | from.count = pool->sectors_per_block; | |
813 | ||
814 | to.bdev = tc->pool_dev->bdev; | |
815 | to.sector = data_dest * pool->sectors_per_block; | |
816 | to.count = pool->sectors_per_block; | |
817 | ||
818 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
819 | 0, copy_complete, m); | |
820 | if (r < 0) { | |
821 | mempool_free(m, pool->mapping_pool); | |
c397741c | 822 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 823 | cell_error(pool, cell); |
991d9fa0 JT |
824 | } |
825 | } | |
826 | } | |
827 | ||
2dd9c257 JT |
828 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
829 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 830 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
831 | { |
832 | schedule_copy(tc, virt_block, tc->pool_dev, | |
833 | data_origin, data_dest, cell, bio); | |
834 | } | |
835 | ||
836 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
837 | dm_block_t data_dest, | |
a24c2569 | 838 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
839 | { |
840 | schedule_copy(tc, virt_block, tc->origin_dev, | |
841 | virt_block, data_dest, cell, bio); | |
842 | } | |
843 | ||
991d9fa0 | 844 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 845 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
846 | struct bio *bio) |
847 | { | |
848 | struct pool *pool = tc->pool; | |
a24c2569 | 849 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 850 | |
7f214665 MS |
851 | m->quiesced = true; |
852 | m->prepared = false; | |
991d9fa0 JT |
853 | m->tc = tc; |
854 | m->virt_block = virt_block; | |
855 | m->data_block = data_block; | |
856 | m->cell = cell; | |
991d9fa0 JT |
857 | |
858 | /* | |
859 | * If the whole block of data is being overwritten or we are not | |
860 | * zeroing pre-existing data, we can issue the bio immediately. | |
861 | * Otherwise we use kcopyd to zero the data first. | |
862 | */ | |
67e2e2b2 | 863 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
864 | process_prepared_mapping(m); |
865 | ||
866 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 867 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 868 | |
eb2aa48d | 869 | h->overwrite_mapping = m; |
991d9fa0 JT |
870 | m->bio = bio; |
871 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 872 | inc_all_io_entry(pool, bio); |
991d9fa0 | 873 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
874 | } else { |
875 | int r; | |
876 | struct dm_io_region to; | |
877 | ||
878 | to.bdev = tc->pool_dev->bdev; | |
879 | to.sector = data_block * pool->sectors_per_block; | |
880 | to.count = pool->sectors_per_block; | |
881 | ||
882 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
883 | if (r < 0) { | |
884 | mempool_free(m, pool->mapping_pool); | |
c397741c | 885 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 886 | cell_error(pool, cell); |
991d9fa0 JT |
887 | } |
888 | } | |
889 | } | |
890 | ||
e49e5829 JT |
891 | /* |
892 | * A non-zero return indicates read_only or fail_io mode. | |
893 | * Many callers don't care about the return value. | |
894 | */ | |
020cc3b5 | 895 | static int commit(struct pool *pool) |
e49e5829 JT |
896 | { |
897 | int r; | |
898 | ||
899 | if (get_pool_mode(pool) != PM_WRITE) | |
900 | return -EINVAL; | |
901 | ||
020cc3b5 JT |
902 | r = dm_pool_commit_metadata(pool->pmd); |
903 | if (r) { | |
904 | DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d", | |
905 | dm_device_name(pool->pool_md), r); | |
e49e5829 | 906 | set_pool_mode(pool, PM_READ_ONLY); |
020cc3b5 | 907 | } |
e49e5829 JT |
908 | |
909 | return r; | |
910 | } | |
911 | ||
88a6621b JT |
912 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
913 | { | |
914 | unsigned long flags; | |
915 | ||
916 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
917 | DMWARN("%s: reached low water mark for data device: sending event.", | |
918 | dm_device_name(pool->pool_md)); | |
919 | spin_lock_irqsave(&pool->lock, flags); | |
920 | pool->low_water_triggered = true; | |
921 | spin_unlock_irqrestore(&pool->lock, flags); | |
922 | dm_table_event(pool->ti->table); | |
923 | } | |
924 | } | |
925 | ||
991d9fa0 JT |
926 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
927 | { | |
928 | int r; | |
929 | dm_block_t free_blocks; | |
930 | unsigned long flags; | |
931 | struct pool *pool = tc->pool; | |
932 | ||
94563bad MS |
933 | /* |
934 | * Once no_free_space is set we must not allow allocation to succeed. | |
935 | * Otherwise it is difficult to explain, debug, test and support. | |
936 | */ | |
937 | if (pool->no_free_space) | |
938 | return -ENOSPC; | |
939 | ||
8d30abff JT |
940 | if (get_pool_mode(pool) != PM_WRITE) |
941 | return -EINVAL; | |
942 | ||
991d9fa0 JT |
943 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
944 | if (r) | |
945 | return r; | |
946 | ||
88a6621b | 947 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
948 | |
949 | if (!free_blocks) { | |
94563bad MS |
950 | /* |
951 | * Try to commit to see if that will free up some | |
952 | * more space. | |
953 | */ | |
020cc3b5 JT |
954 | r = commit(pool); |
955 | if (r) | |
956 | return r; | |
991d9fa0 | 957 | |
94563bad MS |
958 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
959 | if (r) | |
960 | return r; | |
991d9fa0 | 961 | |
94563bad MS |
962 | /* |
963 | * If we still have no space we set a flag to avoid | |
964 | * doing all this checking and return -ENOSPC. This | |
965 | * flag serves as a latch that disallows allocations from | |
966 | * this pool until the admin takes action (e.g. resize or | |
967 | * table reload). | |
968 | */ | |
969 | if (!free_blocks) { | |
4a02b34e | 970 | DMWARN("%s: no free data space available.", |
94563bad MS |
971 | dm_device_name(pool->pool_md)); |
972 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 973 | pool->no_free_space = true; |
94563bad MS |
974 | spin_unlock_irqrestore(&pool->lock, flags); |
975 | return -ENOSPC; | |
991d9fa0 JT |
976 | } |
977 | } | |
978 | ||
979 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e MS |
980 | if (r) { |
981 | if (r == -ENOSPC && | |
982 | !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) && | |
983 | !free_blocks) { | |
984 | DMWARN("%s: no free metadata space available.", | |
985 | dm_device_name(pool->pool_md)); | |
986 | set_pool_mode(pool, PM_READ_ONLY); | |
987 | } | |
991d9fa0 | 988 | return r; |
4a02b34e | 989 | } |
991d9fa0 JT |
990 | |
991 | return 0; | |
992 | } | |
993 | ||
994 | /* | |
995 | * If we have run out of space, queue bios until the device is | |
996 | * resumed, presumably after having been reloaded with more space. | |
997 | */ | |
998 | static void retry_on_resume(struct bio *bio) | |
999 | { | |
59c3d2c6 | 1000 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1001 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1002 | struct pool *pool = tc->pool; |
1003 | unsigned long flags; | |
1004 | ||
1005 | spin_lock_irqsave(&pool->lock, flags); | |
1006 | bio_list_add(&pool->retry_on_resume_list, bio); | |
1007 | spin_unlock_irqrestore(&pool->lock, flags); | |
1008 | } | |
1009 | ||
6beca5eb | 1010 | static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1011 | { |
1012 | struct bio *bio; | |
1013 | struct bio_list bios; | |
1014 | ||
1015 | bio_list_init(&bios); | |
6beca5eb | 1016 | cell_release(pool, cell, &bios); |
991d9fa0 JT |
1017 | |
1018 | while ((bio = bio_list_pop(&bios))) | |
1019 | retry_on_resume(bio); | |
1020 | } | |
1021 | ||
104655fd JT |
1022 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1023 | { | |
1024 | int r; | |
c3a0ce2e | 1025 | unsigned long flags; |
104655fd | 1026 | struct pool *pool = tc->pool; |
a24c2569 | 1027 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1028 | struct dm_cell_key key, key2; |
104655fd JT |
1029 | dm_block_t block = get_bio_block(tc, bio); |
1030 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1031 | struct dm_thin_new_mapping *m; |
104655fd JT |
1032 | |
1033 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1034 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1035 | return; |
1036 | ||
1037 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1038 | switch (r) { | |
1039 | case 0: | |
1040 | /* | |
1041 | * Check nobody is fiddling with this pool block. This can | |
1042 | * happen if someone's in the process of breaking sharing | |
1043 | * on this block. | |
1044 | */ | |
1045 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1046 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1047 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1048 | break; |
1049 | } | |
1050 | ||
1051 | if (io_overlaps_block(pool, bio)) { | |
1052 | /* | |
1053 | * IO may still be going to the destination block. We must | |
1054 | * quiesce before we can do the removal. | |
1055 | */ | |
1056 | m = get_next_mapping(pool); | |
1057 | m->tc = tc; | |
19fa1a67 JT |
1058 | m->pass_discard = pool->pf.discard_passdown; |
1059 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1060 | m->virt_block = block; |
1061 | m->data_block = lookup_result.block; | |
1062 | m->cell = cell; | |
1063 | m->cell2 = cell2; | |
104655fd JT |
1064 | m->bio = bio; |
1065 | ||
44feb387 | 1066 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1067 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1068 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1069 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1070 | wake_worker(pool); |
1071 | } | |
1072 | } else { | |
e8088073 | 1073 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1074 | cell_defer_no_holder(tc, cell); |
1075 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1076 | |
104655fd | 1077 | /* |
49296309 MP |
1078 | * The DM core makes sure that the discard doesn't span |
1079 | * a block boundary. So we submit the discard of a | |
1080 | * partial block appropriately. | |
104655fd | 1081 | */ |
650d2a06 MP |
1082 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1083 | remap_and_issue(tc, bio, lookup_result.block); | |
1084 | else | |
1085 | bio_endio(bio, 0); | |
104655fd JT |
1086 | } |
1087 | break; | |
1088 | ||
1089 | case -ENODATA: | |
1090 | /* | |
1091 | * It isn't provisioned, just forget it. | |
1092 | */ | |
f286ba0e | 1093 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1094 | bio_endio(bio, 0); |
1095 | break; | |
1096 | ||
1097 | default: | |
c397741c MS |
1098 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1099 | __func__, r); | |
f286ba0e | 1100 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1101 | bio_io_error(bio); |
1102 | break; | |
1103 | } | |
1104 | } | |
1105 | ||
991d9fa0 | 1106 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1107 | struct dm_cell_key *key, |
991d9fa0 | 1108 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1109 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1110 | { |
1111 | int r; | |
1112 | dm_block_t data_block; | |
d6fc2042 | 1113 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1114 | |
1115 | r = alloc_data_block(tc, &data_block); | |
1116 | switch (r) { | |
1117 | case 0: | |
2dd9c257 JT |
1118 | schedule_internal_copy(tc, block, lookup_result->block, |
1119 | data_block, cell, bio); | |
991d9fa0 JT |
1120 | break; |
1121 | ||
1122 | case -ENOSPC: | |
d6fc2042 | 1123 | no_space(pool, cell); |
991d9fa0 JT |
1124 | break; |
1125 | ||
1126 | default: | |
c397741c MS |
1127 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1128 | __func__, r); | |
d6fc2042 MS |
1129 | set_pool_mode(pool, PM_READ_ONLY); |
1130 | cell_error(pool, cell); | |
991d9fa0 JT |
1131 | break; |
1132 | } | |
1133 | } | |
1134 | ||
1135 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1136 | dm_block_t block, | |
1137 | struct dm_thin_lookup_result *lookup_result) | |
1138 | { | |
a24c2569 | 1139 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1140 | struct pool *pool = tc->pool; |
44feb387 | 1141 | struct dm_cell_key key; |
991d9fa0 JT |
1142 | |
1143 | /* | |
1144 | * If cell is already occupied, then sharing is already in the process | |
1145 | * of being broken so we have nothing further to do here. | |
1146 | */ | |
1147 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1148 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1149 | return; |
1150 | ||
60049701 | 1151 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1152 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1153 | else { | |
59c3d2c6 | 1154 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1155 | |
44feb387 | 1156 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1157 | inc_all_io_entry(pool, bio); |
f286ba0e | 1158 | cell_defer_no_holder(tc, cell); |
e8088073 | 1159 | |
991d9fa0 JT |
1160 | remap_and_issue(tc, bio, lookup_result->block); |
1161 | } | |
1162 | } | |
1163 | ||
1164 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1165 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1166 | { |
1167 | int r; | |
1168 | dm_block_t data_block; | |
6beca5eb | 1169 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1170 | |
1171 | /* | |
1172 | * Remap empty bios (flushes) immediately, without provisioning. | |
1173 | */ | |
1174 | if (!bio->bi_size) { | |
6beca5eb | 1175 | inc_all_io_entry(pool, bio); |
f286ba0e | 1176 | cell_defer_no_holder(tc, cell); |
e8088073 | 1177 | |
991d9fa0 JT |
1178 | remap_and_issue(tc, bio, 0); |
1179 | return; | |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * Fill read bios with zeroes and complete them immediately. | |
1184 | */ | |
1185 | if (bio_data_dir(bio) == READ) { | |
1186 | zero_fill_bio(bio); | |
f286ba0e | 1187 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1188 | bio_endio(bio, 0); |
1189 | return; | |
1190 | } | |
1191 | ||
1192 | r = alloc_data_block(tc, &data_block); | |
1193 | switch (r) { | |
1194 | case 0: | |
2dd9c257 JT |
1195 | if (tc->origin_dev) |
1196 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1197 | else | |
1198 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1199 | break; |
1200 | ||
1201 | case -ENOSPC: | |
6beca5eb | 1202 | no_space(pool, cell); |
991d9fa0 JT |
1203 | break; |
1204 | ||
1205 | default: | |
c397741c MS |
1206 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1207 | __func__, r); | |
6beca5eb JT |
1208 | set_pool_mode(pool, PM_READ_ONLY); |
1209 | cell_error(pool, cell); | |
991d9fa0 JT |
1210 | break; |
1211 | } | |
1212 | } | |
1213 | ||
1214 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1215 | { | |
1216 | int r; | |
6beca5eb | 1217 | struct pool *pool = tc->pool; |
991d9fa0 | 1218 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1219 | struct dm_bio_prison_cell *cell; |
44feb387 | 1220 | struct dm_cell_key key; |
991d9fa0 JT |
1221 | struct dm_thin_lookup_result lookup_result; |
1222 | ||
1223 | /* | |
1224 | * If cell is already occupied, then the block is already | |
1225 | * being provisioned so we have nothing further to do here. | |
1226 | */ | |
1227 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1228 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1229 | return; |
1230 | ||
1231 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1232 | switch (r) { | |
1233 | case 0: | |
e8088073 | 1234 | if (lookup_result.shared) { |
991d9fa0 | 1235 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1236 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1237 | } else { |
6beca5eb | 1238 | inc_all_io_entry(pool, bio); |
f286ba0e | 1239 | cell_defer_no_holder(tc, cell); |
e8088073 | 1240 | |
991d9fa0 | 1241 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1242 | } |
991d9fa0 JT |
1243 | break; |
1244 | ||
1245 | case -ENODATA: | |
2dd9c257 | 1246 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1247 | inc_all_io_entry(pool, bio); |
f286ba0e | 1248 | cell_defer_no_holder(tc, cell); |
e8088073 | 1249 | |
2dd9c257 JT |
1250 | remap_to_origin_and_issue(tc, bio); |
1251 | } else | |
1252 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1253 | break; |
1254 | ||
1255 | default: | |
c397741c MS |
1256 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1257 | __func__, r); | |
f286ba0e | 1258 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1259 | bio_io_error(bio); |
1260 | break; | |
1261 | } | |
1262 | } | |
1263 | ||
e49e5829 JT |
1264 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1265 | { | |
1266 | int r; | |
1267 | int rw = bio_data_dir(bio); | |
1268 | dm_block_t block = get_bio_block(tc, bio); | |
1269 | struct dm_thin_lookup_result lookup_result; | |
1270 | ||
1271 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1272 | switch (r) { | |
1273 | case 0: | |
1274 | if (lookup_result.shared && (rw == WRITE) && bio->bi_size) | |
1275 | bio_io_error(bio); | |
e8088073 JT |
1276 | else { |
1277 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1278 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1279 | } |
e49e5829 JT |
1280 | break; |
1281 | ||
1282 | case -ENODATA: | |
1283 | if (rw != READ) { | |
1284 | bio_io_error(bio); | |
1285 | break; | |
1286 | } | |
1287 | ||
1288 | if (tc->origin_dev) { | |
e8088073 | 1289 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1290 | remap_to_origin_and_issue(tc, bio); |
1291 | break; | |
1292 | } | |
1293 | ||
1294 | zero_fill_bio(bio); | |
1295 | bio_endio(bio, 0); | |
1296 | break; | |
1297 | ||
1298 | default: | |
c397741c MS |
1299 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1300 | __func__, r); | |
e49e5829 JT |
1301 | bio_io_error(bio); |
1302 | break; | |
1303 | } | |
1304 | } | |
1305 | ||
1306 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) | |
1307 | { | |
1308 | bio_io_error(bio); | |
1309 | } | |
1310 | ||
ac8c3f3d JT |
1311 | /* |
1312 | * FIXME: should we also commit due to size of transaction, measured in | |
1313 | * metadata blocks? | |
1314 | */ | |
905e51b3 JT |
1315 | static int need_commit_due_to_time(struct pool *pool) |
1316 | { | |
1317 | return jiffies < pool->last_commit_jiffies || | |
1318 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1319 | } | |
1320 | ||
991d9fa0 JT |
1321 | static void process_deferred_bios(struct pool *pool) |
1322 | { | |
1323 | unsigned long flags; | |
1324 | struct bio *bio; | |
1325 | struct bio_list bios; | |
991d9fa0 JT |
1326 | |
1327 | bio_list_init(&bios); | |
1328 | ||
1329 | spin_lock_irqsave(&pool->lock, flags); | |
1330 | bio_list_merge(&bios, &pool->deferred_bios); | |
1331 | bio_list_init(&pool->deferred_bios); | |
1332 | spin_unlock_irqrestore(&pool->lock, flags); | |
1333 | ||
1334 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 1335 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1336 | struct thin_c *tc = h->tc; |
1337 | ||
991d9fa0 JT |
1338 | /* |
1339 | * If we've got no free new_mapping structs, and processing | |
1340 | * this bio might require one, we pause until there are some | |
1341 | * prepared mappings to process. | |
1342 | */ | |
1343 | if (ensure_next_mapping(pool)) { | |
1344 | spin_lock_irqsave(&pool->lock, flags); | |
1345 | bio_list_merge(&pool->deferred_bios, &bios); | |
1346 | spin_unlock_irqrestore(&pool->lock, flags); | |
1347 | ||
1348 | break; | |
1349 | } | |
104655fd JT |
1350 | |
1351 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1352 | pool->process_discard(tc, bio); |
104655fd | 1353 | else |
e49e5829 | 1354 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1355 | } |
1356 | ||
1357 | /* | |
1358 | * If there are any deferred flush bios, we must commit | |
1359 | * the metadata before issuing them. | |
1360 | */ | |
1361 | bio_list_init(&bios); | |
1362 | spin_lock_irqsave(&pool->lock, flags); | |
1363 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1364 | bio_list_init(&pool->deferred_flush_bios); | |
1365 | spin_unlock_irqrestore(&pool->lock, flags); | |
1366 | ||
905e51b3 | 1367 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1368 | return; |
1369 | ||
020cc3b5 | 1370 | if (commit(pool)) { |
991d9fa0 JT |
1371 | while ((bio = bio_list_pop(&bios))) |
1372 | bio_io_error(bio); | |
1373 | return; | |
1374 | } | |
905e51b3 | 1375 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1376 | |
1377 | while ((bio = bio_list_pop(&bios))) | |
1378 | generic_make_request(bio); | |
1379 | } | |
1380 | ||
1381 | static void do_worker(struct work_struct *ws) | |
1382 | { | |
1383 | struct pool *pool = container_of(ws, struct pool, worker); | |
1384 | ||
e49e5829 JT |
1385 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1386 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1387 | process_deferred_bios(pool); |
1388 | } | |
1389 | ||
905e51b3 JT |
1390 | /* |
1391 | * We want to commit periodically so that not too much | |
1392 | * unwritten data builds up. | |
1393 | */ | |
1394 | static void do_waker(struct work_struct *ws) | |
1395 | { | |
1396 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1397 | wake_worker(pool); | |
1398 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1399 | } | |
1400 | ||
991d9fa0 JT |
1401 | /*----------------------------------------------------------------*/ |
1402 | ||
e49e5829 JT |
1403 | static enum pool_mode get_pool_mode(struct pool *pool) |
1404 | { | |
1405 | return pool->pf.mode; | |
1406 | } | |
1407 | ||
1408 | static void set_pool_mode(struct pool *pool, enum pool_mode mode) | |
1409 | { | |
1410 | int r; | |
1411 | ||
1412 | pool->pf.mode = mode; | |
1413 | ||
1414 | switch (mode) { | |
1415 | case PM_FAIL: | |
4fa5971a MS |
1416 | DMERR("%s: switching pool to failure mode", |
1417 | dm_device_name(pool->pool_md)); | |
5383ef3a | 1418 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1419 | pool->process_bio = process_bio_fail; |
1420 | pool->process_discard = process_bio_fail; | |
1421 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1422 | pool->process_prepared_discard = process_prepared_discard_fail; | |
1423 | break; | |
1424 | ||
1425 | case PM_READ_ONLY: | |
4fa5971a MS |
1426 | DMERR("%s: switching pool to read-only mode", |
1427 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1428 | r = dm_pool_abort_metadata(pool->pmd); |
1429 | if (r) { | |
4fa5971a MS |
1430 | DMERR("%s: aborting transaction failed", |
1431 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1432 | set_pool_mode(pool, PM_FAIL); |
1433 | } else { | |
1434 | dm_pool_metadata_read_only(pool->pmd); | |
1435 | pool->process_bio = process_bio_read_only; | |
1436 | pool->process_discard = process_discard; | |
1437 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1438 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1439 | } | |
1440 | break; | |
1441 | ||
1442 | case PM_WRITE: | |
9b7aaa64 | 1443 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1444 | pool->process_bio = process_bio; |
1445 | pool->process_discard = process_discard; | |
1446 | pool->process_prepared_mapping = process_prepared_mapping; | |
1447 | pool->process_prepared_discard = process_prepared_discard; | |
1448 | break; | |
1449 | } | |
1450 | } | |
1451 | ||
1452 | /*----------------------------------------------------------------*/ | |
1453 | ||
991d9fa0 JT |
1454 | /* |
1455 | * Mapping functions. | |
1456 | */ | |
1457 | ||
1458 | /* | |
1459 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1460 | */ | |
1461 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1462 | { | |
1463 | unsigned long flags; | |
1464 | struct pool *pool = tc->pool; | |
1465 | ||
1466 | spin_lock_irqsave(&pool->lock, flags); | |
1467 | bio_list_add(&pool->deferred_bios, bio); | |
1468 | spin_unlock_irqrestore(&pool->lock, flags); | |
1469 | ||
1470 | wake_worker(pool); | |
1471 | } | |
1472 | ||
59c3d2c6 | 1473 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1474 | { |
59c3d2c6 | 1475 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1476 | |
1477 | h->tc = tc; | |
1478 | h->shared_read_entry = NULL; | |
e8088073 | 1479 | h->all_io_entry = NULL; |
eb2aa48d | 1480 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1481 | } |
1482 | ||
991d9fa0 JT |
1483 | /* |
1484 | * Non-blocking function called from the thin target's map function. | |
1485 | */ | |
7de3ee57 | 1486 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1487 | { |
1488 | int r; | |
1489 | struct thin_c *tc = ti->private; | |
1490 | dm_block_t block = get_bio_block(tc, bio); | |
1491 | struct dm_thin_device *td = tc->td; | |
1492 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1493 | struct dm_bio_prison_cell cell1, cell2; |
1494 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1495 | struct dm_cell_key key; |
991d9fa0 | 1496 | |
59c3d2c6 | 1497 | thin_hook_bio(tc, bio); |
e49e5829 JT |
1498 | |
1499 | if (get_pool_mode(tc->pool) == PM_FAIL) { | |
1500 | bio_io_error(bio); | |
1501 | return DM_MAPIO_SUBMITTED; | |
1502 | } | |
1503 | ||
104655fd | 1504 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1505 | thin_defer_bio(tc, bio); |
1506 | return DM_MAPIO_SUBMITTED; | |
1507 | } | |
1508 | ||
1509 | r = dm_thin_find_block(td, block, 0, &result); | |
1510 | ||
1511 | /* | |
1512 | * Note that we defer readahead too. | |
1513 | */ | |
1514 | switch (r) { | |
1515 | case 0: | |
1516 | if (unlikely(result.shared)) { | |
1517 | /* | |
1518 | * We have a race condition here between the | |
1519 | * result.shared value returned by the lookup and | |
1520 | * snapshot creation, which may cause new | |
1521 | * sharing. | |
1522 | * | |
1523 | * To avoid this always quiesce the origin before | |
1524 | * taking the snap. You want to do this anyway to | |
1525 | * ensure a consistent application view | |
1526 | * (i.e. lockfs). | |
1527 | * | |
1528 | * More distant ancestors are irrelevant. The | |
1529 | * shared flag will be set in their case. | |
1530 | */ | |
1531 | thin_defer_bio(tc, bio); | |
e8088073 | 1532 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1533 | } |
e8088073 JT |
1534 | |
1535 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1536 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1537 | return DM_MAPIO_SUBMITTED; |
1538 | ||
1539 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1540 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1541 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1542 | return DM_MAPIO_SUBMITTED; |
1543 | } | |
1544 | ||
1545 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1546 | cell_defer_no_holder_no_free(tc, &cell2); |
1547 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1548 | |
1549 | remap(tc, bio, result.block); | |
1550 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1551 | |
1552 | case -ENODATA: | |
e49e5829 JT |
1553 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1554 | /* | |
1555 | * This block isn't provisioned, and we have no way | |
1556 | * of doing so. Just error it. | |
1557 | */ | |
1558 | bio_io_error(bio); | |
2aab3850 | 1559 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1560 | } |
1561 | /* fall through */ | |
1562 | ||
1563 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1564 | /* |
1565 | * In future, the failed dm_thin_find_block above could | |
1566 | * provide the hint to load the metadata into cache. | |
1567 | */ | |
991d9fa0 | 1568 | thin_defer_bio(tc, bio); |
2aab3850 | 1569 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1570 | |
1571 | default: | |
1572 | /* | |
1573 | * Must always call bio_io_error on failure. | |
1574 | * dm_thin_find_block can fail with -EINVAL if the | |
1575 | * pool is switched to fail-io mode. | |
1576 | */ | |
1577 | bio_io_error(bio); | |
2aab3850 | 1578 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1579 | } |
991d9fa0 JT |
1580 | } |
1581 | ||
1582 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1583 | { | |
1584 | int r; | |
1585 | unsigned long flags; | |
1586 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1587 | ||
1588 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1589 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1590 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1591 | ||
1592 | if (!r) { | |
1593 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1594 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1595 | } | |
1596 | ||
1597 | return r; | |
1598 | } | |
1599 | ||
1600 | static void __requeue_bios(struct pool *pool) | |
1601 | { | |
1602 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1603 | bio_list_init(&pool->retry_on_resume_list); | |
1604 | } | |
1605 | ||
1606 | /*---------------------------------------------------------------- | |
1607 | * Binding of control targets to a pool object | |
1608 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1609 | static bool data_dev_supports_discard(struct pool_c *pt) |
1610 | { | |
1611 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1612 | ||
1613 | return q && blk_queue_discard(q); | |
1614 | } | |
1615 | ||
58051b94 JT |
1616 | static bool is_factor(sector_t block_size, uint32_t n) |
1617 | { | |
1618 | return !sector_div(block_size, n); | |
1619 | } | |
1620 | ||
9bc142dd MS |
1621 | /* |
1622 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1623 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1624 | */ |
0424caa1 | 1625 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1626 | { |
0424caa1 MS |
1627 | struct pool *pool = pt->pool; |
1628 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1629 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1630 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1631 | const char *reason = NULL; | |
9bc142dd MS |
1632 | char buf[BDEVNAME_SIZE]; |
1633 | ||
0424caa1 | 1634 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1635 | return; |
1636 | ||
0424caa1 MS |
1637 | if (!data_dev_supports_discard(pt)) |
1638 | reason = "discard unsupported"; | |
1639 | ||
1640 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1641 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1642 | |
0424caa1 MS |
1643 | else if (data_limits->discard_granularity > block_size) |
1644 | reason = "discard granularity larger than a block"; | |
1645 | ||
58051b94 | 1646 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1647 | reason = "discard granularity not a factor of block size"; |
1648 | ||
1649 | if (reason) { | |
1650 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1651 | pt->adjusted_pf.discard_passdown = false; | |
1652 | } | |
9bc142dd MS |
1653 | } |
1654 | ||
991d9fa0 JT |
1655 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1656 | { | |
1657 | struct pool_c *pt = ti->private; | |
1658 | ||
e49e5829 | 1659 | /* |
9b7aaa64 | 1660 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 JT |
1661 | */ |
1662 | enum pool_mode old_mode = pool->pf.mode; | |
0424caa1 | 1663 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1664 | |
9b7aaa64 JT |
1665 | /* |
1666 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1667 | * not going to recover without a thin_repair. So we never let the | |
1668 | * pool move out of the old mode. On the other hand a PM_READ_ONLY | |
1669 | * may have been due to a lack of metadata or data space, and may | |
1670 | * now work (ie. if the underlying devices have been resized). | |
1671 | */ | |
1672 | if (old_mode == PM_FAIL) | |
e49e5829 JT |
1673 | new_mode = old_mode; |
1674 | ||
991d9fa0 JT |
1675 | pool->ti = ti; |
1676 | pool->low_water_blocks = pt->low_water_blocks; | |
0424caa1 | 1677 | pool->pf = pt->adjusted_pf; |
991d9fa0 | 1678 | |
9bc142dd | 1679 | set_pool_mode(pool, new_mode); |
f402693d | 1680 | |
991d9fa0 JT |
1681 | return 0; |
1682 | } | |
1683 | ||
1684 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1685 | { | |
1686 | if (pool->ti == ti) | |
1687 | pool->ti = NULL; | |
1688 | } | |
1689 | ||
1690 | /*---------------------------------------------------------------- | |
1691 | * Pool creation | |
1692 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1693 | /* Initialize pool features. */ |
1694 | static void pool_features_init(struct pool_features *pf) | |
1695 | { | |
e49e5829 | 1696 | pf->mode = PM_WRITE; |
9bc142dd MS |
1697 | pf->zero_new_blocks = true; |
1698 | pf->discard_enabled = true; | |
1699 | pf->discard_passdown = true; | |
67e2e2b2 JT |
1700 | } |
1701 | ||
991d9fa0 JT |
1702 | static void __pool_destroy(struct pool *pool) |
1703 | { | |
1704 | __pool_table_remove(pool); | |
1705 | ||
1706 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1707 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1708 | ||
44feb387 | 1709 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1710 | dm_kcopyd_client_destroy(pool->copier); |
1711 | ||
1712 | if (pool->wq) | |
1713 | destroy_workqueue(pool->wq); | |
1714 | ||
1715 | if (pool->next_mapping) | |
1716 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1717 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1718 | dm_deferred_set_destroy(pool->shared_read_ds); |
1719 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1720 | kfree(pool); |
1721 | } | |
1722 | ||
a24c2569 | 1723 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1724 | |
991d9fa0 JT |
1725 | static struct pool *pool_create(struct mapped_device *pool_md, |
1726 | struct block_device *metadata_dev, | |
e49e5829 JT |
1727 | unsigned long block_size, |
1728 | int read_only, char **error) | |
991d9fa0 JT |
1729 | { |
1730 | int r; | |
1731 | void *err_p; | |
1732 | struct pool *pool; | |
1733 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1734 | bool format_device = read_only ? false : true; |
991d9fa0 | 1735 | |
e49e5829 | 1736 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1737 | if (IS_ERR(pmd)) { |
1738 | *error = "Error creating metadata object"; | |
1739 | return (struct pool *)pmd; | |
1740 | } | |
1741 | ||
1742 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1743 | if (!pool) { | |
1744 | *error = "Error allocating memory for pool"; | |
1745 | err_p = ERR_PTR(-ENOMEM); | |
1746 | goto bad_pool; | |
1747 | } | |
1748 | ||
1749 | pool->pmd = pmd; | |
1750 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1751 | if (block_size & (block_size - 1)) |
1752 | pool->sectors_per_block_shift = -1; | |
1753 | else | |
1754 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1755 | pool->low_water_blocks = 0; |
67e2e2b2 | 1756 | pool_features_init(&pool->pf); |
44feb387 | 1757 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
1758 | if (!pool->prison) { |
1759 | *error = "Error creating pool's bio prison"; | |
1760 | err_p = ERR_PTR(-ENOMEM); | |
1761 | goto bad_prison; | |
1762 | } | |
1763 | ||
df5d2e90 | 1764 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
1765 | if (IS_ERR(pool->copier)) { |
1766 | r = PTR_ERR(pool->copier); | |
1767 | *error = "Error creating pool's kcopyd client"; | |
1768 | err_p = ERR_PTR(r); | |
1769 | goto bad_kcopyd_client; | |
1770 | } | |
1771 | ||
1772 | /* | |
1773 | * Create singlethreaded workqueue that will service all devices | |
1774 | * that use this metadata. | |
1775 | */ | |
1776 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1777 | if (!pool->wq) { | |
1778 | *error = "Error creating pool's workqueue"; | |
1779 | err_p = ERR_PTR(-ENOMEM); | |
1780 | goto bad_wq; | |
1781 | } | |
1782 | ||
1783 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1784 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1785 | spin_lock_init(&pool->lock); |
1786 | bio_list_init(&pool->deferred_bios); | |
1787 | bio_list_init(&pool->deferred_flush_bios); | |
1788 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1789 | INIT_LIST_HEAD(&pool->prepared_discards); |
88a6621b JT |
1790 | pool->low_water_triggered = false; |
1791 | pool->no_free_space = false; | |
991d9fa0 | 1792 | bio_list_init(&pool->retry_on_resume_list); |
44feb387 MS |
1793 | |
1794 | pool->shared_read_ds = dm_deferred_set_create(); | |
1795 | if (!pool->shared_read_ds) { | |
1796 | *error = "Error creating pool's shared read deferred set"; | |
1797 | err_p = ERR_PTR(-ENOMEM); | |
1798 | goto bad_shared_read_ds; | |
1799 | } | |
1800 | ||
1801 | pool->all_io_ds = dm_deferred_set_create(); | |
1802 | if (!pool->all_io_ds) { | |
1803 | *error = "Error creating pool's all io deferred set"; | |
1804 | err_p = ERR_PTR(-ENOMEM); | |
1805 | goto bad_all_io_ds; | |
1806 | } | |
991d9fa0 JT |
1807 | |
1808 | pool->next_mapping = NULL; | |
a24c2569 MS |
1809 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1810 | _new_mapping_cache); | |
991d9fa0 JT |
1811 | if (!pool->mapping_pool) { |
1812 | *error = "Error creating pool's mapping mempool"; | |
1813 | err_p = ERR_PTR(-ENOMEM); | |
1814 | goto bad_mapping_pool; | |
1815 | } | |
1816 | ||
991d9fa0 | 1817 | pool->ref_count = 1; |
905e51b3 | 1818 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1819 | pool->pool_md = pool_md; |
1820 | pool->md_dev = metadata_dev; | |
1821 | __pool_table_insert(pool); | |
1822 | ||
1823 | return pool; | |
1824 | ||
991d9fa0 | 1825 | bad_mapping_pool: |
44feb387 MS |
1826 | dm_deferred_set_destroy(pool->all_io_ds); |
1827 | bad_all_io_ds: | |
1828 | dm_deferred_set_destroy(pool->shared_read_ds); | |
1829 | bad_shared_read_ds: | |
991d9fa0 JT |
1830 | destroy_workqueue(pool->wq); |
1831 | bad_wq: | |
1832 | dm_kcopyd_client_destroy(pool->copier); | |
1833 | bad_kcopyd_client: | |
44feb387 | 1834 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1835 | bad_prison: |
1836 | kfree(pool); | |
1837 | bad_pool: | |
1838 | if (dm_pool_metadata_close(pmd)) | |
1839 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1840 | ||
1841 | return err_p; | |
1842 | } | |
1843 | ||
1844 | static void __pool_inc(struct pool *pool) | |
1845 | { | |
1846 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1847 | pool->ref_count++; | |
1848 | } | |
1849 | ||
1850 | static void __pool_dec(struct pool *pool) | |
1851 | { | |
1852 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1853 | BUG_ON(!pool->ref_count); | |
1854 | if (!--pool->ref_count) | |
1855 | __pool_destroy(pool); | |
1856 | } | |
1857 | ||
1858 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
1859 | struct block_device *metadata_dev, | |
e49e5829 JT |
1860 | unsigned long block_size, int read_only, |
1861 | char **error, int *created) | |
991d9fa0 JT |
1862 | { |
1863 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
1864 | ||
1865 | if (pool) { | |
f09996c9 MS |
1866 | if (pool->pool_md != pool_md) { |
1867 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 1868 | return ERR_PTR(-EBUSY); |
f09996c9 | 1869 | } |
991d9fa0 JT |
1870 | __pool_inc(pool); |
1871 | ||
1872 | } else { | |
1873 | pool = __pool_table_lookup(pool_md); | |
1874 | if (pool) { | |
f09996c9 MS |
1875 | if (pool->md_dev != metadata_dev) { |
1876 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 1877 | return ERR_PTR(-EINVAL); |
f09996c9 | 1878 | } |
991d9fa0 JT |
1879 | __pool_inc(pool); |
1880 | ||
67e2e2b2 | 1881 | } else { |
e49e5829 | 1882 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
1883 | *created = 1; |
1884 | } | |
991d9fa0 JT |
1885 | } |
1886 | ||
1887 | return pool; | |
1888 | } | |
1889 | ||
1890 | /*---------------------------------------------------------------- | |
1891 | * Pool target methods | |
1892 | *--------------------------------------------------------------*/ | |
1893 | static void pool_dtr(struct dm_target *ti) | |
1894 | { | |
1895 | struct pool_c *pt = ti->private; | |
1896 | ||
1897 | mutex_lock(&dm_thin_pool_table.mutex); | |
1898 | ||
1899 | unbind_control_target(pt->pool, ti); | |
1900 | __pool_dec(pt->pool); | |
1901 | dm_put_device(ti, pt->metadata_dev); | |
1902 | dm_put_device(ti, pt->data_dev); | |
1903 | kfree(pt); | |
1904 | ||
1905 | mutex_unlock(&dm_thin_pool_table.mutex); | |
1906 | } | |
1907 | ||
991d9fa0 JT |
1908 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
1909 | struct dm_target *ti) | |
1910 | { | |
1911 | int r; | |
1912 | unsigned argc; | |
1913 | const char *arg_name; | |
1914 | ||
1915 | static struct dm_arg _args[] = { | |
67e2e2b2 | 1916 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
1917 | }; |
1918 | ||
1919 | /* | |
1920 | * No feature arguments supplied. | |
1921 | */ | |
1922 | if (!as->argc) | |
1923 | return 0; | |
1924 | ||
1925 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1926 | if (r) | |
1927 | return -EINVAL; | |
1928 | ||
1929 | while (argc && !r) { | |
1930 | arg_name = dm_shift_arg(as); | |
1931 | argc--; | |
1932 | ||
e49e5829 | 1933 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 1934 | pf->zero_new_blocks = false; |
e49e5829 JT |
1935 | |
1936 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 1937 | pf->discard_enabled = false; |
e49e5829 JT |
1938 | |
1939 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 1940 | pf->discard_passdown = false; |
991d9fa0 | 1941 | |
e49e5829 JT |
1942 | else if (!strcasecmp(arg_name, "read_only")) |
1943 | pf->mode = PM_READ_ONLY; | |
1944 | ||
1945 | else { | |
1946 | ti->error = "Unrecognised pool feature requested"; | |
1947 | r = -EINVAL; | |
1948 | break; | |
1949 | } | |
991d9fa0 JT |
1950 | } |
1951 | ||
1952 | return r; | |
1953 | } | |
1954 | ||
ac8c3f3d JT |
1955 | static void metadata_low_callback(void *context) |
1956 | { | |
1957 | struct pool *pool = context; | |
1958 | ||
1959 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
1960 | dm_device_name(pool->pool_md)); | |
1961 | ||
1962 | dm_table_event(pool->ti->table); | |
1963 | } | |
1964 | ||
b17446df JT |
1965 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
1966 | { | |
1967 | sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
1968 | char buffer[BDEVNAME_SIZE]; | |
1969 | ||
1970 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) { | |
1971 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
1972 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
1973 | metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING; | |
1974 | } | |
1975 | ||
1976 | return metadata_dev_size; | |
1977 | } | |
1978 | ||
24347e95 JT |
1979 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
1980 | { | |
1981 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
1982 | ||
1983 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); | |
1984 | ||
1985 | return metadata_dev_size; | |
1986 | } | |
1987 | ||
ac8c3f3d JT |
1988 | /* |
1989 | * When a metadata threshold is crossed a dm event is triggered, and | |
1990 | * userland should respond by growing the metadata device. We could let | |
1991 | * userland set the threshold, like we do with the data threshold, but I'm | |
1992 | * not sure they know enough to do this well. | |
1993 | */ | |
1994 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
1995 | { | |
1996 | /* | |
1997 | * 4M is ample for all ops with the possible exception of thin | |
1998 | * device deletion which is harmless if it fails (just retry the | |
1999 | * delete after you've grown the device). | |
2000 | */ | |
2001 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2002 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2003 | } | |
2004 | ||
991d9fa0 JT |
2005 | /* |
2006 | * thin-pool <metadata dev> <data dev> | |
2007 | * <data block size (sectors)> | |
2008 | * <low water mark (blocks)> | |
2009 | * [<#feature args> [<arg>]*] | |
2010 | * | |
2011 | * Optional feature arguments are: | |
2012 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2013 | * ignore_discard: disable discard |
2014 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
2015 | */ |
2016 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2017 | { | |
67e2e2b2 | 2018 | int r, pool_created = 0; |
991d9fa0 JT |
2019 | struct pool_c *pt; |
2020 | struct pool *pool; | |
2021 | struct pool_features pf; | |
2022 | struct dm_arg_set as; | |
2023 | struct dm_dev *data_dev; | |
2024 | unsigned long block_size; | |
2025 | dm_block_t low_water_blocks; | |
2026 | struct dm_dev *metadata_dev; | |
5d0db96d | 2027 | fmode_t metadata_mode; |
991d9fa0 JT |
2028 | |
2029 | /* | |
2030 | * FIXME Remove validation from scope of lock. | |
2031 | */ | |
2032 | mutex_lock(&dm_thin_pool_table.mutex); | |
2033 | ||
2034 | if (argc < 4) { | |
2035 | ti->error = "Invalid argument count"; | |
2036 | r = -EINVAL; | |
2037 | goto out_unlock; | |
2038 | } | |
5d0db96d | 2039 | |
991d9fa0 JT |
2040 | as.argc = argc; |
2041 | as.argv = argv; | |
2042 | ||
5d0db96d JT |
2043 | /* |
2044 | * Set default pool features. | |
2045 | */ | |
2046 | pool_features_init(&pf); | |
2047 | ||
2048 | dm_consume_args(&as, 4); | |
2049 | r = parse_pool_features(&as, &pf, ti); | |
2050 | if (r) | |
2051 | goto out_unlock; | |
2052 | ||
2053 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2054 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2055 | if (r) { |
2056 | ti->error = "Error opening metadata block device"; | |
2057 | goto out_unlock; | |
2058 | } | |
2059 | ||
b17446df JT |
2060 | /* |
2061 | * Run for the side-effect of possibly issuing a warning if the | |
2062 | * device is too big. | |
2063 | */ | |
2064 | (void) get_metadata_dev_size(metadata_dev->bdev); | |
991d9fa0 JT |
2065 | |
2066 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2067 | if (r) { | |
2068 | ti->error = "Error getting data device"; | |
2069 | goto out_metadata; | |
2070 | } | |
2071 | ||
2072 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2073 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2074 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2075 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2076 | ti->error = "Invalid block size"; |
2077 | r = -EINVAL; | |
2078 | goto out; | |
2079 | } | |
2080 | ||
2081 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2082 | ti->error = "Invalid low water mark"; | |
2083 | r = -EINVAL; | |
2084 | goto out; | |
2085 | } | |
2086 | ||
991d9fa0 JT |
2087 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2088 | if (!pt) { | |
2089 | r = -ENOMEM; | |
2090 | goto out; | |
2091 | } | |
2092 | ||
2093 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2094 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2095 | if (IS_ERR(pool)) { |
2096 | r = PTR_ERR(pool); | |
2097 | goto out_free_pt; | |
2098 | } | |
2099 | ||
67e2e2b2 JT |
2100 | /* |
2101 | * 'pool_created' reflects whether this is the first table load. | |
2102 | * Top level discard support is not allowed to be changed after | |
2103 | * initial load. This would require a pool reload to trigger thin | |
2104 | * device changes. | |
2105 | */ | |
2106 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2107 | ti->error = "Discard support cannot be disabled once enabled"; | |
2108 | r = -EINVAL; | |
2109 | goto out_flags_changed; | |
2110 | } | |
2111 | ||
991d9fa0 JT |
2112 | pt->pool = pool; |
2113 | pt->ti = ti; | |
2114 | pt->metadata_dev = metadata_dev; | |
2115 | pt->data_dev = data_dev; | |
2116 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2117 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2118 | ti->num_flush_bios = 1; |
9bc142dd | 2119 | |
67e2e2b2 JT |
2120 | /* |
2121 | * Only need to enable discards if the pool should pass | |
2122 | * them down to the data device. The thin device's discard | |
2123 | * processing will cause mappings to be removed from the btree. | |
2124 | */ | |
b60ab990 | 2125 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2126 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2127 | ti->num_discard_bios = 1; |
9bc142dd | 2128 | |
67e2e2b2 JT |
2129 | /* |
2130 | * Setting 'discards_supported' circumvents the normal | |
2131 | * stacking of discard limits (this keeps the pool and | |
2132 | * thin devices' discard limits consistent). | |
2133 | */ | |
0ac55489 | 2134 | ti->discards_supported = true; |
67e2e2b2 | 2135 | } |
991d9fa0 JT |
2136 | ti->private = pt; |
2137 | ||
ac8c3f3d JT |
2138 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2139 | calc_metadata_threshold(pt), | |
2140 | metadata_low_callback, | |
2141 | pool); | |
2142 | if (r) | |
2143 | goto out_free_pt; | |
2144 | ||
991d9fa0 JT |
2145 | pt->callbacks.congested_fn = pool_is_congested; |
2146 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2147 | ||
2148 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2149 | ||
2150 | return 0; | |
2151 | ||
67e2e2b2 JT |
2152 | out_flags_changed: |
2153 | __pool_dec(pool); | |
991d9fa0 JT |
2154 | out_free_pt: |
2155 | kfree(pt); | |
2156 | out: | |
2157 | dm_put_device(ti, data_dev); | |
2158 | out_metadata: | |
2159 | dm_put_device(ti, metadata_dev); | |
2160 | out_unlock: | |
2161 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2162 | ||
2163 | return r; | |
2164 | } | |
2165 | ||
7de3ee57 | 2166 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2167 | { |
2168 | int r; | |
2169 | struct pool_c *pt = ti->private; | |
2170 | struct pool *pool = pt->pool; | |
2171 | unsigned long flags; | |
2172 | ||
2173 | /* | |
2174 | * As this is a singleton target, ti->begin is always zero. | |
2175 | */ | |
2176 | spin_lock_irqsave(&pool->lock, flags); | |
2177 | bio->bi_bdev = pt->data_dev->bdev; | |
2178 | r = DM_MAPIO_REMAPPED; | |
2179 | spin_unlock_irqrestore(&pool->lock, flags); | |
2180 | ||
2181 | return r; | |
2182 | } | |
2183 | ||
b17446df | 2184 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2185 | { |
2186 | int r; | |
2187 | struct pool_c *pt = ti->private; | |
2188 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2189 | sector_t data_size = ti->len; |
2190 | dm_block_t sb_data_size; | |
991d9fa0 | 2191 | |
b17446df | 2192 | *need_commit = false; |
991d9fa0 | 2193 | |
55f2b8bd MS |
2194 | (void) sector_div(data_size, pool->sectors_per_block); |
2195 | ||
991d9fa0 JT |
2196 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2197 | if (r) { | |
4fa5971a MS |
2198 | DMERR("%s: failed to retrieve data device size", |
2199 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2200 | return r; |
2201 | } | |
2202 | ||
2203 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2204 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2205 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2206 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2207 | return -EINVAL; |
2208 | ||
2209 | } else if (data_size > sb_data_size) { | |
2210 | r = dm_pool_resize_data_dev(pool->pmd, data_size); | |
2211 | if (r) { | |
4fa5971a MS |
2212 | DMERR("%s: failed to resize data device", |
2213 | dm_device_name(pool->pool_md)); | |
e49e5829 | 2214 | set_pool_mode(pool, PM_READ_ONLY); |
991d9fa0 JT |
2215 | return r; |
2216 | } | |
2217 | ||
b17446df | 2218 | *need_commit = true; |
991d9fa0 JT |
2219 | } |
2220 | ||
2221 | return 0; | |
2222 | } | |
2223 | ||
24347e95 JT |
2224 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2225 | { | |
2226 | int r; | |
2227 | struct pool_c *pt = ti->private; | |
2228 | struct pool *pool = pt->pool; | |
2229 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2230 | ||
2231 | *need_commit = false; | |
2232 | ||
610bba8b | 2233 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2234 | |
2235 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2236 | if (r) { | |
4fa5971a MS |
2237 | DMERR("%s: failed to retrieve metadata device size", |
2238 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2239 | return r; |
2240 | } | |
2241 | ||
2242 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2243 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2244 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2245 | metadata_dev_size, sb_metadata_dev_size); |
2246 | return -EINVAL; | |
2247 | ||
2248 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
2249 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); | |
2250 | if (r) { | |
4fa5971a MS |
2251 | DMERR("%s: failed to resize metadata device", |
2252 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2253 | return r; |
2254 | } | |
2255 | ||
2256 | *need_commit = true; | |
2257 | } | |
2258 | ||
2259 | return 0; | |
2260 | } | |
2261 | ||
b17446df JT |
2262 | /* |
2263 | * Retrieves the number of blocks of the data device from | |
2264 | * the superblock and compares it to the actual device size, | |
2265 | * thus resizing the data device in case it has grown. | |
2266 | * | |
2267 | * This both copes with opening preallocated data devices in the ctr | |
2268 | * being followed by a resume | |
2269 | * -and- | |
2270 | * calling the resume method individually after userspace has | |
2271 | * grown the data device in reaction to a table event. | |
2272 | */ | |
2273 | static int pool_preresume(struct dm_target *ti) | |
2274 | { | |
2275 | int r; | |
24347e95 | 2276 | bool need_commit1, need_commit2; |
b17446df JT |
2277 | struct pool_c *pt = ti->private; |
2278 | struct pool *pool = pt->pool; | |
2279 | ||
2280 | /* | |
2281 | * Take control of the pool object. | |
2282 | */ | |
2283 | r = bind_control_target(pool, ti); | |
2284 | if (r) | |
2285 | return r; | |
2286 | ||
2287 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2288 | if (r) | |
2289 | return r; | |
2290 | ||
24347e95 JT |
2291 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2292 | if (r) | |
2293 | return r; | |
2294 | ||
2295 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2296 | (void) commit(pool); |
b17446df JT |
2297 | |
2298 | return 0; | |
2299 | } | |
2300 | ||
991d9fa0 JT |
2301 | static void pool_resume(struct dm_target *ti) |
2302 | { | |
2303 | struct pool_c *pt = ti->private; | |
2304 | struct pool *pool = pt->pool; | |
2305 | unsigned long flags; | |
2306 | ||
2307 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b JT |
2308 | pool->low_water_triggered = false; |
2309 | pool->no_free_space = false; | |
991d9fa0 JT |
2310 | __requeue_bios(pool); |
2311 | spin_unlock_irqrestore(&pool->lock, flags); | |
2312 | ||
905e51b3 | 2313 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2314 | } |
2315 | ||
2316 | static void pool_postsuspend(struct dm_target *ti) | |
2317 | { | |
991d9fa0 JT |
2318 | struct pool_c *pt = ti->private; |
2319 | struct pool *pool = pt->pool; | |
2320 | ||
905e51b3 | 2321 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2322 | flush_workqueue(pool->wq); |
020cc3b5 | 2323 | (void) commit(pool); |
991d9fa0 JT |
2324 | } |
2325 | ||
2326 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2327 | { | |
2328 | if (argc != args_required) { | |
2329 | DMWARN("Message received with %u arguments instead of %u.", | |
2330 | argc, args_required); | |
2331 | return -EINVAL; | |
2332 | } | |
2333 | ||
2334 | return 0; | |
2335 | } | |
2336 | ||
2337 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2338 | { | |
2339 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2340 | *dev_id <= MAX_DEV_ID) | |
2341 | return 0; | |
2342 | ||
2343 | if (warning) | |
2344 | DMWARN("Message received with invalid device id: %s", arg); | |
2345 | ||
2346 | return -EINVAL; | |
2347 | } | |
2348 | ||
2349 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2350 | { | |
2351 | dm_thin_id dev_id; | |
2352 | int r; | |
2353 | ||
2354 | r = check_arg_count(argc, 2); | |
2355 | if (r) | |
2356 | return r; | |
2357 | ||
2358 | r = read_dev_id(argv[1], &dev_id, 1); | |
2359 | if (r) | |
2360 | return r; | |
2361 | ||
2362 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2363 | if (r) { | |
2364 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2365 | argv[1]); | |
2366 | return r; | |
2367 | } | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
2372 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2373 | { | |
2374 | dm_thin_id dev_id; | |
2375 | dm_thin_id origin_dev_id; | |
2376 | int r; | |
2377 | ||
2378 | r = check_arg_count(argc, 3); | |
2379 | if (r) | |
2380 | return r; | |
2381 | ||
2382 | r = read_dev_id(argv[1], &dev_id, 1); | |
2383 | if (r) | |
2384 | return r; | |
2385 | ||
2386 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2387 | if (r) | |
2388 | return r; | |
2389 | ||
2390 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2391 | if (r) { | |
2392 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2393 | argv[1], argv[2]); | |
2394 | return r; | |
2395 | } | |
2396 | ||
2397 | return 0; | |
2398 | } | |
2399 | ||
2400 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2401 | { | |
2402 | dm_thin_id dev_id; | |
2403 | int r; | |
2404 | ||
2405 | r = check_arg_count(argc, 2); | |
2406 | if (r) | |
2407 | return r; | |
2408 | ||
2409 | r = read_dev_id(argv[1], &dev_id, 1); | |
2410 | if (r) | |
2411 | return r; | |
2412 | ||
2413 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2414 | if (r) | |
2415 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2416 | ||
2417 | return r; | |
2418 | } | |
2419 | ||
2420 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2421 | { | |
2422 | dm_thin_id old_id, new_id; | |
2423 | int r; | |
2424 | ||
2425 | r = check_arg_count(argc, 3); | |
2426 | if (r) | |
2427 | return r; | |
2428 | ||
2429 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2430 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2431 | return -EINVAL; | |
2432 | } | |
2433 | ||
2434 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2435 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2436 | return -EINVAL; | |
2437 | } | |
2438 | ||
2439 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2440 | if (r) { | |
2441 | DMWARN("Failed to change transaction id from %s to %s.", | |
2442 | argv[1], argv[2]); | |
2443 | return r; | |
2444 | } | |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
cc8394d8 JT |
2449 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2450 | { | |
2451 | int r; | |
2452 | ||
2453 | r = check_arg_count(argc, 1); | |
2454 | if (r) | |
2455 | return r; | |
2456 | ||
020cc3b5 | 2457 | (void) commit(pool); |
0d200aef | 2458 | |
cc8394d8 JT |
2459 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2460 | if (r) | |
2461 | DMWARN("reserve_metadata_snap message failed."); | |
2462 | ||
2463 | return r; | |
2464 | } | |
2465 | ||
2466 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2467 | { | |
2468 | int r; | |
2469 | ||
2470 | r = check_arg_count(argc, 1); | |
2471 | if (r) | |
2472 | return r; | |
2473 | ||
2474 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2475 | if (r) | |
2476 | DMWARN("release_metadata_snap message failed."); | |
2477 | ||
2478 | return r; | |
2479 | } | |
2480 | ||
991d9fa0 JT |
2481 | /* |
2482 | * Messages supported: | |
2483 | * create_thin <dev_id> | |
2484 | * create_snap <dev_id> <origin_id> | |
2485 | * delete <dev_id> | |
2486 | * trim <dev_id> <new_size_in_sectors> | |
2487 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2488 | * reserve_metadata_snap |
2489 | * release_metadata_snap | |
991d9fa0 JT |
2490 | */ |
2491 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2492 | { | |
2493 | int r = -EINVAL; | |
2494 | struct pool_c *pt = ti->private; | |
2495 | struct pool *pool = pt->pool; | |
2496 | ||
2497 | if (!strcasecmp(argv[0], "create_thin")) | |
2498 | r = process_create_thin_mesg(argc, argv, pool); | |
2499 | ||
2500 | else if (!strcasecmp(argv[0], "create_snap")) | |
2501 | r = process_create_snap_mesg(argc, argv, pool); | |
2502 | ||
2503 | else if (!strcasecmp(argv[0], "delete")) | |
2504 | r = process_delete_mesg(argc, argv, pool); | |
2505 | ||
2506 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2507 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2508 | ||
cc8394d8 JT |
2509 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2510 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2511 | ||
2512 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2513 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2514 | ||
991d9fa0 JT |
2515 | else |
2516 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2517 | ||
e49e5829 | 2518 | if (!r) |
020cc3b5 | 2519 | (void) commit(pool); |
991d9fa0 JT |
2520 | |
2521 | return r; | |
2522 | } | |
2523 | ||
e49e5829 JT |
2524 | static void emit_flags(struct pool_features *pf, char *result, |
2525 | unsigned sz, unsigned maxlen) | |
2526 | { | |
2527 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
2528 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY); | |
2529 | DMEMIT("%u ", count); | |
2530 | ||
2531 | if (!pf->zero_new_blocks) | |
2532 | DMEMIT("skip_block_zeroing "); | |
2533 | ||
2534 | if (!pf->discard_enabled) | |
2535 | DMEMIT("ignore_discard "); | |
2536 | ||
2537 | if (!pf->discard_passdown) | |
2538 | DMEMIT("no_discard_passdown "); | |
2539 | ||
2540 | if (pf->mode == PM_READ_ONLY) | |
2541 | DMEMIT("read_only "); | |
2542 | } | |
2543 | ||
991d9fa0 JT |
2544 | /* |
2545 | * Status line is: | |
2546 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2547 | * <used data sectors>/<total data sectors> <held metadata root> | |
2548 | */ | |
fd7c092e MP |
2549 | static void pool_status(struct dm_target *ti, status_type_t type, |
2550 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2551 | { |
e49e5829 | 2552 | int r; |
991d9fa0 JT |
2553 | unsigned sz = 0; |
2554 | uint64_t transaction_id; | |
2555 | dm_block_t nr_free_blocks_data; | |
2556 | dm_block_t nr_free_blocks_metadata; | |
2557 | dm_block_t nr_blocks_data; | |
2558 | dm_block_t nr_blocks_metadata; | |
2559 | dm_block_t held_root; | |
2560 | char buf[BDEVNAME_SIZE]; | |
2561 | char buf2[BDEVNAME_SIZE]; | |
2562 | struct pool_c *pt = ti->private; | |
2563 | struct pool *pool = pt->pool; | |
2564 | ||
2565 | switch (type) { | |
2566 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2567 | if (get_pool_mode(pool) == PM_FAIL) { |
2568 | DMEMIT("Fail"); | |
2569 | break; | |
2570 | } | |
2571 | ||
1f4e0ff0 AK |
2572 | /* Commit to ensure statistics aren't out-of-date */ |
2573 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2574 | (void) commit(pool); |
1f4e0ff0 | 2575 | |
fd7c092e MP |
2576 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2577 | if (r) { | |
4fa5971a MS |
2578 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2579 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2580 | goto err; |
2581 | } | |
991d9fa0 | 2582 | |
fd7c092e MP |
2583 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2584 | if (r) { | |
4fa5971a MS |
2585 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2586 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2587 | goto err; |
2588 | } | |
991d9fa0 JT |
2589 | |
2590 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2591 | if (r) { |
4fa5971a MS |
2592 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2593 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2594 | goto err; |
2595 | } | |
991d9fa0 | 2596 | |
fd7c092e MP |
2597 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2598 | if (r) { | |
4fa5971a MS |
2599 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2600 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2601 | goto err; |
2602 | } | |
991d9fa0 JT |
2603 | |
2604 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2605 | if (r) { |
4fa5971a MS |
2606 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2607 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2608 | goto err; |
2609 | } | |
991d9fa0 | 2610 | |
cc8394d8 | 2611 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2612 | if (r) { |
4fa5971a MS |
2613 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2614 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2615 | goto err; |
2616 | } | |
991d9fa0 JT |
2617 | |
2618 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2619 | (unsigned long long)transaction_id, | |
2620 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2621 | (unsigned long long)nr_blocks_metadata, | |
2622 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2623 | (unsigned long long)nr_blocks_data); | |
2624 | ||
2625 | if (held_root) | |
e49e5829 JT |
2626 | DMEMIT("%llu ", held_root); |
2627 | else | |
2628 | DMEMIT("- "); | |
2629 | ||
2630 | if (pool->pf.mode == PM_READ_ONLY) | |
2631 | DMEMIT("ro "); | |
991d9fa0 | 2632 | else |
e49e5829 JT |
2633 | DMEMIT("rw "); |
2634 | ||
018debea MS |
2635 | if (!pool->pf.discard_enabled) |
2636 | DMEMIT("ignore_discard"); | |
2637 | else if (pool->pf.discard_passdown) | |
e49e5829 JT |
2638 | DMEMIT("discard_passdown"); |
2639 | else | |
2640 | DMEMIT("no_discard_passdown"); | |
991d9fa0 JT |
2641 | |
2642 | break; | |
2643 | ||
2644 | case STATUSTYPE_TABLE: | |
2645 | DMEMIT("%s %s %lu %llu ", | |
2646 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2647 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2648 | (unsigned long)pool->sectors_per_block, | |
2649 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2650 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2651 | break; |
2652 | } | |
fd7c092e | 2653 | return; |
991d9fa0 | 2654 | |
fd7c092e MP |
2655 | err: |
2656 | DMEMIT("Error"); | |
991d9fa0 JT |
2657 | } |
2658 | ||
2659 | static int pool_iterate_devices(struct dm_target *ti, | |
2660 | iterate_devices_callout_fn fn, void *data) | |
2661 | { | |
2662 | struct pool_c *pt = ti->private; | |
2663 | ||
2664 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2665 | } | |
2666 | ||
2667 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2668 | struct bio_vec *biovec, int max_size) | |
2669 | { | |
2670 | struct pool_c *pt = ti->private; | |
2671 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2672 | ||
2673 | if (!q->merge_bvec_fn) | |
2674 | return max_size; | |
2675 | ||
2676 | bvm->bi_bdev = pt->data_dev->bdev; | |
2677 | ||
2678 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2679 | } | |
2680 | ||
0424caa1 | 2681 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2682 | { |
0424caa1 MS |
2683 | struct pool *pool = pt->pool; |
2684 | struct queue_limits *data_limits; | |
2685 | ||
104655fd JT |
2686 | limits->max_discard_sectors = pool->sectors_per_block; |
2687 | ||
2688 | /* | |
0424caa1 | 2689 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2690 | */ |
0424caa1 MS |
2691 | if (pt->adjusted_pf.discard_passdown) { |
2692 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
2693 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 2694 | } else |
0424caa1 | 2695 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
2696 | } |
2697 | ||
991d9fa0 JT |
2698 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2699 | { | |
2700 | struct pool_c *pt = ti->private; | |
2701 | struct pool *pool = pt->pool; | |
0cc67cd9 | 2702 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 2703 | |
0cc67cd9 MS |
2704 | /* |
2705 | * If the system-determined stacked limits are compatible with the | |
2706 | * pool's blocksize (io_opt is a factor) do not override them. | |
2707 | */ | |
2708 | if (io_opt_sectors < pool->sectors_per_block || | |
2709 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
2710 | blk_limits_io_min(limits, 0); | |
2711 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
2712 | } | |
0424caa1 MS |
2713 | |
2714 | /* | |
2715 | * pt->adjusted_pf is a staging area for the actual features to use. | |
2716 | * They get transferred to the live pool in bind_control_target() | |
2717 | * called from pool_preresume(). | |
2718 | */ | |
b60ab990 MS |
2719 | if (!pt->adjusted_pf.discard_enabled) { |
2720 | /* | |
2721 | * Must explicitly disallow stacking discard limits otherwise the | |
2722 | * block layer will stack them if pool's data device has support. | |
2723 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
2724 | * user to see that, so make sure to set all discard limits to 0. | |
2725 | */ | |
2726 | limits->discard_granularity = 0; | |
0424caa1 | 2727 | return; |
b60ab990 | 2728 | } |
0424caa1 MS |
2729 | |
2730 | disable_passdown_if_not_supported(pt); | |
2731 | ||
2732 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
2733 | } |
2734 | ||
2735 | static struct target_type pool_target = { | |
2736 | .name = "thin-pool", | |
2737 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2738 | DM_TARGET_IMMUTABLE, | |
94563bad | 2739 | .version = {1, 9, 0}, |
991d9fa0 JT |
2740 | .module = THIS_MODULE, |
2741 | .ctr = pool_ctr, | |
2742 | .dtr = pool_dtr, | |
2743 | .map = pool_map, | |
2744 | .postsuspend = pool_postsuspend, | |
2745 | .preresume = pool_preresume, | |
2746 | .resume = pool_resume, | |
2747 | .message = pool_message, | |
2748 | .status = pool_status, | |
2749 | .merge = pool_merge, | |
2750 | .iterate_devices = pool_iterate_devices, | |
2751 | .io_hints = pool_io_hints, | |
2752 | }; | |
2753 | ||
2754 | /*---------------------------------------------------------------- | |
2755 | * Thin target methods | |
2756 | *--------------------------------------------------------------*/ | |
2757 | static void thin_dtr(struct dm_target *ti) | |
2758 | { | |
2759 | struct thin_c *tc = ti->private; | |
2760 | ||
2761 | mutex_lock(&dm_thin_pool_table.mutex); | |
2762 | ||
2763 | __pool_dec(tc->pool); | |
2764 | dm_pool_close_thin_device(tc->td); | |
2765 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2766 | if (tc->origin_dev) |
2767 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2768 | kfree(tc); |
2769 | ||
2770 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2771 | } | |
2772 | ||
2773 | /* | |
2774 | * Thin target parameters: | |
2775 | * | |
2dd9c257 | 2776 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2777 | * |
2778 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2779 | * dev_id: the internal device identifier | |
2dd9c257 | 2780 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2781 | * |
2782 | * If the pool device has discards disabled, they get disabled for the thin | |
2783 | * device as well. | |
991d9fa0 JT |
2784 | */ |
2785 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2786 | { | |
2787 | int r; | |
2788 | struct thin_c *tc; | |
2dd9c257 | 2789 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2790 | struct mapped_device *pool_md; |
2791 | ||
2792 | mutex_lock(&dm_thin_pool_table.mutex); | |
2793 | ||
2dd9c257 | 2794 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2795 | ti->error = "Invalid argument count"; |
2796 | r = -EINVAL; | |
2797 | goto out_unlock; | |
2798 | } | |
2799 | ||
2800 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2801 | if (!tc) { | |
2802 | ti->error = "Out of memory"; | |
2803 | r = -ENOMEM; | |
2804 | goto out_unlock; | |
2805 | } | |
2806 | ||
2dd9c257 JT |
2807 | if (argc == 3) { |
2808 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2809 | if (r) { | |
2810 | ti->error = "Error opening origin device"; | |
2811 | goto bad_origin_dev; | |
2812 | } | |
2813 | tc->origin_dev = origin_dev; | |
2814 | } | |
2815 | ||
991d9fa0 JT |
2816 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2817 | if (r) { | |
2818 | ti->error = "Error opening pool device"; | |
2819 | goto bad_pool_dev; | |
2820 | } | |
2821 | tc->pool_dev = pool_dev; | |
2822 | ||
2823 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2824 | ti->error = "Invalid device id"; | |
2825 | r = -EINVAL; | |
2826 | goto bad_common; | |
2827 | } | |
2828 | ||
2829 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2830 | if (!pool_md) { | |
2831 | ti->error = "Couldn't get pool mapped device"; | |
2832 | r = -EINVAL; | |
2833 | goto bad_common; | |
2834 | } | |
2835 | ||
2836 | tc->pool = __pool_table_lookup(pool_md); | |
2837 | if (!tc->pool) { | |
2838 | ti->error = "Couldn't find pool object"; | |
2839 | r = -EINVAL; | |
2840 | goto bad_pool_lookup; | |
2841 | } | |
2842 | __pool_inc(tc->pool); | |
2843 | ||
e49e5829 JT |
2844 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2845 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
2846 | goto bad_thin_open; | |
2847 | } | |
2848 | ||
991d9fa0 JT |
2849 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
2850 | if (r) { | |
2851 | ti->error = "Couldn't open thin internal device"; | |
2852 | goto bad_thin_open; | |
2853 | } | |
2854 | ||
542f9038 MS |
2855 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2856 | if (r) | |
2857 | goto bad_thin_open; | |
2858 | ||
55a62eef | 2859 | ti->num_flush_bios = 1; |
16ad3d10 | 2860 | ti->flush_supported = true; |
59c3d2c6 | 2861 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
2862 | |
2863 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 2864 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2865 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 2866 | ti->discards_supported = true; |
55a62eef | 2867 | ti->num_discard_bios = 1; |
55a62eef AK |
2868 | /* Discard bios must be split on a block boundary */ |
2869 | ti->split_discard_bios = true; | |
67e2e2b2 | 2870 | } |
991d9fa0 JT |
2871 | |
2872 | dm_put(pool_md); | |
2873 | ||
2874 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2875 | ||
2876 | return 0; | |
2877 | ||
2878 | bad_thin_open: | |
2879 | __pool_dec(tc->pool); | |
2880 | bad_pool_lookup: | |
2881 | dm_put(pool_md); | |
2882 | bad_common: | |
2883 | dm_put_device(ti, tc->pool_dev); | |
2884 | bad_pool_dev: | |
2dd9c257 JT |
2885 | if (tc->origin_dev) |
2886 | dm_put_device(ti, tc->origin_dev); | |
2887 | bad_origin_dev: | |
991d9fa0 JT |
2888 | kfree(tc); |
2889 | out_unlock: | |
2890 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2891 | ||
2892 | return r; | |
2893 | } | |
2894 | ||
7de3ee57 | 2895 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 2896 | { |
6efd6e83 | 2897 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 | 2898 | |
7de3ee57 | 2899 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
2900 | } |
2901 | ||
7de3ee57 | 2902 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
2903 | { |
2904 | unsigned long flags; | |
59c3d2c6 | 2905 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 2906 | struct list_head work; |
a24c2569 | 2907 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2908 | struct pool *pool = h->tc->pool; |
2909 | ||
2910 | if (h->shared_read_entry) { | |
2911 | INIT_LIST_HEAD(&work); | |
44feb387 | 2912 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
2913 | |
2914 | spin_lock_irqsave(&pool->lock, flags); | |
2915 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2916 | list_del(&m->list); | |
7f214665 | 2917 | m->quiesced = true; |
eb2aa48d JT |
2918 | __maybe_add_mapping(m); |
2919 | } | |
2920 | spin_unlock_irqrestore(&pool->lock, flags); | |
2921 | } | |
2922 | ||
104655fd JT |
2923 | if (h->all_io_entry) { |
2924 | INIT_LIST_HEAD(&work); | |
44feb387 | 2925 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
2926 | if (!list_empty(&work)) { |
2927 | spin_lock_irqsave(&pool->lock, flags); | |
2928 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 2929 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
2930 | spin_unlock_irqrestore(&pool->lock, flags); |
2931 | wake_worker(pool); | |
2932 | } | |
104655fd JT |
2933 | } |
2934 | ||
eb2aa48d JT |
2935 | return 0; |
2936 | } | |
2937 | ||
991d9fa0 JT |
2938 | static void thin_postsuspend(struct dm_target *ti) |
2939 | { | |
2940 | if (dm_noflush_suspending(ti)) | |
2941 | requeue_io((struct thin_c *)ti->private); | |
2942 | } | |
2943 | ||
2944 | /* | |
2945 | * <nr mapped sectors> <highest mapped sector> | |
2946 | */ | |
fd7c092e MP |
2947 | static void thin_status(struct dm_target *ti, status_type_t type, |
2948 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
2949 | { |
2950 | int r; | |
2951 | ssize_t sz = 0; | |
2952 | dm_block_t mapped, highest; | |
2953 | char buf[BDEVNAME_SIZE]; | |
2954 | struct thin_c *tc = ti->private; | |
2955 | ||
e49e5829 JT |
2956 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2957 | DMEMIT("Fail"); | |
fd7c092e | 2958 | return; |
e49e5829 JT |
2959 | } |
2960 | ||
991d9fa0 JT |
2961 | if (!tc->td) |
2962 | DMEMIT("-"); | |
2963 | else { | |
2964 | switch (type) { | |
2965 | case STATUSTYPE_INFO: | |
2966 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
2967 | if (r) { |
2968 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
2969 | goto err; | |
2970 | } | |
991d9fa0 JT |
2971 | |
2972 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
2973 | if (r < 0) { |
2974 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
2975 | goto err; | |
2976 | } | |
991d9fa0 JT |
2977 | |
2978 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
2979 | if (r) | |
2980 | DMEMIT("%llu", ((highest + 1) * | |
2981 | tc->pool->sectors_per_block) - 1); | |
2982 | else | |
2983 | DMEMIT("-"); | |
2984 | break; | |
2985 | ||
2986 | case STATUSTYPE_TABLE: | |
2987 | DMEMIT("%s %lu", | |
2988 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
2989 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
2990 | if (tc->origin_dev) |
2991 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
2992 | break; |
2993 | } | |
2994 | } | |
2995 | ||
fd7c092e MP |
2996 | return; |
2997 | ||
2998 | err: | |
2999 | DMEMIT("Error"); | |
991d9fa0 JT |
3000 | } |
3001 | ||
3002 | static int thin_iterate_devices(struct dm_target *ti, | |
3003 | iterate_devices_callout_fn fn, void *data) | |
3004 | { | |
55f2b8bd | 3005 | sector_t blocks; |
991d9fa0 | 3006 | struct thin_c *tc = ti->private; |
55f2b8bd | 3007 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3008 | |
3009 | /* | |
3010 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3011 | * we follow a more convoluted path through to the pool's target. | |
3012 | */ | |
55f2b8bd | 3013 | if (!pool->ti) |
991d9fa0 JT |
3014 | return 0; /* nothing is bound */ |
3015 | ||
55f2b8bd MS |
3016 | blocks = pool->ti->len; |
3017 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3018 | if (blocks) |
55f2b8bd | 3019 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3020 | |
3021 | return 0; | |
3022 | } | |
3023 | ||
991d9fa0 JT |
3024 | static struct target_type thin_target = { |
3025 | .name = "thin", | |
94563bad | 3026 | .version = {1, 9, 0}, |
991d9fa0 JT |
3027 | .module = THIS_MODULE, |
3028 | .ctr = thin_ctr, | |
3029 | .dtr = thin_dtr, | |
3030 | .map = thin_map, | |
eb2aa48d | 3031 | .end_io = thin_endio, |
991d9fa0 JT |
3032 | .postsuspend = thin_postsuspend, |
3033 | .status = thin_status, | |
3034 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3035 | }; |
3036 | ||
3037 | /*----------------------------------------------------------------*/ | |
3038 | ||
3039 | static int __init dm_thin_init(void) | |
3040 | { | |
3041 | int r; | |
3042 | ||
3043 | pool_table_init(); | |
3044 | ||
3045 | r = dm_register_target(&thin_target); | |
3046 | if (r) | |
3047 | return r; | |
3048 | ||
3049 | r = dm_register_target(&pool_target); | |
3050 | if (r) | |
a24c2569 MS |
3051 | goto bad_pool_target; |
3052 | ||
3053 | r = -ENOMEM; | |
3054 | ||
a24c2569 MS |
3055 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3056 | if (!_new_mapping_cache) | |
3057 | goto bad_new_mapping_cache; | |
3058 | ||
a24c2569 MS |
3059 | return 0; |
3060 | ||
a24c2569 | 3061 | bad_new_mapping_cache: |
a24c2569 MS |
3062 | dm_unregister_target(&pool_target); |
3063 | bad_pool_target: | |
3064 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3065 | |
3066 | return r; | |
3067 | } | |
3068 | ||
3069 | static void dm_thin_exit(void) | |
3070 | { | |
3071 | dm_unregister_target(&thin_target); | |
3072 | dm_unregister_target(&pool_target); | |
a24c2569 | 3073 | |
a24c2569 | 3074 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3075 | } |
3076 | ||
3077 | module_init(dm_thin_init); | |
3078 | module_exit(dm_thin_exit); | |
3079 | ||
7cab8bf1 | 3080 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3081 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3082 | MODULE_LICENSE("GPL"); |