]>
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> | |
c140e1c4 | 15 | #include <linux/rculist.h> |
991d9fa0 JT |
16 | #include <linux/init.h> |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
67324ea1 | 19 | #include <linux/rbtree.h> |
991d9fa0 JT |
20 | |
21 | #define DM_MSG_PREFIX "thin" | |
22 | ||
23 | /* | |
24 | * Tunable constants | |
25 | */ | |
7768ed33 | 26 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 JT |
27 | #define MAPPING_POOL_SIZE 1024 |
28 | #define PRISON_CELLS 1024 | |
905e51b3 | 29 | #define COMMIT_PERIOD HZ |
80c57893 MS |
30 | #define NO_SPACE_TIMEOUT_SECS 60 |
31 | ||
32 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 33 | |
df5d2e90 MP |
34 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
35 | "A percentage of time allocated for copy on write"); | |
36 | ||
991d9fa0 JT |
37 | /* |
38 | * The block size of the device holding pool data must be | |
39 | * between 64KB and 1GB. | |
40 | */ | |
41 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
42 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
43 | ||
991d9fa0 JT |
44 | /* |
45 | * Device id is restricted to 24 bits. | |
46 | */ | |
47 | #define MAX_DEV_ID ((1 << 24) - 1) | |
48 | ||
49 | /* | |
50 | * How do we handle breaking sharing of data blocks? | |
51 | * ================================================= | |
52 | * | |
53 | * We use a standard copy-on-write btree to store the mappings for the | |
54 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
55 | * the data). When you take an internal snapshot you clone the root node | |
56 | * of the origin btree. After this there is no concept of an origin or a | |
57 | * snapshot. They are just two device trees that happen to point to the | |
58 | * same data blocks. | |
59 | * | |
60 | * When we get a write in we decide if it's to a shared data block using | |
61 | * some timestamp magic. If it is, we have to break sharing. | |
62 | * | |
63 | * Let's say we write to a shared block in what was the origin. The | |
64 | * steps are: | |
65 | * | |
66 | * i) plug io further to this physical block. (see bio_prison code). | |
67 | * | |
68 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 69 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
70 | * |
71 | * iii) copy the data block to a newly allocate block. This step can be | |
72 | * missed out if the io covers the block. (schedule_copy). | |
73 | * | |
74 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 75 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
76 | * sharing of btree nodes between the two devices. Breaking sharing only |
77 | * effects the btree of that specific device. Btrees for the other | |
78 | * devices that share the block never change. The btree for the origin | |
79 | * device as it was after the last commit is untouched, ie. we're using | |
80 | * persistent data structures in the functional programming sense. | |
81 | * | |
82 | * v) unplug io to this physical block, including the io that triggered | |
83 | * the breaking of sharing. | |
84 | * | |
85 | * Steps (ii) and (iii) occur in parallel. | |
86 | * | |
87 | * The metadata _doesn't_ need to be committed before the io continues. We | |
88 | * get away with this because the io is always written to a _new_ block. | |
89 | * If there's a crash, then: | |
90 | * | |
91 | * - The origin mapping will point to the old origin block (the shared | |
92 | * one). This will contain the data as it was before the io that triggered | |
93 | * the breaking of sharing came in. | |
94 | * | |
95 | * - The snap mapping still points to the old block. As it would after | |
96 | * the commit. | |
97 | * | |
98 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
99 | * will continue to think that data block in the snapshot device is shared | |
100 | * even after the write to the origin has broken sharing. I suspect data | |
101 | * blocks will typically be shared by many different devices, so we're | |
102 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
103 | * devices that reference this data block. At the moment I think the | |
104 | * benefits far, far outweigh the disadvantages. | |
105 | */ | |
106 | ||
107 | /*----------------------------------------------------------------*/ | |
108 | ||
991d9fa0 JT |
109 | /* |
110 | * Key building. | |
111 | */ | |
112 | static void build_data_key(struct dm_thin_device *td, | |
44feb387 | 113 | dm_block_t b, struct dm_cell_key *key) |
991d9fa0 JT |
114 | { |
115 | key->virtual = 0; | |
116 | key->dev = dm_thin_dev_id(td); | |
117 | key->block = b; | |
118 | } | |
119 | ||
120 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 121 | struct dm_cell_key *key) |
991d9fa0 JT |
122 | { |
123 | key->virtual = 1; | |
124 | key->dev = dm_thin_dev_id(td); | |
125 | key->block = b; | |
126 | } | |
127 | ||
128 | /*----------------------------------------------------------------*/ | |
129 | ||
130 | /* | |
131 | * A pool device ties together a metadata device and a data device. It | |
132 | * also provides the interface for creating and destroying internal | |
133 | * devices. | |
134 | */ | |
a24c2569 | 135 | struct dm_thin_new_mapping; |
67e2e2b2 | 136 | |
e49e5829 | 137 | /* |
3e1a0699 | 138 | * The pool runs in 4 modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
139 | */ |
140 | enum pool_mode { | |
141 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 142 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
e49e5829 JT |
143 | PM_READ_ONLY, /* metadata may not be changed */ |
144 | PM_FAIL, /* all I/O fails */ | |
145 | }; | |
146 | ||
67e2e2b2 | 147 | struct pool_features { |
e49e5829 JT |
148 | enum pool_mode mode; |
149 | ||
9bc142dd MS |
150 | bool zero_new_blocks:1; |
151 | bool discard_enabled:1; | |
152 | bool discard_passdown:1; | |
787a996c | 153 | bool error_if_no_space:1; |
67e2e2b2 JT |
154 | }; |
155 | ||
e49e5829 JT |
156 | struct thin_c; |
157 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
158 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); | |
159 | ||
991d9fa0 JT |
160 | struct pool { |
161 | struct list_head list; | |
162 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
163 | ||
164 | struct mapped_device *pool_md; | |
165 | struct block_device *md_dev; | |
166 | struct dm_pool_metadata *pmd; | |
167 | ||
991d9fa0 | 168 | dm_block_t low_water_blocks; |
55f2b8bd | 169 | uint32_t sectors_per_block; |
f9a8e0cd | 170 | int sectors_per_block_shift; |
991d9fa0 | 171 | |
67e2e2b2 | 172 | struct pool_features pf; |
88a6621b | 173 | bool low_water_triggered:1; /* A dm event has been sent */ |
991d9fa0 | 174 | |
44feb387 | 175 | struct dm_bio_prison *prison; |
991d9fa0 JT |
176 | struct dm_kcopyd_client *copier; |
177 | ||
178 | struct workqueue_struct *wq; | |
179 | struct work_struct worker; | |
905e51b3 | 180 | struct delayed_work waker; |
85ad643b | 181 | struct delayed_work no_space_timeout; |
991d9fa0 | 182 | |
905e51b3 | 183 | unsigned long last_commit_jiffies; |
55f2b8bd | 184 | unsigned ref_count; |
991d9fa0 JT |
185 | |
186 | spinlock_t lock; | |
991d9fa0 JT |
187 | struct bio_list deferred_flush_bios; |
188 | struct list_head prepared_mappings; | |
104655fd | 189 | struct list_head prepared_discards; |
c140e1c4 | 190 | struct list_head active_thins; |
991d9fa0 | 191 | |
44feb387 MS |
192 | struct dm_deferred_set *shared_read_ds; |
193 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 194 | |
a24c2569 | 195 | struct dm_thin_new_mapping *next_mapping; |
991d9fa0 | 196 | mempool_t *mapping_pool; |
e49e5829 JT |
197 | |
198 | process_bio_fn process_bio; | |
199 | process_bio_fn process_discard; | |
200 | ||
201 | process_mapping_fn process_prepared_mapping; | |
202 | process_mapping_fn process_prepared_discard; | |
991d9fa0 JT |
203 | }; |
204 | ||
e49e5829 | 205 | static enum pool_mode get_pool_mode(struct pool *pool); |
b5330655 | 206 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 207 | |
991d9fa0 JT |
208 | /* |
209 | * Target context for a pool. | |
210 | */ | |
211 | struct pool_c { | |
212 | struct dm_target *ti; | |
213 | struct pool *pool; | |
214 | struct dm_dev *data_dev; | |
215 | struct dm_dev *metadata_dev; | |
216 | struct dm_target_callbacks callbacks; | |
217 | ||
218 | dm_block_t low_water_blocks; | |
0424caa1 MS |
219 | struct pool_features requested_pf; /* Features requested during table load */ |
220 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
221 | }; |
222 | ||
223 | /* | |
224 | * Target context for a thin. | |
225 | */ | |
226 | struct thin_c { | |
c140e1c4 | 227 | struct list_head list; |
991d9fa0 | 228 | struct dm_dev *pool_dev; |
2dd9c257 | 229 | struct dm_dev *origin_dev; |
991d9fa0 JT |
230 | dm_thin_id dev_id; |
231 | ||
232 | struct pool *pool; | |
233 | struct dm_thin_device *td; | |
738211f7 | 234 | bool requeue_mode:1; |
c140e1c4 MS |
235 | spinlock_t lock; |
236 | struct bio_list deferred_bio_list; | |
237 | struct bio_list retry_on_resume_list; | |
67324ea1 | 238 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
239 | |
240 | /* | |
241 | * Ensures the thin is not destroyed until the worker has finished | |
242 | * iterating the active_thins list. | |
243 | */ | |
244 | atomic_t refcount; | |
245 | struct completion can_destroy; | |
991d9fa0 JT |
246 | }; |
247 | ||
248 | /*----------------------------------------------------------------*/ | |
249 | ||
025b9685 JT |
250 | /* |
251 | * wake_worker() is used when new work is queued and when pool_resume is | |
252 | * ready to continue deferred IO processing. | |
253 | */ | |
254 | static void wake_worker(struct pool *pool) | |
255 | { | |
256 | queue_work(pool->wq, &pool->worker); | |
257 | } | |
258 | ||
259 | /*----------------------------------------------------------------*/ | |
260 | ||
6beca5eb JT |
261 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
262 | struct dm_bio_prison_cell **cell_result) | |
263 | { | |
264 | int r; | |
265 | struct dm_bio_prison_cell *cell_prealloc; | |
266 | ||
267 | /* | |
268 | * Allocate a cell from the prison's mempool. | |
269 | * This might block but it can't fail. | |
270 | */ | |
271 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
272 | ||
273 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
274 | if (r) | |
275 | /* | |
276 | * We reused an old cell; we can get rid of | |
277 | * the new one. | |
278 | */ | |
279 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
280 | ||
281 | return r; | |
282 | } | |
283 | ||
284 | static void cell_release(struct pool *pool, | |
285 | struct dm_bio_prison_cell *cell, | |
286 | struct bio_list *bios) | |
287 | { | |
288 | dm_cell_release(pool->prison, cell, bios); | |
289 | dm_bio_prison_free_cell(pool->prison, cell); | |
290 | } | |
291 | ||
292 | static void cell_release_no_holder(struct pool *pool, | |
293 | struct dm_bio_prison_cell *cell, | |
294 | struct bio_list *bios) | |
295 | { | |
296 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
297 | dm_bio_prison_free_cell(pool->prison, cell); | |
298 | } | |
299 | ||
025b9685 JT |
300 | static void cell_defer_no_holder_no_free(struct thin_c *tc, |
301 | struct dm_bio_prison_cell *cell) | |
302 | { | |
303 | struct pool *pool = tc->pool; | |
304 | unsigned long flags; | |
305 | ||
c140e1c4 MS |
306 | spin_lock_irqsave(&tc->lock, flags); |
307 | dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list); | |
308 | spin_unlock_irqrestore(&tc->lock, flags); | |
025b9685 JT |
309 | |
310 | wake_worker(pool); | |
311 | } | |
312 | ||
af91805a MS |
313 | static void cell_error_with_code(struct pool *pool, |
314 | struct dm_bio_prison_cell *cell, int error_code) | |
6beca5eb | 315 | { |
af91805a | 316 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
317 | dm_bio_prison_free_cell(pool->prison, cell); |
318 | } | |
319 | ||
af91805a MS |
320 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
321 | { | |
322 | cell_error_with_code(pool, cell, -EIO); | |
323 | } | |
324 | ||
6beca5eb JT |
325 | /*----------------------------------------------------------------*/ |
326 | ||
991d9fa0 JT |
327 | /* |
328 | * A global list of pools that uses a struct mapped_device as a key. | |
329 | */ | |
330 | static struct dm_thin_pool_table { | |
331 | struct mutex mutex; | |
332 | struct list_head pools; | |
333 | } dm_thin_pool_table; | |
334 | ||
335 | static void pool_table_init(void) | |
336 | { | |
337 | mutex_init(&dm_thin_pool_table.mutex); | |
338 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
339 | } | |
340 | ||
341 | static void __pool_table_insert(struct pool *pool) | |
342 | { | |
343 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
344 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
345 | } | |
346 | ||
347 | static void __pool_table_remove(struct pool *pool) | |
348 | { | |
349 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
350 | list_del(&pool->list); | |
351 | } | |
352 | ||
353 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
354 | { | |
355 | struct pool *pool = NULL, *tmp; | |
356 | ||
357 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
358 | ||
359 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
360 | if (tmp->pool_md == md) { | |
361 | pool = tmp; | |
362 | break; | |
363 | } | |
364 | } | |
365 | ||
366 | return pool; | |
367 | } | |
368 | ||
369 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
370 | { | |
371 | struct pool *pool = NULL, *tmp; | |
372 | ||
373 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
374 | ||
375 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
376 | if (tmp->md_dev == md_dev) { | |
377 | pool = tmp; | |
378 | break; | |
379 | } | |
380 | } | |
381 | ||
382 | return pool; | |
383 | } | |
384 | ||
385 | /*----------------------------------------------------------------*/ | |
386 | ||
a24c2569 | 387 | struct dm_thin_endio_hook { |
eb2aa48d | 388 | struct thin_c *tc; |
44feb387 MS |
389 | struct dm_deferred_entry *shared_read_entry; |
390 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 391 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 392 | struct rb_node rb_node; |
eb2aa48d JT |
393 | }; |
394 | ||
18adc577 | 395 | static void requeue_bio_list(struct thin_c *tc, struct bio_list *master) |
991d9fa0 JT |
396 | { |
397 | struct bio *bio; | |
398 | struct bio_list bios; | |
18adc577 | 399 | unsigned long flags; |
991d9fa0 JT |
400 | |
401 | bio_list_init(&bios); | |
18adc577 | 402 | |
c140e1c4 | 403 | spin_lock_irqsave(&tc->lock, flags); |
991d9fa0 JT |
404 | bio_list_merge(&bios, master); |
405 | bio_list_init(master); | |
c140e1c4 | 406 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 407 | |
c140e1c4 MS |
408 | while ((bio = bio_list_pop(&bios))) |
409 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
991d9fa0 JT |
410 | } |
411 | ||
412 | static void requeue_io(struct thin_c *tc) | |
413 | { | |
c140e1c4 MS |
414 | requeue_bio_list(tc, &tc->deferred_bio_list); |
415 | requeue_bio_list(tc, &tc->retry_on_resume_list); | |
991d9fa0 JT |
416 | } |
417 | ||
c140e1c4 | 418 | static void error_thin_retry_list(struct thin_c *tc) |
3e1a0699 JT |
419 | { |
420 | struct bio *bio; | |
421 | unsigned long flags; | |
422 | struct bio_list bios; | |
423 | ||
424 | bio_list_init(&bios); | |
425 | ||
c140e1c4 MS |
426 | spin_lock_irqsave(&tc->lock, flags); |
427 | bio_list_merge(&bios, &tc->retry_on_resume_list); | |
428 | bio_list_init(&tc->retry_on_resume_list); | |
429 | spin_unlock_irqrestore(&tc->lock, flags); | |
3e1a0699 JT |
430 | |
431 | while ((bio = bio_list_pop(&bios))) | |
432 | bio_io_error(bio); | |
433 | } | |
434 | ||
c140e1c4 MS |
435 | static void error_retry_list(struct pool *pool) |
436 | { | |
437 | struct thin_c *tc; | |
438 | ||
439 | rcu_read_lock(); | |
440 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
441 | error_thin_retry_list(tc); | |
442 | rcu_read_unlock(); | |
443 | } | |
444 | ||
991d9fa0 JT |
445 | /* |
446 | * This section of code contains the logic for processing a thin device's IO. | |
447 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
448 | * but most is exclusively called from the thin target rather than the thin-pool | |
449 | * target. | |
450 | */ | |
451 | ||
58f77a21 MS |
452 | static bool block_size_is_power_of_two(struct pool *pool) |
453 | { | |
454 | return pool->sectors_per_block_shift >= 0; | |
455 | } | |
456 | ||
991d9fa0 JT |
457 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) |
458 | { | |
58f77a21 | 459 | struct pool *pool = tc->pool; |
4f024f37 | 460 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 461 | |
58f77a21 MS |
462 | if (block_size_is_power_of_two(pool)) |
463 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 464 | else |
58f77a21 | 465 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
466 | |
467 | return block_nr; | |
991d9fa0 JT |
468 | } |
469 | ||
470 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) | |
471 | { | |
472 | struct pool *pool = tc->pool; | |
4f024f37 | 473 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 JT |
474 | |
475 | bio->bi_bdev = tc->pool_dev->bdev; | |
58f77a21 | 476 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
477 | bio->bi_iter.bi_sector = |
478 | (block << pool->sectors_per_block_shift) | | |
479 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 480 | else |
4f024f37 | 481 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 482 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
483 | } |
484 | ||
2dd9c257 JT |
485 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
486 | { | |
487 | bio->bi_bdev = tc->origin_dev->bdev; | |
488 | } | |
489 | ||
4afdd680 JT |
490 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
491 | { | |
492 | return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && | |
493 | dm_thin_changed_this_transaction(tc->td); | |
494 | } | |
495 | ||
e8088073 JT |
496 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
497 | { | |
498 | struct dm_thin_endio_hook *h; | |
499 | ||
500 | if (bio->bi_rw & REQ_DISCARD) | |
501 | return; | |
502 | ||
59c3d2c6 | 503 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
504 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
505 | } | |
506 | ||
2dd9c257 | 507 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
508 | { |
509 | struct pool *pool = tc->pool; | |
510 | unsigned long flags; | |
511 | ||
e49e5829 JT |
512 | if (!bio_triggers_commit(tc, bio)) { |
513 | generic_make_request(bio); | |
514 | return; | |
515 | } | |
516 | ||
991d9fa0 | 517 | /* |
e49e5829 JT |
518 | * Complete bio with an error if earlier I/O caused changes to |
519 | * the metadata that can't be committed e.g, due to I/O errors | |
520 | * on the metadata device. | |
991d9fa0 | 521 | */ |
e49e5829 JT |
522 | if (dm_thin_aborted_changes(tc->td)) { |
523 | bio_io_error(bio); | |
524 | return; | |
525 | } | |
526 | ||
527 | /* | |
528 | * Batch together any bios that trigger commits and then issue a | |
529 | * single commit for them in process_deferred_bios(). | |
530 | */ | |
531 | spin_lock_irqsave(&pool->lock, flags); | |
532 | bio_list_add(&pool->deferred_flush_bios, bio); | |
533 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
534 | } |
535 | ||
2dd9c257 JT |
536 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
537 | { | |
538 | remap_to_origin(tc, bio); | |
539 | issue(tc, bio); | |
540 | } | |
541 | ||
542 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
543 | dm_block_t block) | |
544 | { | |
545 | remap(tc, bio, block); | |
546 | issue(tc, bio); | |
547 | } | |
548 | ||
991d9fa0 JT |
549 | /*----------------------------------------------------------------*/ |
550 | ||
551 | /* | |
552 | * Bio endio functions. | |
553 | */ | |
a24c2569 | 554 | struct dm_thin_new_mapping { |
991d9fa0 JT |
555 | struct list_head list; |
556 | ||
7f214665 MS |
557 | bool pass_discard:1; |
558 | bool definitely_not_shared:1; | |
991d9fa0 | 559 | |
50f3c3ef JT |
560 | /* |
561 | * Track quiescing, copying and zeroing preparation actions. When this | |
562 | * counter hits zero the block is prepared and can be inserted into the | |
563 | * btree. | |
564 | */ | |
565 | atomic_t prepare_actions; | |
566 | ||
7f214665 | 567 | int err; |
991d9fa0 JT |
568 | struct thin_c *tc; |
569 | dm_block_t virt_block; | |
570 | dm_block_t data_block; | |
a24c2569 | 571 | struct dm_bio_prison_cell *cell, *cell2; |
991d9fa0 JT |
572 | |
573 | /* | |
574 | * If the bio covers the whole area of a block then we can avoid | |
575 | * zeroing or copying. Instead this bio is hooked. The bio will | |
576 | * still be in the cell, so care has to be taken to avoid issuing | |
577 | * the bio twice. | |
578 | */ | |
579 | struct bio *bio; | |
580 | bio_end_io_t *saved_bi_end_io; | |
581 | }; | |
582 | ||
50f3c3ef | 583 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
584 | { |
585 | struct pool *pool = m->tc->pool; | |
586 | ||
50f3c3ef | 587 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 588 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
589 | wake_worker(pool); |
590 | } | |
591 | } | |
592 | ||
593 | static void copy_complete(int read_err, unsigned long write_err, void *context) | |
594 | { | |
595 | unsigned long flags; | |
a24c2569 | 596 | struct dm_thin_new_mapping *m = context; |
991d9fa0 JT |
597 | struct pool *pool = m->tc->pool; |
598 | ||
599 | m->err = read_err || write_err ? -EIO : 0; | |
600 | ||
601 | spin_lock_irqsave(&pool->lock, flags); | |
50f3c3ef | 602 | __complete_mapping_preparation(m); |
991d9fa0 JT |
603 | spin_unlock_irqrestore(&pool->lock, flags); |
604 | } | |
605 | ||
606 | static void overwrite_endio(struct bio *bio, int err) | |
607 | { | |
608 | unsigned long flags; | |
59c3d2c6 | 609 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 610 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 JT |
611 | struct pool *pool = m->tc->pool; |
612 | ||
613 | m->err = err; | |
614 | ||
615 | spin_lock_irqsave(&pool->lock, flags); | |
50f3c3ef | 616 | __complete_mapping_preparation(m); |
991d9fa0 JT |
617 | spin_unlock_irqrestore(&pool->lock, flags); |
618 | } | |
619 | ||
991d9fa0 JT |
620 | /*----------------------------------------------------------------*/ |
621 | ||
622 | /* | |
623 | * Workqueue. | |
624 | */ | |
625 | ||
626 | /* | |
627 | * Prepared mapping jobs. | |
628 | */ | |
629 | ||
630 | /* | |
631 | * This sends the bios in the cell back to the deferred_bios list. | |
632 | */ | |
2aab3850 | 633 | static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
634 | { |
635 | struct pool *pool = tc->pool; | |
636 | unsigned long flags; | |
637 | ||
c140e1c4 MS |
638 | spin_lock_irqsave(&tc->lock, flags); |
639 | cell_release(pool, cell, &tc->deferred_bio_list); | |
640 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
641 | |
642 | wake_worker(pool); | |
643 | } | |
644 | ||
645 | /* | |
6beca5eb | 646 | * Same as cell_defer above, except it omits the original holder of the cell. |
991d9fa0 | 647 | */ |
f286ba0e | 648 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 649 | { |
991d9fa0 JT |
650 | struct pool *pool = tc->pool; |
651 | unsigned long flags; | |
652 | ||
c140e1c4 MS |
653 | spin_lock_irqsave(&tc->lock, flags); |
654 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
655 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
656 | |
657 | wake_worker(pool); | |
658 | } | |
659 | ||
e49e5829 JT |
660 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
661 | { | |
196d38bc | 662 | if (m->bio) { |
e49e5829 | 663 | m->bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
664 | atomic_inc(&m->bio->bi_remaining); |
665 | } | |
6beca5eb | 666 | cell_error(m->tc->pool, m->cell); |
e49e5829 JT |
667 | list_del(&m->list); |
668 | mempool_free(m, m->tc->pool->mapping_pool); | |
669 | } | |
025b9685 | 670 | |
a24c2569 | 671 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
672 | { |
673 | struct thin_c *tc = m->tc; | |
6beca5eb | 674 | struct pool *pool = tc->pool; |
991d9fa0 JT |
675 | struct bio *bio; |
676 | int r; | |
677 | ||
678 | bio = m->bio; | |
196d38bc | 679 | if (bio) { |
991d9fa0 | 680 | bio->bi_end_io = m->saved_bi_end_io; |
196d38bc KO |
681 | atomic_inc(&bio->bi_remaining); |
682 | } | |
991d9fa0 JT |
683 | |
684 | if (m->err) { | |
6beca5eb | 685 | cell_error(pool, m->cell); |
905386f8 | 686 | goto out; |
991d9fa0 JT |
687 | } |
688 | ||
689 | /* | |
690 | * Commit the prepared block into the mapping btree. | |
691 | * Any I/O for this block arriving after this point will get | |
692 | * remapped to it directly. | |
693 | */ | |
694 | r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block); | |
695 | if (r) { | |
b5330655 | 696 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 697 | cell_error(pool, m->cell); |
905386f8 | 698 | goto out; |
991d9fa0 JT |
699 | } |
700 | ||
701 | /* | |
702 | * Release any bios held while the block was being provisioned. | |
703 | * If we are processing a write bio that completely covers the block, | |
704 | * we already processed it so can ignore it now when processing | |
705 | * the bios in the cell. | |
706 | */ | |
707 | if (bio) { | |
f286ba0e | 708 | cell_defer_no_holder(tc, m->cell); |
991d9fa0 JT |
709 | bio_endio(bio, 0); |
710 | } else | |
2aab3850 | 711 | cell_defer(tc, m->cell); |
991d9fa0 | 712 | |
905386f8 | 713 | out: |
991d9fa0 | 714 | list_del(&m->list); |
6beca5eb | 715 | mempool_free(m, pool->mapping_pool); |
991d9fa0 JT |
716 | } |
717 | ||
e49e5829 | 718 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
104655fd | 719 | { |
104655fd JT |
720 | struct thin_c *tc = m->tc; |
721 | ||
e49e5829 | 722 | bio_io_error(m->bio); |
f286ba0e JT |
723 | cell_defer_no_holder(tc, m->cell); |
724 | cell_defer_no_holder(tc, m->cell2); | |
e49e5829 JT |
725 | mempool_free(m, tc->pool->mapping_pool); |
726 | } | |
727 | ||
728 | static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m) | |
729 | { | |
730 | struct thin_c *tc = m->tc; | |
104655fd | 731 | |
e8088073 | 732 | inc_all_io_entry(tc->pool, m->bio); |
f286ba0e JT |
733 | cell_defer_no_holder(tc, m->cell); |
734 | cell_defer_no_holder(tc, m->cell2); | |
e8088073 | 735 | |
104655fd | 736 | if (m->pass_discard) |
19fa1a67 JT |
737 | if (m->definitely_not_shared) |
738 | remap_and_issue(tc, m->bio, m->data_block); | |
739 | else { | |
740 | bool used = false; | |
741 | if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used) | |
742 | bio_endio(m->bio, 0); | |
743 | else | |
744 | remap_and_issue(tc, m->bio, m->data_block); | |
745 | } | |
104655fd JT |
746 | else |
747 | bio_endio(m->bio, 0); | |
748 | ||
104655fd JT |
749 | mempool_free(m, tc->pool->mapping_pool); |
750 | } | |
751 | ||
e49e5829 JT |
752 | static void process_prepared_discard(struct dm_thin_new_mapping *m) |
753 | { | |
754 | int r; | |
755 | struct thin_c *tc = m->tc; | |
756 | ||
757 | r = dm_thin_remove_block(tc->td, m->virt_block); | |
758 | if (r) | |
c397741c | 759 | DMERR_LIMIT("dm_thin_remove_block() failed"); |
e49e5829 JT |
760 | |
761 | process_prepared_discard_passdown(m); | |
762 | } | |
763 | ||
104655fd | 764 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 765 | process_mapping_fn *fn) |
991d9fa0 JT |
766 | { |
767 | unsigned long flags; | |
768 | struct list_head maps; | |
a24c2569 | 769 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
770 | |
771 | INIT_LIST_HEAD(&maps); | |
772 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 773 | list_splice_init(head, &maps); |
991d9fa0 JT |
774 | spin_unlock_irqrestore(&pool->lock, flags); |
775 | ||
776 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 777 | (*fn)(m); |
991d9fa0 JT |
778 | } |
779 | ||
780 | /* | |
781 | * Deferred bio jobs. | |
782 | */ | |
104655fd | 783 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 784 | { |
4f024f37 KO |
785 | return bio->bi_iter.bi_size == |
786 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
787 | } |
788 | ||
789 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
790 | { | |
791 | return (bio_data_dir(bio) == WRITE) && | |
792 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
793 | } |
794 | ||
795 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
796 | bio_end_io_t *fn) | |
797 | { | |
798 | *save = bio->bi_end_io; | |
799 | bio->bi_end_io = fn; | |
800 | } | |
801 | ||
802 | static int ensure_next_mapping(struct pool *pool) | |
803 | { | |
804 | if (pool->next_mapping) | |
805 | return 0; | |
806 | ||
807 | pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC); | |
808 | ||
809 | return pool->next_mapping ? 0 : -ENOMEM; | |
810 | } | |
811 | ||
a24c2569 | 812 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 813 | { |
16961b04 | 814 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
815 | |
816 | BUG_ON(!pool->next_mapping); | |
817 | ||
16961b04 MS |
818 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
819 | INIT_LIST_HEAD(&m->list); | |
820 | m->bio = NULL; | |
821 | ||
991d9fa0 JT |
822 | pool->next_mapping = NULL; |
823 | ||
16961b04 | 824 | return m; |
991d9fa0 JT |
825 | } |
826 | ||
827 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, | |
2dd9c257 JT |
828 | struct dm_dev *origin, dm_block_t data_origin, |
829 | dm_block_t data_dest, | |
a24c2569 | 830 | struct dm_bio_prison_cell *cell, struct bio *bio) |
991d9fa0 JT |
831 | { |
832 | int r; | |
833 | struct pool *pool = tc->pool; | |
a24c2569 | 834 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 835 | |
991d9fa0 JT |
836 | m->tc = tc; |
837 | m->virt_block = virt_block; | |
838 | m->data_block = data_dest; | |
839 | m->cell = cell; | |
991d9fa0 | 840 | |
44feb387 | 841 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
50f3c3ef JT |
842 | atomic_set(&m->prepare_actions, 1); /* copy only */ |
843 | else | |
844 | atomic_set(&m->prepare_actions, 2); /* quiesce + copy */ | |
991d9fa0 JT |
845 | |
846 | /* | |
847 | * IO to pool_dev remaps to the pool target's data_dev. | |
848 | * | |
849 | * If the whole block of data is being overwritten, we can issue the | |
850 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
851 | */ | |
852 | if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 853 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 854 | |
eb2aa48d | 855 | h->overwrite_mapping = m; |
991d9fa0 JT |
856 | m->bio = bio; |
857 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 858 | inc_all_io_entry(pool, bio); |
991d9fa0 JT |
859 | remap_and_issue(tc, bio, data_dest); |
860 | } else { | |
861 | struct dm_io_region from, to; | |
862 | ||
2dd9c257 | 863 | from.bdev = origin->bdev; |
991d9fa0 JT |
864 | from.sector = data_origin * pool->sectors_per_block; |
865 | from.count = pool->sectors_per_block; | |
866 | ||
867 | to.bdev = tc->pool_dev->bdev; | |
868 | to.sector = data_dest * pool->sectors_per_block; | |
869 | to.count = pool->sectors_per_block; | |
870 | ||
871 | r = dm_kcopyd_copy(pool->copier, &from, 1, &to, | |
872 | 0, copy_complete, m); | |
873 | if (r < 0) { | |
874 | mempool_free(m, pool->mapping_pool); | |
c397741c | 875 | DMERR_LIMIT("dm_kcopyd_copy() failed"); |
6beca5eb | 876 | cell_error(pool, cell); |
991d9fa0 JT |
877 | } |
878 | } | |
879 | } | |
880 | ||
2dd9c257 JT |
881 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
882 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 883 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
884 | { |
885 | schedule_copy(tc, virt_block, tc->pool_dev, | |
886 | data_origin, data_dest, cell, bio); | |
887 | } | |
888 | ||
889 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, | |
890 | dm_block_t data_dest, | |
a24c2569 | 891 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
892 | { |
893 | schedule_copy(tc, virt_block, tc->origin_dev, | |
894 | virt_block, data_dest, cell, bio); | |
895 | } | |
896 | ||
991d9fa0 | 897 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 898 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
899 | struct bio *bio) |
900 | { | |
901 | struct pool *pool = tc->pool; | |
a24c2569 | 902 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 903 | |
50f3c3ef | 904 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 JT |
905 | m->tc = tc; |
906 | m->virt_block = virt_block; | |
907 | m->data_block = data_block; | |
908 | m->cell = cell; | |
991d9fa0 JT |
909 | |
910 | /* | |
911 | * If the whole block of data is being overwritten or we are not | |
912 | * zeroing pre-existing data, we can issue the bio immediately. | |
913 | * Otherwise we use kcopyd to zero the data first. | |
914 | */ | |
67e2e2b2 | 915 | if (!pool->pf.zero_new_blocks) |
991d9fa0 JT |
916 | process_prepared_mapping(m); |
917 | ||
918 | else if (io_overwrites_block(pool, bio)) { | |
59c3d2c6 | 919 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 920 | |
eb2aa48d | 921 | h->overwrite_mapping = m; |
991d9fa0 JT |
922 | m->bio = bio; |
923 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
e8088073 | 924 | inc_all_io_entry(pool, bio); |
991d9fa0 | 925 | remap_and_issue(tc, bio, data_block); |
991d9fa0 JT |
926 | } else { |
927 | int r; | |
928 | struct dm_io_region to; | |
929 | ||
930 | to.bdev = tc->pool_dev->bdev; | |
931 | to.sector = data_block * pool->sectors_per_block; | |
932 | to.count = pool->sectors_per_block; | |
933 | ||
934 | r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m); | |
935 | if (r < 0) { | |
936 | mempool_free(m, pool->mapping_pool); | |
c397741c | 937 | DMERR_LIMIT("dm_kcopyd_zero() failed"); |
6beca5eb | 938 | cell_error(pool, cell); |
991d9fa0 JT |
939 | } |
940 | } | |
941 | } | |
942 | ||
e49e5829 JT |
943 | /* |
944 | * A non-zero return indicates read_only or fail_io mode. | |
945 | * Many callers don't care about the return value. | |
946 | */ | |
020cc3b5 | 947 | static int commit(struct pool *pool) |
e49e5829 JT |
948 | { |
949 | int r; | |
950 | ||
8d07e8a5 | 951 | if (get_pool_mode(pool) >= PM_READ_ONLY) |
e49e5829 JT |
952 | return -EINVAL; |
953 | ||
020cc3b5 | 954 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
955 | if (r) |
956 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
e49e5829 JT |
957 | |
958 | return r; | |
959 | } | |
960 | ||
88a6621b JT |
961 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
962 | { | |
963 | unsigned long flags; | |
964 | ||
965 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
966 | DMWARN("%s: reached low water mark for data device: sending event.", | |
967 | dm_device_name(pool->pool_md)); | |
968 | spin_lock_irqsave(&pool->lock, flags); | |
969 | pool->low_water_triggered = true; | |
970 | spin_unlock_irqrestore(&pool->lock, flags); | |
971 | dm_table_event(pool->ti->table); | |
972 | } | |
973 | } | |
974 | ||
3e1a0699 JT |
975 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
976 | ||
991d9fa0 JT |
977 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
978 | { | |
979 | int r; | |
980 | dm_block_t free_blocks; | |
991d9fa0 JT |
981 | struct pool *pool = tc->pool; |
982 | ||
3e1a0699 | 983 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
984 | return -EINVAL; |
985 | ||
991d9fa0 | 986 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
987 | if (r) { |
988 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 989 | return r; |
b5330655 | 990 | } |
991d9fa0 | 991 | |
88a6621b | 992 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
993 | |
994 | if (!free_blocks) { | |
94563bad MS |
995 | /* |
996 | * Try to commit to see if that will free up some | |
997 | * more space. | |
998 | */ | |
020cc3b5 JT |
999 | r = commit(pool); |
1000 | if (r) | |
1001 | return r; | |
991d9fa0 | 1002 | |
94563bad | 1003 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1004 | if (r) { |
1005 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1006 | return r; |
b5330655 | 1007 | } |
991d9fa0 | 1008 | |
94563bad | 1009 | if (!free_blocks) { |
3e1a0699 | 1010 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1011 | return -ENOSPC; |
991d9fa0 JT |
1012 | } |
1013 | } | |
1014 | ||
1015 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1016 | if (r) { |
b5330655 | 1017 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); |
991d9fa0 | 1018 | return r; |
4a02b34e | 1019 | } |
991d9fa0 JT |
1020 | |
1021 | return 0; | |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * If we have run out of space, queue bios until the device is | |
1026 | * resumed, presumably after having been reloaded with more space. | |
1027 | */ | |
1028 | static void retry_on_resume(struct bio *bio) | |
1029 | { | |
59c3d2c6 | 1030 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1031 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1032 | unsigned long flags; |
1033 | ||
c140e1c4 MS |
1034 | spin_lock_irqsave(&tc->lock, flags); |
1035 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1036 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1037 | } |
1038 | ||
af91805a | 1039 | static int should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1040 | { |
3e1a0699 JT |
1041 | enum pool_mode m = get_pool_mode(pool); |
1042 | ||
1043 | switch (m) { | |
1044 | case PM_WRITE: | |
1045 | /* Shouldn't get here */ | |
1046 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
af91805a | 1047 | return -EIO; |
3e1a0699 JT |
1048 | |
1049 | case PM_OUT_OF_DATA_SPACE: | |
af91805a | 1050 | return pool->pf.error_if_no_space ? -ENOSPC : 0; |
3e1a0699 JT |
1051 | |
1052 | case PM_READ_ONLY: | |
1053 | case PM_FAIL: | |
af91805a | 1054 | return -EIO; |
3e1a0699 JT |
1055 | default: |
1056 | /* Shouldn't get here */ | |
1057 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
af91805a | 1058 | return -EIO; |
3e1a0699 JT |
1059 | } |
1060 | } | |
8c0f0e8c | 1061 | |
3e1a0699 JT |
1062 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1063 | { | |
af91805a MS |
1064 | int error = should_error_unserviceable_bio(pool); |
1065 | ||
1066 | if (error) | |
1067 | bio_endio(bio, error); | |
6d16202b MS |
1068 | else |
1069 | retry_on_resume(bio); | |
8c0f0e8c MS |
1070 | } |
1071 | ||
399caddf | 1072 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1073 | { |
1074 | struct bio *bio; | |
1075 | struct bio_list bios; | |
af91805a | 1076 | int error; |
991d9fa0 | 1077 | |
af91805a MS |
1078 | error = should_error_unserviceable_bio(pool); |
1079 | if (error) { | |
1080 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1081 | return; |
1082 | } | |
1083 | ||
991d9fa0 | 1084 | bio_list_init(&bios); |
6beca5eb | 1085 | cell_release(pool, cell, &bios); |
991d9fa0 | 1086 | |
af91805a MS |
1087 | error = should_error_unserviceable_bio(pool); |
1088 | if (error) | |
3e1a0699 | 1089 | while ((bio = bio_list_pop(&bios))) |
af91805a | 1090 | bio_endio(bio, error); |
3e1a0699 JT |
1091 | else |
1092 | while ((bio = bio_list_pop(&bios))) | |
1093 | retry_on_resume(bio); | |
991d9fa0 JT |
1094 | } |
1095 | ||
104655fd JT |
1096 | static void process_discard(struct thin_c *tc, struct bio *bio) |
1097 | { | |
1098 | int r; | |
c3a0ce2e | 1099 | unsigned long flags; |
104655fd | 1100 | struct pool *pool = tc->pool; |
a24c2569 | 1101 | struct dm_bio_prison_cell *cell, *cell2; |
44feb387 | 1102 | struct dm_cell_key key, key2; |
104655fd JT |
1103 | dm_block_t block = get_bio_block(tc, bio); |
1104 | struct dm_thin_lookup_result lookup_result; | |
a24c2569 | 1105 | struct dm_thin_new_mapping *m; |
104655fd JT |
1106 | |
1107 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1108 | if (bio_detain(tc->pool, &key, bio, &cell)) |
104655fd JT |
1109 | return; |
1110 | ||
1111 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1112 | switch (r) { | |
1113 | case 0: | |
1114 | /* | |
1115 | * Check nobody is fiddling with this pool block. This can | |
1116 | * happen if someone's in the process of breaking sharing | |
1117 | * on this block. | |
1118 | */ | |
1119 | build_data_key(tc->td, lookup_result.block, &key2); | |
6beca5eb | 1120 | if (bio_detain(tc->pool, &key2, bio, &cell2)) { |
f286ba0e | 1121 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1122 | break; |
1123 | } | |
1124 | ||
1125 | if (io_overlaps_block(pool, bio)) { | |
1126 | /* | |
1127 | * IO may still be going to the destination block. We must | |
1128 | * quiesce before we can do the removal. | |
1129 | */ | |
1130 | m = get_next_mapping(pool); | |
1131 | m->tc = tc; | |
19fa1a67 JT |
1132 | m->pass_discard = pool->pf.discard_passdown; |
1133 | m->definitely_not_shared = !lookup_result.shared; | |
104655fd JT |
1134 | m->virt_block = block; |
1135 | m->data_block = lookup_result.block; | |
1136 | m->cell = cell; | |
1137 | m->cell2 = cell2; | |
104655fd JT |
1138 | m->bio = bio; |
1139 | ||
44feb387 | 1140 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) { |
c3a0ce2e | 1141 | spin_lock_irqsave(&pool->lock, flags); |
daec338b | 1142 | list_add_tail(&m->list, &pool->prepared_discards); |
c3a0ce2e | 1143 | spin_unlock_irqrestore(&pool->lock, flags); |
104655fd JT |
1144 | wake_worker(pool); |
1145 | } | |
1146 | } else { | |
e8088073 | 1147 | inc_all_io_entry(pool, bio); |
f286ba0e JT |
1148 | cell_defer_no_holder(tc, cell); |
1149 | cell_defer_no_holder(tc, cell2); | |
e8088073 | 1150 | |
104655fd | 1151 | /* |
49296309 MP |
1152 | * The DM core makes sure that the discard doesn't span |
1153 | * a block boundary. So we submit the discard of a | |
1154 | * partial block appropriately. | |
104655fd | 1155 | */ |
650d2a06 MP |
1156 | if ((!lookup_result.shared) && pool->pf.discard_passdown) |
1157 | remap_and_issue(tc, bio, lookup_result.block); | |
1158 | else | |
1159 | bio_endio(bio, 0); | |
104655fd JT |
1160 | } |
1161 | break; | |
1162 | ||
1163 | case -ENODATA: | |
1164 | /* | |
1165 | * It isn't provisioned, just forget it. | |
1166 | */ | |
f286ba0e | 1167 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1168 | bio_endio(bio, 0); |
1169 | break; | |
1170 | ||
1171 | default: | |
c397741c MS |
1172 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1173 | __func__, r); | |
f286ba0e | 1174 | cell_defer_no_holder(tc, cell); |
104655fd JT |
1175 | bio_io_error(bio); |
1176 | break; | |
1177 | } | |
1178 | } | |
1179 | ||
991d9fa0 | 1180 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1181 | struct dm_cell_key *key, |
991d9fa0 | 1182 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1183 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1184 | { |
1185 | int r; | |
1186 | dm_block_t data_block; | |
d6fc2042 | 1187 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1188 | |
1189 | r = alloc_data_block(tc, &data_block); | |
1190 | switch (r) { | |
1191 | case 0: | |
2dd9c257 JT |
1192 | schedule_internal_copy(tc, block, lookup_result->block, |
1193 | data_block, cell, bio); | |
991d9fa0 JT |
1194 | break; |
1195 | ||
1196 | case -ENOSPC: | |
399caddf | 1197 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1198 | break; |
1199 | ||
1200 | default: | |
c397741c MS |
1201 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1202 | __func__, r); | |
d6fc2042 | 1203 | cell_error(pool, cell); |
991d9fa0 JT |
1204 | break; |
1205 | } | |
1206 | } | |
1207 | ||
1208 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, | |
1209 | dm_block_t block, | |
1210 | struct dm_thin_lookup_result *lookup_result) | |
1211 | { | |
a24c2569 | 1212 | struct dm_bio_prison_cell *cell; |
991d9fa0 | 1213 | struct pool *pool = tc->pool; |
44feb387 | 1214 | struct dm_cell_key key; |
991d9fa0 JT |
1215 | |
1216 | /* | |
1217 | * If cell is already occupied, then sharing is already in the process | |
1218 | * of being broken so we have nothing further to do here. | |
1219 | */ | |
1220 | build_data_key(tc->td, lookup_result->block, &key); | |
6beca5eb | 1221 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1222 | return; |
1223 | ||
4f024f37 | 1224 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) |
991d9fa0 JT |
1225 | break_sharing(tc, bio, block, &key, lookup_result, cell); |
1226 | else { | |
59c3d2c6 | 1227 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1228 | |
44feb387 | 1229 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1230 | inc_all_io_entry(pool, bio); |
f286ba0e | 1231 | cell_defer_no_holder(tc, cell); |
e8088073 | 1232 | |
991d9fa0 JT |
1233 | remap_and_issue(tc, bio, lookup_result->block); |
1234 | } | |
1235 | } | |
1236 | ||
1237 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1238 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1239 | { |
1240 | int r; | |
1241 | dm_block_t data_block; | |
6beca5eb | 1242 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1243 | |
1244 | /* | |
1245 | * Remap empty bios (flushes) immediately, without provisioning. | |
1246 | */ | |
4f024f37 | 1247 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1248 | inc_all_io_entry(pool, bio); |
f286ba0e | 1249 | cell_defer_no_holder(tc, cell); |
e8088073 | 1250 | |
991d9fa0 JT |
1251 | remap_and_issue(tc, bio, 0); |
1252 | return; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * Fill read bios with zeroes and complete them immediately. | |
1257 | */ | |
1258 | if (bio_data_dir(bio) == READ) { | |
1259 | zero_fill_bio(bio); | |
f286ba0e | 1260 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1261 | bio_endio(bio, 0); |
1262 | return; | |
1263 | } | |
1264 | ||
1265 | r = alloc_data_block(tc, &data_block); | |
1266 | switch (r) { | |
1267 | case 0: | |
2dd9c257 JT |
1268 | if (tc->origin_dev) |
1269 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1270 | else | |
1271 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1272 | break; |
1273 | ||
1274 | case -ENOSPC: | |
399caddf | 1275 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1276 | break; |
1277 | ||
1278 | default: | |
c397741c MS |
1279 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1280 | __func__, r); | |
6beca5eb | 1281 | cell_error(pool, cell); |
991d9fa0 JT |
1282 | break; |
1283 | } | |
1284 | } | |
1285 | ||
1286 | static void process_bio(struct thin_c *tc, struct bio *bio) | |
1287 | { | |
1288 | int r; | |
6beca5eb | 1289 | struct pool *pool = tc->pool; |
991d9fa0 | 1290 | dm_block_t block = get_bio_block(tc, bio); |
a24c2569 | 1291 | struct dm_bio_prison_cell *cell; |
44feb387 | 1292 | struct dm_cell_key key; |
991d9fa0 JT |
1293 | struct dm_thin_lookup_result lookup_result; |
1294 | ||
1295 | /* | |
1296 | * If cell is already occupied, then the block is already | |
1297 | * being provisioned so we have nothing further to do here. | |
1298 | */ | |
1299 | build_virtual_key(tc->td, block, &key); | |
6beca5eb | 1300 | if (bio_detain(pool, &key, bio, &cell)) |
991d9fa0 JT |
1301 | return; |
1302 | ||
1303 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1304 | switch (r) { | |
1305 | case 0: | |
e8088073 | 1306 | if (lookup_result.shared) { |
991d9fa0 | 1307 | process_shared_bio(tc, bio, block, &lookup_result); |
6beca5eb | 1308 | cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */ |
e8088073 | 1309 | } else { |
6beca5eb | 1310 | inc_all_io_entry(pool, bio); |
f286ba0e | 1311 | cell_defer_no_holder(tc, cell); |
e8088073 | 1312 | |
991d9fa0 | 1313 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1314 | } |
991d9fa0 JT |
1315 | break; |
1316 | ||
1317 | case -ENODATA: | |
2dd9c257 | 1318 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1319 | inc_all_io_entry(pool, bio); |
f286ba0e | 1320 | cell_defer_no_holder(tc, cell); |
e8088073 | 1321 | |
2dd9c257 JT |
1322 | remap_to_origin_and_issue(tc, bio); |
1323 | } else | |
1324 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1325 | break; |
1326 | ||
1327 | default: | |
c397741c MS |
1328 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1329 | __func__, r); | |
f286ba0e | 1330 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1331 | bio_io_error(bio); |
1332 | break; | |
1333 | } | |
1334 | } | |
1335 | ||
e49e5829 JT |
1336 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
1337 | { | |
1338 | int r; | |
1339 | int rw = bio_data_dir(bio); | |
1340 | dm_block_t block = get_bio_block(tc, bio); | |
1341 | struct dm_thin_lookup_result lookup_result; | |
1342 | ||
1343 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1344 | switch (r) { | |
1345 | case 0: | |
4f024f37 | 1346 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) |
8c0f0e8c | 1347 | handle_unserviceable_bio(tc->pool, bio); |
e8088073 JT |
1348 | else { |
1349 | inc_all_io_entry(tc->pool, bio); | |
e49e5829 | 1350 | remap_and_issue(tc, bio, lookup_result.block); |
e8088073 | 1351 | } |
e49e5829 JT |
1352 | break; |
1353 | ||
1354 | case -ENODATA: | |
1355 | if (rw != READ) { | |
8c0f0e8c | 1356 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
1357 | break; |
1358 | } | |
1359 | ||
1360 | if (tc->origin_dev) { | |
e8088073 | 1361 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
1362 | remap_to_origin_and_issue(tc, bio); |
1363 | break; | |
1364 | } | |
1365 | ||
1366 | zero_fill_bio(bio); | |
1367 | bio_endio(bio, 0); | |
1368 | break; | |
1369 | ||
1370 | default: | |
c397741c MS |
1371 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1372 | __func__, r); | |
e49e5829 JT |
1373 | bio_io_error(bio); |
1374 | break; | |
1375 | } | |
1376 | } | |
1377 | ||
3e1a0699 JT |
1378 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
1379 | { | |
1380 | bio_endio(bio, 0); | |
1381 | } | |
1382 | ||
e49e5829 JT |
1383 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
1384 | { | |
1385 | bio_io_error(bio); | |
1386 | } | |
1387 | ||
ac8c3f3d JT |
1388 | /* |
1389 | * FIXME: should we also commit due to size of transaction, measured in | |
1390 | * metadata blocks? | |
1391 | */ | |
905e51b3 JT |
1392 | static int need_commit_due_to_time(struct pool *pool) |
1393 | { | |
1394 | return jiffies < pool->last_commit_jiffies || | |
1395 | jiffies > pool->last_commit_jiffies + COMMIT_PERIOD; | |
1396 | } | |
1397 | ||
67324ea1 MS |
1398 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
1399 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
1400 | ||
1401 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
1402 | { | |
1403 | struct rb_node **rbp, *parent; | |
1404 | struct dm_thin_endio_hook *pbd; | |
1405 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
1406 | ||
1407 | rbp = &tc->sort_bio_list.rb_node; | |
1408 | parent = NULL; | |
1409 | while (*rbp) { | |
1410 | parent = *rbp; | |
1411 | pbd = thin_pbd(parent); | |
1412 | ||
1413 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
1414 | rbp = &(*rbp)->rb_left; | |
1415 | else | |
1416 | rbp = &(*rbp)->rb_right; | |
1417 | } | |
1418 | ||
1419 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1420 | rb_link_node(&pbd->rb_node, parent, rbp); | |
1421 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
1422 | } | |
1423 | ||
1424 | static void __extract_sorted_bios(struct thin_c *tc) | |
1425 | { | |
1426 | struct rb_node *node; | |
1427 | struct dm_thin_endio_hook *pbd; | |
1428 | struct bio *bio; | |
1429 | ||
1430 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
1431 | pbd = thin_pbd(node); | |
1432 | bio = thin_bio(pbd); | |
1433 | ||
1434 | bio_list_add(&tc->deferred_bio_list, bio); | |
1435 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
1436 | } | |
1437 | ||
1438 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
1439 | } | |
1440 | ||
1441 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
1442 | { | |
1443 | struct bio *bio; | |
1444 | struct bio_list bios; | |
1445 | ||
1446 | bio_list_init(&bios); | |
1447 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
1448 | bio_list_init(&tc->deferred_bio_list); | |
1449 | ||
1450 | /* Sort deferred_bio_list using rb-tree */ | |
1451 | while ((bio = bio_list_pop(&bios))) | |
1452 | __thin_bio_rb_add(tc, bio); | |
1453 | ||
1454 | /* | |
1455 | * Transfer the sorted bios in sort_bio_list back to | |
1456 | * deferred_bio_list to allow lockless submission of | |
1457 | * all bios. | |
1458 | */ | |
1459 | __extract_sorted_bios(tc); | |
1460 | } | |
1461 | ||
c140e1c4 | 1462 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 1463 | { |
c140e1c4 | 1464 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1465 | unsigned long flags; |
1466 | struct bio *bio; | |
1467 | struct bio_list bios; | |
67324ea1 | 1468 | struct blk_plug plug; |
991d9fa0 | 1469 | |
c140e1c4 MS |
1470 | if (tc->requeue_mode) { |
1471 | requeue_bio_list(tc, &tc->deferred_bio_list); | |
1472 | return; | |
1473 | } | |
1474 | ||
991d9fa0 JT |
1475 | bio_list_init(&bios); |
1476 | ||
c140e1c4 | 1477 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
1478 | |
1479 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
1480 | spin_unlock_irqrestore(&tc->lock, flags); | |
1481 | return; | |
1482 | } | |
1483 | ||
1484 | __sort_thin_deferred_bios(tc); | |
1485 | ||
c140e1c4 MS |
1486 | bio_list_merge(&bios, &tc->deferred_bio_list); |
1487 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 1488 | |
c140e1c4 | 1489 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 1490 | |
67324ea1 | 1491 | blk_start_plug(&plug); |
991d9fa0 | 1492 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
1493 | /* |
1494 | * If we've got no free new_mapping structs, and processing | |
1495 | * this bio might require one, we pause until there are some | |
1496 | * prepared mappings to process. | |
1497 | */ | |
1498 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
1499 | spin_lock_irqsave(&tc->lock, flags); |
1500 | bio_list_add(&tc->deferred_bio_list, bio); | |
1501 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
1502 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1503 | break; |
1504 | } | |
104655fd JT |
1505 | |
1506 | if (bio->bi_rw & REQ_DISCARD) | |
e49e5829 | 1507 | pool->process_discard(tc, bio); |
104655fd | 1508 | else |
e49e5829 | 1509 | pool->process_bio(tc, bio); |
991d9fa0 | 1510 | } |
67324ea1 | 1511 | blk_finish_plug(&plug); |
c140e1c4 MS |
1512 | } |
1513 | ||
b10ebd34 JT |
1514 | static void thin_get(struct thin_c *tc); |
1515 | static void thin_put(struct thin_c *tc); | |
1516 | ||
1517 | /* | |
1518 | * We can't hold rcu_read_lock() around code that can block. So we | |
1519 | * find a thin with the rcu lock held; bump a refcount; then drop | |
1520 | * the lock. | |
1521 | */ | |
1522 | static struct thin_c *get_first_thin(struct pool *pool) | |
1523 | { | |
1524 | struct thin_c *tc = NULL; | |
1525 | ||
1526 | rcu_read_lock(); | |
1527 | if (!list_empty(&pool->active_thins)) { | |
1528 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
1529 | thin_get(tc); | |
1530 | } | |
1531 | rcu_read_unlock(); | |
1532 | ||
1533 | return tc; | |
1534 | } | |
1535 | ||
1536 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
1537 | { | |
1538 | struct thin_c *old_tc = tc; | |
1539 | ||
1540 | rcu_read_lock(); | |
1541 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
1542 | thin_get(tc); | |
1543 | thin_put(old_tc); | |
1544 | rcu_read_unlock(); | |
1545 | return tc; | |
1546 | } | |
1547 | thin_put(old_tc); | |
1548 | rcu_read_unlock(); | |
1549 | ||
1550 | return NULL; | |
1551 | } | |
1552 | ||
c140e1c4 MS |
1553 | static void process_deferred_bios(struct pool *pool) |
1554 | { | |
1555 | unsigned long flags; | |
1556 | struct bio *bio; | |
1557 | struct bio_list bios; | |
1558 | struct thin_c *tc; | |
1559 | ||
b10ebd34 JT |
1560 | tc = get_first_thin(pool); |
1561 | while (tc) { | |
c140e1c4 | 1562 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
1563 | tc = get_next_thin(pool, tc); |
1564 | } | |
991d9fa0 JT |
1565 | |
1566 | /* | |
1567 | * If there are any deferred flush bios, we must commit | |
1568 | * the metadata before issuing them. | |
1569 | */ | |
1570 | bio_list_init(&bios); | |
1571 | spin_lock_irqsave(&pool->lock, flags); | |
1572 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
1573 | bio_list_init(&pool->deferred_flush_bios); | |
1574 | spin_unlock_irqrestore(&pool->lock, flags); | |
1575 | ||
4d1662a3 MS |
1576 | if (bio_list_empty(&bios) && |
1577 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) | |
991d9fa0 JT |
1578 | return; |
1579 | ||
020cc3b5 | 1580 | if (commit(pool)) { |
991d9fa0 JT |
1581 | while ((bio = bio_list_pop(&bios))) |
1582 | bio_io_error(bio); | |
1583 | return; | |
1584 | } | |
905e51b3 | 1585 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
1586 | |
1587 | while ((bio = bio_list_pop(&bios))) | |
1588 | generic_make_request(bio); | |
1589 | } | |
1590 | ||
1591 | static void do_worker(struct work_struct *ws) | |
1592 | { | |
1593 | struct pool *pool = container_of(ws, struct pool, worker); | |
1594 | ||
e49e5829 JT |
1595 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
1596 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); | |
991d9fa0 JT |
1597 | process_deferred_bios(pool); |
1598 | } | |
1599 | ||
905e51b3 JT |
1600 | /* |
1601 | * We want to commit periodically so that not too much | |
1602 | * unwritten data builds up. | |
1603 | */ | |
1604 | static void do_waker(struct work_struct *ws) | |
1605 | { | |
1606 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
1607 | wake_worker(pool); | |
1608 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
1609 | } | |
1610 | ||
85ad643b JT |
1611 | /* |
1612 | * We're holding onto IO to allow userland time to react. After the | |
1613 | * timeout either the pool will have been resized (and thus back in | |
1614 | * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO. | |
1615 | */ | |
1616 | static void do_no_space_timeout(struct work_struct *ws) | |
1617 | { | |
1618 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
1619 | no_space_timeout); | |
1620 | ||
1621 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) | |
1622 | set_pool_mode(pool, PM_READ_ONLY); | |
1623 | } | |
1624 | ||
991d9fa0 JT |
1625 | /*----------------------------------------------------------------*/ |
1626 | ||
e7a3e871 | 1627 | struct pool_work { |
738211f7 | 1628 | struct work_struct worker; |
e7a3e871 JT |
1629 | struct completion complete; |
1630 | }; | |
1631 | ||
1632 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
1633 | { | |
1634 | return container_of(ws, struct pool_work, worker); | |
1635 | } | |
1636 | ||
1637 | static void pool_work_complete(struct pool_work *pw) | |
1638 | { | |
1639 | complete(&pw->complete); | |
1640 | } | |
738211f7 | 1641 | |
e7a3e871 JT |
1642 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
1643 | void (*fn)(struct work_struct *)) | |
1644 | { | |
1645 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
1646 | init_completion(&pw->complete); | |
1647 | queue_work(pool->wq, &pw->worker); | |
1648 | wait_for_completion(&pw->complete); | |
1649 | } | |
1650 | ||
1651 | /*----------------------------------------------------------------*/ | |
1652 | ||
1653 | struct noflush_work { | |
1654 | struct pool_work pw; | |
1655 | struct thin_c *tc; | |
738211f7 JT |
1656 | }; |
1657 | ||
e7a3e871 | 1658 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 1659 | { |
e7a3e871 | 1660 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
1661 | } |
1662 | ||
1663 | static void do_noflush_start(struct work_struct *ws) | |
1664 | { | |
e7a3e871 | 1665 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
1666 | w->tc->requeue_mode = true; |
1667 | requeue_io(w->tc); | |
e7a3e871 | 1668 | pool_work_complete(&w->pw); |
738211f7 JT |
1669 | } |
1670 | ||
1671 | static void do_noflush_stop(struct work_struct *ws) | |
1672 | { | |
e7a3e871 | 1673 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 1674 | w->tc->requeue_mode = false; |
e7a3e871 | 1675 | pool_work_complete(&w->pw); |
738211f7 JT |
1676 | } |
1677 | ||
1678 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
1679 | { | |
1680 | struct noflush_work w; | |
1681 | ||
738211f7 | 1682 | w.tc = tc; |
e7a3e871 | 1683 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
1684 | } |
1685 | ||
1686 | /*----------------------------------------------------------------*/ | |
1687 | ||
e49e5829 JT |
1688 | static enum pool_mode get_pool_mode(struct pool *pool) |
1689 | { | |
1690 | return pool->pf.mode; | |
1691 | } | |
1692 | ||
3e1a0699 JT |
1693 | static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode) |
1694 | { | |
1695 | dm_table_event(pool->ti->table); | |
1696 | DMINFO("%s: switching pool to %s mode", | |
1697 | dm_device_name(pool->pool_md), new_mode); | |
1698 | } | |
1699 | ||
8b64e881 | 1700 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 1701 | { |
cdc2b415 | 1702 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
1703 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
1704 | enum pool_mode old_mode = get_pool_mode(pool); | |
80c57893 | 1705 | unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
1706 | |
1707 | /* | |
1708 | * Never allow the pool to transition to PM_WRITE mode if user | |
1709 | * intervention is required to verify metadata and data consistency. | |
1710 | */ | |
1711 | if (new_mode == PM_WRITE && needs_check) { | |
1712 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
1713 | dm_device_name(pool->pool_md)); | |
1714 | if (old_mode != new_mode) | |
1715 | new_mode = old_mode; | |
1716 | else | |
1717 | new_mode = PM_READ_ONLY; | |
1718 | } | |
1719 | /* | |
1720 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
1721 | * not going to recover without a thin_repair. So we never let the | |
1722 | * pool move out of the old mode. | |
1723 | */ | |
1724 | if (old_mode == PM_FAIL) | |
1725 | new_mode = old_mode; | |
e49e5829 | 1726 | |
8b64e881 | 1727 | switch (new_mode) { |
e49e5829 | 1728 | case PM_FAIL: |
8b64e881 | 1729 | if (old_mode != new_mode) |
3e1a0699 | 1730 | notify_of_pool_mode_change(pool, "failure"); |
5383ef3a | 1731 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
1732 | pool->process_bio = process_bio_fail; |
1733 | pool->process_discard = process_bio_fail; | |
1734 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1735 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
1736 | |
1737 | error_retry_list(pool); | |
e49e5829 JT |
1738 | break; |
1739 | ||
1740 | case PM_READ_ONLY: | |
8b64e881 | 1741 | if (old_mode != new_mode) |
3e1a0699 JT |
1742 | notify_of_pool_mode_change(pool, "read-only"); |
1743 | dm_pool_metadata_read_only(pool->pmd); | |
1744 | pool->process_bio = process_bio_read_only; | |
1745 | pool->process_discard = process_bio_success; | |
1746 | pool->process_prepared_mapping = process_prepared_mapping_fail; | |
1747 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
1748 | ||
1749 | error_retry_list(pool); | |
1750 | break; | |
1751 | ||
1752 | case PM_OUT_OF_DATA_SPACE: | |
1753 | /* | |
1754 | * Ideally we'd never hit this state; the low water mark | |
1755 | * would trigger userland to extend the pool before we | |
1756 | * completely run out of data space. However, many small | |
1757 | * IOs to unprovisioned space can consume data space at an | |
1758 | * alarming rate. Adjust your low water mark if you're | |
1759 | * frequently seeing this mode. | |
1760 | */ | |
1761 | if (old_mode != new_mode) | |
1762 | notify_of_pool_mode_change(pool, "out-of-data-space"); | |
1763 | pool->process_bio = process_bio_read_only; | |
1764 | pool->process_discard = process_discard; | |
1765 | pool->process_prepared_mapping = process_prepared_mapping; | |
1766 | pool->process_prepared_discard = process_prepared_discard_passdown; | |
85ad643b | 1767 | |
80c57893 MS |
1768 | if (!pool->pf.error_if_no_space && no_space_timeout) |
1769 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
1770 | break; |
1771 | ||
1772 | case PM_WRITE: | |
8b64e881 | 1773 | if (old_mode != new_mode) |
3e1a0699 | 1774 | notify_of_pool_mode_change(pool, "write"); |
9b7aaa64 | 1775 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 JT |
1776 | pool->process_bio = process_bio; |
1777 | pool->process_discard = process_discard; | |
1778 | pool->process_prepared_mapping = process_prepared_mapping; | |
1779 | pool->process_prepared_discard = process_prepared_discard; | |
1780 | break; | |
1781 | } | |
8b64e881 MS |
1782 | |
1783 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
1784 | /* |
1785 | * The pool mode may have changed, sync it so bind_control_target() | |
1786 | * doesn't cause an unexpected mode transition on resume. | |
1787 | */ | |
1788 | pt->adjusted_pf.mode = new_mode; | |
e49e5829 JT |
1789 | } |
1790 | ||
07f2b6e0 | 1791 | static void abort_transaction(struct pool *pool) |
b5330655 | 1792 | { |
07f2b6e0 MS |
1793 | const char *dev_name = dm_device_name(pool->pool_md); |
1794 | ||
1795 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
1796 | if (dm_pool_abort_metadata(pool->pmd)) { | |
1797 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
1798 | set_pool_mode(pool, PM_FAIL); | |
1799 | } | |
1800 | ||
1801 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
1802 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
1803 | set_pool_mode(pool, PM_FAIL); | |
1804 | } | |
1805 | } | |
399caddf | 1806 | |
07f2b6e0 MS |
1807 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
1808 | { | |
b5330655 JT |
1809 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
1810 | dm_device_name(pool->pool_md), op, r); | |
1811 | ||
07f2b6e0 | 1812 | abort_transaction(pool); |
b5330655 JT |
1813 | set_pool_mode(pool, PM_READ_ONLY); |
1814 | } | |
1815 | ||
e49e5829 JT |
1816 | /*----------------------------------------------------------------*/ |
1817 | ||
991d9fa0 JT |
1818 | /* |
1819 | * Mapping functions. | |
1820 | */ | |
1821 | ||
1822 | /* | |
1823 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
1824 | */ | |
1825 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
1826 | { | |
1827 | unsigned long flags; | |
1828 | struct pool *pool = tc->pool; | |
1829 | ||
c140e1c4 MS |
1830 | spin_lock_irqsave(&tc->lock, flags); |
1831 | bio_list_add(&tc->deferred_bio_list, bio); | |
1832 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1833 | |
1834 | wake_worker(pool); | |
1835 | } | |
1836 | ||
59c3d2c6 | 1837 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 1838 | { |
59c3d2c6 | 1839 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
1840 | |
1841 | h->tc = tc; | |
1842 | h->shared_read_entry = NULL; | |
e8088073 | 1843 | h->all_io_entry = NULL; |
eb2aa48d | 1844 | h->overwrite_mapping = NULL; |
eb2aa48d JT |
1845 | } |
1846 | ||
991d9fa0 JT |
1847 | /* |
1848 | * Non-blocking function called from the thin target's map function. | |
1849 | */ | |
7de3ee57 | 1850 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
1851 | { |
1852 | int r; | |
1853 | struct thin_c *tc = ti->private; | |
1854 | dm_block_t block = get_bio_block(tc, bio); | |
1855 | struct dm_thin_device *td = tc->td; | |
1856 | struct dm_thin_lookup_result result; | |
025b9685 JT |
1857 | struct dm_bio_prison_cell cell1, cell2; |
1858 | struct dm_bio_prison_cell *cell_result; | |
e8088073 | 1859 | struct dm_cell_key key; |
991d9fa0 | 1860 | |
59c3d2c6 | 1861 | thin_hook_bio(tc, bio); |
e49e5829 | 1862 | |
738211f7 JT |
1863 | if (tc->requeue_mode) { |
1864 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1865 | return DM_MAPIO_SUBMITTED; | |
1866 | } | |
1867 | ||
e49e5829 JT |
1868 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
1869 | bio_io_error(bio); | |
1870 | return DM_MAPIO_SUBMITTED; | |
1871 | } | |
1872 | ||
104655fd | 1873 | if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) { |
991d9fa0 JT |
1874 | thin_defer_bio(tc, bio); |
1875 | return DM_MAPIO_SUBMITTED; | |
1876 | } | |
1877 | ||
1878 | r = dm_thin_find_block(td, block, 0, &result); | |
1879 | ||
1880 | /* | |
1881 | * Note that we defer readahead too. | |
1882 | */ | |
1883 | switch (r) { | |
1884 | case 0: | |
1885 | if (unlikely(result.shared)) { | |
1886 | /* | |
1887 | * We have a race condition here between the | |
1888 | * result.shared value returned by the lookup and | |
1889 | * snapshot creation, which may cause new | |
1890 | * sharing. | |
1891 | * | |
1892 | * To avoid this always quiesce the origin before | |
1893 | * taking the snap. You want to do this anyway to | |
1894 | * ensure a consistent application view | |
1895 | * (i.e. lockfs). | |
1896 | * | |
1897 | * More distant ancestors are irrelevant. The | |
1898 | * shared flag will be set in their case. | |
1899 | */ | |
1900 | thin_defer_bio(tc, bio); | |
e8088073 | 1901 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1902 | } |
e8088073 JT |
1903 | |
1904 | build_virtual_key(tc->td, block, &key); | |
025b9685 | 1905 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result)) |
e8088073 JT |
1906 | return DM_MAPIO_SUBMITTED; |
1907 | ||
1908 | build_data_key(tc->td, result.block, &key); | |
025b9685 JT |
1909 | if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) { |
1910 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1911 | return DM_MAPIO_SUBMITTED; |
1912 | } | |
1913 | ||
1914 | inc_all_io_entry(tc->pool, bio); | |
025b9685 JT |
1915 | cell_defer_no_holder_no_free(tc, &cell2); |
1916 | cell_defer_no_holder_no_free(tc, &cell1); | |
e8088073 JT |
1917 | |
1918 | remap(tc, bio, result.block); | |
1919 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
1920 | |
1921 | case -ENODATA: | |
e49e5829 JT |
1922 | if (get_pool_mode(tc->pool) == PM_READ_ONLY) { |
1923 | /* | |
1924 | * This block isn't provisioned, and we have no way | |
8c0f0e8c | 1925 | * of doing so. |
e49e5829 | 1926 | */ |
8c0f0e8c | 1927 | handle_unserviceable_bio(tc->pool, bio); |
2aab3850 | 1928 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1929 | } |
1930 | /* fall through */ | |
1931 | ||
1932 | case -EWOULDBLOCK: | |
991d9fa0 JT |
1933 | /* |
1934 | * In future, the failed dm_thin_find_block above could | |
1935 | * provide the hint to load the metadata into cache. | |
1936 | */ | |
991d9fa0 | 1937 | thin_defer_bio(tc, bio); |
2aab3850 | 1938 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
1939 | |
1940 | default: | |
1941 | /* | |
1942 | * Must always call bio_io_error on failure. | |
1943 | * dm_thin_find_block can fail with -EINVAL if the | |
1944 | * pool is switched to fail-io mode. | |
1945 | */ | |
1946 | bio_io_error(bio); | |
2aab3850 | 1947 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 1948 | } |
991d9fa0 JT |
1949 | } |
1950 | ||
1951 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1952 | { | |
991d9fa0 | 1953 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 1954 | struct request_queue *q; |
991d9fa0 | 1955 | |
760fe67e MS |
1956 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
1957 | return 1; | |
991d9fa0 | 1958 | |
760fe67e MS |
1959 | q = bdev_get_queue(pt->data_dev->bdev); |
1960 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
991d9fa0 JT |
1961 | } |
1962 | ||
c140e1c4 | 1963 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 1964 | { |
c140e1c4 MS |
1965 | unsigned long flags; |
1966 | struct thin_c *tc; | |
1967 | ||
1968 | rcu_read_lock(); | |
1969 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
1970 | spin_lock_irqsave(&tc->lock, flags); | |
1971 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
1972 | bio_list_init(&tc->retry_on_resume_list); | |
1973 | spin_unlock_irqrestore(&tc->lock, flags); | |
1974 | } | |
1975 | rcu_read_unlock(); | |
991d9fa0 JT |
1976 | } |
1977 | ||
1978 | /*---------------------------------------------------------------- | |
1979 | * Binding of control targets to a pool object | |
1980 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
1981 | static bool data_dev_supports_discard(struct pool_c *pt) |
1982 | { | |
1983 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
1984 | ||
1985 | return q && blk_queue_discard(q); | |
1986 | } | |
1987 | ||
58051b94 JT |
1988 | static bool is_factor(sector_t block_size, uint32_t n) |
1989 | { | |
1990 | return !sector_div(block_size, n); | |
1991 | } | |
1992 | ||
9bc142dd MS |
1993 | /* |
1994 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 1995 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 1996 | */ |
0424caa1 | 1997 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 1998 | { |
0424caa1 MS |
1999 | struct pool *pool = pt->pool; |
2000 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2001 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
2002 | sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT; | |
2003 | const char *reason = NULL; | |
9bc142dd MS |
2004 | char buf[BDEVNAME_SIZE]; |
2005 | ||
0424caa1 | 2006 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2007 | return; |
2008 | ||
0424caa1 MS |
2009 | if (!data_dev_supports_discard(pt)) |
2010 | reason = "discard unsupported"; | |
2011 | ||
2012 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2013 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2014 | |
0424caa1 MS |
2015 | else if (data_limits->discard_granularity > block_size) |
2016 | reason = "discard granularity larger than a block"; | |
2017 | ||
58051b94 | 2018 | else if (!is_factor(block_size, data_limits->discard_granularity)) |
0424caa1 MS |
2019 | reason = "discard granularity not a factor of block size"; |
2020 | ||
2021 | if (reason) { | |
2022 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2023 | pt->adjusted_pf.discard_passdown = false; | |
2024 | } | |
9bc142dd MS |
2025 | } |
2026 | ||
991d9fa0 JT |
2027 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2028 | { | |
2029 | struct pool_c *pt = ti->private; | |
2030 | ||
e49e5829 | 2031 | /* |
9b7aaa64 | 2032 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2033 | */ |
07f2b6e0 | 2034 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2035 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2036 | |
8b64e881 MS |
2037 | /* |
2038 | * Don't change the pool's mode until set_pool_mode() below. | |
2039 | * Otherwise the pool's process_* function pointers may | |
2040 | * not match the desired pool mode. | |
2041 | */ | |
2042 | pt->adjusted_pf.mode = old_mode; | |
2043 | ||
2044 | pool->ti = ti; | |
2045 | pool->pf = pt->adjusted_pf; | |
2046 | pool->low_water_blocks = pt->low_water_blocks; | |
2047 | ||
9bc142dd | 2048 | set_pool_mode(pool, new_mode); |
f402693d | 2049 | |
991d9fa0 JT |
2050 | return 0; |
2051 | } | |
2052 | ||
2053 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2054 | { | |
2055 | if (pool->ti == ti) | |
2056 | pool->ti = NULL; | |
2057 | } | |
2058 | ||
2059 | /*---------------------------------------------------------------- | |
2060 | * Pool creation | |
2061 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2062 | /* Initialize pool features. */ |
2063 | static void pool_features_init(struct pool_features *pf) | |
2064 | { | |
e49e5829 | 2065 | pf->mode = PM_WRITE; |
9bc142dd MS |
2066 | pf->zero_new_blocks = true; |
2067 | pf->discard_enabled = true; | |
2068 | pf->discard_passdown = true; | |
787a996c | 2069 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2070 | } |
2071 | ||
991d9fa0 JT |
2072 | static void __pool_destroy(struct pool *pool) |
2073 | { | |
2074 | __pool_table_remove(pool); | |
2075 | ||
2076 | if (dm_pool_metadata_close(pool->pmd) < 0) | |
2077 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2078 | ||
44feb387 | 2079 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2080 | dm_kcopyd_client_destroy(pool->copier); |
2081 | ||
2082 | if (pool->wq) | |
2083 | destroy_workqueue(pool->wq); | |
2084 | ||
2085 | if (pool->next_mapping) | |
2086 | mempool_free(pool->next_mapping, pool->mapping_pool); | |
2087 | mempool_destroy(pool->mapping_pool); | |
44feb387 MS |
2088 | dm_deferred_set_destroy(pool->shared_read_ds); |
2089 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2090 | kfree(pool); |
2091 | } | |
2092 | ||
a24c2569 | 2093 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2094 | |
991d9fa0 JT |
2095 | static struct pool *pool_create(struct mapped_device *pool_md, |
2096 | struct block_device *metadata_dev, | |
e49e5829 JT |
2097 | unsigned long block_size, |
2098 | int read_only, char **error) | |
991d9fa0 JT |
2099 | { |
2100 | int r; | |
2101 | void *err_p; | |
2102 | struct pool *pool; | |
2103 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2104 | bool format_device = read_only ? false : true; |
991d9fa0 | 2105 | |
e49e5829 | 2106 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2107 | if (IS_ERR(pmd)) { |
2108 | *error = "Error creating metadata object"; | |
2109 | return (struct pool *)pmd; | |
2110 | } | |
2111 | ||
2112 | pool = kmalloc(sizeof(*pool), GFP_KERNEL); | |
2113 | if (!pool) { | |
2114 | *error = "Error allocating memory for pool"; | |
2115 | err_p = ERR_PTR(-ENOMEM); | |
2116 | goto bad_pool; | |
2117 | } | |
2118 | ||
2119 | pool->pmd = pmd; | |
2120 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2121 | if (block_size & (block_size - 1)) |
2122 | pool->sectors_per_block_shift = -1; | |
2123 | else | |
2124 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2125 | pool->low_water_blocks = 0; |
67e2e2b2 | 2126 | pool_features_init(&pool->pf); |
44feb387 | 2127 | pool->prison = dm_bio_prison_create(PRISON_CELLS); |
991d9fa0 JT |
2128 | if (!pool->prison) { |
2129 | *error = "Error creating pool's bio prison"; | |
2130 | err_p = ERR_PTR(-ENOMEM); | |
2131 | goto bad_prison; | |
2132 | } | |
2133 | ||
df5d2e90 | 2134 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2135 | if (IS_ERR(pool->copier)) { |
2136 | r = PTR_ERR(pool->copier); | |
2137 | *error = "Error creating pool's kcopyd client"; | |
2138 | err_p = ERR_PTR(r); | |
2139 | goto bad_kcopyd_client; | |
2140 | } | |
2141 | ||
2142 | /* | |
2143 | * Create singlethreaded workqueue that will service all devices | |
2144 | * that use this metadata. | |
2145 | */ | |
2146 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2147 | if (!pool->wq) { | |
2148 | *error = "Error creating pool's workqueue"; | |
2149 | err_p = ERR_PTR(-ENOMEM); | |
2150 | goto bad_wq; | |
2151 | } | |
2152 | ||
2153 | INIT_WORK(&pool->worker, do_worker); | |
905e51b3 | 2154 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 2155 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 2156 | spin_lock_init(&pool->lock); |
991d9fa0 JT |
2157 | bio_list_init(&pool->deferred_flush_bios); |
2158 | INIT_LIST_HEAD(&pool->prepared_mappings); | |
104655fd | 2159 | INIT_LIST_HEAD(&pool->prepared_discards); |
c140e1c4 | 2160 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 2161 | pool->low_water_triggered = false; |
44feb387 MS |
2162 | |
2163 | pool->shared_read_ds = dm_deferred_set_create(); | |
2164 | if (!pool->shared_read_ds) { | |
2165 | *error = "Error creating pool's shared read deferred set"; | |
2166 | err_p = ERR_PTR(-ENOMEM); | |
2167 | goto bad_shared_read_ds; | |
2168 | } | |
2169 | ||
2170 | pool->all_io_ds = dm_deferred_set_create(); | |
2171 | if (!pool->all_io_ds) { | |
2172 | *error = "Error creating pool's all io deferred set"; | |
2173 | err_p = ERR_PTR(-ENOMEM); | |
2174 | goto bad_all_io_ds; | |
2175 | } | |
991d9fa0 JT |
2176 | |
2177 | pool->next_mapping = NULL; | |
a24c2569 MS |
2178 | pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE, |
2179 | _new_mapping_cache); | |
991d9fa0 JT |
2180 | if (!pool->mapping_pool) { |
2181 | *error = "Error creating pool's mapping mempool"; | |
2182 | err_p = ERR_PTR(-ENOMEM); | |
2183 | goto bad_mapping_pool; | |
2184 | } | |
2185 | ||
991d9fa0 | 2186 | pool->ref_count = 1; |
905e51b3 | 2187 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
2188 | pool->pool_md = pool_md; |
2189 | pool->md_dev = metadata_dev; | |
2190 | __pool_table_insert(pool); | |
2191 | ||
2192 | return pool; | |
2193 | ||
991d9fa0 | 2194 | bad_mapping_pool: |
44feb387 MS |
2195 | dm_deferred_set_destroy(pool->all_io_ds); |
2196 | bad_all_io_ds: | |
2197 | dm_deferred_set_destroy(pool->shared_read_ds); | |
2198 | bad_shared_read_ds: | |
991d9fa0 JT |
2199 | destroy_workqueue(pool->wq); |
2200 | bad_wq: | |
2201 | dm_kcopyd_client_destroy(pool->copier); | |
2202 | bad_kcopyd_client: | |
44feb387 | 2203 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2204 | bad_prison: |
2205 | kfree(pool); | |
2206 | bad_pool: | |
2207 | if (dm_pool_metadata_close(pmd)) | |
2208 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2209 | ||
2210 | return err_p; | |
2211 | } | |
2212 | ||
2213 | static void __pool_inc(struct pool *pool) | |
2214 | { | |
2215 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2216 | pool->ref_count++; | |
2217 | } | |
2218 | ||
2219 | static void __pool_dec(struct pool *pool) | |
2220 | { | |
2221 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
2222 | BUG_ON(!pool->ref_count); | |
2223 | if (!--pool->ref_count) | |
2224 | __pool_destroy(pool); | |
2225 | } | |
2226 | ||
2227 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
2228 | struct block_device *metadata_dev, | |
e49e5829 JT |
2229 | unsigned long block_size, int read_only, |
2230 | char **error, int *created) | |
991d9fa0 JT |
2231 | { |
2232 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
2233 | ||
2234 | if (pool) { | |
f09996c9 MS |
2235 | if (pool->pool_md != pool_md) { |
2236 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 2237 | return ERR_PTR(-EBUSY); |
f09996c9 | 2238 | } |
991d9fa0 JT |
2239 | __pool_inc(pool); |
2240 | ||
2241 | } else { | |
2242 | pool = __pool_table_lookup(pool_md); | |
2243 | if (pool) { | |
f09996c9 MS |
2244 | if (pool->md_dev != metadata_dev) { |
2245 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 2246 | return ERR_PTR(-EINVAL); |
f09996c9 | 2247 | } |
991d9fa0 JT |
2248 | __pool_inc(pool); |
2249 | ||
67e2e2b2 | 2250 | } else { |
e49e5829 | 2251 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
2252 | *created = 1; |
2253 | } | |
991d9fa0 JT |
2254 | } |
2255 | ||
2256 | return pool; | |
2257 | } | |
2258 | ||
2259 | /*---------------------------------------------------------------- | |
2260 | * Pool target methods | |
2261 | *--------------------------------------------------------------*/ | |
2262 | static void pool_dtr(struct dm_target *ti) | |
2263 | { | |
2264 | struct pool_c *pt = ti->private; | |
2265 | ||
2266 | mutex_lock(&dm_thin_pool_table.mutex); | |
2267 | ||
2268 | unbind_control_target(pt->pool, ti); | |
2269 | __pool_dec(pt->pool); | |
2270 | dm_put_device(ti, pt->metadata_dev); | |
2271 | dm_put_device(ti, pt->data_dev); | |
2272 | kfree(pt); | |
2273 | ||
2274 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2275 | } | |
2276 | ||
991d9fa0 JT |
2277 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
2278 | struct dm_target *ti) | |
2279 | { | |
2280 | int r; | |
2281 | unsigned argc; | |
2282 | const char *arg_name; | |
2283 | ||
2284 | static struct dm_arg _args[] = { | |
74aa45c3 | 2285 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
2286 | }; |
2287 | ||
2288 | /* | |
2289 | * No feature arguments supplied. | |
2290 | */ | |
2291 | if (!as->argc) | |
2292 | return 0; | |
2293 | ||
2294 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
2295 | if (r) | |
2296 | return -EINVAL; | |
2297 | ||
2298 | while (argc && !r) { | |
2299 | arg_name = dm_shift_arg(as); | |
2300 | argc--; | |
2301 | ||
e49e5829 | 2302 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 2303 | pf->zero_new_blocks = false; |
e49e5829 JT |
2304 | |
2305 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 2306 | pf->discard_enabled = false; |
e49e5829 JT |
2307 | |
2308 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 2309 | pf->discard_passdown = false; |
991d9fa0 | 2310 | |
e49e5829 JT |
2311 | else if (!strcasecmp(arg_name, "read_only")) |
2312 | pf->mode = PM_READ_ONLY; | |
2313 | ||
787a996c MS |
2314 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
2315 | pf->error_if_no_space = true; | |
2316 | ||
e49e5829 JT |
2317 | else { |
2318 | ti->error = "Unrecognised pool feature requested"; | |
2319 | r = -EINVAL; | |
2320 | break; | |
2321 | } | |
991d9fa0 JT |
2322 | } |
2323 | ||
2324 | return r; | |
2325 | } | |
2326 | ||
ac8c3f3d JT |
2327 | static void metadata_low_callback(void *context) |
2328 | { | |
2329 | struct pool *pool = context; | |
2330 | ||
2331 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
2332 | dm_device_name(pool->pool_md)); | |
2333 | ||
2334 | dm_table_event(pool->ti->table); | |
2335 | } | |
2336 | ||
7d48935e MS |
2337 | static sector_t get_dev_size(struct block_device *bdev) |
2338 | { | |
2339 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
2340 | } | |
2341 | ||
2342 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 2343 | { |
7d48935e | 2344 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
2345 | char buffer[BDEVNAME_SIZE]; |
2346 | ||
7d48935e | 2347 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
2348 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
2349 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
2350 | } |
2351 | ||
2352 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
2353 | { | |
2354 | sector_t metadata_dev_size = get_dev_size(bdev); | |
2355 | ||
2356 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
2357 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
2358 | |
2359 | return metadata_dev_size; | |
2360 | } | |
2361 | ||
24347e95 JT |
2362 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
2363 | { | |
2364 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
2365 | ||
7d48935e | 2366 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
2367 | |
2368 | return metadata_dev_size; | |
2369 | } | |
2370 | ||
ac8c3f3d JT |
2371 | /* |
2372 | * When a metadata threshold is crossed a dm event is triggered, and | |
2373 | * userland should respond by growing the metadata device. We could let | |
2374 | * userland set the threshold, like we do with the data threshold, but I'm | |
2375 | * not sure they know enough to do this well. | |
2376 | */ | |
2377 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
2378 | { | |
2379 | /* | |
2380 | * 4M is ample for all ops with the possible exception of thin | |
2381 | * device deletion which is harmless if it fails (just retry the | |
2382 | * delete after you've grown the device). | |
2383 | */ | |
2384 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
2385 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
2386 | } | |
2387 | ||
991d9fa0 JT |
2388 | /* |
2389 | * thin-pool <metadata dev> <data dev> | |
2390 | * <data block size (sectors)> | |
2391 | * <low water mark (blocks)> | |
2392 | * [<#feature args> [<arg>]*] | |
2393 | * | |
2394 | * Optional feature arguments are: | |
2395 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
2396 | * ignore_discard: disable discard |
2397 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
2398 | * read_only: Don't allow any changes to be made to the pool metadata. |
2399 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
2400 | */ |
2401 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
2402 | { | |
67e2e2b2 | 2403 | int r, pool_created = 0; |
991d9fa0 JT |
2404 | struct pool_c *pt; |
2405 | struct pool *pool; | |
2406 | struct pool_features pf; | |
2407 | struct dm_arg_set as; | |
2408 | struct dm_dev *data_dev; | |
2409 | unsigned long block_size; | |
2410 | dm_block_t low_water_blocks; | |
2411 | struct dm_dev *metadata_dev; | |
5d0db96d | 2412 | fmode_t metadata_mode; |
991d9fa0 JT |
2413 | |
2414 | /* | |
2415 | * FIXME Remove validation from scope of lock. | |
2416 | */ | |
2417 | mutex_lock(&dm_thin_pool_table.mutex); | |
2418 | ||
2419 | if (argc < 4) { | |
2420 | ti->error = "Invalid argument count"; | |
2421 | r = -EINVAL; | |
2422 | goto out_unlock; | |
2423 | } | |
5d0db96d | 2424 | |
991d9fa0 JT |
2425 | as.argc = argc; |
2426 | as.argv = argv; | |
2427 | ||
5d0db96d JT |
2428 | /* |
2429 | * Set default pool features. | |
2430 | */ | |
2431 | pool_features_init(&pf); | |
2432 | ||
2433 | dm_consume_args(&as, 4); | |
2434 | r = parse_pool_features(&as, &pf, ti); | |
2435 | if (r) | |
2436 | goto out_unlock; | |
2437 | ||
2438 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
2439 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
2440 | if (r) { |
2441 | ti->error = "Error opening metadata block device"; | |
2442 | goto out_unlock; | |
2443 | } | |
7d48935e | 2444 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
2445 | |
2446 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
2447 | if (r) { | |
2448 | ti->error = "Error getting data device"; | |
2449 | goto out_metadata; | |
2450 | } | |
2451 | ||
2452 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
2453 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
2454 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 2455 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
2456 | ti->error = "Invalid block size"; |
2457 | r = -EINVAL; | |
2458 | goto out; | |
2459 | } | |
2460 | ||
2461 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
2462 | ti->error = "Invalid low water mark"; | |
2463 | r = -EINVAL; | |
2464 | goto out; | |
2465 | } | |
2466 | ||
991d9fa0 JT |
2467 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
2468 | if (!pt) { | |
2469 | r = -ENOMEM; | |
2470 | goto out; | |
2471 | } | |
2472 | ||
2473 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 2474 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
2475 | if (IS_ERR(pool)) { |
2476 | r = PTR_ERR(pool); | |
2477 | goto out_free_pt; | |
2478 | } | |
2479 | ||
67e2e2b2 JT |
2480 | /* |
2481 | * 'pool_created' reflects whether this is the first table load. | |
2482 | * Top level discard support is not allowed to be changed after | |
2483 | * initial load. This would require a pool reload to trigger thin | |
2484 | * device changes. | |
2485 | */ | |
2486 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
2487 | ti->error = "Discard support cannot be disabled once enabled"; | |
2488 | r = -EINVAL; | |
2489 | goto out_flags_changed; | |
2490 | } | |
2491 | ||
991d9fa0 JT |
2492 | pt->pool = pool; |
2493 | pt->ti = ti; | |
2494 | pt->metadata_dev = metadata_dev; | |
2495 | pt->data_dev = data_dev; | |
2496 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 2497 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 2498 | ti->num_flush_bios = 1; |
9bc142dd | 2499 | |
67e2e2b2 JT |
2500 | /* |
2501 | * Only need to enable discards if the pool should pass | |
2502 | * them down to the data device. The thin device's discard | |
2503 | * processing will cause mappings to be removed from the btree. | |
2504 | */ | |
b60ab990 | 2505 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 2506 | if (pf.discard_enabled && pf.discard_passdown) { |
55a62eef | 2507 | ti->num_discard_bios = 1; |
9bc142dd | 2508 | |
67e2e2b2 JT |
2509 | /* |
2510 | * Setting 'discards_supported' circumvents the normal | |
2511 | * stacking of discard limits (this keeps the pool and | |
2512 | * thin devices' discard limits consistent). | |
2513 | */ | |
0ac55489 | 2514 | ti->discards_supported = true; |
67e2e2b2 | 2515 | } |
991d9fa0 JT |
2516 | ti->private = pt; |
2517 | ||
ac8c3f3d JT |
2518 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
2519 | calc_metadata_threshold(pt), | |
2520 | metadata_low_callback, | |
2521 | pool); | |
2522 | if (r) | |
2523 | goto out_free_pt; | |
2524 | ||
991d9fa0 JT |
2525 | pt->callbacks.congested_fn = pool_is_congested; |
2526 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
2527 | ||
2528 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2529 | ||
2530 | return 0; | |
2531 | ||
67e2e2b2 JT |
2532 | out_flags_changed: |
2533 | __pool_dec(pool); | |
991d9fa0 JT |
2534 | out_free_pt: |
2535 | kfree(pt); | |
2536 | out: | |
2537 | dm_put_device(ti, data_dev); | |
2538 | out_metadata: | |
2539 | dm_put_device(ti, metadata_dev); | |
2540 | out_unlock: | |
2541 | mutex_unlock(&dm_thin_pool_table.mutex); | |
2542 | ||
2543 | return r; | |
2544 | } | |
2545 | ||
7de3ee57 | 2546 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2547 | { |
2548 | int r; | |
2549 | struct pool_c *pt = ti->private; | |
2550 | struct pool *pool = pt->pool; | |
2551 | unsigned long flags; | |
2552 | ||
2553 | /* | |
2554 | * As this is a singleton target, ti->begin is always zero. | |
2555 | */ | |
2556 | spin_lock_irqsave(&pool->lock, flags); | |
2557 | bio->bi_bdev = pt->data_dev->bdev; | |
2558 | r = DM_MAPIO_REMAPPED; | |
2559 | spin_unlock_irqrestore(&pool->lock, flags); | |
2560 | ||
2561 | return r; | |
2562 | } | |
2563 | ||
b17446df | 2564 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
2565 | { |
2566 | int r; | |
2567 | struct pool_c *pt = ti->private; | |
2568 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
2569 | sector_t data_size = ti->len; |
2570 | dm_block_t sb_data_size; | |
991d9fa0 | 2571 | |
b17446df | 2572 | *need_commit = false; |
991d9fa0 | 2573 | |
55f2b8bd MS |
2574 | (void) sector_div(data_size, pool->sectors_per_block); |
2575 | ||
991d9fa0 JT |
2576 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
2577 | if (r) { | |
4fa5971a MS |
2578 | DMERR("%s: failed to retrieve data device size", |
2579 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
2580 | return r; |
2581 | } | |
2582 | ||
2583 | if (data_size < sb_data_size) { | |
4fa5971a MS |
2584 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
2585 | dm_device_name(pool->pool_md), | |
55f2b8bd | 2586 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
2587 | return -EINVAL; |
2588 | ||
2589 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
2590 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2591 | DMERR("%s: unable to grow the data device until repaired.", | |
2592 | dm_device_name(pool->pool_md)); | |
2593 | return 0; | |
2594 | } | |
2595 | ||
6f7f51d4 MS |
2596 | if (sb_data_size) |
2597 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
2598 | dm_device_name(pool->pool_md), | |
2599 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
2600 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
2601 | if (r) { | |
b5330655 | 2602 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
2603 | return r; |
2604 | } | |
2605 | ||
b17446df | 2606 | *need_commit = true; |
991d9fa0 JT |
2607 | } |
2608 | ||
2609 | return 0; | |
2610 | } | |
2611 | ||
24347e95 JT |
2612 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
2613 | { | |
2614 | int r; | |
2615 | struct pool_c *pt = ti->private; | |
2616 | struct pool *pool = pt->pool; | |
2617 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
2618 | ||
2619 | *need_commit = false; | |
2620 | ||
610bba8b | 2621 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
2622 | |
2623 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
2624 | if (r) { | |
4fa5971a MS |
2625 | DMERR("%s: failed to retrieve metadata device size", |
2626 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
2627 | return r; |
2628 | } | |
2629 | ||
2630 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
2631 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
2632 | dm_device_name(pool->pool_md), | |
24347e95 JT |
2633 | metadata_dev_size, sb_metadata_dev_size); |
2634 | return -EINVAL; | |
2635 | ||
2636 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
2637 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
2638 | DMERR("%s: unable to grow the metadata device until repaired.", | |
2639 | dm_device_name(pool->pool_md)); | |
2640 | return 0; | |
2641 | } | |
2642 | ||
7d48935e | 2643 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
2644 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
2645 | dm_device_name(pool->pool_md), | |
2646 | sb_metadata_dev_size, metadata_dev_size); | |
24347e95 JT |
2647 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
2648 | if (r) { | |
b5330655 | 2649 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
2650 | return r; |
2651 | } | |
2652 | ||
2653 | *need_commit = true; | |
2654 | } | |
2655 | ||
2656 | return 0; | |
2657 | } | |
2658 | ||
b17446df JT |
2659 | /* |
2660 | * Retrieves the number of blocks of the data device from | |
2661 | * the superblock and compares it to the actual device size, | |
2662 | * thus resizing the data device in case it has grown. | |
2663 | * | |
2664 | * This both copes with opening preallocated data devices in the ctr | |
2665 | * being followed by a resume | |
2666 | * -and- | |
2667 | * calling the resume method individually after userspace has | |
2668 | * grown the data device in reaction to a table event. | |
2669 | */ | |
2670 | static int pool_preresume(struct dm_target *ti) | |
2671 | { | |
2672 | int r; | |
24347e95 | 2673 | bool need_commit1, need_commit2; |
b17446df JT |
2674 | struct pool_c *pt = ti->private; |
2675 | struct pool *pool = pt->pool; | |
2676 | ||
2677 | /* | |
2678 | * Take control of the pool object. | |
2679 | */ | |
2680 | r = bind_control_target(pool, ti); | |
2681 | if (r) | |
2682 | return r; | |
2683 | ||
2684 | r = maybe_resize_data_dev(ti, &need_commit1); | |
2685 | if (r) | |
2686 | return r; | |
2687 | ||
24347e95 JT |
2688 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
2689 | if (r) | |
2690 | return r; | |
2691 | ||
2692 | if (need_commit1 || need_commit2) | |
020cc3b5 | 2693 | (void) commit(pool); |
b17446df JT |
2694 | |
2695 | return 0; | |
2696 | } | |
2697 | ||
991d9fa0 JT |
2698 | static void pool_resume(struct dm_target *ti) |
2699 | { | |
2700 | struct pool_c *pt = ti->private; | |
2701 | struct pool *pool = pt->pool; | |
2702 | unsigned long flags; | |
2703 | ||
2704 | spin_lock_irqsave(&pool->lock, flags); | |
88a6621b | 2705 | pool->low_water_triggered = false; |
991d9fa0 | 2706 | spin_unlock_irqrestore(&pool->lock, flags); |
c140e1c4 | 2707 | requeue_bios(pool); |
991d9fa0 | 2708 | |
905e51b3 | 2709 | do_waker(&pool->waker.work); |
991d9fa0 JT |
2710 | } |
2711 | ||
2712 | static void pool_postsuspend(struct dm_target *ti) | |
2713 | { | |
991d9fa0 JT |
2714 | struct pool_c *pt = ti->private; |
2715 | struct pool *pool = pt->pool; | |
2716 | ||
905e51b3 | 2717 | cancel_delayed_work(&pool->waker); |
85ad643b | 2718 | cancel_delayed_work(&pool->no_space_timeout); |
991d9fa0 | 2719 | flush_workqueue(pool->wq); |
020cc3b5 | 2720 | (void) commit(pool); |
991d9fa0 JT |
2721 | } |
2722 | ||
2723 | static int check_arg_count(unsigned argc, unsigned args_required) | |
2724 | { | |
2725 | if (argc != args_required) { | |
2726 | DMWARN("Message received with %u arguments instead of %u.", | |
2727 | argc, args_required); | |
2728 | return -EINVAL; | |
2729 | } | |
2730 | ||
2731 | return 0; | |
2732 | } | |
2733 | ||
2734 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
2735 | { | |
2736 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
2737 | *dev_id <= MAX_DEV_ID) | |
2738 | return 0; | |
2739 | ||
2740 | if (warning) | |
2741 | DMWARN("Message received with invalid device id: %s", arg); | |
2742 | ||
2743 | return -EINVAL; | |
2744 | } | |
2745 | ||
2746 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
2747 | { | |
2748 | dm_thin_id dev_id; | |
2749 | int r; | |
2750 | ||
2751 | r = check_arg_count(argc, 2); | |
2752 | if (r) | |
2753 | return r; | |
2754 | ||
2755 | r = read_dev_id(argv[1], &dev_id, 1); | |
2756 | if (r) | |
2757 | return r; | |
2758 | ||
2759 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
2760 | if (r) { | |
2761 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
2762 | argv[1]); | |
2763 | return r; | |
2764 | } | |
2765 | ||
2766 | return 0; | |
2767 | } | |
2768 | ||
2769 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2770 | { | |
2771 | dm_thin_id dev_id; | |
2772 | dm_thin_id origin_dev_id; | |
2773 | int r; | |
2774 | ||
2775 | r = check_arg_count(argc, 3); | |
2776 | if (r) | |
2777 | return r; | |
2778 | ||
2779 | r = read_dev_id(argv[1], &dev_id, 1); | |
2780 | if (r) | |
2781 | return r; | |
2782 | ||
2783 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
2784 | if (r) | |
2785 | return r; | |
2786 | ||
2787 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
2788 | if (r) { | |
2789 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
2790 | argv[1], argv[2]); | |
2791 | return r; | |
2792 | } | |
2793 | ||
2794 | return 0; | |
2795 | } | |
2796 | ||
2797 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
2798 | { | |
2799 | dm_thin_id dev_id; | |
2800 | int r; | |
2801 | ||
2802 | r = check_arg_count(argc, 2); | |
2803 | if (r) | |
2804 | return r; | |
2805 | ||
2806 | r = read_dev_id(argv[1], &dev_id, 1); | |
2807 | if (r) | |
2808 | return r; | |
2809 | ||
2810 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
2811 | if (r) | |
2812 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
2813 | ||
2814 | return r; | |
2815 | } | |
2816 | ||
2817 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
2818 | { | |
2819 | dm_thin_id old_id, new_id; | |
2820 | int r; | |
2821 | ||
2822 | r = check_arg_count(argc, 3); | |
2823 | if (r) | |
2824 | return r; | |
2825 | ||
2826 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
2827 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
2828 | return -EINVAL; | |
2829 | } | |
2830 | ||
2831 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
2832 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
2833 | return -EINVAL; | |
2834 | } | |
2835 | ||
2836 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
2837 | if (r) { | |
2838 | DMWARN("Failed to change transaction id from %s to %s.", | |
2839 | argv[1], argv[2]); | |
2840 | return r; | |
2841 | } | |
2842 | ||
2843 | return 0; | |
2844 | } | |
2845 | ||
cc8394d8 JT |
2846 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
2847 | { | |
2848 | int r; | |
2849 | ||
2850 | r = check_arg_count(argc, 1); | |
2851 | if (r) | |
2852 | return r; | |
2853 | ||
020cc3b5 | 2854 | (void) commit(pool); |
0d200aef | 2855 | |
cc8394d8 JT |
2856 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
2857 | if (r) | |
2858 | DMWARN("reserve_metadata_snap message failed."); | |
2859 | ||
2860 | return r; | |
2861 | } | |
2862 | ||
2863 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
2864 | { | |
2865 | int r; | |
2866 | ||
2867 | r = check_arg_count(argc, 1); | |
2868 | if (r) | |
2869 | return r; | |
2870 | ||
2871 | r = dm_pool_release_metadata_snap(pool->pmd); | |
2872 | if (r) | |
2873 | DMWARN("release_metadata_snap message failed."); | |
2874 | ||
2875 | return r; | |
2876 | } | |
2877 | ||
991d9fa0 JT |
2878 | /* |
2879 | * Messages supported: | |
2880 | * create_thin <dev_id> | |
2881 | * create_snap <dev_id> <origin_id> | |
2882 | * delete <dev_id> | |
2883 | * trim <dev_id> <new_size_in_sectors> | |
2884 | * set_transaction_id <current_trans_id> <new_trans_id> | |
cc8394d8 JT |
2885 | * reserve_metadata_snap |
2886 | * release_metadata_snap | |
991d9fa0 JT |
2887 | */ |
2888 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv) | |
2889 | { | |
2890 | int r = -EINVAL; | |
2891 | struct pool_c *pt = ti->private; | |
2892 | struct pool *pool = pt->pool; | |
2893 | ||
2894 | if (!strcasecmp(argv[0], "create_thin")) | |
2895 | r = process_create_thin_mesg(argc, argv, pool); | |
2896 | ||
2897 | else if (!strcasecmp(argv[0], "create_snap")) | |
2898 | r = process_create_snap_mesg(argc, argv, pool); | |
2899 | ||
2900 | else if (!strcasecmp(argv[0], "delete")) | |
2901 | r = process_delete_mesg(argc, argv, pool); | |
2902 | ||
2903 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
2904 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
2905 | ||
cc8394d8 JT |
2906 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
2907 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
2908 | ||
2909 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
2910 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
2911 | ||
991d9fa0 JT |
2912 | else |
2913 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
2914 | ||
e49e5829 | 2915 | if (!r) |
020cc3b5 | 2916 | (void) commit(pool); |
991d9fa0 JT |
2917 | |
2918 | return r; | |
2919 | } | |
2920 | ||
e49e5829 JT |
2921 | static void emit_flags(struct pool_features *pf, char *result, |
2922 | unsigned sz, unsigned maxlen) | |
2923 | { | |
2924 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
2925 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
2926 | pf->error_if_no_space; | |
e49e5829 JT |
2927 | DMEMIT("%u ", count); |
2928 | ||
2929 | if (!pf->zero_new_blocks) | |
2930 | DMEMIT("skip_block_zeroing "); | |
2931 | ||
2932 | if (!pf->discard_enabled) | |
2933 | DMEMIT("ignore_discard "); | |
2934 | ||
2935 | if (!pf->discard_passdown) | |
2936 | DMEMIT("no_discard_passdown "); | |
2937 | ||
2938 | if (pf->mode == PM_READ_ONLY) | |
2939 | DMEMIT("read_only "); | |
787a996c MS |
2940 | |
2941 | if (pf->error_if_no_space) | |
2942 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
2943 | } |
2944 | ||
991d9fa0 JT |
2945 | /* |
2946 | * Status line is: | |
2947 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
2948 | * <used data sectors>/<total data sectors> <held metadata root> | |
2949 | */ | |
fd7c092e MP |
2950 | static void pool_status(struct dm_target *ti, status_type_t type, |
2951 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 2952 | { |
e49e5829 | 2953 | int r; |
991d9fa0 JT |
2954 | unsigned sz = 0; |
2955 | uint64_t transaction_id; | |
2956 | dm_block_t nr_free_blocks_data; | |
2957 | dm_block_t nr_free_blocks_metadata; | |
2958 | dm_block_t nr_blocks_data; | |
2959 | dm_block_t nr_blocks_metadata; | |
2960 | dm_block_t held_root; | |
2961 | char buf[BDEVNAME_SIZE]; | |
2962 | char buf2[BDEVNAME_SIZE]; | |
2963 | struct pool_c *pt = ti->private; | |
2964 | struct pool *pool = pt->pool; | |
2965 | ||
2966 | switch (type) { | |
2967 | case STATUSTYPE_INFO: | |
e49e5829 JT |
2968 | if (get_pool_mode(pool) == PM_FAIL) { |
2969 | DMEMIT("Fail"); | |
2970 | break; | |
2971 | } | |
2972 | ||
1f4e0ff0 AK |
2973 | /* Commit to ensure statistics aren't out-of-date */ |
2974 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 2975 | (void) commit(pool); |
1f4e0ff0 | 2976 | |
fd7c092e MP |
2977 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
2978 | if (r) { | |
4fa5971a MS |
2979 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
2980 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2981 | goto err; |
2982 | } | |
991d9fa0 | 2983 | |
fd7c092e MP |
2984 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
2985 | if (r) { | |
4fa5971a MS |
2986 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
2987 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2988 | goto err; |
2989 | } | |
991d9fa0 JT |
2990 | |
2991 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 2992 | if (r) { |
4fa5971a MS |
2993 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
2994 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
2995 | goto err; |
2996 | } | |
991d9fa0 | 2997 | |
fd7c092e MP |
2998 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
2999 | if (r) { | |
4fa5971a MS |
3000 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3001 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3002 | goto err; |
3003 | } | |
991d9fa0 JT |
3004 | |
3005 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3006 | if (r) { |
4fa5971a MS |
3007 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3008 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3009 | goto err; |
3010 | } | |
991d9fa0 | 3011 | |
cc8394d8 | 3012 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3013 | if (r) { |
4fa5971a MS |
3014 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3015 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3016 | goto err; |
3017 | } | |
991d9fa0 JT |
3018 | |
3019 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3020 | (unsigned long long)transaction_id, | |
3021 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3022 | (unsigned long long)nr_blocks_metadata, | |
3023 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3024 | (unsigned long long)nr_blocks_data); | |
3025 | ||
3026 | if (held_root) | |
e49e5829 JT |
3027 | DMEMIT("%llu ", held_root); |
3028 | else | |
3029 | DMEMIT("- "); | |
3030 | ||
3e1a0699 JT |
3031 | if (pool->pf.mode == PM_OUT_OF_DATA_SPACE) |
3032 | DMEMIT("out_of_data_space "); | |
3033 | else if (pool->pf.mode == PM_READ_ONLY) | |
e49e5829 | 3034 | DMEMIT("ro "); |
991d9fa0 | 3035 | else |
e49e5829 JT |
3036 | DMEMIT("rw "); |
3037 | ||
018debea | 3038 | if (!pool->pf.discard_enabled) |
787a996c | 3039 | DMEMIT("ignore_discard "); |
018debea | 3040 | else if (pool->pf.discard_passdown) |
787a996c MS |
3041 | DMEMIT("discard_passdown "); |
3042 | else | |
3043 | DMEMIT("no_discard_passdown "); | |
3044 | ||
3045 | if (pool->pf.error_if_no_space) | |
3046 | DMEMIT("error_if_no_space "); | |
e49e5829 | 3047 | else |
787a996c | 3048 | DMEMIT("queue_if_no_space "); |
991d9fa0 JT |
3049 | |
3050 | break; | |
3051 | ||
3052 | case STATUSTYPE_TABLE: | |
3053 | DMEMIT("%s %s %lu %llu ", | |
3054 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
3055 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
3056 | (unsigned long)pool->sectors_per_block, | |
3057 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 3058 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
3059 | break; |
3060 | } | |
fd7c092e | 3061 | return; |
991d9fa0 | 3062 | |
fd7c092e MP |
3063 | err: |
3064 | DMEMIT("Error"); | |
991d9fa0 JT |
3065 | } |
3066 | ||
3067 | static int pool_iterate_devices(struct dm_target *ti, | |
3068 | iterate_devices_callout_fn fn, void *data) | |
3069 | { | |
3070 | struct pool_c *pt = ti->private; | |
3071 | ||
3072 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
3073 | } | |
3074 | ||
3075 | static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
3076 | struct bio_vec *biovec, int max_size) | |
3077 | { | |
3078 | struct pool_c *pt = ti->private; | |
3079 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
3080 | ||
3081 | if (!q->merge_bvec_fn) | |
3082 | return max_size; | |
3083 | ||
3084 | bvm->bi_bdev = pt->data_dev->bdev; | |
3085 | ||
3086 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
3087 | } | |
3088 | ||
0424caa1 | 3089 | static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits) |
104655fd | 3090 | { |
0424caa1 MS |
3091 | struct pool *pool = pt->pool; |
3092 | struct queue_limits *data_limits; | |
3093 | ||
104655fd JT |
3094 | limits->max_discard_sectors = pool->sectors_per_block; |
3095 | ||
3096 | /* | |
0424caa1 | 3097 | * discard_granularity is just a hint, and not enforced. |
104655fd | 3098 | */ |
0424caa1 MS |
3099 | if (pt->adjusted_pf.discard_passdown) { |
3100 | data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits; | |
09869de5 LC |
3101 | limits->discard_granularity = max(data_limits->discard_granularity, |
3102 | pool->sectors_per_block << SECTOR_SHIFT); | |
f13945d7 | 3103 | } else |
0424caa1 | 3104 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; |
104655fd JT |
3105 | } |
3106 | ||
991d9fa0 JT |
3107 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3108 | { | |
3109 | struct pool_c *pt = ti->private; | |
3110 | struct pool *pool = pt->pool; | |
0cc67cd9 | 3111 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
991d9fa0 | 3112 | |
0cc67cd9 MS |
3113 | /* |
3114 | * If the system-determined stacked limits are compatible with the | |
3115 | * pool's blocksize (io_opt is a factor) do not override them. | |
3116 | */ | |
3117 | if (io_opt_sectors < pool->sectors_per_block || | |
3118 | do_div(io_opt_sectors, pool->sectors_per_block)) { | |
3119 | blk_limits_io_min(limits, 0); | |
3120 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
3121 | } | |
0424caa1 MS |
3122 | |
3123 | /* | |
3124 | * pt->adjusted_pf is a staging area for the actual features to use. | |
3125 | * They get transferred to the live pool in bind_control_target() | |
3126 | * called from pool_preresume(). | |
3127 | */ | |
b60ab990 MS |
3128 | if (!pt->adjusted_pf.discard_enabled) { |
3129 | /* | |
3130 | * Must explicitly disallow stacking discard limits otherwise the | |
3131 | * block layer will stack them if pool's data device has support. | |
3132 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
3133 | * user to see that, so make sure to set all discard limits to 0. | |
3134 | */ | |
3135 | limits->discard_granularity = 0; | |
0424caa1 | 3136 | return; |
b60ab990 | 3137 | } |
0424caa1 MS |
3138 | |
3139 | disable_passdown_if_not_supported(pt); | |
3140 | ||
3141 | set_discard_limits(pt, limits); | |
991d9fa0 JT |
3142 | } |
3143 | ||
3144 | static struct target_type pool_target = { | |
3145 | .name = "thin-pool", | |
3146 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
3147 | DM_TARGET_IMMUTABLE, | |
67324ea1 | 3148 | .version = {1, 12, 0}, |
991d9fa0 JT |
3149 | .module = THIS_MODULE, |
3150 | .ctr = pool_ctr, | |
3151 | .dtr = pool_dtr, | |
3152 | .map = pool_map, | |
3153 | .postsuspend = pool_postsuspend, | |
3154 | .preresume = pool_preresume, | |
3155 | .resume = pool_resume, | |
3156 | .message = pool_message, | |
3157 | .status = pool_status, | |
3158 | .merge = pool_merge, | |
3159 | .iterate_devices = pool_iterate_devices, | |
3160 | .io_hints = pool_io_hints, | |
3161 | }; | |
3162 | ||
3163 | /*---------------------------------------------------------------- | |
3164 | * Thin target methods | |
3165 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
3166 | static void thin_get(struct thin_c *tc) |
3167 | { | |
3168 | atomic_inc(&tc->refcount); | |
3169 | } | |
3170 | ||
3171 | static void thin_put(struct thin_c *tc) | |
3172 | { | |
3173 | if (atomic_dec_and_test(&tc->refcount)) | |
3174 | complete(&tc->can_destroy); | |
3175 | } | |
3176 | ||
991d9fa0 JT |
3177 | static void thin_dtr(struct dm_target *ti) |
3178 | { | |
3179 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
3180 | unsigned long flags; |
3181 | ||
b10ebd34 JT |
3182 | thin_put(tc); |
3183 | wait_for_completion(&tc->can_destroy); | |
3184 | ||
c140e1c4 MS |
3185 | spin_lock_irqsave(&tc->pool->lock, flags); |
3186 | list_del_rcu(&tc->list); | |
3187 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
3188 | synchronize_rcu(); | |
991d9fa0 JT |
3189 | |
3190 | mutex_lock(&dm_thin_pool_table.mutex); | |
3191 | ||
3192 | __pool_dec(tc->pool); | |
3193 | dm_pool_close_thin_device(tc->td); | |
3194 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
3195 | if (tc->origin_dev) |
3196 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
3197 | kfree(tc); |
3198 | ||
3199 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * Thin target parameters: | |
3204 | * | |
2dd9c257 | 3205 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
3206 | * |
3207 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
3208 | * dev_id: the internal device identifier | |
2dd9c257 | 3209 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
3210 | * |
3211 | * If the pool device has discards disabled, they get disabled for the thin | |
3212 | * device as well. | |
991d9fa0 JT |
3213 | */ |
3214 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3215 | { | |
3216 | int r; | |
3217 | struct thin_c *tc; | |
2dd9c257 | 3218 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 3219 | struct mapped_device *pool_md; |
5e3283e2 | 3220 | unsigned long flags; |
991d9fa0 JT |
3221 | |
3222 | mutex_lock(&dm_thin_pool_table.mutex); | |
3223 | ||
2dd9c257 | 3224 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
3225 | ti->error = "Invalid argument count"; |
3226 | r = -EINVAL; | |
3227 | goto out_unlock; | |
3228 | } | |
3229 | ||
3230 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
3231 | if (!tc) { | |
3232 | ti->error = "Out of memory"; | |
3233 | r = -ENOMEM; | |
3234 | goto out_unlock; | |
3235 | } | |
c140e1c4 MS |
3236 | spin_lock_init(&tc->lock); |
3237 | bio_list_init(&tc->deferred_bio_list); | |
3238 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 3239 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 3240 | |
2dd9c257 JT |
3241 | if (argc == 3) { |
3242 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); | |
3243 | if (r) { | |
3244 | ti->error = "Error opening origin device"; | |
3245 | goto bad_origin_dev; | |
3246 | } | |
3247 | tc->origin_dev = origin_dev; | |
3248 | } | |
3249 | ||
991d9fa0 JT |
3250 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
3251 | if (r) { | |
3252 | ti->error = "Error opening pool device"; | |
3253 | goto bad_pool_dev; | |
3254 | } | |
3255 | tc->pool_dev = pool_dev; | |
3256 | ||
3257 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
3258 | ti->error = "Invalid device id"; | |
3259 | r = -EINVAL; | |
3260 | goto bad_common; | |
3261 | } | |
3262 | ||
3263 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
3264 | if (!pool_md) { | |
3265 | ti->error = "Couldn't get pool mapped device"; | |
3266 | r = -EINVAL; | |
3267 | goto bad_common; | |
3268 | } | |
3269 | ||
3270 | tc->pool = __pool_table_lookup(pool_md); | |
3271 | if (!tc->pool) { | |
3272 | ti->error = "Couldn't find pool object"; | |
3273 | r = -EINVAL; | |
3274 | goto bad_pool_lookup; | |
3275 | } | |
3276 | __pool_inc(tc->pool); | |
3277 | ||
e49e5829 JT |
3278 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3279 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 3280 | r = -EINVAL; |
e49e5829 JT |
3281 | goto bad_thin_open; |
3282 | } | |
3283 | ||
991d9fa0 JT |
3284 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
3285 | if (r) { | |
3286 | ti->error = "Couldn't open thin internal device"; | |
3287 | goto bad_thin_open; | |
3288 | } | |
3289 | ||
542f9038 MS |
3290 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
3291 | if (r) | |
1acacc07 | 3292 | goto bad_target_max_io_len; |
542f9038 | 3293 | |
55a62eef | 3294 | ti->num_flush_bios = 1; |
16ad3d10 | 3295 | ti->flush_supported = true; |
59c3d2c6 | 3296 | ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
3297 | |
3298 | /* In case the pool supports discards, pass them on. */ | |
b60ab990 | 3299 | ti->discard_zeroes_data_unsupported = true; |
67e2e2b2 | 3300 | if (tc->pool->pf.discard_enabled) { |
0ac55489 | 3301 | ti->discards_supported = true; |
55a62eef | 3302 | ti->num_discard_bios = 1; |
55a62eef AK |
3303 | /* Discard bios must be split on a block boundary */ |
3304 | ti->split_discard_bios = true; | |
67e2e2b2 | 3305 | } |
991d9fa0 JT |
3306 | |
3307 | dm_put(pool_md); | |
3308 | ||
3309 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3310 | ||
b10ebd34 JT |
3311 | atomic_set(&tc->refcount, 1); |
3312 | init_completion(&tc->can_destroy); | |
3313 | ||
5e3283e2 | 3314 | spin_lock_irqsave(&tc->pool->lock, flags); |
c140e1c4 | 3315 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 3316 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
3317 | /* |
3318 | * This synchronize_rcu() call is needed here otherwise we risk a | |
3319 | * wake_worker() call finding no bios to process (because the newly | |
3320 | * added tc isn't yet visible). So this reduces latency since we | |
3321 | * aren't then dependent on the periodic commit to wake_worker(). | |
3322 | */ | |
3323 | synchronize_rcu(); | |
3324 | ||
991d9fa0 JT |
3325 | return 0; |
3326 | ||
1acacc07 MS |
3327 | bad_target_max_io_len: |
3328 | dm_pool_close_thin_device(tc->td); | |
991d9fa0 JT |
3329 | bad_thin_open: |
3330 | __pool_dec(tc->pool); | |
3331 | bad_pool_lookup: | |
3332 | dm_put(pool_md); | |
3333 | bad_common: | |
3334 | dm_put_device(ti, tc->pool_dev); | |
3335 | bad_pool_dev: | |
2dd9c257 JT |
3336 | if (tc->origin_dev) |
3337 | dm_put_device(ti, tc->origin_dev); | |
3338 | bad_origin_dev: | |
991d9fa0 JT |
3339 | kfree(tc); |
3340 | out_unlock: | |
3341 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3342 | ||
3343 | return r; | |
3344 | } | |
3345 | ||
7de3ee57 | 3346 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 3347 | { |
4f024f37 | 3348 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 3349 | |
7de3ee57 | 3350 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
3351 | } |
3352 | ||
7de3ee57 | 3353 | static int thin_endio(struct dm_target *ti, struct bio *bio, int err) |
eb2aa48d JT |
3354 | { |
3355 | unsigned long flags; | |
59c3d2c6 | 3356 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 3357 | struct list_head work; |
a24c2569 | 3358 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
3359 | struct pool *pool = h->tc->pool; |
3360 | ||
3361 | if (h->shared_read_entry) { | |
3362 | INIT_LIST_HEAD(&work); | |
44feb387 | 3363 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
3364 | |
3365 | spin_lock_irqsave(&pool->lock, flags); | |
3366 | list_for_each_entry_safe(m, tmp, &work, list) { | |
3367 | list_del(&m->list); | |
50f3c3ef | 3368 | __complete_mapping_preparation(m); |
eb2aa48d JT |
3369 | } |
3370 | spin_unlock_irqrestore(&pool->lock, flags); | |
3371 | } | |
3372 | ||
104655fd JT |
3373 | if (h->all_io_entry) { |
3374 | INIT_LIST_HEAD(&work); | |
44feb387 | 3375 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
3376 | if (!list_empty(&work)) { |
3377 | spin_lock_irqsave(&pool->lock, flags); | |
3378 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 3379 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
3380 | spin_unlock_irqrestore(&pool->lock, flags); |
3381 | wake_worker(pool); | |
3382 | } | |
104655fd JT |
3383 | } |
3384 | ||
eb2aa48d JT |
3385 | return 0; |
3386 | } | |
3387 | ||
738211f7 | 3388 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 3389 | { |
738211f7 JT |
3390 | struct thin_c *tc = ti->private; |
3391 | ||
991d9fa0 | 3392 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
3393 | noflush_work(tc, do_noflush_start); |
3394 | } | |
3395 | ||
3396 | static void thin_postsuspend(struct dm_target *ti) | |
3397 | { | |
3398 | struct thin_c *tc = ti->private; | |
3399 | ||
3400 | /* | |
3401 | * The dm_noflush_suspending flag has been cleared by now, so | |
3402 | * unfortunately we must always run this. | |
3403 | */ | |
3404 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
3405 | } |
3406 | ||
3407 | /* | |
3408 | * <nr mapped sectors> <highest mapped sector> | |
3409 | */ | |
fd7c092e MP |
3410 | static void thin_status(struct dm_target *ti, status_type_t type, |
3411 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
3412 | { |
3413 | int r; | |
3414 | ssize_t sz = 0; | |
3415 | dm_block_t mapped, highest; | |
3416 | char buf[BDEVNAME_SIZE]; | |
3417 | struct thin_c *tc = ti->private; | |
3418 | ||
e49e5829 JT |
3419 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
3420 | DMEMIT("Fail"); | |
fd7c092e | 3421 | return; |
e49e5829 JT |
3422 | } |
3423 | ||
991d9fa0 JT |
3424 | if (!tc->td) |
3425 | DMEMIT("-"); | |
3426 | else { | |
3427 | switch (type) { | |
3428 | case STATUSTYPE_INFO: | |
3429 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
3430 | if (r) { |
3431 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
3432 | goto err; | |
3433 | } | |
991d9fa0 JT |
3434 | |
3435 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
3436 | if (r < 0) { |
3437 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
3438 | goto err; | |
3439 | } | |
991d9fa0 JT |
3440 | |
3441 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
3442 | if (r) | |
3443 | DMEMIT("%llu", ((highest + 1) * | |
3444 | tc->pool->sectors_per_block) - 1); | |
3445 | else | |
3446 | DMEMIT("-"); | |
3447 | break; | |
3448 | ||
3449 | case STATUSTYPE_TABLE: | |
3450 | DMEMIT("%s %lu", | |
3451 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
3452 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
3453 | if (tc->origin_dev) |
3454 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
3455 | break; |
3456 | } | |
3457 | } | |
3458 | ||
fd7c092e MP |
3459 | return; |
3460 | ||
3461 | err: | |
3462 | DMEMIT("Error"); | |
991d9fa0 JT |
3463 | } |
3464 | ||
3465 | static int thin_iterate_devices(struct dm_target *ti, | |
3466 | iterate_devices_callout_fn fn, void *data) | |
3467 | { | |
55f2b8bd | 3468 | sector_t blocks; |
991d9fa0 | 3469 | struct thin_c *tc = ti->private; |
55f2b8bd | 3470 | struct pool *pool = tc->pool; |
991d9fa0 JT |
3471 | |
3472 | /* | |
3473 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
3474 | * we follow a more convoluted path through to the pool's target. | |
3475 | */ | |
55f2b8bd | 3476 | if (!pool->ti) |
991d9fa0 JT |
3477 | return 0; /* nothing is bound */ |
3478 | ||
55f2b8bd MS |
3479 | blocks = pool->ti->len; |
3480 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 3481 | if (blocks) |
55f2b8bd | 3482 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
3483 | |
3484 | return 0; | |
3485 | } | |
3486 | ||
991d9fa0 JT |
3487 | static struct target_type thin_target = { |
3488 | .name = "thin", | |
67324ea1 | 3489 | .version = {1, 12, 0}, |
991d9fa0 JT |
3490 | .module = THIS_MODULE, |
3491 | .ctr = thin_ctr, | |
3492 | .dtr = thin_dtr, | |
3493 | .map = thin_map, | |
eb2aa48d | 3494 | .end_io = thin_endio, |
738211f7 | 3495 | .presuspend = thin_presuspend, |
991d9fa0 JT |
3496 | .postsuspend = thin_postsuspend, |
3497 | .status = thin_status, | |
3498 | .iterate_devices = thin_iterate_devices, | |
991d9fa0 JT |
3499 | }; |
3500 | ||
3501 | /*----------------------------------------------------------------*/ | |
3502 | ||
3503 | static int __init dm_thin_init(void) | |
3504 | { | |
3505 | int r; | |
3506 | ||
3507 | pool_table_init(); | |
3508 | ||
3509 | r = dm_register_target(&thin_target); | |
3510 | if (r) | |
3511 | return r; | |
3512 | ||
3513 | r = dm_register_target(&pool_target); | |
3514 | if (r) | |
a24c2569 MS |
3515 | goto bad_pool_target; |
3516 | ||
3517 | r = -ENOMEM; | |
3518 | ||
a24c2569 MS |
3519 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
3520 | if (!_new_mapping_cache) | |
3521 | goto bad_new_mapping_cache; | |
3522 | ||
a24c2569 MS |
3523 | return 0; |
3524 | ||
a24c2569 | 3525 | bad_new_mapping_cache: |
a24c2569 MS |
3526 | dm_unregister_target(&pool_target); |
3527 | bad_pool_target: | |
3528 | dm_unregister_target(&thin_target); | |
991d9fa0 JT |
3529 | |
3530 | return r; | |
3531 | } | |
3532 | ||
3533 | static void dm_thin_exit(void) | |
3534 | { | |
3535 | dm_unregister_target(&thin_target); | |
3536 | dm_unregister_target(&pool_target); | |
a24c2569 | 3537 | |
a24c2569 | 3538 | kmem_cache_destroy(_new_mapping_cache); |
991d9fa0 JT |
3539 | } |
3540 | ||
3541 | module_init(dm_thin_init); | |
3542 | module_exit(dm_thin_exit); | |
3543 | ||
80c57893 MS |
3544 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
3545 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
3546 | ||
7cab8bf1 | 3547 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
3548 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
3549 | MODULE_LICENSE("GPL"); |