]>
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 | 200 | static enum pool_mode get_pool_mode(struct pool *pool); |
399caddf | 201 | static void out_of_data_space(struct pool *pool); |
b5330655 | 202 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 203 | |
991d9fa0 JT |
204 | /* |
205 | * Target context for a pool. | |
206 | */ | |
207 | struct pool_c { | |
208 | struct dm_target *ti; | |
209 | struct pool *pool; | |
210 | struct dm_dev *data_dev; | |
211 | struct dm_dev *metadata_dev; | |
212 | struct dm_target_callbacks callbacks; | |
213 | ||
214 | dm_block_t low_water_blocks; | |
0424caa1 MS |
215 | struct pool_features requested_pf; /* Features requested during table load */ |
216 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
217 | }; |
218 | ||
219 | /* | |
220 | * Target context for a thin. | |
221 | */ | |
222 | struct thin_c { | |
223 | struct dm_dev *pool_dev; | |
2dd9c257 | 224 | struct dm_dev *origin_dev; |
991d9fa0 JT |
225 | dm_thin_id dev_id; |
226 | ||
227 | struct pool *pool; | |
228 | struct dm_thin_device *td; | |
229 | }; | |
230 | ||
231 | /*----------------------------------------------------------------*/ | |
232 | ||
025b9685 JT |
233 | /* |
234 | * wake_worker() is used when new work is queued and when pool_resume is | |
235 | * ready to continue deferred IO processing. | |
236 | */ | |
237 | static void wake_worker(struct pool *pool) | |
238 | { | |
239 | queue_work(pool->wq, &pool->worker); | |
240 | } | |
241 | ||
242 | /*----------------------------------------------------------------*/ | |
243 | ||
6beca5eb JT |
244 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
245 | struct dm_bio_prison_cell **cell_result) | |
246 | { | |
247 | int r; | |
248 | struct dm_bio_prison_cell *cell_prealloc; | |
249 | ||
250 | /* | |
251 | * Allocate a cell from the prison's mempool. | |
252 | * This might block but it can't fail. | |
253 | */ | |
254 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
255 | ||
256 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
257 | if (r) | |
258 | /* | |
259 | * We reused an old cell; we can get rid of | |
260 | * the new one. | |
261 | */ | |
262 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
263 | ||
264 | return r; | |
265 | } | |
266 | ||
267 | static void cell_release(struct pool *pool, | |
268 | struct dm_bio_prison_cell *cell, | |
269 | struct bio_list *bios) | |
270 | { | |
271 | dm_cell_release(pool->prison, cell, bios); | |
272 | dm_bio_prison_free_cell(pool->prison, cell); | |
273 | } | |
274 | ||
275 | static void cell_release_no_holder(struct pool *pool, | |
276 | struct dm_bio_prison_cell *cell, | |
277 | struct bio_list *bios) | |
278 | { | |
279 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
280 | dm_bio_prison_free_cell(pool->prison, cell); | |
281 | } | |
282 | ||
025b9685 JT |
283 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
284 | struct dm_bio_prison_cell *cell) | |
285 | { | |
286 | struct pool *pool = tc->pool; | |
287 | unsigned long flags; | |
288 | ||
289 | spin_lock_irqsave(&pool->lock, flags); | |
290 | dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios); | |
291 | spin_unlock_irqrestore(&pool->lock, flags); | |
292 | ||
293 | wake_worker(pool); | |
294 | } | |
295 | ||
6beca5eb JT |
296 | static void cell_error(struct pool *pool, |
297 | struct dm_bio_prison_cell *cell) | |
298 | { | |
299 | dm_cell_error(pool->prison, cell); | |
300 | dm_bio_prison_free_cell(pool->prison, cell); | |
301 | } | |
302 | ||
303 | /*----------------------------------------------------------------*/ | |
304 | ||
991d9fa0 JT |
305 | /* |
306 | * A global list of pools that uses a struct mapped_device as a key. | |
307 | */ | |
308 | static struct dm_thin_pool_table { | |
309 | struct mutex mutex; | |
310 | struct list_head pools; | |
311 | } dm_thin_pool_table; | |
312 | ||
313 | static void pool_table_init(void) | |
314 | { | |
315 | mutex_init(&dm_thin_pool_table.mutex); | |
316 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
317 | } | |
318 | ||
319 | static void __pool_table_insert(struct pool *pool) | |
320 | { | |
321 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
322 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
323 | } | |
324 | ||
325 | static void __pool_table_remove(struct pool *pool) | |
326 | { | |
327 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
328 | list_del(&pool->list); | |
329 | } | |
330 | ||
331 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
332 | { | |
333 | struct pool *pool = NULL, *tmp; | |
334 | ||
335 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
336 | ||
337 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
338 | if (tmp->pool_md == md) { | |
339 | pool = tmp; | |
340 | break; | |
341 | } | |
342 | } | |
343 | ||
344 | return pool; | |
345 | } | |
346 | ||
347 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
348 | { | |
349 | struct pool *pool = NULL, *tmp; | |
350 | ||
351 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
352 | ||
353 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
354 | if (tmp->md_dev == md_dev) { | |
355 | pool = tmp; | |
356 | break; | |
357 | } | |
358 | } | |
359 | ||
360 | return pool; | |
361 | } | |
362 | ||
363 | /*----------------------------------------------------------------*/ | |
364 | ||
a24c2569 | 365 | struct dm_thin_endio_hook { |
eb2aa48d | 366 | struct thin_c *tc; |
44feb387 MS |
367 | struct dm_deferred_entry *shared_read_entry; |
368 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 369 | struct dm_thin_new_mapping *overwrite_mapping; |
eb2aa48d JT |
370 | }; |
371 | ||
991d9fa0 JT |
372 | static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
373 | { | |
374 | struct bio *bio; | |
375 | struct bio_list bios; | |
376 | ||
377 | bio_list_init(&bios); | |
378 | bio_list_merge(&bios, master); | |
379 | bio_list_init(master); | |
380 | ||
381 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 382 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 383 | |
eb2aa48d | 384 | if (h->tc == tc) |
991d9fa0 JT |
385 | bio_endio(bio, DM_ENDIO_REQUEUE); |
386 | else | |
387 | bio_list_add(master, bio); | |
388 | } | |
389 | } | |
390 | ||
391 | static void requeue_io(struct thin_c *tc) | |
392 | { | |
393 | struct pool *pool = tc->pool; | |
394 | unsigned long flags; | |
395 | ||
396 | spin_lock_irqsave(&pool->lock, flags); | |
397 | __requeue_bio_list(tc, &pool->deferred_bios); | |
398 | __requeue_bio_list(tc, &pool->retry_on_resume_list); | |
399 | spin_unlock_irqrestore(&pool->lock, flags); | |
400 | } | |
401 | ||
402 | /* | |
403 | * This section of code contains the logic for processing a thin device's IO. | |
404 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
405 | * but most is exclusively called from the thin target rather than the thin-pool | |
406 | * target. | |
407 | */ | |
408 | ||
58f77a21 MS |
409 | static bool block_size_is_power_of_two(struct pool *pool) |
410 | { | |
411 | return pool->sectors_per_block_shift >= 0; | |
412 | } | |
413 | ||
991d9fa0 JT |
414 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
415 | { | |
58f77a21 | 416 | struct pool *pool = tc->pool; |
55f2b8bd MS |
417 | sector_t block_nr = bio->bi_sector; |
418 | ||
58f77a21 MS |
419 | if (block_size_is_power_of_two(pool)) |
420 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 421 | else |
58f77a21 | 422 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
423 | |
424 | return block_nr; | |
991d9fa0 JT |
425 | } |
426 | ||
427 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
428 | { | |
429 | struct pool *pool = tc->pool; | |
55f2b8bd | 430 | sector_t bi_sector = bio->bi_sector; |
991d9fa0 JT |
431 | |
432 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 433 | if (block_size_is_power_of_two(pool)) |
f9a8e0cd MP |
434 | bio->bi_sector = (block << pool->sectors_per_block_shift) | |
435 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 MS |
436 | else |
437 | bio->bi_sector = (block * pool->sectors_per_block) + | |
438 | sector_div(bi_sector, pool->sectors_per_block); | |
991d9fa0 JT |
439 | } |
440 | ||
2dd9c257 JT |
441 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
442 | { | |
443 | bio->bi_bdev = tc->origin_dev->bdev; | |
444 | } | |
445 | ||
4afdd680 JT |
446 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
447 | { | |
448 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
449 | dm_thin_changed_this_transaction(tc->td); | |
450 | } | |
451 | ||
e8088073 JT |
452 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
453 | { | |
454 | struct dm_thin_endio_hook *h; | |
455 | ||
456 | if (bio->bi_rw & REQ_DISCARD) | |
457 | return; | |
458 | ||
59c3d2c6 | 459 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
460 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
461 | } | |
462 | ||
2dd9c257 | 463 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
464 | { |
465 | struct pool *pool = tc->pool; | |
466 | unsigned long flags; | |
467 | ||
e49e5829 JT |
468 | if (!bio_triggers_commit(tc, bio)) { |
469 | generic_make_request(bio); | |
470 | return; | |
471 | } | |
472 | ||
991d9fa0 | 473 | /* |
e49e5829 JT |
474 | * Complete bio with an error if earlier I/O caused changes to |
475 | * the metadata that can't be committed e.g, due to I/O errors | |
476 | * on the metadata device. | |
991d9fa0 | 477 | */ |
e49e5829 JT |
478 | if (dm_thin_aborted_changes(tc->td)) { |
479 | bio_io_error(bio); | |
480 | return; | |
481 | } | |
482 | ||
483 | /* | |
484 | * Batch together any bios that trigger commits and then issue a | |
485 | * single commit for them in process_deferred_bios(). | |
486 | */ | |
487 | spin_lock_irqsave(&pool->lock, flags); | |
488 | bio_list_add(&pool->deferred_flush_bios, bio); | |
489 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
490 | } |
491 | ||
2dd9c257 JT |
492 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
493 | { | |
494 | remap_to_origin(tc, bio); | |
495 | issue(tc, bio); | |
496 | } | |
497 | ||
498 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
499 | dm_block_t block) | |
500 | { | |
501 | remap(tc, bio, block); | |
502 | issue(tc, bio); | |
503 | } | |
504 | ||
991d9fa0 JT |
505 | /*----------------------------------------------------------------*/ |
506 | ||
507 | /* | |
508 | * Bio endio functions. | |
509 | */ | |
a24c2569 | 510 | struct dm_thin_new_mapping { |
991d9fa0 JT |
511 | struct list_head list; |
512 | ||
7f214665 MS |
513 | bool quiesced:1; |
514 | bool prepared:1; | |
515 | bool pass_discard:1; | |
516 | bool definitely_not_shared:1; | |
991d9fa0 | 517 | |
7f214665 | 518 | int err; |
991d9fa0 JT |
519 | struct thin_c *tc; |
520 | dm_block_t virt_block; | |
521 | dm_block_t data_block; | |
a24c2569 | 522 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
523 | |
524 | /* | |
525 | * If the bio covers the whole area of a block then we can avoid | |
526 | * zeroing or copying. Instead this bio is hooked. The bio will | |
527 | * still be in the cell, so care has to be taken to avoid issuing | |
528 | * the bio twice. | |
529 | */ | |
530 | struct bio *bio; | |
531 | bio_end_io_t *saved_bi_end_io; | |
532 | }; | |
533 | ||
a24c2569 | 534 | static void __maybe_add_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
535 | { |
536 | struct pool *pool = m->tc->pool; | |
537 | ||
eb2aa48d | 538 | if (m->quiesced && m->prepared) { |
daec338b | 539 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
540 | wake_worker(pool); |
541 | } | |
542 | } | |
543 | ||
544 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
545 | { | |
546 | unsigned long flags; | |
a24c2569 | 547 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
548 | struct pool *pool = m->tc->pool; |
549 | ||
550 | m->err = read_err || write_err ? -EIO : 0; | |
551 | ||
552 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 553 | m->prepared = true; |
991d9fa0 JT |
554 | __maybe_add_mapping(m); |
555 | spin_unlock_irqrestore(&pool->lock, flags); | |
556 | } | |
557 | ||
558 | static void overwrite_endio(struct bio *bio, int err) | |
559 | { | |
560 | unsigned long flags; | |
59c3d2c6 | 561 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 562 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
563 | struct pool *pool = m->tc->pool; |
564 | ||
565 | m->err = err; | |
566 | ||
567 | spin_lock_irqsave(&pool->lock, flags); | |
7f214665 | 568 | m->prepared = true; |
991d9fa0 JT |
569 | __maybe_add_mapping(m); |
570 | spin_unlock_irqrestore(&pool->lock, flags); | |
571 | } | |
572 | ||
991d9fa0 JT |
573 | /*----------------------------------------------------------------*/ |
574 | ||
575 | /* | |
576 | * Workqueue. | |
577 | */ | |
578 | ||
579 | /* | |
580 | * Prepared mapping jobs. | |
581 | */ | |
582 | ||
583 | /* | |
584 | * This sends the bios in the cell back to the deferred_bios list. | |
585 | */ | |
2aab3850 | 586 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
587 | { |
588 | struct pool *pool = tc->pool; | |
589 | unsigned long flags; | |
590 | ||
591 | spin_lock_irqsave(&pool->lock, flags); | |
6beca5eb | 592 | cell_release(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
593 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
594 | ||
595 | wake_worker(pool); | |
596 | } | |
597 | ||
598 | /* | |
6beca5eb | 599 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 600 | */ |
f286ba0e | 601 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 602 | { |
991d9fa0 JT |
603 | struct pool *pool = tc->pool; |
604 | unsigned long flags; | |
605 | ||
991d9fa0 | 606 | spin_lock_irqsave(&pool->lock, flags); |
6beca5eb | 607 | cell_release_no_holder(pool, cell, &pool->deferred_bios); |
991d9fa0 JT |
608 | spin_unlock_irqrestore(&pool->lock, flags); |
609 | ||
610 | wake_worker(pool); | |
611 | } | |
612 | ||
e49e5829 JT |
613 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
614 | { | |
615 | if (m->bio) | |
616 | m->bio->bi_end_io = m->saved_bi_end_io; | |
6beca5eb | 617 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
618 | list_del(&m->list); |
619 | mempool_free(m, m->tc->pool->mapping_pool); | |
620 | } | |
025b9685 | 621 | |
a24c2569 | 622 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
623 | { |
624 | struct thin_c *tc = m->tc; | |
6beca5eb | 625 | struct pool *pool = tc->pool; |
991d9fa0 JT |
626 | struct bio *bio; |
627 | int r; | |
628 | ||
629 | bio = m->bio; | |
630 | if (bio) | |
631 | bio->bi_end_io = m->saved_bi_end_io; | |
632 | ||
633 | if (m->err) { | |
6beca5eb | 634 | cell_error(pool, m->cell); |
905386f8 | 635 | goto out; |
991d9fa0 JT |
636 | } |
637 | ||
638 | /* | |
639 | * Commit the prepared block into the mapping btree. | |
640 | * Any I/O for this block arriving after this point will get | |
641 | * remapped to it directly. | |
642 | */ | |
643 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
644 | if (r) { | |
b5330655 | 645 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 646 | cell_error(pool, m->cell); |
905386f8 | 647 | goto out; |
991d9fa0 JT |
648 | } |
649 | ||
650 | /* | |
651 | * Release any bios held while the block was being provisioned. | |
652 | * If we are processing a write bio that completely covers the block, | |
653 | * we already processed it so can ignore it now when processing | |
654 | * the bios in the cell. | |
655 | */ | |
656 | if (bio) { | |
f286ba0e | 657 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
658 | bio_endio(bio, 0); |
659 | } else | |
2aab3850 | 660 | cell_defer(tc, m->cell); |
991d9fa0 | 661 | |
905386f8 | 662 | out: |
991d9fa0 | 663 | list_del(&m->list); |
6beca5eb | 664 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
665 | } |
666 | ||
e49e5829 | 667 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 668 | { |
104655fd JT |
669 | struct thin_c *tc = m->tc; |
670 | ||
e49e5829 | 671 | bio_io_error(m->bio); |
f286ba0e JT |
672 | cell_defer_no_holder(tc, m->cell); |
673 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
674 | mempool_free(m, tc->pool->mapping_pool); |
675 | } | |
676 | ||
677 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
678 | { | |
679 | struct thin_c *tc = m->tc; | |
104655fd | 680 | |
e8088073 | 681 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
682 | cell_defer_no_holder(tc, m->cell); |
683 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 684 | |
104655fd | 685 | if (m->pass_discard) |
19fa1a67 JT |
686 | if (m->definitely_not_shared) |
687 | remap_and_issue(tc, m->bio, m->data_block); | |
688 | else { | |
689 | bool used = false; | |
690 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
691 | bio_endio(m->bio, 0); | |
692 | else | |
693 | remap_and_issue(tc, m->bio, m->data_block); | |
694 | } | |
104655fd JT |
695 | else |
696 | bio_endio(m->bio, 0); | |
697 | ||
104655fd JT |
698 | mempool_free(m, tc->pool->mapping_pool); |
699 | } | |
700 | ||
e49e5829 JT |
701 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
702 | { | |
703 | int r; | |
704 | struct thin_c *tc = m->tc; | |
705 | ||
706 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
707 | if (r) | |
c397741c | 708 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
709 | |
710 | process_prepared_discard_passdown(m); | |
711 | } | |
712 | ||
104655fd | 713 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 714 | process_mapping_fn *fn) |
991d9fa0 JT |
715 | { |
716 | unsigned long flags; | |
717 | struct list_head maps; | |
a24c2569 | 718 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
719 | |
720 | INIT_LIST_HEAD(&maps); | |
721 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 722 | list_splice_init(head, &maps); |
991d9fa0 JT |
723 | spin_unlock_irqrestore(&pool->lock, flags); |
724 | ||
725 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 726 | (*fn)(m); |
991d9fa0 JT |
727 | } |
728 | ||
729 | /* | |
730 | * Deferred bio jobs. | |
731 | */ | |
104655fd | 732 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 733 | { |
f9a8e0cd | 734 | return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT); |
104655fd JT |
735 | } |
736 | ||
737 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
738 | { | |
739 | return (bio_data_dir(bio) == WRITE) && | |
740 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
741 | } |
742 | ||
743 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
744 | bio_end_io_t *fn) | |
745 | { | |
746 | *save = bio->bi_end_io; | |
747 | bio->bi_end_io = fn; | |
748 | } | |
749 | ||
750 | static int ensure_next_mapping(struct pool *pool) | |
751 | { | |
752 | if (pool->next_mapping) | |
753 | return 0; | |
754 | ||
755 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
756 | ||
757 | return pool->next_mapping ? 0 : -ENOMEM; | |
758 | } | |
759 | ||
a24c2569 | 760 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 761 | { |
16961b04 | 762 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
763 | |
764 | BUG_ON(!pool->next_mapping); | |
765 | ||
16961b04 MS |
766 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
767 | INIT_LIST_HEAD(&m->list); | |
768 | m->bio = NULL; | |
769 | ||
991d9fa0 JT |
770 | pool->next_mapping = NULL; |
771 | ||
16961b04 | 772 | return m; |
991d9fa0 JT |
773 | } |
774 | ||
775 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
776 | struct dm_dev *origin, dm_block_t data_origin, |
777 | dm_block_t data_dest, | |
a24c2569 | 778 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
779 | { |
780 | int r; | |
781 | struct pool *pool = tc->pool; | |
a24c2569 | 782 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 783 | |
991d9fa0 JT |
784 | m->tc = tc; |
785 | m->virt_block = virt_block; | |
786 | m->data_block = data_dest; | |
787 | m->cell = cell; | |
991d9fa0 | 788 | |
44feb387 | 789 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
7f214665 | 790 | m->quiesced = true; |
991d9fa0 JT |
791 | |
792 | /* | |
793 | * IO to pool_dev remaps to the pool target's data_dev. | |
794 | * | |
795 | * If the whole block of data is being overwritten, we can issue the | |
796 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
797 | */ | |
798 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 799 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 800 | |
eb2aa48d | 801 | h->overwrite_mapping = m; |
991d9fa0 JT |
802 | m->bio = bio; |
803 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 804 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
805 | remap_and_issue(tc, bio, data_dest); |
806 | } else { | |
807 | struct dm_io_region from, to; | |
808 | ||
2dd9c257 | 809 | from.bdev = origin->bdev; |
991d9fa0 JT |
810 | from.sector = data_origin * pool->sectors_per_block; |
811 | from.count = pool->sectors_per_block; | |
812 | ||
813 | to.bdev = tc->pool_dev->bdev; | |
814 | to.sector = data_dest * pool->sectors_per_block; | |
815 | to.count = pool->sectors_per_block; | |
816 | ||
817 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
818 | 0, copy_complete, m); | |
819 | if (r < 0) { | |
820 | mempool_free(m, pool->mapping_pool); | |
c397741c | 821 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 822 | cell_error(pool, cell); |
991d9fa0 JT |
823 | } |
824 | } | |
825 | } | |
826 | ||
2dd9c257 JT |
827 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
828 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 829 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
830 | { |
831 | schedule_copy(tc, virt_block, tc->pool_dev, | |
832 | data_origin, data_dest, cell, bio); | |
833 | } | |
834 | ||
835 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
836 | dm_block_t data_dest, | |
a24c2569 | 837 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
838 | { |
839 | schedule_copy(tc, virt_block, tc->origin_dev, | |
840 | virt_block, data_dest, cell, bio); | |
841 | } | |
842 | ||
991d9fa0 | 843 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 844 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
845 | struct bio *bio) |
846 | { | |
847 | struct pool *pool = tc->pool; | |
a24c2569 | 848 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 849 | |
7f214665 MS |
850 | m->quiesced = true; |
851 | m->prepared = false; | |
991d9fa0 JT |
852 | m->tc = tc; |
853 | m->virt_block = virt_block; | |
854 | m->data_block = data_block; | |
855 | m->cell = cell; | |
991d9fa0 JT |
856 | |
857 | /* | |
858 | * If the whole block of data is being overwritten or we are not | |
859 | * zeroing pre-existing data, we can issue the bio immediately. | |
860 | * Otherwise we use kcopyd to zero the data first. | |
861 | */ | |
67e2e2b2 | 862 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
863 | process_prepared_mapping(m); |
864 | ||
865 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 866 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 867 | |
eb2aa48d | 868 | h->overwrite_mapping = m; |
991d9fa0 JT |
869 | m->bio = bio; |
870 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 871 | inc_all_io_entry(pool, bio); |
991d9fa0 | 872 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
873 | } else { |
874 | int r; | |
875 | struct dm_io_region to; | |
876 | ||
877 | to.bdev = tc->pool_dev->bdev; | |
878 | to.sector = data_block * pool->sectors_per_block; | |
879 | to.count = pool->sectors_per_block; | |
880 | ||
881 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
882 | if (r < 0) { | |
883 | mempool_free(m, pool->mapping_pool); | |
c397741c | 884 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 885 | cell_error(pool, cell); |
991d9fa0 JT |
886 | } |
887 | } | |
888 | } | |
889 | ||
e49e5829 JT |
890 | /* |
891 | * A non-zero return indicates read_only or fail_io mode. | |
892 | * Many callers don't care about the return value. | |
893 | */ | |
020cc3b5 | 894 | static int commit(struct pool *pool) |
e49e5829 JT |
895 | { |
896 | int r; | |
897 | ||
898 | if (get_pool_mode(pool) != PM_WRITE) | |
899 | return -EINVAL; | |
900 | ||
020cc3b5 | 901 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
902 | if (r) |
903 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
904 | |
905 | return r; | |
906 | } | |
907 | ||
88a6621b JT |
908 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
909 | { | |
910 | unsigned long flags; | |
911 | ||
912 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
913 | DMWARN("%s: reached low water mark for data device: sending event.", | |
914 | dm_device_name(pool->pool_md)); | |
915 | spin_lock_irqsave(&pool->lock, flags); | |
916 | pool->low_water_triggered = true; | |
917 | spin_unlock_irqrestore(&pool->lock, flags); | |
918 | dm_table_event(pool->ti->table); | |
919 | } | |
920 | } | |
921 | ||
991d9fa0 JT |
922 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
923 | { | |
924 | int r; | |
925 | dm_block_t free_blocks; | |
991d9fa0 JT |
926 | struct pool *pool = tc->pool; |
927 | ||
8d30abff JT |
928 | if (get_pool_mode(pool) != PM_WRITE) |
929 | return -EINVAL; | |
930 | ||
991d9fa0 | 931 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
932 | if (r) { |
933 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 934 | return r; |
b5330655 | 935 | } |
991d9fa0 | 936 | |
88a6621b | 937 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
938 | |
939 | if (!free_blocks) { | |
94563bad MS |
940 | /* |
941 | * Try to commit to see if that will free up some | |
942 | * more space. | |
943 | */ | |
020cc3b5 JT |
944 | r = commit(pool); |
945 | if (r) | |
946 | return r; | |
991d9fa0 | 947 | |
94563bad | 948 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
949 | if (r) { |
950 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 951 | return r; |
b5330655 | 952 | } |
991d9fa0 | 953 | |
94563bad | 954 | if (!free_blocks) { |
399caddf | 955 | out_of_data_space(pool); |
94563bad | 956 | return -ENOSPC; |
991d9fa0 JT |
957 | } |
958 | } | |
959 | ||
960 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 961 | if (r) { |
b5330655 | 962 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 963 | return r; |
4a02b34e | 964 | } |
991d9fa0 JT |
965 | |
966 | return 0; | |
967 | } | |
968 | ||
969 | /* | |
970 | * If we have run out of space, queue bios until the device is | |
971 | * resumed, presumably after having been reloaded with more space. | |
972 | */ | |
973 | static void retry_on_resume(struct bio *bio) | |
974 | { | |
59c3d2c6 | 975 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 976 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
977 | struct pool *pool = tc->pool; |
978 | unsigned long flags; | |
979 | ||
980 | spin_lock_irqsave(&pool->lock, flags); | |
981 | bio_list_add(&pool->retry_on_resume_list, bio); | |
982 | spin_unlock_irqrestore(&pool->lock, flags); | |
983 | } | |
984 | ||
399caddf | 985 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
986 | { |
987 | struct bio *bio; | |
988 | struct bio_list bios; | |
989 | ||
990 | bio_list_init(&bios); | |
6beca5eb | 991 | cell_release(pool, cell, &bios); |
991d9fa0 JT |
992 | |
993 | while ((bio = bio_list_pop(&bios))) | |
994 | retry_on_resume(bio); | |
995 | } | |
996 | ||
104655fd JT |
997 | static void process_discard(struct thin_c *tc, struct bio *bio) |
998 | { | |
999 | int r; | |
c3a0ce2e | 1000 | unsigned long flags; |
104655fd | 1001 | struct pool *pool = tc->pool; |
a24c2569 | 1002 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1003 | struct dm_cell_key key, key2; |
104655fd JT |
1004 | dm_block_t block = get_bio_block(tc, bio); |
1005 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1006 | struct dm_thin_new_mapping *m; |
104655fd JT |
1007 | |
1008 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1009 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1010 | return; |
1011 | ||
1012 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1013 | switch (r) { | |
1014 | case 0: | |
1015 | /* | |
1016 | * Check nobody is fiddling with this pool block. This can | |
1017 | * happen if someone's in the process of breaking sharing | |
1018 | * on this block. | |
1019 | */ | |
1020 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1021 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1022 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1023 | break; |
1024 | } | |
1025 | ||
1026 | if (io_overlaps_block(pool, bio)) { | |
1027 | /* | |
1028 | * IO may still be going to the destination block. We must | |
1029 | * quiesce before we can do the removal. | |
1030 | */ | |
1031 | m = get_next_mapping(pool); | |
1032 | m->tc = tc; | |
19fa1a67 JT |
1033 | m->pass_discard = pool->pf.discard_passdown; |
1034 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1035 | m->virt_block = block; |
1036 | m->data_block = lookup_result.block; | |
1037 | m->cell = cell; | |
1038 | m->cell2 = cell2; | |
104655fd JT |
1039 | m->bio = bio; |
1040 | ||
44feb387 | 1041 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1042 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1043 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1044 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1045 | wake_worker(pool); |
1046 | } | |
1047 | } else { | |
e8088073 | 1048 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1049 | cell_defer_no_holder(tc, cell); |
1050 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1051 | |
104655fd | 1052 | /* |
49296309 MP |
1053 | * The DM core makes sure that the discard doesn't span |
1054 | * a block boundary. So we submit the discard of a | |
1055 | * partial block appropriately. | |
104655fd | 1056 | */ |
650d2a06 MP |
1057 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1058 | remap_and_issue(tc, bio, lookup_result.block); | |
1059 | else | |
1060 | bio_endio(bio, 0); | |
104655fd JT |
1061 | } |
1062 | break; | |
1063 | ||
1064 | case -ENODATA: | |
1065 | /* | |
1066 | * It isn't provisioned, just forget it. | |
1067 | */ | |
f286ba0e | 1068 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1069 | bio_endio(bio, 0); |
1070 | break; | |
1071 | ||
1072 | default: | |
c397741c MS |
1073 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1074 | __func__, r); | |
f286ba0e | 1075 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1076 | bio_io_error(bio); |
1077 | break; | |
1078 | } | |
1079 | } | |
1080 | ||
991d9fa0 | 1081 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1082 | struct dm_cell_key *key, |
991d9fa0 | 1083 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1084 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1085 | { |
1086 | int r; | |
1087 | dm_block_t data_block; | |
d6fc2042 | 1088 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1089 | |
1090 | r = alloc_data_block(tc, &data_block); | |
1091 | switch (r) { | |
1092 | case 0: | |
2dd9c257 JT |
1093 | schedule_internal_copy(tc, block, lookup_result->block, |
1094 | data_block, cell, bio); | |
991d9fa0 JT |
1095 | break; |
1096 | ||
1097 | case -ENOSPC: | |
399caddf | 1098 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1099 | break; |
1100 | ||
1101 | default: | |
c397741c MS |
1102 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1103 | __func__, r); | |
d6fc2042 | 1104 | cell_error(pool, cell); |
991d9fa0 JT |
1105 | break; |
1106 | } | |
1107 | } | |
1108 | ||
1109 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1110 | dm_block_t block, | |
1111 | struct dm_thin_lookup_result *lookup_result) | |
1112 | { | |
a24c2569 | 1113 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1114 | struct pool *pool = tc->pool; |
44feb387 | 1115 | struct dm_cell_key key; |
991d9fa0 JT |
1116 | |
1117 | /* | |
1118 | * If cell is already occupied, then sharing is already in the process | |
1119 | * of being broken so we have nothing further to do here. | |
1120 | */ | |
1121 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1122 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1123 | return; |
1124 | ||
60049701 | 1125 | if (bio_data_dir(bio) == WRITE && bio->bi_size) |
991d9fa0 JT |
1126 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1127 | else { | |
59c3d2c6 | 1128 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1129 | |
44feb387 | 1130 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1131 | inc_all_io_entry(pool, bio); |
f286ba0e | 1132 | cell_defer_no_holder(tc, cell); |
e8088073 | 1133 | |
991d9fa0 JT |
1134 | remap_and_issue(tc, bio, lookup_result->block); |
1135 | } | |
1136 | } | |
1137 | ||
1138 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1139 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1140 | { |
1141 | int r; | |
1142 | dm_block_t data_block; | |
6beca5eb | 1143 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1144 | |
1145 | /* | |
1146 | * Remap empty bios (flushes) immediately, without provisioning. | |
1147 | */ | |
1148 | if (!bio->bi_size) { | |
6beca5eb | 1149 | inc_all_io_entry(pool, bio); |
f286ba0e | 1150 | cell_defer_no_holder(tc, cell); |
e8088073 | 1151 | |
991d9fa0 JT |
1152 | remap_and_issue(tc, bio, 0); |
1153 | return; | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * Fill read bios with zeroes and complete them immediately. | |
1158 | */ | |
1159 | if (bio_data_dir(bio) == READ) { | |
1160 | zero_fill_bio(bio); | |
f286ba0e | 1161 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1162 | bio_endio(bio, 0); |
1163 | return; | |
1164 | } | |
1165 | ||
1166 | r = alloc_data_block(tc, &data_block); | |
1167 | switch (r) { | |
1168 | case 0: | |
2dd9c257 JT |
1169 | if (tc->origin_dev) |
1170 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1171 | else | |
1172 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1173 | break; |
1174 | ||
1175 | case -ENOSPC: | |
399caddf | 1176 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1177 | break; |
1178 | ||
1179 | default: | |
c397741c MS |
1180 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1181 | __func__, r); | |
6beca5eb | 1182 | cell_error(pool, cell); |
991d9fa0 JT |
1183 | break; |
1184 | } | |
1185 | } | |
1186 | ||
1187 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1188 | { | |
1189 | int r; | |
6beca5eb | 1190 | struct pool *pool = tc->pool; |
991d9fa0 | 1191 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1192 | struct dm_bio_prison_cell *cell; |
44feb387 | 1193 | struct dm_cell_key key; |
991d9fa0 JT |
1194 | struct dm_thin_lookup_result lookup_result; |
1195 | ||
1196 | /* | |
1197 | * If cell is already occupied, then the block is already | |
1198 | * being provisioned so we have nothing further to do here. | |
1199 | */ | |
1200 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1201 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1202 | return; |
1203 | ||
1204 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1205 | switch (r) { | |
1206 | case 0: | |
e8088073 | 1207 | if (lookup_result.shared) { |
991d9fa0 | 1208 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1209 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1210 | } else { |
6beca5eb | 1211 | inc_all_io_entry(pool, bio); |
f286ba0e | 1212 | cell_defer_no_holder(tc, cell); |
e8088073 | 1213 | |
991d9fa0 | 1214 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1215 | } |
991d9fa0 JT |
1216 | break; |
1217 | ||
1218 | case -ENODATA: | |
2dd9c257 | 1219 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1220 | inc_all_io_entry(pool, bio); |
f286ba0e | 1221 | cell_defer_no_holder(tc, cell); |
e8088073 | 1222 | |
2dd9c257 JT |
1223 | remap_to_origin_and_issue(tc, bio); |
1224 | } else | |
1225 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1226 | break; |
1227 | ||
1228 | default: | |
c397741c MS |
1229 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1230 | __func__, r); | |
f286ba0e | 1231 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1232 | bio_io_error(bio); |
1233 | break; | |
1234 | } | |
1235 | } | |
1236 | ||
e49e5829 JT |
1237 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1238 | { | |
1239 | int r; | |
1240 | int rw = bio_data_dir(bio); | |
1241 | dm_block_t block = get_bio_block(tc, bio); | |
1242 | struct dm_thin_lookup_result lookup_result; | |
1243 | ||
1244 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1245 | switch (r) { | |
1246 | case 0: | |
1247 | if (lookup_result.shared && (rw == WRITE) && bio->bi_size) | |
1248 | bio_io_error(bio); | |
e8088073 JT |
1249 | else { |
1250 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1251 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1252 | } |
e49e5829 JT |
1253 | break; |
1254 | ||
1255 | case -ENODATA: | |
1256 | if (rw != READ) { | |
1257 | bio_io_error(bio); | |
1258 | break; | |
1259 | } | |
1260 | ||
1261 | if (tc->origin_dev) { | |
e8088073 | 1262 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1263 | remap_to_origin_and_issue(tc, bio); |
1264 | break; | |
1265 | } | |
1266 | ||
1267 | zero_fill_bio(bio); | |
1268 | bio_endio(bio, 0); | |
1269 | break; | |
1270 | ||
1271 | default: | |
c397741c MS |
1272 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1273 | __func__, r); | |
e49e5829 JT |
1274 | bio_io_error(bio); |
1275 | break; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) | |
1280 | { | |
1281 | bio_io_error(bio); | |
1282 | } | |
1283 | ||
ac8c3f3d JT |
1284 | /* |
1285 | * FIXME: should we also commit due to size of transaction, measured in | |
1286 | * metadata blocks? | |
1287 | */ | |
905e51b3 JT |
1288 | static int need_commit_due_to_time(struct pool *pool) |
1289 | { | |
1290 | return jiffies < pool->last_commit_jiffies || | |
1291 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1292 | } | |
1293 | ||
991d9fa0 JT |
1294 | static void process_deferred_bios(struct pool *pool) |
1295 | { | |
1296 | unsigned long flags; | |
1297 | struct bio *bio; | |
1298 | struct bio_list bios; | |
991d9fa0 JT |
1299 | |
1300 | bio_list_init(&bios); | |
1301 | ||
1302 | spin_lock_irqsave(&pool->lock, flags); | |
1303 | bio_list_merge(&bios, &pool->deferred_bios); | |
1304 | bio_list_init(&pool->deferred_bios); | |
1305 | spin_unlock_irqrestore(&pool->lock, flags); | |
1306 | ||
1307 | while ((bio = bio_list_pop(&bios))) { | |
59c3d2c6 | 1308 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1309 | struct thin_c *tc = h->tc; |
1310 | ||
991d9fa0 JT |
1311 | /* |
1312 | * If we've got no free new_mapping structs, and processing | |
1313 | * this bio might require one, we pause until there are some | |
1314 | * prepared mappings to process. | |
1315 | */ | |
1316 | if (ensure_next_mapping(pool)) { | |
1317 | spin_lock_irqsave(&pool->lock, flags); | |
1318 | bio_list_merge(&pool->deferred_bios, &bios); | |
1319 | spin_unlock_irqrestore(&pool->lock, flags); | |
1320 | ||
1321 | break; | |
1322 | } | |
104655fd JT |
1323 | |
1324 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1325 | pool->process_discard(tc, bio); |
104655fd | 1326 | else |
e49e5829 | 1327 | pool->process_bio(tc, bio); |
991d9fa0 JT |
1328 | } |
1329 | ||
1330 | /* | |
1331 | * If there are any deferred flush bios, we must commit | |
1332 | * the metadata before issuing them. | |
1333 | */ | |
1334 | bio_list_init(&bios); | |
1335 | spin_lock_irqsave(&pool->lock, flags); | |
1336 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1337 | bio_list_init(&pool->deferred_flush_bios); | |
1338 | spin_unlock_irqrestore(&pool->lock, flags); | |
1339 | ||
905e51b3 | 1340 | if (bio_list_empty(&bios) && !need_commit_due_to_time(pool)) |
991d9fa0 JT |
1341 | return; |
1342 | ||
020cc3b5 | 1343 | if (commit(pool)) { |
991d9fa0 JT |
1344 | while ((bio = bio_list_pop(&bios))) |
1345 | bio_io_error(bio); | |
1346 | return; | |
1347 | } | |
905e51b3 | 1348 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1349 | |
1350 | while ((bio = bio_list_pop(&bios))) | |
1351 | generic_make_request(bio); | |
1352 | } | |
1353 | ||
1354 | static void do_worker(struct work_struct *ws) | |
1355 | { | |
1356 | struct pool *pool = container_of(ws, struct pool, worker); | |
1357 | ||
e49e5829 JT |
1358 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1359 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1360 | process_deferred_bios(pool); |
1361 | } | |
1362 | ||
905e51b3 JT |
1363 | /* |
1364 | * We want to commit periodically so that not too much | |
1365 | * unwritten data builds up. | |
1366 | */ | |
1367 | static void do_waker(struct work_struct *ws) | |
1368 | { | |
1369 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1370 | wake_worker(pool); | |
1371 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1372 | } | |
1373 | ||
991d9fa0 JT |
1374 | /*----------------------------------------------------------------*/ |
1375 | ||
e49e5829 JT |
1376 | static enum pool_mode get_pool_mode(struct pool *pool) |
1377 | { | |
1378 | return pool->pf.mode; | |
1379 | } | |
1380 | ||
1381 | static void set_pool_mode(struct pool *pool, enum pool_mode mode) | |
1382 | { | |
1383 | int r; | |
1384 | ||
1385 | pool->pf.mode = mode; | |
1386 | ||
1387 | switch (mode) { | |
1388 | case PM_FAIL: | |
4fa5971a MS |
1389 | DMERR("%s: switching pool to failure mode", |
1390 | dm_device_name(pool->pool_md)); | |
5383ef3a | 1391 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1392 | pool->process_bio = process_bio_fail; |
1393 | pool->process_discard = process_bio_fail; | |
1394 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1395 | pool->process_prepared_discard = process_prepared_discard_fail; | |
1396 | break; | |
1397 | ||
1398 | case PM_READ_ONLY: | |
4fa5971a MS |
1399 | DMERR("%s: switching pool to read-only mode", |
1400 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1401 | r = dm_pool_abort_metadata(pool->pmd); |
1402 | if (r) { | |
4fa5971a MS |
1403 | DMERR("%s: aborting transaction failed", |
1404 | dm_device_name(pool->pool_md)); | |
e49e5829 JT |
1405 | set_pool_mode(pool, PM_FAIL); |
1406 | } else { | |
1407 | dm_pool_metadata_read_only(pool->pmd); | |
1408 | pool->process_bio = process_bio_read_only; | |
1409 | pool->process_discard = process_discard; | |
1410 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1411 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1412 | } | |
1413 | break; | |
1414 | ||
1415 | case PM_WRITE: | |
9b7aaa64 | 1416 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1417 | pool->process_bio = process_bio; |
1418 | pool->process_discard = process_discard; | |
1419 | pool->process_prepared_mapping = process_prepared_mapping; | |
1420 | pool->process_prepared_discard = process_prepared_discard; | |
1421 | break; | |
1422 | } | |
1423 | } | |
1424 | ||
399caddf MS |
1425 | static void set_no_free_space(struct pool *pool) |
1426 | { | |
1427 | unsigned long flags; | |
1428 | ||
1429 | spin_lock_irqsave(&pool->lock, flags); | |
1430 | pool->no_free_space = true; | |
1431 | spin_unlock_irqrestore(&pool->lock, flags); | |
1432 | } | |
1433 | ||
b5330655 JT |
1434 | /* |
1435 | * Rather than calling set_pool_mode directly, use these which describe the | |
1436 | * reason for mode degradation. | |
1437 | */ | |
399caddf MS |
1438 | static void out_of_data_space(struct pool *pool) |
1439 | { | |
1440 | DMERR_LIMIT("%s: no free data space available.", | |
1441 | dm_device_name(pool->pool_md)); | |
1442 | set_no_free_space(pool); | |
1443 | set_pool_mode(pool, PM_READ_ONLY); | |
1444 | } | |
1445 | ||
b5330655 JT |
1446 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1447 | { | |
399caddf MS |
1448 | dm_block_t free_blocks; |
1449 | ||
b5330655 JT |
1450 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1451 | dm_device_name(pool->pool_md), op, r); | |
1452 | ||
399caddf MS |
1453 | if (r == -ENOSPC && |
1454 | !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) && | |
1455 | !free_blocks) { | |
1456 | DMERR_LIMIT("%s: no free metadata space available.", | |
1457 | dm_device_name(pool->pool_md)); | |
1458 | set_no_free_space(pool); | |
1459 | } | |
1460 | ||
b5330655 JT |
1461 | set_pool_mode(pool, PM_READ_ONLY); |
1462 | } | |
1463 | ||
e49e5829 JT |
1464 | /*----------------------------------------------------------------*/ |
1465 | ||
991d9fa0 JT |
1466 | /* |
1467 | * Mapping functions. | |
1468 | */ | |
1469 | ||
1470 | /* | |
1471 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1472 | */ | |
1473 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1474 | { | |
1475 | unsigned long flags; | |
1476 | struct pool *pool = tc->pool; | |
1477 | ||
1478 | spin_lock_irqsave(&pool->lock, flags); | |
1479 | bio_list_add(&pool->deferred_bios, bio); | |
1480 | spin_unlock_irqrestore(&pool->lock, flags); | |
1481 | ||
1482 | wake_worker(pool); | |
1483 | } | |
1484 | ||
59c3d2c6 | 1485 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1486 | { |
59c3d2c6 | 1487 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1488 | |
1489 | h->tc = tc; | |
1490 | h->shared_read_entry = NULL; | |
e8088073 | 1491 | h->all_io_entry = NULL; |
eb2aa48d | 1492 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1493 | } |
1494 | ||
991d9fa0 JT |
1495 | /* |
1496 | * Non-blocking function called from the thin target's map function. | |
1497 | */ | |
7de3ee57 | 1498 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1499 | { |
1500 | int r; | |
1501 | struct thin_c *tc = ti->private; | |
1502 | dm_block_t block = get_bio_block(tc, bio); | |
1503 | struct dm_thin_device *td = tc->td; | |
1504 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1505 | struct dm_bio_prison_cell cell1, cell2; |
1506 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1507 | struct dm_cell_key key; |
991d9fa0 | 1508 | |
59c3d2c6 | 1509 | thin_hook_bio(tc, bio); |
e49e5829 JT |
1510 | |
1511 | if (get_pool_mode(tc->pool) == PM_FAIL) { | |
1512 | bio_io_error(bio); | |
1513 | return DM_MAPIO_SUBMITTED; | |
1514 | } | |
1515 | ||
104655fd | 1516 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1517 | thin_defer_bio(tc, bio); |
1518 | return DM_MAPIO_SUBMITTED; | |
1519 | } | |
1520 | ||
1521 | r = dm_thin_find_block(td, block, 0, &result); | |
1522 | ||
1523 | /* | |
1524 | * Note that we defer readahead too. | |
1525 | */ | |
1526 | switch (r) { | |
1527 | case 0: | |
1528 | if (unlikely(result.shared)) { | |
1529 | /* | |
1530 | * We have a race condition here between the | |
1531 | * result.shared value returned by the lookup and | |
1532 | * snapshot creation, which may cause new | |
1533 | * sharing. | |
1534 | * | |
1535 | * To avoid this always quiesce the origin before | |
1536 | * taking the snap. You want to do this anyway to | |
1537 | * ensure a consistent application view | |
1538 | * (i.e. lockfs). | |
1539 | * | |
1540 | * More distant ancestors are irrelevant. The | |
1541 | * shared flag will be set in their case. | |
1542 | */ | |
1543 | thin_defer_bio(tc, bio); | |
e8088073 | 1544 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1545 | } |
e8088073 JT |
1546 | |
1547 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1548 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1549 | return DM_MAPIO_SUBMITTED; |
1550 | ||
1551 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1552 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1553 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1554 | return DM_MAPIO_SUBMITTED; |
1555 | } | |
1556 | ||
1557 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1558 | cell_defer_no_holder_no_free(tc, &cell2); |
1559 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1560 | |
1561 | remap(tc, bio, result.block); | |
1562 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1563 | |
1564 | case -ENODATA: | |
e49e5829 JT |
1565 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1566 | /* | |
1567 | * This block isn't provisioned, and we have no way | |
1568 | * of doing so. Just error it. | |
1569 | */ | |
1570 | bio_io_error(bio); | |
2aab3850 | 1571 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1572 | } |
1573 | /* fall through */ | |
1574 | ||
1575 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1576 | /* |
1577 | * In future, the failed dm_thin_find_block above could | |
1578 | * provide the hint to load the metadata into cache. | |
1579 | */ | |
991d9fa0 | 1580 | thin_defer_bio(tc, bio); |
2aab3850 | 1581 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1582 | |
1583 | default: | |
1584 | /* | |
1585 | * Must always call bio_io_error on failure. | |
1586 | * dm_thin_find_block can fail with -EINVAL if the | |
1587 | * pool is switched to fail-io mode. | |
1588 | */ | |
1589 | bio_io_error(bio); | |
2aab3850 | 1590 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1591 | } |
991d9fa0 JT |
1592 | } |
1593 | ||
1594 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1595 | { | |
1596 | int r; | |
1597 | unsigned long flags; | |
1598 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); | |
1599 | ||
1600 | spin_lock_irqsave(&pt->pool->lock, flags); | |
1601 | r = !bio_list_empty(&pt->pool->retry_on_resume_list); | |
1602 | spin_unlock_irqrestore(&pt->pool->lock, flags); | |
1603 | ||
1604 | if (!r) { | |
1605 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1606 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
1607 | } | |
1608 | ||
1609 | return r; | |
1610 | } | |
1611 | ||
1612 | static void __requeue_bios(struct pool *pool) | |
1613 | { | |
1614 | bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list); | |
1615 | bio_list_init(&pool->retry_on_resume_list); | |
1616 | } | |
1617 | ||
1618 | /*---------------------------------------------------------------- | |
1619 | * Binding of control targets to a pool object | |
1620 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1621 | static bool data_dev_supports_discard(struct pool_c *pt) |
1622 | { | |
1623 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1624 | ||
1625 | return q && blk_queue_discard(q); | |
1626 | } | |
1627 | ||
58051b94 JT |
1628 | static bool is_factor(sector_t block_size, uint32_t n) |
1629 | { | |
1630 | return !sector_div(block_size, n); | |
1631 | } | |
1632 | ||
9bc142dd MS |
1633 | /* |
1634 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1635 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1636 | */ |
0424caa1 | 1637 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1638 | { |
0424caa1 MS |
1639 | struct pool *pool = pt->pool; |
1640 | struct block_device *data_bdev = pt->data_dev->bdev; | |
1641 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
1642 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
1643 | const char *reason = NULL; | |
9bc142dd MS |
1644 | char buf[BDEVNAME_SIZE]; |
1645 | ||
0424caa1 | 1646 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
1647 | return; |
1648 | ||
0424caa1 MS |
1649 | if (!data_dev_supports_discard(pt)) |
1650 | reason = "discard unsupported"; | |
1651 | ||
1652 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
1653 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 1654 | |
0424caa1 MS |
1655 | else if (data_limits->discard_granularity > block_size) |
1656 | reason = "discard granularity larger than a block"; | |
1657 | ||
58051b94 | 1658 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
1659 | reason = "discard granularity not a factor of block size"; |
1660 | ||
1661 | if (reason) { | |
1662 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
1663 | pt->adjusted_pf.discard_passdown = false; | |
1664 | } | |
9bc142dd MS |
1665 | } |
1666 | ||
991d9fa0 JT |
1667 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
1668 | { | |
1669 | struct pool_c *pt = ti->private; | |
1670 | ||
e49e5829 | 1671 | /* |
9b7aaa64 | 1672 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 JT |
1673 | */ |
1674 | enum pool_mode old_mode = pool->pf.mode; | |
0424caa1 | 1675 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 1676 | |
9b7aaa64 JT |
1677 | /* |
1678 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1679 | * not going to recover without a thin_repair. So we never let the | |
1680 | * pool move out of the old mode. On the other hand a PM_READ_ONLY | |
1681 | * may have been due to a lack of metadata or data space, and may | |
1682 | * now work (ie. if the underlying devices have been resized). | |
1683 | */ | |
1684 | if (old_mode == PM_FAIL) | |
e49e5829 JT |
1685 | new_mode = old_mode; |
1686 | ||
991d9fa0 JT |
1687 | pool->ti = ti; |
1688 | pool->low_water_blocks = pt->low_water_blocks; | |
0424caa1 | 1689 | pool->pf = pt->adjusted_pf; |
991d9fa0 | 1690 | |
9bc142dd | 1691 | set_pool_mode(pool, new_mode); |
f402693d | 1692 | |
991d9fa0 JT |
1693 | return 0; |
1694 | } | |
1695 | ||
1696 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
1697 | { | |
1698 | if (pool->ti == ti) | |
1699 | pool->ti = NULL; | |
1700 | } | |
1701 | ||
1702 | /*---------------------------------------------------------------- | |
1703 | * Pool creation | |
1704 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
1705 | /* Initialize pool features. */ |
1706 | static void pool_features_init(struct pool_features *pf) | |
1707 | { | |
e49e5829 | 1708 | pf->mode = PM_WRITE; |
9bc142dd MS |
1709 | pf->zero_new_blocks = true; |
1710 | pf->discard_enabled = true; | |
1711 | pf->discard_passdown = true; | |
67e2e2b2 JT |
1712 | } |
1713 | ||
991d9fa0 JT |
1714 | static void __pool_destroy(struct pool *pool) |
1715 | { | |
1716 | __pool_table_remove(pool); | |
1717 | ||
1718 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
1719 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1720 | ||
44feb387 | 1721 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1722 | dm_kcopyd_client_destroy(pool->copier); |
1723 | ||
1724 | if (pool->wq) | |
1725 | destroy_workqueue(pool->wq); | |
1726 | ||
1727 | if (pool->next_mapping) | |
1728 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
1729 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
1730 | dm_deferred_set_destroy(pool->shared_read_ds); |
1731 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
1732 | kfree(pool); |
1733 | } | |
1734 | ||
a24c2569 | 1735 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 1736 | |
991d9fa0 JT |
1737 | static struct pool *pool_create(struct mapped_device *pool_md, |
1738 | struct block_device *metadata_dev, | |
e49e5829 JT |
1739 | unsigned long block_size, |
1740 | int read_only, char **error) | |
991d9fa0 JT |
1741 | { |
1742 | int r; | |
1743 | void *err_p; | |
1744 | struct pool *pool; | |
1745 | struct dm_pool_metadata *pmd; | |
e49e5829 | 1746 | bool format_device = read_only ? false : true; |
991d9fa0 | 1747 | |
e49e5829 | 1748 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
1749 | if (IS_ERR(pmd)) { |
1750 | *error = "Error creating metadata object"; | |
1751 | return (struct pool *)pmd; | |
1752 | } | |
1753 | ||
1754 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
1755 | if (!pool) { | |
1756 | *error = "Error allocating memory for pool"; | |
1757 | err_p = ERR_PTR(-ENOMEM); | |
1758 | goto bad_pool; | |
1759 | } | |
1760 | ||
1761 | pool->pmd = pmd; | |
1762 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
1763 | if (block_size & (block_size - 1)) |
1764 | pool->sectors_per_block_shift = -1; | |
1765 | else | |
1766 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 1767 | pool->low_water_blocks = 0; |
67e2e2b2 | 1768 | pool_features_init(&pool->pf); |
44feb387 | 1769 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
1770 | if (!pool->prison) { |
1771 | *error = "Error creating pool's bio prison"; | |
1772 | err_p = ERR_PTR(-ENOMEM); | |
1773 | goto bad_prison; | |
1774 | } | |
1775 | ||
df5d2e90 | 1776 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
1777 | if (IS_ERR(pool->copier)) { |
1778 | r = PTR_ERR(pool->copier); | |
1779 | *error = "Error creating pool's kcopyd client"; | |
1780 | err_p = ERR_PTR(r); | |
1781 | goto bad_kcopyd_client; | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Create singlethreaded workqueue that will service all devices | |
1786 | * that use this metadata. | |
1787 | */ | |
1788 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1789 | if (!pool->wq) { | |
1790 | *error = "Error creating pool's workqueue"; | |
1791 | err_p = ERR_PTR(-ENOMEM); | |
1792 | goto bad_wq; | |
1793 | } | |
1794 | ||
1795 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 1796 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
991d9fa0 JT |
1797 | spin_lock_init(&pool->lock); |
1798 | bio_list_init(&pool->deferred_bios); | |
1799 | bio_list_init(&pool->deferred_flush_bios); | |
1800 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 1801 | INIT_LIST_HEAD(&pool->prepared_discards); |
88a6621b JT |
1802 | pool->low_water_triggered = false; |
1803 | pool->no_free_space = false; | |
991d9fa0 | 1804 | bio_list_init(&pool->retry_on_resume_list); |
44feb387 MS |
1805 | |
1806 | pool->shared_read_ds = dm_deferred_set_create(); | |
1807 | if (!pool->shared_read_ds) { | |
1808 | *error = "Error creating pool's shared read deferred set"; | |
1809 | err_p = ERR_PTR(-ENOMEM); | |
1810 | goto bad_shared_read_ds; | |
1811 | } | |
1812 | ||
1813 | pool->all_io_ds = dm_deferred_set_create(); | |
1814 | if (!pool->all_io_ds) { | |
1815 | *error = "Error creating pool's all io deferred set"; | |
1816 | err_p = ERR_PTR(-ENOMEM); | |
1817 | goto bad_all_io_ds; | |
1818 | } | |
991d9fa0 JT |
1819 | |
1820 | pool->next_mapping = NULL; | |
a24c2569 MS |
1821 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
1822 | _new_mapping_cache); | |
991d9fa0 JT |
1823 | if (!pool->mapping_pool) { |
1824 | *error = "Error creating pool's mapping mempool"; | |
1825 | err_p = ERR_PTR(-ENOMEM); | |
1826 | goto bad_mapping_pool; | |
1827 | } | |
1828 | ||
991d9fa0 | 1829 | pool->ref_count = 1; |
905e51b3 | 1830 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1831 | pool->pool_md = pool_md; |
1832 | pool->md_dev = metadata_dev; | |
1833 | __pool_table_insert(pool); | |
1834 | ||
1835 | return pool; | |
1836 | ||
991d9fa0 | 1837 | bad_mapping_pool: |
44feb387 MS |
1838 | dm_deferred_set_destroy(pool->all_io_ds); |
1839 | bad_all_io_ds: | |
1840 | dm_deferred_set_destroy(pool->shared_read_ds); | |
1841 | bad_shared_read_ds: | |
991d9fa0 JT |
1842 | destroy_workqueue(pool->wq); |
1843 | bad_wq: | |
1844 | dm_kcopyd_client_destroy(pool->copier); | |
1845 | bad_kcopyd_client: | |
44feb387 | 1846 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
1847 | bad_prison: |
1848 | kfree(pool); | |
1849 | bad_pool: | |
1850 | if (dm_pool_metadata_close(pmd)) | |
1851 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
1852 | ||
1853 | return err_p; | |
1854 | } | |
1855 | ||
1856 | static void __pool_inc(struct pool *pool) | |
1857 | { | |
1858 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1859 | pool->ref_count++; | |
1860 | } | |
1861 | ||
1862 | static void __pool_dec(struct pool *pool) | |
1863 | { | |
1864 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
1865 | BUG_ON(!pool->ref_count); | |
1866 | if (!--pool->ref_count) | |
1867 | __pool_destroy(pool); | |
1868 | } | |
1869 | ||
1870 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
1871 | struct block_device *metadata_dev, | |
e49e5829 JT |
1872 | unsigned long block_size, int read_only, |
1873 | char **error, int *created) | |
991d9fa0 JT |
1874 | { |
1875 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
1876 | ||
1877 | if (pool) { | |
f09996c9 MS |
1878 | if (pool->pool_md != pool_md) { |
1879 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 1880 | return ERR_PTR(-EBUSY); |
f09996c9 | 1881 | } |
991d9fa0 JT |
1882 | __pool_inc(pool); |
1883 | ||
1884 | } else { | |
1885 | pool = __pool_table_lookup(pool_md); | |
1886 | if (pool) { | |
f09996c9 MS |
1887 | if (pool->md_dev != metadata_dev) { |
1888 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 1889 | return ERR_PTR(-EINVAL); |
f09996c9 | 1890 | } |
991d9fa0 JT |
1891 | __pool_inc(pool); |
1892 | ||
67e2e2b2 | 1893 | } else { |
e49e5829 | 1894 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
1895 | *created = 1; |
1896 | } | |
991d9fa0 JT |
1897 | } |
1898 | ||
1899 | return pool; | |
1900 | } | |
1901 | ||
1902 | /*---------------------------------------------------------------- | |
1903 | * Pool target methods | |
1904 | *--------------------------------------------------------------*/ | |
1905 | static void pool_dtr(struct dm_target *ti) | |
1906 | { | |
1907 | struct pool_c *pt = ti->private; | |
1908 | ||
1909 | mutex_lock(&dm_thin_pool_table.mutex); | |
1910 | ||
1911 | unbind_control_target(pt->pool, ti); | |
1912 | __pool_dec(pt->pool); | |
1913 | dm_put_device(ti, pt->metadata_dev); | |
1914 | dm_put_device(ti, pt->data_dev); | |
1915 | kfree(pt); | |
1916 | ||
1917 | mutex_unlock(&dm_thin_pool_table.mutex); | |
1918 | } | |
1919 | ||
991d9fa0 JT |
1920 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
1921 | struct dm_target *ti) | |
1922 | { | |
1923 | int r; | |
1924 | unsigned argc; | |
1925 | const char *arg_name; | |
1926 | ||
1927 | static struct dm_arg _args[] = { | |
67e2e2b2 | 1928 | {0, 3, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
1929 | }; |
1930 | ||
1931 | /* | |
1932 | * No feature arguments supplied. | |
1933 | */ | |
1934 | if (!as->argc) | |
1935 | return 0; | |
1936 | ||
1937 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1938 | if (r) | |
1939 | return -EINVAL; | |
1940 | ||
1941 | while (argc && !r) { | |
1942 | arg_name = dm_shift_arg(as); | |
1943 | argc--; | |
1944 | ||
e49e5829 | 1945 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 1946 | pf->zero_new_blocks = false; |
e49e5829 JT |
1947 | |
1948 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 1949 | pf->discard_enabled = false; |
e49e5829 JT |
1950 | |
1951 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 1952 | pf->discard_passdown = false; |
991d9fa0 | 1953 | |
e49e5829 JT |
1954 | else if (!strcasecmp(arg_name, "read_only")) |
1955 | pf->mode = PM_READ_ONLY; | |
1956 | ||
1957 | else { | |
1958 | ti->error = "Unrecognised pool feature requested"; | |
1959 | r = -EINVAL; | |
1960 | break; | |
1961 | } | |
991d9fa0 JT |
1962 | } |
1963 | ||
1964 | return r; | |
1965 | } | |
1966 | ||
ac8c3f3d JT |
1967 | static void metadata_low_callback(void *context) |
1968 | { | |
1969 | struct pool *pool = context; | |
1970 | ||
1971 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
1972 | dm_device_name(pool->pool_md)); | |
1973 | ||
1974 | dm_table_event(pool->ti->table); | |
1975 | } | |
1976 | ||
b17446df JT |
1977 | static sector_t get_metadata_dev_size(struct block_device *bdev) |
1978 | { | |
1979 | sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
1980 | char buffer[BDEVNAME_SIZE]; | |
1981 | ||
1982 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) { | |
1983 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", | |
1984 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
1985 | metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING; | |
1986 | } | |
1987 | ||
1988 | return metadata_dev_size; | |
1989 | } | |
1990 | ||
24347e95 JT |
1991 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
1992 | { | |
1993 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
1994 | ||
1995 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); | |
1996 | ||
1997 | return metadata_dev_size; | |
1998 | } | |
1999 | ||
ac8c3f3d JT |
2000 | /* |
2001 | * When a metadata threshold is crossed a dm event is triggered, and | |
2002 | * userland should respond by growing the metadata device. We could let | |
2003 | * userland set the threshold, like we do with the data threshold, but I'm | |
2004 | * not sure they know enough to do this well. | |
2005 | */ | |
2006 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2007 | { | |
2008 | /* | |
2009 | * 4M is ample for all ops with the possible exception of thin | |
2010 | * device deletion which is harmless if it fails (just retry the | |
2011 | * delete after you've grown the device). | |
2012 | */ | |
2013 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2014 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2015 | } | |
2016 | ||
991d9fa0 JT |
2017 | /* |
2018 | * thin-pool <metadata dev> <data dev> | |
2019 | * <data block size (sectors)> | |
2020 | * <low water mark (blocks)> | |
2021 | * [<#feature args> [<arg>]*] | |
2022 | * | |
2023 | * Optional feature arguments are: | |
2024 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2025 | * ignore_discard: disable discard |
2026 | * no_discard_passdown: don't pass discards down to the data device | |
991d9fa0 JT |
2027 | */ |
2028 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2029 | { | |
67e2e2b2 | 2030 | int r, pool_created = 0; |
991d9fa0 JT |
2031 | struct pool_c *pt; |
2032 | struct pool *pool; | |
2033 | struct pool_features pf; | |
2034 | struct dm_arg_set as; | |
2035 | struct dm_dev *data_dev; | |
2036 | unsigned long block_size; | |
2037 | dm_block_t low_water_blocks; | |
2038 | struct dm_dev *metadata_dev; | |
5d0db96d | 2039 | fmode_t metadata_mode; |
991d9fa0 JT |
2040 | |
2041 | /* | |
2042 | * FIXME Remove validation from scope of lock. | |
2043 | */ | |
2044 | mutex_lock(&dm_thin_pool_table.mutex); | |
2045 | ||
2046 | if (argc < 4) { | |
2047 | ti->error = "Invalid argument count"; | |
2048 | r = -EINVAL; | |
2049 | goto out_unlock; | |
2050 | } | |
5d0db96d | 2051 | |
991d9fa0 JT |
2052 | as.argc = argc; |
2053 | as.argv = argv; | |
2054 | ||
5d0db96d JT |
2055 | /* |
2056 | * Set default pool features. | |
2057 | */ | |
2058 | pool_features_init(&pf); | |
2059 | ||
2060 | dm_consume_args(&as, 4); | |
2061 | r = parse_pool_features(&as, &pf, ti); | |
2062 | if (r) | |
2063 | goto out_unlock; | |
2064 | ||
2065 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2066 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2067 | if (r) { |
2068 | ti->error = "Error opening metadata block device"; | |
2069 | goto out_unlock; | |
2070 | } | |
2071 | ||
b17446df JT |
2072 | /* |
2073 | * Run for the side-effect of possibly issuing a warning if the | |
2074 | * device is too big. | |
2075 | */ | |
2076 | (void) get_metadata_dev_size(metadata_dev->bdev); | |
991d9fa0 JT |
2077 | |
2078 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2079 | if (r) { | |
2080 | ti->error = "Error getting data device"; | |
2081 | goto out_metadata; | |
2082 | } | |
2083 | ||
2084 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2085 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2086 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2087 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2088 | ti->error = "Invalid block size"; |
2089 | r = -EINVAL; | |
2090 | goto out; | |
2091 | } | |
2092 | ||
2093 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2094 | ti->error = "Invalid low water mark"; | |
2095 | r = -EINVAL; | |
2096 | goto out; | |
2097 | } | |
2098 | ||
991d9fa0 JT |
2099 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2100 | if (!pt) { | |
2101 | r = -ENOMEM; | |
2102 | goto out; | |
2103 | } | |
2104 | ||
2105 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2106 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2107 | if (IS_ERR(pool)) { |
2108 | r = PTR_ERR(pool); | |
2109 | goto out_free_pt; | |
2110 | } | |
2111 | ||
67e2e2b2 JT |
2112 | /* |
2113 | * 'pool_created' reflects whether this is the first table load. | |
2114 | * Top level discard support is not allowed to be changed after | |
2115 | * initial load. This would require a pool reload to trigger thin | |
2116 | * device changes. | |
2117 | */ | |
2118 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2119 | ti->error = "Discard support cannot be disabled once enabled"; | |
2120 | r = -EINVAL; | |
2121 | goto out_flags_changed; | |
2122 | } | |
2123 | ||
991d9fa0 JT |
2124 | pt->pool = pool; |
2125 | pt->ti = ti; | |
2126 | pt->metadata_dev = metadata_dev; | |
2127 | pt->data_dev = data_dev; | |
2128 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2129 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2130 | ti->num_flush_bios = 1; |
9bc142dd | 2131 | |
67e2e2b2 JT |
2132 | /* |
2133 | * Only need to enable discards if the pool should pass | |
2134 | * them down to the data device. The thin device's discard | |
2135 | * processing will cause mappings to be removed from the btree. | |
2136 | */ | |
b60ab990 | 2137 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2138 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2139 | ti->num_discard_bios = 1; |
9bc142dd | 2140 | |
67e2e2b2 JT |
2141 | /* |
2142 | * Setting 'discards_supported' circumvents the normal | |
2143 | * stacking of discard limits (this keeps the pool and | |
2144 | * thin devices' discard limits consistent). | |
2145 | */ | |
0ac55489 | 2146 | ti->discards_supported = true; |
67e2e2b2 | 2147 | } |
991d9fa0 JT |
2148 | ti->private = pt; |
2149 | ||
ac8c3f3d JT |
2150 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2151 | calc_metadata_threshold(pt), | |
2152 | metadata_low_callback, | |
2153 | pool); | |
2154 | if (r) | |
2155 | goto out_free_pt; | |
2156 | ||
991d9fa0 JT |
2157 | pt->callbacks.congested_fn = pool_is_congested; |
2158 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2159 | ||
2160 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2161 | ||
2162 | return 0; | |
2163 | ||
67e2e2b2 JT |
2164 | out_flags_changed: |
2165 | __pool_dec(pool); | |
991d9fa0 JT |
2166 | out_free_pt: |
2167 | kfree(pt); | |
2168 | out: | |
2169 | dm_put_device(ti, data_dev); | |
2170 | out_metadata: | |
2171 | dm_put_device(ti, metadata_dev); | |
2172 | out_unlock: | |
2173 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2174 | ||
2175 | return r; | |
2176 | } | |
2177 | ||
7de3ee57 | 2178 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2179 | { |
2180 | int r; | |
2181 | struct pool_c *pt = ti->private; | |
2182 | struct pool *pool = pt->pool; | |
2183 | unsigned long flags; | |
2184 | ||
2185 | /* | |
2186 | * As this is a singleton target, ti->begin is always zero. | |
2187 | */ | |
2188 | spin_lock_irqsave(&pool->lock, flags); | |
2189 | bio->bi_bdev = pt->data_dev->bdev; | |
2190 | r = DM_MAPIO_REMAPPED; | |
2191 | spin_unlock_irqrestore(&pool->lock, flags); | |
2192 | ||
2193 | return r; | |
2194 | } | |
2195 | ||
b17446df | 2196 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2197 | { |
2198 | int r; | |
2199 | struct pool_c *pt = ti->private; | |
2200 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2201 | sector_t data_size = ti->len; |
2202 | dm_block_t sb_data_size; | |
991d9fa0 | 2203 | |
b17446df | 2204 | *need_commit = false; |
991d9fa0 | 2205 | |
55f2b8bd MS |
2206 | (void) sector_div(data_size, pool->sectors_per_block); |
2207 | ||
991d9fa0 JT |
2208 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2209 | if (r) { | |
4fa5971a MS |
2210 | DMERR("%s: failed to retrieve data device size", |
2211 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2212 | return r; |
2213 | } | |
2214 | ||
2215 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2216 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2217 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2218 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2219 | return -EINVAL; |
2220 | ||
2221 | } else if (data_size > sb_data_size) { | |
6f7f51d4 MS |
2222 | if (sb_data_size) |
2223 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2224 | dm_device_name(pool->pool_md), | |
2225 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2226 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2227 | if (r) { | |
b5330655 | 2228 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2229 | return r; |
2230 | } | |
2231 | ||
b17446df | 2232 | *need_commit = true; |
991d9fa0 JT |
2233 | } |
2234 | ||
2235 | return 0; | |
2236 | } | |
2237 | ||
24347e95 JT |
2238 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2239 | { | |
2240 | int r; | |
2241 | struct pool_c *pt = ti->private; | |
2242 | struct pool *pool = pt->pool; | |
2243 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2244 | ||
2245 | *need_commit = false; | |
2246 | ||
610bba8b | 2247 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2248 | |
2249 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2250 | if (r) { | |
4fa5971a MS |
2251 | DMERR("%s: failed to retrieve metadata device size", |
2252 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2253 | return r; |
2254 | } | |
2255 | ||
2256 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2257 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2258 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2259 | metadata_dev_size, sb_metadata_dev_size); |
2260 | return -EINVAL; | |
2261 | ||
2262 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
6f7f51d4 MS |
2263 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2264 | dm_device_name(pool->pool_md), | |
2265 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2266 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2267 | if (r) { | |
b5330655 | 2268 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2269 | return r; |
2270 | } | |
2271 | ||
2272 | *need_commit = true; | |
2273 | } | |
2274 | ||
2275 | return 0; | |
2276 | } | |
2277 | ||
b17446df JT |
2278 | /* |
2279 | * Retrieves the number of blocks of the data device from | |
2280 | * the superblock and compares it to the actual device size, | |
2281 | * thus resizing the data device in case it has grown. | |
2282 | * | |
2283 | * This both copes with opening preallocated data devices in the ctr | |
2284 | * being followed by a resume | |
2285 | * -and- | |
2286 | * calling the resume method individually after userspace has | |
2287 | * grown the data device in reaction to a table event. | |
2288 | */ | |
2289 | static int pool_preresume(struct dm_target *ti) | |
2290 | { | |
2291 | int r; | |
24347e95 | 2292 | bool need_commit1, need_commit2; |
b17446df JT |
2293 | struct pool_c *pt = ti->private; |
2294 | struct pool *pool = pt->pool; | |
2295 | ||
2296 | /* | |
2297 | * Take control of the pool object. | |
2298 | */ | |
2299 | r = bind_control_target(pool, ti); | |
2300 | if (r) | |
2301 | return r; | |
2302 | ||
2303 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2304 | if (r) | |
2305 | return r; | |
2306 | ||
24347e95 JT |
2307 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2308 | if (r) | |
2309 | return r; | |
2310 | ||
2311 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2312 | (void) commit(pool); |
b17446df JT |
2313 | |
2314 | return 0; | |
2315 | } | |
2316 | ||
991d9fa0 JT |
2317 | static void pool_resume(struct dm_target *ti) |
2318 | { | |
2319 | struct pool_c *pt = ti->private; | |
2320 | struct pool *pool = pt->pool; | |
2321 | unsigned long flags; | |
2322 | ||
2323 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b JT |
2324 | pool->low_water_triggered = false; |
2325 | pool->no_free_space = false; | |
991d9fa0 JT |
2326 | __requeue_bios(pool); |
2327 | spin_unlock_irqrestore(&pool->lock, flags); | |
2328 | ||
905e51b3 | 2329 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2330 | } |
2331 | ||
2332 | static void pool_postsuspend(struct dm_target *ti) | |
2333 | { | |
991d9fa0 JT |
2334 | struct pool_c *pt = ti->private; |
2335 | struct pool *pool = pt->pool; | |
2336 | ||
905e51b3 | 2337 | cancel_delayed_work(&pool->waker); |
991d9fa0 | 2338 | flush_workqueue(pool->wq); |
020cc3b5 | 2339 | (void) commit(pool); |
991d9fa0 JT |
2340 | } |
2341 | ||
2342 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2343 | { | |
2344 | if (argc != args_required) { | |
2345 | DMWARN("Message received with %u arguments instead of %u.", | |
2346 | argc, args_required); | |
2347 | return -EINVAL; | |
2348 | } | |
2349 | ||
2350 | return 0; | |
2351 | } | |
2352 | ||
2353 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2354 | { | |
2355 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2356 | *dev_id <= MAX_DEV_ID) | |
2357 | return 0; | |
2358 | ||
2359 | if (warning) | |
2360 | DMWARN("Message received with invalid device id: %s", arg); | |
2361 | ||
2362 | return -EINVAL; | |
2363 | } | |
2364 | ||
2365 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2366 | { | |
2367 | dm_thin_id dev_id; | |
2368 | int r; | |
2369 | ||
2370 | r = check_arg_count(argc, 2); | |
2371 | if (r) | |
2372 | return r; | |
2373 | ||
2374 | r = read_dev_id(argv[1], &dev_id, 1); | |
2375 | if (r) | |
2376 | return r; | |
2377 | ||
2378 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2379 | if (r) { | |
2380 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2381 | argv[1]); | |
2382 | return r; | |
2383 | } | |
2384 | ||
2385 | return 0; | |
2386 | } | |
2387 | ||
2388 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2389 | { | |
2390 | dm_thin_id dev_id; | |
2391 | dm_thin_id origin_dev_id; | |
2392 | int r; | |
2393 | ||
2394 | r = check_arg_count(argc, 3); | |
2395 | if (r) | |
2396 | return r; | |
2397 | ||
2398 | r = read_dev_id(argv[1], &dev_id, 1); | |
2399 | if (r) | |
2400 | return r; | |
2401 | ||
2402 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2403 | if (r) | |
2404 | return r; | |
2405 | ||
2406 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2407 | if (r) { | |
2408 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2409 | argv[1], argv[2]); | |
2410 | return r; | |
2411 | } | |
2412 | ||
2413 | return 0; | |
2414 | } | |
2415 | ||
2416 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2417 | { | |
2418 | dm_thin_id dev_id; | |
2419 | int r; | |
2420 | ||
2421 | r = check_arg_count(argc, 2); | |
2422 | if (r) | |
2423 | return r; | |
2424 | ||
2425 | r = read_dev_id(argv[1], &dev_id, 1); | |
2426 | if (r) | |
2427 | return r; | |
2428 | ||
2429 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2430 | if (r) | |
2431 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2432 | ||
2433 | return r; | |
2434 | } | |
2435 | ||
2436 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2437 | { | |
2438 | dm_thin_id old_id, new_id; | |
2439 | int r; | |
2440 | ||
2441 | r = check_arg_count(argc, 3); | |
2442 | if (r) | |
2443 | return r; | |
2444 | ||
2445 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2446 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2447 | return -EINVAL; | |
2448 | } | |
2449 | ||
2450 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2451 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2452 | return -EINVAL; | |
2453 | } | |
2454 | ||
2455 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2456 | if (r) { | |
2457 | DMWARN("Failed to change transaction id from %s to %s.", | |
2458 | argv[1], argv[2]); | |
2459 | return r; | |
2460 | } | |
2461 | ||
2462 | return 0; | |
2463 | } | |
2464 | ||
cc8394d8 JT |
2465 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2466 | { | |
2467 | int r; | |
2468 | ||
2469 | r = check_arg_count(argc, 1); | |
2470 | if (r) | |
2471 | return r; | |
2472 | ||
020cc3b5 | 2473 | (void) commit(pool); |
0d200aef | 2474 | |
cc8394d8 JT |
2475 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2476 | if (r) | |
2477 | DMWARN("reserve_metadata_snap message failed."); | |
2478 | ||
2479 | return r; | |
2480 | } | |
2481 | ||
2482 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2483 | { | |
2484 | int r; | |
2485 | ||
2486 | r = check_arg_count(argc, 1); | |
2487 | if (r) | |
2488 | return r; | |
2489 | ||
2490 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2491 | if (r) | |
2492 | DMWARN("release_metadata_snap message failed."); | |
2493 | ||
2494 | return r; | |
2495 | } | |
2496 | ||
991d9fa0 JT |
2497 | /* |
2498 | * Messages supported: | |
2499 | * create_thin <dev_id> | |
2500 | * create_snap <dev_id> <origin_id> | |
2501 | * delete <dev_id> | |
2502 | * trim <dev_id> <new_size_in_sectors> | |
2503 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2504 | * reserve_metadata_snap |
2505 | * release_metadata_snap | |
991d9fa0 JT |
2506 | */ |
2507 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2508 | { | |
2509 | int r = -EINVAL; | |
2510 | struct pool_c *pt = ti->private; | |
2511 | struct pool *pool = pt->pool; | |
2512 | ||
2513 | if (!strcasecmp(argv[0], "create_thin")) | |
2514 | r = process_create_thin_mesg(argc, argv, pool); | |
2515 | ||
2516 | else if (!strcasecmp(argv[0], "create_snap")) | |
2517 | r = process_create_snap_mesg(argc, argv, pool); | |
2518 | ||
2519 | else if (!strcasecmp(argv[0], "delete")) | |
2520 | r = process_delete_mesg(argc, argv, pool); | |
2521 | ||
2522 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2523 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2524 | ||
cc8394d8 JT |
2525 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2526 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2527 | ||
2528 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2529 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2530 | ||
991d9fa0 JT |
2531 | else |
2532 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2533 | ||
e49e5829 | 2534 | if (!r) |
020cc3b5 | 2535 | (void) commit(pool); |
991d9fa0 JT |
2536 | |
2537 | return r; | |
2538 | } | |
2539 | ||
e49e5829 JT |
2540 | static void emit_flags(struct pool_features *pf, char *result, |
2541 | unsigned sz, unsigned maxlen) | |
2542 | { | |
2543 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
2544 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY); | |
2545 | DMEMIT("%u ", count); | |
2546 | ||
2547 | if (!pf->zero_new_blocks) | |
2548 | DMEMIT("skip_block_zeroing "); | |
2549 | ||
2550 | if (!pf->discard_enabled) | |
2551 | DMEMIT("ignore_discard "); | |
2552 | ||
2553 | if (!pf->discard_passdown) | |
2554 | DMEMIT("no_discard_passdown "); | |
2555 | ||
2556 | if (pf->mode == PM_READ_ONLY) | |
2557 | DMEMIT("read_only "); | |
2558 | } | |
2559 | ||
991d9fa0 JT |
2560 | /* |
2561 | * Status line is: | |
2562 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2563 | * <used data sectors>/<total data sectors> <held metadata root> | |
2564 | */ | |
fd7c092e MP |
2565 | static void pool_status(struct dm_target *ti, status_type_t type, |
2566 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2567 | { |
e49e5829 | 2568 | int r; |
991d9fa0 JT |
2569 | unsigned sz = 0; |
2570 | uint64_t transaction_id; | |
2571 | dm_block_t nr_free_blocks_data; | |
2572 | dm_block_t nr_free_blocks_metadata; | |
2573 | dm_block_t nr_blocks_data; | |
2574 | dm_block_t nr_blocks_metadata; | |
2575 | dm_block_t held_root; | |
2576 | char buf[BDEVNAME_SIZE]; | |
2577 | char buf2[BDEVNAME_SIZE]; | |
2578 | struct pool_c *pt = ti->private; | |
2579 | struct pool *pool = pt->pool; | |
2580 | ||
2581 | switch (type) { | |
2582 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2583 | if (get_pool_mode(pool) == PM_FAIL) { |
2584 | DMEMIT("Fail"); | |
2585 | break; | |
2586 | } | |
2587 | ||
1f4e0ff0 AK |
2588 | /* Commit to ensure statistics aren't out-of-date */ |
2589 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2590 | (void) commit(pool); |
1f4e0ff0 | 2591 | |
fd7c092e MP |
2592 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2593 | if (r) { | |
4fa5971a MS |
2594 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2595 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2596 | goto err; |
2597 | } | |
991d9fa0 | 2598 | |
fd7c092e MP |
2599 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2600 | if (r) { | |
4fa5971a MS |
2601 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2602 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2603 | goto err; |
2604 | } | |
991d9fa0 JT |
2605 | |
2606 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2607 | if (r) { |
4fa5971a MS |
2608 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2609 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2610 | goto err; |
2611 | } | |
991d9fa0 | 2612 | |
fd7c092e MP |
2613 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2614 | if (r) { | |
4fa5971a MS |
2615 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
2616 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2617 | goto err; |
2618 | } | |
991d9fa0 JT |
2619 | |
2620 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 2621 | if (r) { |
4fa5971a MS |
2622 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
2623 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2624 | goto err; |
2625 | } | |
991d9fa0 | 2626 | |
cc8394d8 | 2627 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 2628 | if (r) { |
4fa5971a MS |
2629 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
2630 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2631 | goto err; |
2632 | } | |
991d9fa0 JT |
2633 | |
2634 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
2635 | (unsigned long long)transaction_id, | |
2636 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
2637 | (unsigned long long)nr_blocks_metadata, | |
2638 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
2639 | (unsigned long long)nr_blocks_data); | |
2640 | ||
2641 | if (held_root) | |
e49e5829 JT |
2642 | DMEMIT("%llu ", held_root); |
2643 | else | |
2644 | DMEMIT("- "); | |
2645 | ||
2646 | if (pool->pf.mode == PM_READ_ONLY) | |
2647 | DMEMIT("ro "); | |
991d9fa0 | 2648 | else |
e49e5829 JT |
2649 | DMEMIT("rw "); |
2650 | ||
018debea MS |
2651 | if (!pool->pf.discard_enabled) |
2652 | DMEMIT("ignore_discard"); | |
2653 | else if (pool->pf.discard_passdown) | |
e49e5829 JT |
2654 | DMEMIT("discard_passdown"); |
2655 | else | |
2656 | DMEMIT("no_discard_passdown"); | |
991d9fa0 JT |
2657 | |
2658 | break; | |
2659 | ||
2660 | case STATUSTYPE_TABLE: | |
2661 | DMEMIT("%s %s %lu %llu ", | |
2662 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
2663 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
2664 | (unsigned long)pool->sectors_per_block, | |
2665 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 2666 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
2667 | break; |
2668 | } | |
fd7c092e | 2669 | return; |
991d9fa0 | 2670 | |
fd7c092e MP |
2671 | err: |
2672 | DMEMIT("Error"); | |
991d9fa0 JT |
2673 | } |
2674 | ||
2675 | static int pool_iterate_devices(struct dm_target *ti, | |
2676 | iterate_devices_callout_fn fn, void *data) | |
2677 | { | |
2678 | struct pool_c *pt = ti->private; | |
2679 | ||
2680 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
2681 | } | |
2682 | ||
2683 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
2684 | struct bio_vec *biovec, int max_size) | |
2685 | { | |
2686 | struct pool_c *pt = ti->private; | |
2687 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2688 | ||
2689 | if (!q->merge_bvec_fn) | |
2690 | return max_size; | |
2691 | ||
2692 | bvm->bi_bdev = pt->data_dev->bdev; | |
2693 | ||
2694 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2695 | } | |
2696 | ||
0424caa1 | 2697 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 2698 | { |
0424caa1 MS |
2699 | struct pool *pool = pt->pool; |
2700 | struct queue_limits *data_limits; | |
2701 | ||
104655fd JT |
2702 | limits->max_discard_sectors = pool->sectors_per_block; |
2703 | ||
2704 | /* | |
0424caa1 | 2705 | * discard_granularity is just a hint, and not enforced. |
104655fd | 2706 | */ |
0424caa1 MS |
2707 | if (pt->adjusted_pf.discard_passdown) { |
2708 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
2709 | limits->discard_granularity = data_limits->discard_granularity; | |
f13945d7 | 2710 | } else |
0424caa1 | 2711 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
2712 | } |
2713 | ||
991d9fa0 JT |
2714 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2715 | { | |
2716 | struct pool_c *pt = ti->private; | |
2717 | struct pool *pool = pt->pool; | |
0cc67cd9 | 2718 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 2719 | |
0cc67cd9 MS |
2720 | /* |
2721 | * If the system-determined stacked limits are compatible with the | |
2722 | * pool's blocksize (io_opt is a factor) do not override them. | |
2723 | */ | |
2724 | if (io_opt_sectors < pool->sectors_per_block || | |
2725 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
2726 | blk_limits_io_min(limits, 0); | |
2727 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
2728 | } | |
0424caa1 MS |
2729 | |
2730 | /* | |
2731 | * pt->adjusted_pf is a staging area for the actual features to use. | |
2732 | * They get transferred to the live pool in bind_control_target() | |
2733 | * called from pool_preresume(). | |
2734 | */ | |
b60ab990 MS |
2735 | if (!pt->adjusted_pf.discard_enabled) { |
2736 | /* | |
2737 | * Must explicitly disallow stacking discard limits otherwise the | |
2738 | * block layer will stack them if pool's data device has support. | |
2739 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
2740 | * user to see that, so make sure to set all discard limits to 0. | |
2741 | */ | |
2742 | limits->discard_granularity = 0; | |
0424caa1 | 2743 | return; |
b60ab990 | 2744 | } |
0424caa1 MS |
2745 | |
2746 | disable_passdown_if_not_supported(pt); | |
2747 | ||
2748 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
2749 | } |
2750 | ||
2751 | static struct target_type pool_target = { | |
2752 | .name = "thin-pool", | |
2753 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
2754 | DM_TARGET_IMMUTABLE, | |
94563bad | 2755 | .version = {1, 9, 0}, |
991d9fa0 JT |
2756 | .module = THIS_MODULE, |
2757 | .ctr = pool_ctr, | |
2758 | .dtr = pool_dtr, | |
2759 | .map = pool_map, | |
2760 | .postsuspend = pool_postsuspend, | |
2761 | .preresume = pool_preresume, | |
2762 | .resume = pool_resume, | |
2763 | .message = pool_message, | |
2764 | .status = pool_status, | |
2765 | .merge = pool_merge, | |
2766 | .iterate_devices = pool_iterate_devices, | |
2767 | .io_hints = pool_io_hints, | |
2768 | }; | |
2769 | ||
2770 | /*---------------------------------------------------------------- | |
2771 | * Thin target methods | |
2772 | *--------------------------------------------------------------*/ | |
2773 | static void thin_dtr(struct dm_target *ti) | |
2774 | { | |
2775 | struct thin_c *tc = ti->private; | |
2776 | ||
2777 | mutex_lock(&dm_thin_pool_table.mutex); | |
2778 | ||
2779 | __pool_dec(tc->pool); | |
2780 | dm_pool_close_thin_device(tc->td); | |
2781 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
2782 | if (tc->origin_dev) |
2783 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
2784 | kfree(tc); |
2785 | ||
2786 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2787 | } | |
2788 | ||
2789 | /* | |
2790 | * Thin target parameters: | |
2791 | * | |
2dd9c257 | 2792 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
2793 | * |
2794 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
2795 | * dev_id: the internal device identifier | |
2dd9c257 | 2796 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
2797 | * |
2798 | * If the pool device has discards disabled, they get disabled for the thin | |
2799 | * device as well. | |
991d9fa0 JT |
2800 | */ |
2801 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2802 | { | |
2803 | int r; | |
2804 | struct thin_c *tc; | |
2dd9c257 | 2805 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
2806 | struct mapped_device *pool_md; |
2807 | ||
2808 | mutex_lock(&dm_thin_pool_table.mutex); | |
2809 | ||
2dd9c257 | 2810 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
2811 | ti->error = "Invalid argument count"; |
2812 | r = -EINVAL; | |
2813 | goto out_unlock; | |
2814 | } | |
2815 | ||
2816 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
2817 | if (!tc) { | |
2818 | ti->error = "Out of memory"; | |
2819 | r = -ENOMEM; | |
2820 | goto out_unlock; | |
2821 | } | |
2822 | ||
2dd9c257 JT |
2823 | if (argc == 3) { |
2824 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
2825 | if (r) { | |
2826 | ti->error = "Error opening origin device"; | |
2827 | goto bad_origin_dev; | |
2828 | } | |
2829 | tc->origin_dev = origin_dev; | |
2830 | } | |
2831 | ||
991d9fa0 JT |
2832 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
2833 | if (r) { | |
2834 | ti->error = "Error opening pool device"; | |
2835 | goto bad_pool_dev; | |
2836 | } | |
2837 | tc->pool_dev = pool_dev; | |
2838 | ||
2839 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
2840 | ti->error = "Invalid device id"; | |
2841 | r = -EINVAL; | |
2842 | goto bad_common; | |
2843 | } | |
2844 | ||
2845 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
2846 | if (!pool_md) { | |
2847 | ti->error = "Couldn't get pool mapped device"; | |
2848 | r = -EINVAL; | |
2849 | goto bad_common; | |
2850 | } | |
2851 | ||
2852 | tc->pool = __pool_table_lookup(pool_md); | |
2853 | if (!tc->pool) { | |
2854 | ti->error = "Couldn't find pool object"; | |
2855 | r = -EINVAL; | |
2856 | goto bad_pool_lookup; | |
2857 | } | |
2858 | __pool_inc(tc->pool); | |
2859 | ||
e49e5829 JT |
2860 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2861 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
2862 | goto bad_thin_open; | |
2863 | } | |
2864 | ||
991d9fa0 JT |
2865 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
2866 | if (r) { | |
2867 | ti->error = "Couldn't open thin internal device"; | |
2868 | goto bad_thin_open; | |
2869 | } | |
2870 | ||
542f9038 MS |
2871 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
2872 | if (r) | |
2873 | goto bad_thin_open; | |
2874 | ||
55a62eef | 2875 | ti->num_flush_bios = 1; |
16ad3d10 | 2876 | ti->flush_supported = true; |
59c3d2c6 | 2877 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
2878 | |
2879 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 2880 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2881 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 2882 | ti->discards_supported = true; |
55a62eef | 2883 | ti->num_discard_bios = 1; |
55a62eef AK |
2884 | /* Discard bios must be split on a block boundary */ |
2885 | ti->split_discard_bios = true; | |
67e2e2b2 | 2886 | } |
991d9fa0 JT |
2887 | |
2888 | dm_put(pool_md); | |
2889 | ||
2890 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2891 | ||
2892 | return 0; | |
2893 | ||
2894 | bad_thin_open: | |
2895 | __pool_dec(tc->pool); | |
2896 | bad_pool_lookup: | |
2897 | dm_put(pool_md); | |
2898 | bad_common: | |
2899 | dm_put_device(ti, tc->pool_dev); | |
2900 | bad_pool_dev: | |
2dd9c257 JT |
2901 | if (tc->origin_dev) |
2902 | dm_put_device(ti, tc->origin_dev); | |
2903 | bad_origin_dev: | |
991d9fa0 JT |
2904 | kfree(tc); |
2905 | out_unlock: | |
2906 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2907 | ||
2908 | return r; | |
2909 | } | |
2910 | ||
7de3ee57 | 2911 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 2912 | { |
6efd6e83 | 2913 | bio->bi_sector = dm_target_offset(ti, bio->bi_sector); |
991d9fa0 | 2914 | |
7de3ee57 | 2915 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
2916 | } |
2917 | ||
7de3ee57 | 2918 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
2919 | { |
2920 | unsigned long flags; | |
59c3d2c6 | 2921 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 2922 | struct list_head work; |
a24c2569 | 2923 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
2924 | struct pool *pool = h->tc->pool; |
2925 | ||
2926 | if (h->shared_read_entry) { | |
2927 | INIT_LIST_HEAD(&work); | |
44feb387 | 2928 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
2929 | |
2930 | spin_lock_irqsave(&pool->lock, flags); | |
2931 | list_for_each_entry_safe(m, tmp, &work, list) { | |
2932 | list_del(&m->list); | |
7f214665 | 2933 | m->quiesced = true; |
eb2aa48d JT |
2934 | __maybe_add_mapping(m); |
2935 | } | |
2936 | spin_unlock_irqrestore(&pool->lock, flags); | |
2937 | } | |
2938 | ||
104655fd JT |
2939 | if (h->all_io_entry) { |
2940 | INIT_LIST_HEAD(&work); | |
44feb387 | 2941 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
2942 | if (!list_empty(&work)) { |
2943 | spin_lock_irqsave(&pool->lock, flags); | |
2944 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 2945 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
2946 | spin_unlock_irqrestore(&pool->lock, flags); |
2947 | wake_worker(pool); | |
2948 | } | |
104655fd JT |
2949 | } |
2950 | ||
eb2aa48d JT |
2951 | return 0; |
2952 | } | |
2953 | ||
991d9fa0 JT |
2954 | static void thin_postsuspend(struct dm_target *ti) |
2955 | { | |
2956 | if (dm_noflush_suspending(ti)) | |
2957 | requeue_io((struct thin_c *)ti->private); | |
2958 | } | |
2959 | ||
2960 | /* | |
2961 | * <nr mapped sectors> <highest mapped sector> | |
2962 | */ | |
fd7c092e MP |
2963 | static void thin_status(struct dm_target *ti, status_type_t type, |
2964 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
2965 | { |
2966 | int r; | |
2967 | ssize_t sz = 0; | |
2968 | dm_block_t mapped, highest; | |
2969 | char buf[BDEVNAME_SIZE]; | |
2970 | struct thin_c *tc = ti->private; | |
2971 | ||
e49e5829 JT |
2972 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2973 | DMEMIT("Fail"); | |
fd7c092e | 2974 | return; |
e49e5829 JT |
2975 | } |
2976 | ||
991d9fa0 JT |
2977 | if (!tc->td) |
2978 | DMEMIT("-"); | |
2979 | else { | |
2980 | switch (type) { | |
2981 | case STATUSTYPE_INFO: | |
2982 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
2983 | if (r) { |
2984 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
2985 | goto err; | |
2986 | } | |
991d9fa0 JT |
2987 | |
2988 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
2989 | if (r < 0) { |
2990 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
2991 | goto err; | |
2992 | } | |
991d9fa0 JT |
2993 | |
2994 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
2995 | if (r) | |
2996 | DMEMIT("%llu", ((highest + 1) * | |
2997 | tc->pool->sectors_per_block) - 1); | |
2998 | else | |
2999 | DMEMIT("-"); | |
3000 | break; | |
3001 | ||
3002 | case STATUSTYPE_TABLE: | |
3003 | DMEMIT("%s %lu", | |
3004 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3005 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3006 | if (tc->origin_dev) |
3007 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3008 | break; |
3009 | } | |
3010 | } | |
3011 | ||
fd7c092e MP |
3012 | return; |
3013 | ||
3014 | err: | |
3015 | DMEMIT("Error"); | |
991d9fa0 JT |
3016 | } |
3017 | ||
3018 | static int thin_iterate_devices(struct dm_target *ti, | |
3019 | iterate_devices_callout_fn fn, void *data) | |
3020 | { | |
55f2b8bd | 3021 | sector_t blocks; |
991d9fa0 | 3022 | struct thin_c *tc = ti->private; |
55f2b8bd | 3023 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3024 | |
3025 | /* | |
3026 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3027 | * we follow a more convoluted path through to the pool's target. | |
3028 | */ | |
55f2b8bd | 3029 | if (!pool->ti) |
991d9fa0 JT |
3030 | return 0; /* nothing is bound */ |
3031 | ||
55f2b8bd MS |
3032 | blocks = pool->ti->len; |
3033 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3034 | if (blocks) |
55f2b8bd | 3035 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3036 | |
3037 | return 0; | |
3038 | } | |
3039 | ||
991d9fa0 JT |
3040 | static struct target_type thin_target = { |
3041 | .name = "thin", | |
94563bad | 3042 | .version = {1, 9, 0}, |
991d9fa0 JT |
3043 | .module = THIS_MODULE, |
3044 | .ctr = thin_ctr, | |
3045 | .dtr = thin_dtr, | |
3046 | .map = thin_map, | |
eb2aa48d | 3047 | .end_io = thin_endio, |
991d9fa0 JT |
3048 | .postsuspend = thin_postsuspend, |
3049 | .status = thin_status, | |
3050 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3051 | }; |
3052 | ||
3053 | /*----------------------------------------------------------------*/ | |
3054 | ||
3055 | static int __init dm_thin_init(void) | |
3056 | { | |
3057 | int r; | |
3058 | ||
3059 | pool_table_init(); | |
3060 | ||
3061 | r = dm_register_target(&thin_target); | |
3062 | if (r) | |
3063 | return r; | |
3064 | ||
3065 | r = dm_register_target(&pool_target); | |
3066 | if (r) | |
a24c2569 MS |
3067 | goto bad_pool_target; |
3068 | ||
3069 | r = -ENOMEM; | |
3070 | ||
a24c2569 MS |
3071 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3072 | if (!_new_mapping_cache) | |
3073 | goto bad_new_mapping_cache; | |
3074 | ||
a24c2569 MS |
3075 | return 0; |
3076 | ||
a24c2569 | 3077 | bad_new_mapping_cache: |
a24c2569 MS |
3078 | dm_unregister_target(&pool_target); |
3079 | bad_pool_target: | |
3080 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3081 | |
3082 | return r; | |
3083 | } | |
3084 | ||
3085 | static void dm_thin_exit(void) | |
3086 | { | |
3087 | dm_unregister_target(&thin_target); | |
3088 | dm_unregister_target(&pool_target); | |
a24c2569 | 3089 | |
a24c2569 | 3090 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3091 | } |
3092 | ||
3093 | module_init(dm_thin_init); | |
3094 | module_exit(dm_thin_exit); | |
3095 | ||
7cab8bf1 | 3096 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3097 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3098 | MODULE_LICENSE("GPL"); |