]>
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" | |
742c8fdc | 8 | #include "dm-bio-prison-v1.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> | |
0f30af98 | 14 | #include <linux/jiffies.h> |
604ea906 | 15 | #include <linux/log2.h> |
991d9fa0 | 16 | #include <linux/list.h> |
c140e1c4 | 17 | #include <linux/rculist.h> |
991d9fa0 JT |
18 | #include <linux/init.h> |
19 | #include <linux/module.h> | |
20 | #include <linux/slab.h> | |
a822c83e | 21 | #include <linux/vmalloc.h> |
ac4c3f34 | 22 | #include <linux/sort.h> |
67324ea1 | 23 | #include <linux/rbtree.h> |
991d9fa0 JT |
24 | |
25 | #define DM_MSG_PREFIX "thin" | |
26 | ||
27 | /* | |
28 | * Tunable constants | |
29 | */ | |
7768ed33 | 30 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 31 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 32 | #define COMMIT_PERIOD HZ |
80c57893 MS |
33 | #define NO_SPACE_TIMEOUT_SECS 60 |
34 | ||
35 | static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; | |
991d9fa0 | 36 | |
df5d2e90 MP |
37 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
38 | "A percentage of time allocated for copy on write"); | |
39 | ||
991d9fa0 JT |
40 | /* |
41 | * The block size of the device holding pool data must be | |
42 | * between 64KB and 1GB. | |
43 | */ | |
44 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
45 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
46 | ||
991d9fa0 JT |
47 | /* |
48 | * Device id is restricted to 24 bits. | |
49 | */ | |
50 | #define MAX_DEV_ID ((1 << 24) - 1) | |
51 | ||
52 | /* | |
53 | * How do we handle breaking sharing of data blocks? | |
54 | * ================================================= | |
55 | * | |
56 | * We use a standard copy-on-write btree to store the mappings for the | |
57 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
58 | * the data). When you take an internal snapshot you clone the root node | |
59 | * of the origin btree. After this there is no concept of an origin or a | |
60 | * snapshot. They are just two device trees that happen to point to the | |
61 | * same data blocks. | |
62 | * | |
63 | * When we get a write in we decide if it's to a shared data block using | |
64 | * some timestamp magic. If it is, we have to break sharing. | |
65 | * | |
66 | * Let's say we write to a shared block in what was the origin. The | |
67 | * steps are: | |
68 | * | |
69 | * i) plug io further to this physical block. (see bio_prison code). | |
70 | * | |
71 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 72 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
73 | * |
74 | * iii) copy the data block to a newly allocate block. This step can be | |
75 | * missed out if the io covers the block. (schedule_copy). | |
76 | * | |
77 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 78 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
79 | * sharing of btree nodes between the two devices. Breaking sharing only |
80 | * effects the btree of that specific device. Btrees for the other | |
81 | * devices that share the block never change. The btree for the origin | |
82 | * device as it was after the last commit is untouched, ie. we're using | |
83 | * persistent data structures in the functional programming sense. | |
84 | * | |
85 | * v) unplug io to this physical block, including the io that triggered | |
86 | * the breaking of sharing. | |
87 | * | |
88 | * Steps (ii) and (iii) occur in parallel. | |
89 | * | |
90 | * The metadata _doesn't_ need to be committed before the io continues. We | |
91 | * get away with this because the io is always written to a _new_ block. | |
92 | * If there's a crash, then: | |
93 | * | |
94 | * - The origin mapping will point to the old origin block (the shared | |
95 | * one). This will contain the data as it was before the io that triggered | |
96 | * the breaking of sharing came in. | |
97 | * | |
98 | * - The snap mapping still points to the old block. As it would after | |
99 | * the commit. | |
100 | * | |
101 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
102 | * will continue to think that data block in the snapshot device is shared | |
103 | * even after the write to the origin has broken sharing. I suspect data | |
104 | * blocks will typically be shared by many different devices, so we're | |
105 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
106 | * devices that reference this data block. At the moment I think the | |
107 | * benefits far, far outweigh the disadvantages. | |
108 | */ | |
109 | ||
110 | /*----------------------------------------------------------------*/ | |
111 | ||
991d9fa0 JT |
112 | /* |
113 | * Key building. | |
114 | */ | |
34fbcf62 JT |
115 | enum lock_space { |
116 | VIRTUAL, | |
117 | PHYSICAL | |
118 | }; | |
119 | ||
120 | static void build_key(struct dm_thin_device *td, enum lock_space ls, | |
121 | dm_block_t b, dm_block_t e, struct dm_cell_key *key) | |
991d9fa0 | 122 | { |
34fbcf62 | 123 | key->virtual = (ls == VIRTUAL); |
991d9fa0 | 124 | key->dev = dm_thin_dev_id(td); |
5f274d88 | 125 | key->block_begin = b; |
34fbcf62 JT |
126 | key->block_end = e; |
127 | } | |
128 | ||
129 | static void build_data_key(struct dm_thin_device *td, dm_block_t b, | |
130 | struct dm_cell_key *key) | |
131 | { | |
132 | build_key(td, PHYSICAL, b, b + 1llu, key); | |
991d9fa0 JT |
133 | } |
134 | ||
135 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 136 | struct dm_cell_key *key) |
991d9fa0 | 137 | { |
34fbcf62 | 138 | build_key(td, VIRTUAL, b, b + 1llu, key); |
991d9fa0 JT |
139 | } |
140 | ||
141 | /*----------------------------------------------------------------*/ | |
142 | ||
7d327fe0 JT |
143 | #define THROTTLE_THRESHOLD (1 * HZ) |
144 | ||
145 | struct throttle { | |
146 | struct rw_semaphore lock; | |
147 | unsigned long threshold; | |
148 | bool throttle_applied; | |
149 | }; | |
150 | ||
151 | static void throttle_init(struct throttle *t) | |
152 | { | |
153 | init_rwsem(&t->lock); | |
154 | t->throttle_applied = false; | |
155 | } | |
156 | ||
157 | static void throttle_work_start(struct throttle *t) | |
158 | { | |
159 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
160 | } | |
161 | ||
162 | static void throttle_work_update(struct throttle *t) | |
163 | { | |
164 | if (!t->throttle_applied && jiffies > t->threshold) { | |
165 | down_write(&t->lock); | |
166 | t->throttle_applied = true; | |
167 | } | |
168 | } | |
169 | ||
170 | static void throttle_work_complete(struct throttle *t) | |
171 | { | |
172 | if (t->throttle_applied) { | |
173 | t->throttle_applied = false; | |
174 | up_write(&t->lock); | |
175 | } | |
176 | } | |
177 | ||
178 | static void throttle_lock(struct throttle *t) | |
179 | { | |
180 | down_read(&t->lock); | |
181 | } | |
182 | ||
183 | static void throttle_unlock(struct throttle *t) | |
184 | { | |
185 | up_read(&t->lock); | |
186 | } | |
187 | ||
188 | /*----------------------------------------------------------------*/ | |
189 | ||
991d9fa0 JT |
190 | /* |
191 | * A pool device ties together a metadata device and a data device. It | |
192 | * also provides the interface for creating and destroying internal | |
193 | * devices. | |
194 | */ | |
a24c2569 | 195 | struct dm_thin_new_mapping; |
67e2e2b2 | 196 | |
e49e5829 | 197 | /* |
f6c36758 | 198 | * The pool runs in various modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
199 | */ |
200 | enum pool_mode { | |
201 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 202 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
3ab91828 JT |
203 | |
204 | /* | |
205 | * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY. | |
206 | */ | |
207 | PM_OUT_OF_METADATA_SPACE, | |
e49e5829 | 208 | PM_READ_ONLY, /* metadata may not be changed */ |
3ab91828 | 209 | |
e49e5829 JT |
210 | PM_FAIL, /* all I/O fails */ |
211 | }; | |
212 | ||
67e2e2b2 | 213 | struct pool_features { |
e49e5829 JT |
214 | enum pool_mode mode; |
215 | ||
9bc142dd MS |
216 | bool zero_new_blocks:1; |
217 | bool discard_enabled:1; | |
218 | bool discard_passdown:1; | |
787a996c | 219 | bool error_if_no_space:1; |
67e2e2b2 JT |
220 | }; |
221 | ||
e49e5829 JT |
222 | struct thin_c; |
223 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 224 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
225 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
226 | ||
ac4c3f34 JT |
227 | #define CELL_SORT_ARRAY_SIZE 8192 |
228 | ||
991d9fa0 JT |
229 | struct pool { |
230 | struct list_head list; | |
231 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
232 | ||
233 | struct mapped_device *pool_md; | |
234 | struct block_device *md_dev; | |
235 | struct dm_pool_metadata *pmd; | |
236 | ||
991d9fa0 | 237 | dm_block_t low_water_blocks; |
55f2b8bd | 238 | uint32_t sectors_per_block; |
f9a8e0cd | 239 | int sectors_per_block_shift; |
991d9fa0 | 240 | |
67e2e2b2 | 241 | struct pool_features pf; |
88a6621b | 242 | bool low_water_triggered:1; /* A dm event has been sent */ |
80e96c54 | 243 | bool suspended:1; |
c3667cc6 | 244 | bool out_of_data_space:1; |
991d9fa0 | 245 | |
44feb387 | 246 | struct dm_bio_prison *prison; |
991d9fa0 JT |
247 | struct dm_kcopyd_client *copier; |
248 | ||
72d711c8 | 249 | struct work_struct worker; |
991d9fa0 | 250 | struct workqueue_struct *wq; |
7d327fe0 | 251 | struct throttle throttle; |
905e51b3 | 252 | struct delayed_work waker; |
85ad643b | 253 | struct delayed_work no_space_timeout; |
991d9fa0 | 254 | |
905e51b3 | 255 | unsigned long last_commit_jiffies; |
55f2b8bd | 256 | unsigned ref_count; |
991d9fa0 JT |
257 | |
258 | spinlock_t lock; | |
991d9fa0 | 259 | struct bio_list deferred_flush_bios; |
4ae280b4 | 260 | struct bio_list deferred_flush_completions; |
991d9fa0 | 261 | struct list_head prepared_mappings; |
104655fd | 262 | struct list_head prepared_discards; |
2a0fbffb | 263 | struct list_head prepared_discards_pt2; |
c140e1c4 | 264 | struct list_head active_thins; |
991d9fa0 | 265 | |
44feb387 MS |
266 | struct dm_deferred_set *shared_read_ds; |
267 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 268 | |
a24c2569 | 269 | struct dm_thin_new_mapping *next_mapping; |
e49e5829 JT |
270 | |
271 | process_bio_fn process_bio; | |
272 | process_bio_fn process_discard; | |
273 | ||
a374bb21 JT |
274 | process_cell_fn process_cell; |
275 | process_cell_fn process_discard_cell; | |
276 | ||
e49e5829 JT |
277 | process_mapping_fn process_prepared_mapping; |
278 | process_mapping_fn process_prepared_discard; | |
2a0fbffb | 279 | process_mapping_fn process_prepared_discard_pt2; |
ac4c3f34 | 280 | |
a822c83e | 281 | struct dm_bio_prison_cell **cell_sort_array; |
72d711c8 MS |
282 | |
283 | mempool_t mapping_pool; | |
991d9fa0 JT |
284 | }; |
285 | ||
b5330655 | 286 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 287 | |
f6c36758 MS |
288 | static enum pool_mode get_pool_mode(struct pool *pool) |
289 | { | |
290 | return pool->pf.mode; | |
291 | } | |
292 | ||
293 | static void notify_of_pool_mode_change(struct pool *pool) | |
294 | { | |
295 | const char *descs[] = { | |
296 | "write", | |
297 | "out-of-data-space", | |
298 | "read-only", | |
299 | "read-only", | |
300 | "fail" | |
301 | }; | |
302 | const char *extra_desc = NULL; | |
303 | enum pool_mode mode = get_pool_mode(pool); | |
304 | ||
305 | if (mode == PM_OUT_OF_DATA_SPACE) { | |
306 | if (!pool->pf.error_if_no_space) | |
307 | extra_desc = " (queue IO)"; | |
308 | else | |
309 | extra_desc = " (error IO)"; | |
310 | } | |
311 | ||
312 | dm_table_event(pool->ti->table); | |
313 | DMINFO("%s: switching pool to %s%s mode", | |
314 | dm_device_name(pool->pool_md), | |
315 | descs[(int)mode], extra_desc ? : ""); | |
316 | } | |
317 | ||
991d9fa0 JT |
318 | /* |
319 | * Target context for a pool. | |
320 | */ | |
321 | struct pool_c { | |
322 | struct dm_target *ti; | |
323 | struct pool *pool; | |
324 | struct dm_dev *data_dev; | |
325 | struct dm_dev *metadata_dev; | |
326 | struct dm_target_callbacks callbacks; | |
327 | ||
328 | dm_block_t low_water_blocks; | |
0424caa1 MS |
329 | struct pool_features requested_pf; /* Features requested during table load */ |
330 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
ddd2185e | 331 | struct bio flush_bio; |
991d9fa0 JT |
332 | }; |
333 | ||
334 | /* | |
335 | * Target context for a thin. | |
336 | */ | |
337 | struct thin_c { | |
c140e1c4 | 338 | struct list_head list; |
991d9fa0 | 339 | struct dm_dev *pool_dev; |
2dd9c257 | 340 | struct dm_dev *origin_dev; |
e5aea7b4 | 341 | sector_t origin_size; |
991d9fa0 JT |
342 | dm_thin_id dev_id; |
343 | ||
344 | struct pool *pool; | |
345 | struct dm_thin_device *td; | |
583024d2 MS |
346 | struct mapped_device *thin_md; |
347 | ||
738211f7 | 348 | bool requeue_mode:1; |
c140e1c4 | 349 | spinlock_t lock; |
a374bb21 | 350 | struct list_head deferred_cells; |
c140e1c4 MS |
351 | struct bio_list deferred_bio_list; |
352 | struct bio_list retry_on_resume_list; | |
67324ea1 | 353 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
354 | |
355 | /* | |
356 | * Ensures the thin is not destroyed until the worker has finished | |
357 | * iterating the active_thins list. | |
358 | */ | |
22d4c291 | 359 | refcount_t refcount; |
b10ebd34 | 360 | struct completion can_destroy; |
991d9fa0 JT |
361 | }; |
362 | ||
363 | /*----------------------------------------------------------------*/ | |
364 | ||
34fbcf62 JT |
365 | static bool block_size_is_power_of_two(struct pool *pool) |
366 | { | |
367 | return pool->sectors_per_block_shift >= 0; | |
368 | } | |
369 | ||
370 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) | |
371 | { | |
372 | return block_size_is_power_of_two(pool) ? | |
373 | (b << pool->sectors_per_block_shift) : | |
374 | (b * pool->sectors_per_block); | |
375 | } | |
376 | ||
202bae52 JT |
377 | /*----------------------------------------------------------------*/ |
378 | ||
379 | struct discard_op { | |
380 | struct thin_c *tc; | |
381 | struct blk_plug plug; | |
382 | struct bio *parent_bio; | |
383 | struct bio *bio; | |
384 | }; | |
385 | ||
386 | static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent) | |
387 | { | |
388 | BUG_ON(!parent); | |
389 | ||
390 | op->tc = tc; | |
391 | blk_start_plug(&op->plug); | |
392 | op->parent_bio = parent; | |
393 | op->bio = NULL; | |
394 | } | |
395 | ||
396 | static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e) | |
34fbcf62 | 397 | { |
202bae52 | 398 | struct thin_c *tc = op->tc; |
34fbcf62 JT |
399 | sector_t s = block_to_sectors(tc->pool, data_b); |
400 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); | |
3dba53a9 | 401 | |
202bae52 | 402 | return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, |
469e3216 | 403 | GFP_NOWAIT, 0, &op->bio); |
202bae52 JT |
404 | } |
405 | ||
406 | static void end_discard(struct discard_op *op, int r) | |
407 | { | |
408 | if (op->bio) { | |
409 | /* | |
410 | * Even if one of the calls to issue_discard failed, we | |
411 | * need to wait for the chain to complete. | |
412 | */ | |
413 | bio_chain(op->bio, op->parent_bio); | |
e6047149 | 414 | bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0); |
4e49ea4a | 415 | submit_bio(op->bio); |
3dba53a9 | 416 | } |
34fbcf62 | 417 | |
202bae52 JT |
418 | blk_finish_plug(&op->plug); |
419 | ||
420 | /* | |
421 | * Even if r is set, there could be sub discards in flight that we | |
422 | * need to wait for. | |
423 | */ | |
4e4cbee9 CH |
424 | if (r && !op->parent_bio->bi_status) |
425 | op->parent_bio->bi_status = errno_to_blk_status(r); | |
202bae52 | 426 | bio_endio(op->parent_bio); |
34fbcf62 JT |
427 | } |
428 | ||
429 | /*----------------------------------------------------------------*/ | |
430 | ||
025b9685 JT |
431 | /* |
432 | * wake_worker() is used when new work is queued and when pool_resume is | |
433 | * ready to continue deferred IO processing. | |
434 | */ | |
435 | static void wake_worker(struct pool *pool) | |
436 | { | |
437 | queue_work(pool->wq, &pool->worker); | |
438 | } | |
439 | ||
440 | /*----------------------------------------------------------------*/ | |
441 | ||
6beca5eb JT |
442 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
443 | struct dm_bio_prison_cell **cell_result) | |
444 | { | |
445 | int r; | |
446 | struct dm_bio_prison_cell *cell_prealloc; | |
447 | ||
448 | /* | |
449 | * Allocate a cell from the prison's mempool. | |
450 | * This might block but it can't fail. | |
451 | */ | |
452 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
453 | ||
454 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
455 | if (r) | |
456 | /* | |
457 | * We reused an old cell; we can get rid of | |
458 | * the new one. | |
459 | */ | |
460 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
461 | ||
462 | return r; | |
463 | } | |
464 | ||
465 | static void cell_release(struct pool *pool, | |
466 | struct dm_bio_prison_cell *cell, | |
467 | struct bio_list *bios) | |
468 | { | |
469 | dm_cell_release(pool->prison, cell, bios); | |
470 | dm_bio_prison_free_cell(pool->prison, cell); | |
471 | } | |
472 | ||
2d759a46 JT |
473 | static void cell_visit_release(struct pool *pool, |
474 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
475 | void *context, | |
476 | struct dm_bio_prison_cell *cell) | |
477 | { | |
478 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
479 | dm_bio_prison_free_cell(pool->prison, cell); | |
480 | } | |
481 | ||
6beca5eb JT |
482 | static void cell_release_no_holder(struct pool *pool, |
483 | struct dm_bio_prison_cell *cell, | |
484 | struct bio_list *bios) | |
485 | { | |
486 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
487 | dm_bio_prison_free_cell(pool->prison, cell); | |
488 | } | |
489 | ||
af91805a | 490 | static void cell_error_with_code(struct pool *pool, |
4e4cbee9 | 491 | struct dm_bio_prison_cell *cell, blk_status_t error_code) |
6beca5eb | 492 | { |
af91805a | 493 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
494 | dm_bio_prison_free_cell(pool->prison, cell); |
495 | } | |
496 | ||
4e4cbee9 | 497 | static blk_status_t get_pool_io_error_code(struct pool *pool) |
c3667cc6 | 498 | { |
4e4cbee9 | 499 | return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR; |
c3667cc6 MS |
500 | } |
501 | ||
af91805a MS |
502 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
503 | { | |
4e4cbee9 | 504 | cell_error_with_code(pool, cell, get_pool_io_error_code(pool)); |
af91805a MS |
505 | } |
506 | ||
a374bb21 JT |
507 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
508 | { | |
509 | cell_error_with_code(pool, cell, 0); | |
510 | } | |
511 | ||
512 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
513 | { | |
4e4cbee9 | 514 | cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE); |
a374bb21 JT |
515 | } |
516 | ||
6beca5eb JT |
517 | /*----------------------------------------------------------------*/ |
518 | ||
991d9fa0 JT |
519 | /* |
520 | * A global list of pools that uses a struct mapped_device as a key. | |
521 | */ | |
522 | static struct dm_thin_pool_table { | |
523 | struct mutex mutex; | |
524 | struct list_head pools; | |
525 | } dm_thin_pool_table; | |
526 | ||
527 | static void pool_table_init(void) | |
528 | { | |
529 | mutex_init(&dm_thin_pool_table.mutex); | |
530 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
531 | } | |
532 | ||
d5ffebdd MS |
533 | static void pool_table_exit(void) |
534 | { | |
535 | mutex_destroy(&dm_thin_pool_table.mutex); | |
536 | } | |
537 | ||
991d9fa0 JT |
538 | static void __pool_table_insert(struct pool *pool) |
539 | { | |
540 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
541 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
542 | } | |
543 | ||
544 | static void __pool_table_remove(struct pool *pool) | |
545 | { | |
546 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
547 | list_del(&pool->list); | |
548 | } | |
549 | ||
550 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
551 | { | |
552 | struct pool *pool = NULL, *tmp; | |
553 | ||
554 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
555 | ||
556 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
557 | if (tmp->pool_md == md) { | |
558 | pool = tmp; | |
559 | break; | |
560 | } | |
561 | } | |
562 | ||
563 | return pool; | |
564 | } | |
565 | ||
566 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
567 | { | |
568 | struct pool *pool = NULL, *tmp; | |
569 | ||
570 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
571 | ||
572 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
573 | if (tmp->md_dev == md_dev) { | |
574 | pool = tmp; | |
575 | break; | |
576 | } | |
577 | } | |
578 | ||
579 | return pool; | |
580 | } | |
581 | ||
582 | /*----------------------------------------------------------------*/ | |
583 | ||
a24c2569 | 584 | struct dm_thin_endio_hook { |
eb2aa48d | 585 | struct thin_c *tc; |
44feb387 MS |
586 | struct dm_deferred_entry *shared_read_entry; |
587 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 588 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 589 | struct rb_node rb_node; |
34fbcf62 | 590 | struct dm_bio_prison_cell *cell; |
eb2aa48d JT |
591 | }; |
592 | ||
42d6a8ce MS |
593 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
594 | { | |
595 | bio_list_merge(bios, master); | |
596 | bio_list_init(master); | |
597 | } | |
598 | ||
4e4cbee9 | 599 | static void error_bio_list(struct bio_list *bios, blk_status_t error) |
991d9fa0 JT |
600 | { |
601 | struct bio *bio; | |
42d6a8ce | 602 | |
4246a0b6 | 603 | while ((bio = bio_list_pop(bios))) { |
4e4cbee9 | 604 | bio->bi_status = error; |
4246a0b6 CH |
605 | bio_endio(bio); |
606 | } | |
42d6a8ce MS |
607 | } |
608 | ||
4e4cbee9 CH |
609 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, |
610 | blk_status_t error) | |
42d6a8ce | 611 | { |
991d9fa0 | 612 | struct bio_list bios; |
18adc577 | 613 | unsigned long flags; |
991d9fa0 JT |
614 | |
615 | bio_list_init(&bios); | |
18adc577 | 616 | |
c140e1c4 | 617 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce | 618 | __merge_bio_list(&bios, master); |
c140e1c4 | 619 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 620 | |
42d6a8ce | 621 | error_bio_list(&bios, error); |
991d9fa0 JT |
622 | } |
623 | ||
a374bb21 JT |
624 | static void requeue_deferred_cells(struct thin_c *tc) |
625 | { | |
626 | struct pool *pool = tc->pool; | |
627 | unsigned long flags; | |
628 | struct list_head cells; | |
629 | struct dm_bio_prison_cell *cell, *tmp; | |
630 | ||
631 | INIT_LIST_HEAD(&cells); | |
632 | ||
633 | spin_lock_irqsave(&tc->lock, flags); | |
634 | list_splice_init(&tc->deferred_cells, &cells); | |
635 | spin_unlock_irqrestore(&tc->lock, flags); | |
636 | ||
637 | list_for_each_entry_safe(cell, tmp, &cells, user_list) | |
638 | cell_requeue(pool, cell); | |
639 | } | |
640 | ||
991d9fa0 JT |
641 | static void requeue_io(struct thin_c *tc) |
642 | { | |
3e1a0699 | 643 | struct bio_list bios; |
42d6a8ce | 644 | unsigned long flags; |
3e1a0699 JT |
645 | |
646 | bio_list_init(&bios); | |
647 | ||
c140e1c4 | 648 | spin_lock_irqsave(&tc->lock, flags); |
42d6a8ce MS |
649 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
650 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
c140e1c4 | 651 | spin_unlock_irqrestore(&tc->lock, flags); |
3e1a0699 | 652 | |
4e4cbee9 | 653 | error_bio_list(&bios, BLK_STS_DM_REQUEUE); |
42d6a8ce | 654 | requeue_deferred_cells(tc); |
3e1a0699 JT |
655 | } |
656 | ||
4e4cbee9 | 657 | static void error_retry_list_with_code(struct pool *pool, blk_status_t error) |
c140e1c4 MS |
658 | { |
659 | struct thin_c *tc; | |
660 | ||
661 | rcu_read_lock(); | |
662 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
0a927c2f | 663 | error_thin_bio_list(tc, &tc->retry_on_resume_list, error); |
c140e1c4 MS |
664 | rcu_read_unlock(); |
665 | } | |
666 | ||
0a927c2f MS |
667 | static void error_retry_list(struct pool *pool) |
668 | { | |
4e4cbee9 | 669 | error_retry_list_with_code(pool, get_pool_io_error_code(pool)); |
0a927c2f MS |
670 | } |
671 | ||
991d9fa0 JT |
672 | /* |
673 | * This section of code contains the logic for processing a thin device's IO. | |
674 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
675 | * but most is exclusively called from the thin target rather than the thin-pool | |
676 | * target. | |
677 | */ | |
678 | ||
679 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
680 | { | |
58f77a21 | 681 | struct pool *pool = tc->pool; |
4f024f37 | 682 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 683 | |
58f77a21 MS |
684 | if (block_size_is_power_of_two(pool)) |
685 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 686 | else |
58f77a21 | 687 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
688 | |
689 | return block_nr; | |
991d9fa0 JT |
690 | } |
691 | ||
34fbcf62 JT |
692 | /* |
693 | * Returns the _complete_ blocks that this bio covers. | |
694 | */ | |
695 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, | |
696 | dm_block_t *begin, dm_block_t *end) | |
697 | { | |
698 | struct pool *pool = tc->pool; | |
699 | sector_t b = bio->bi_iter.bi_sector; | |
700 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
701 | ||
702 | b += pool->sectors_per_block - 1ull; /* so we round up */ | |
703 | ||
704 | if (block_size_is_power_of_two(pool)) { | |
705 | b >>= pool->sectors_per_block_shift; | |
706 | e >>= pool->sectors_per_block_shift; | |
707 | } else { | |
708 | (void) sector_div(b, pool->sectors_per_block); | |
709 | (void) sector_div(e, pool->sectors_per_block); | |
710 | } | |
711 | ||
712 | if (e < b) | |
713 | /* Can happen if the bio is within a single block. */ | |
714 | e = b; | |
715 | ||
716 | *begin = b; | |
717 | *end = e; | |
718 | } | |
719 | ||
991d9fa0 JT |
720 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
721 | { | |
722 | struct pool *pool = tc->pool; | |
4f024f37 | 723 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 | 724 | |
74d46992 | 725 | bio_set_dev(bio, tc->pool_dev->bdev); |
58f77a21 | 726 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
727 | bio->bi_iter.bi_sector = |
728 | (block << pool->sectors_per_block_shift) | | |
729 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 730 | else |
4f024f37 | 731 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 732 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
733 | } |
734 | ||
2dd9c257 JT |
735 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
736 | { | |
74d46992 | 737 | bio_set_dev(bio, tc->origin_dev->bdev); |
2dd9c257 JT |
738 | } |
739 | ||
4afdd680 JT |
740 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
741 | { | |
f73f44eb | 742 | return op_is_flush(bio->bi_opf) && |
4afdd680 JT |
743 | dm_thin_changed_this_transaction(tc->td); |
744 | } | |
745 | ||
e8088073 JT |
746 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
747 | { | |
748 | struct dm_thin_endio_hook *h; | |
749 | ||
e6047149 | 750 | if (bio_op(bio) == REQ_OP_DISCARD) |
e8088073 JT |
751 | return; |
752 | ||
59c3d2c6 | 753 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
754 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
755 | } | |
756 | ||
2dd9c257 | 757 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
758 | { |
759 | struct pool *pool = tc->pool; | |
760 | unsigned long flags; | |
761 | ||
e49e5829 JT |
762 | if (!bio_triggers_commit(tc, bio)) { |
763 | generic_make_request(bio); | |
764 | return; | |
765 | } | |
766 | ||
991d9fa0 | 767 | /* |
e49e5829 JT |
768 | * Complete bio with an error if earlier I/O caused changes to |
769 | * the metadata that can't be committed e.g, due to I/O errors | |
770 | * on the metadata device. | |
991d9fa0 | 771 | */ |
e49e5829 JT |
772 | if (dm_thin_aborted_changes(tc->td)) { |
773 | bio_io_error(bio); | |
774 | return; | |
775 | } | |
776 | ||
777 | /* | |
778 | * Batch together any bios that trigger commits and then issue a | |
779 | * single commit for them in process_deferred_bios(). | |
780 | */ | |
781 | spin_lock_irqsave(&pool->lock, flags); | |
782 | bio_list_add(&pool->deferred_flush_bios, bio); | |
783 | spin_unlock_irqrestore(&pool->lock, flags); | |
991d9fa0 JT |
784 | } |
785 | ||
2dd9c257 JT |
786 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
787 | { | |
788 | remap_to_origin(tc, bio); | |
789 | issue(tc, bio); | |
790 | } | |
791 | ||
792 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
793 | dm_block_t block) | |
794 | { | |
795 | remap(tc, bio, block); | |
796 | issue(tc, bio); | |
797 | } | |
798 | ||
991d9fa0 JT |
799 | /*----------------------------------------------------------------*/ |
800 | ||
801 | /* | |
802 | * Bio endio functions. | |
803 | */ | |
a24c2569 | 804 | struct dm_thin_new_mapping { |
991d9fa0 JT |
805 | struct list_head list; |
806 | ||
7f214665 | 807 | bool pass_discard:1; |
34fbcf62 | 808 | bool maybe_shared:1; |
991d9fa0 | 809 | |
50f3c3ef JT |
810 | /* |
811 | * Track quiescing, copying and zeroing preparation actions. When this | |
812 | * counter hits zero the block is prepared and can be inserted into the | |
813 | * btree. | |
814 | */ | |
815 | atomic_t prepare_actions; | |
816 | ||
4e4cbee9 | 817 | blk_status_t status; |
991d9fa0 | 818 | struct thin_c *tc; |
34fbcf62 | 819 | dm_block_t virt_begin, virt_end; |
991d9fa0 | 820 | dm_block_t data_block; |
34fbcf62 | 821 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
822 | |
823 | /* | |
824 | * If the bio covers the whole area of a block then we can avoid | |
825 | * zeroing or copying. Instead this bio is hooked. The bio will | |
826 | * still be in the cell, so care has to be taken to avoid issuing | |
827 | * the bio twice. | |
828 | */ | |
829 | struct bio *bio; | |
830 | bio_end_io_t *saved_bi_end_io; | |
831 | }; | |
832 | ||
50f3c3ef | 833 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
834 | { |
835 | struct pool *pool = m->tc->pool; | |
836 | ||
50f3c3ef | 837 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 838 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
839 | wake_worker(pool); |
840 | } | |
841 | } | |
842 | ||
e5aea7b4 | 843 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
844 | { |
845 | unsigned long flags; | |
991d9fa0 JT |
846 | struct pool *pool = m->tc->pool; |
847 | ||
991d9fa0 | 848 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 849 | __complete_mapping_preparation(m); |
991d9fa0 JT |
850 | spin_unlock_irqrestore(&pool->lock, flags); |
851 | } | |
852 | ||
e5aea7b4 JT |
853 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
854 | { | |
855 | struct dm_thin_new_mapping *m = context; | |
856 | ||
4e4cbee9 | 857 | m->status = read_err || write_err ? BLK_STS_IOERR : 0; |
e5aea7b4 JT |
858 | complete_mapping_preparation(m); |
859 | } | |
860 | ||
4246a0b6 | 861 | static void overwrite_endio(struct bio *bio) |
991d9fa0 | 862 | { |
59c3d2c6 | 863 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 864 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 | 865 | |
8b908f8e MS |
866 | bio->bi_end_io = m->saved_bi_end_io; |
867 | ||
4e4cbee9 | 868 | m->status = bio->bi_status; |
e5aea7b4 | 869 | complete_mapping_preparation(m); |
991d9fa0 JT |
870 | } |
871 | ||
991d9fa0 JT |
872 | /*----------------------------------------------------------------*/ |
873 | ||
874 | /* | |
875 | * Workqueue. | |
876 | */ | |
877 | ||
878 | /* | |
879 | * Prepared mapping jobs. | |
880 | */ | |
881 | ||
882 | /* | |
2d759a46 JT |
883 | * This sends the bios in the cell, except the original holder, back |
884 | * to the deferred_bios list. | |
991d9fa0 | 885 | */ |
f286ba0e | 886 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 887 | { |
991d9fa0 JT |
888 | struct pool *pool = tc->pool; |
889 | unsigned long flags; | |
890 | ||
c140e1c4 MS |
891 | spin_lock_irqsave(&tc->lock, flags); |
892 | cell_release_no_holder(pool, cell, &tc->deferred_bio_list); | |
893 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
894 | |
895 | wake_worker(pool); | |
896 | } | |
897 | ||
a374bb21 JT |
898 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
899 | ||
2d759a46 JT |
900 | struct remap_info { |
901 | struct thin_c *tc; | |
902 | struct bio_list defer_bios; | |
903 | struct bio_list issue_bios; | |
904 | }; | |
905 | ||
906 | static void __inc_remap_and_issue_cell(void *context, | |
907 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 908 | { |
2d759a46 | 909 | struct remap_info *info = context; |
a374bb21 | 910 | struct bio *bio; |
a374bb21 | 911 | |
2d759a46 | 912 | while ((bio = bio_list_pop(&cell->bios))) { |
f73f44eb | 913 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) |
2d759a46 | 914 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 915 | else { |
2d759a46 JT |
916 | inc_all_io_entry(info->tc->pool, bio); |
917 | ||
918 | /* | |
919 | * We can't issue the bios with the bio prison lock | |
920 | * held, so we add them to a list to issue on | |
921 | * return from this function. | |
922 | */ | |
923 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
924 | } |
925 | } | |
926 | } | |
927 | ||
2d759a46 JT |
928 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
929 | struct dm_bio_prison_cell *cell, | |
930 | dm_block_t block) | |
931 | { | |
932 | struct bio *bio; | |
933 | struct remap_info info; | |
934 | ||
935 | info.tc = tc; | |
936 | bio_list_init(&info.defer_bios); | |
937 | bio_list_init(&info.issue_bios); | |
938 | ||
939 | /* | |
940 | * We have to be careful to inc any bios we're about to issue | |
941 | * before the cell is released, and avoid a race with new bios | |
942 | * being added to the cell. | |
943 | */ | |
944 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
945 | &info, cell); | |
946 | ||
947 | while ((bio = bio_list_pop(&info.defer_bios))) | |
948 | thin_defer_bio(tc, bio); | |
949 | ||
950 | while ((bio = bio_list_pop(&info.issue_bios))) | |
951 | remap_and_issue(info.tc, bio, block); | |
952 | } | |
953 | ||
e49e5829 JT |
954 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
955 | { | |
6beca5eb | 956 | cell_error(m->tc->pool, m->cell); |
e49e5829 | 957 | list_del(&m->list); |
6f1c819c | 958 | mempool_free(m, &m->tc->pool->mapping_pool); |
e49e5829 | 959 | } |
025b9685 | 960 | |
4ae280b4 NT |
961 | static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio) |
962 | { | |
963 | struct pool *pool = tc->pool; | |
964 | unsigned long flags; | |
965 | ||
966 | /* | |
967 | * If the bio has the REQ_FUA flag set we must commit the metadata | |
968 | * before signaling its completion. | |
969 | */ | |
970 | if (!bio_triggers_commit(tc, bio)) { | |
971 | bio_endio(bio); | |
972 | return; | |
973 | } | |
974 | ||
975 | /* | |
976 | * Complete bio with an error if earlier I/O caused changes to the | |
977 | * metadata that can't be committed, e.g, due to I/O errors on the | |
978 | * metadata device. | |
979 | */ | |
980 | if (dm_thin_aborted_changes(tc->td)) { | |
981 | bio_io_error(bio); | |
982 | return; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Batch together any bios that trigger commits and then issue a | |
987 | * single commit for them in process_deferred_bios(). | |
988 | */ | |
989 | spin_lock_irqsave(&pool->lock, flags); | |
990 | bio_list_add(&pool->deferred_flush_completions, bio); | |
991 | spin_unlock_irqrestore(&pool->lock, flags); | |
992 | } | |
993 | ||
a24c2569 | 994 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
995 | { |
996 | struct thin_c *tc = m->tc; | |
6beca5eb | 997 | struct pool *pool = tc->pool; |
8b908f8e | 998 | struct bio *bio = m->bio; |
991d9fa0 JT |
999 | int r; |
1000 | ||
4e4cbee9 | 1001 | if (m->status) { |
6beca5eb | 1002 | cell_error(pool, m->cell); |
905386f8 | 1003 | goto out; |
991d9fa0 JT |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * Commit the prepared block into the mapping btree. | |
1008 | * Any I/O for this block arriving after this point will get | |
1009 | * remapped to it directly. | |
1010 | */ | |
34fbcf62 | 1011 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
991d9fa0 | 1012 | if (r) { |
b5330655 | 1013 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 1014 | cell_error(pool, m->cell); |
905386f8 | 1015 | goto out; |
991d9fa0 JT |
1016 | } |
1017 | ||
1018 | /* | |
1019 | * Release any bios held while the block was being provisioned. | |
1020 | * If we are processing a write bio that completely covers the block, | |
1021 | * we already processed it so can ignore it now when processing | |
1022 | * the bios in the cell. | |
1023 | */ | |
1024 | if (bio) { | |
2d759a46 | 1025 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
4ae280b4 | 1026 | complete_overwrite_bio(tc, bio); |
2d759a46 JT |
1027 | } else { |
1028 | inc_all_io_entry(tc->pool, m->cell->holder); | |
1029 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
1030 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
1031 | } | |
991d9fa0 | 1032 | |
905386f8 | 1033 | out: |
991d9fa0 | 1034 | list_del(&m->list); |
6f1c819c | 1035 | mempool_free(m, &pool->mapping_pool); |
991d9fa0 JT |
1036 | } |
1037 | ||
34fbcf62 JT |
1038 | /*----------------------------------------------------------------*/ |
1039 | ||
1040 | static void free_discard_mapping(struct dm_thin_new_mapping *m) | |
104655fd | 1041 | { |
104655fd | 1042 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1043 | if (m->cell) |
1044 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1045 | mempool_free(m, &tc->pool->mapping_pool); |
34fbcf62 | 1046 | } |
104655fd | 1047 | |
34fbcf62 JT |
1048 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
1049 | { | |
e49e5829 | 1050 | bio_io_error(m->bio); |
34fbcf62 JT |
1051 | free_discard_mapping(m); |
1052 | } | |
1053 | ||
1054 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) | |
1055 | { | |
4246a0b6 | 1056 | bio_endio(m->bio); |
34fbcf62 JT |
1057 | free_discard_mapping(m); |
1058 | } | |
1059 | ||
1060 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) | |
1061 | { | |
1062 | int r; | |
1063 | struct thin_c *tc = m->tc; | |
1064 | ||
1065 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); | |
1066 | if (r) { | |
1067 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); | |
1068 | bio_io_error(m->bio); | |
1069 | } else | |
4246a0b6 | 1070 | bio_endio(m->bio); |
34fbcf62 | 1071 | |
f286ba0e | 1072 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1073 | mempool_free(m, &tc->pool->mapping_pool); |
e49e5829 JT |
1074 | } |
1075 | ||
202bae52 JT |
1076 | /*----------------------------------------------------------------*/ |
1077 | ||
2a0fbffb JT |
1078 | static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m, |
1079 | struct bio *discard_parent) | |
e49e5829 | 1080 | { |
34fbcf62 JT |
1081 | /* |
1082 | * We've already unmapped this range of blocks, but before we | |
1083 | * passdown we have to check that these blocks are now unused. | |
1084 | */ | |
202bae52 | 1085 | int r = 0; |
d445bd9c | 1086 | bool shared = true; |
e49e5829 | 1087 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1088 | struct pool *pool = tc->pool; |
1089 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; | |
202bae52 | 1090 | struct discard_op op; |
104655fd | 1091 | |
2a0fbffb | 1092 | begin_discard(&op, tc, discard_parent); |
34fbcf62 JT |
1093 | while (b != end) { |
1094 | /* find start of unmapped run */ | |
1095 | for (; b < end; b++) { | |
d445bd9c | 1096 | r = dm_pool_block_is_shared(pool->pmd, b, &shared); |
34fbcf62 | 1097 | if (r) |
202bae52 | 1098 | goto out; |
e8088073 | 1099 | |
d445bd9c | 1100 | if (!shared) |
34fbcf62 | 1101 | break; |
19fa1a67 | 1102 | } |
104655fd | 1103 | |
34fbcf62 JT |
1104 | if (b == end) |
1105 | break; | |
1106 | ||
1107 | /* find end of run */ | |
1108 | for (e = b + 1; e != end; e++) { | |
d445bd9c | 1109 | r = dm_pool_block_is_shared(pool->pmd, e, &shared); |
34fbcf62 | 1110 | if (r) |
202bae52 | 1111 | goto out; |
34fbcf62 | 1112 | |
d445bd9c | 1113 | if (shared) |
34fbcf62 JT |
1114 | break; |
1115 | } | |
1116 | ||
202bae52 | 1117 | r = issue_discard(&op, b, e); |
34fbcf62 | 1118 | if (r) |
202bae52 | 1119 | goto out; |
34fbcf62 JT |
1120 | |
1121 | b = e; | |
1122 | } | |
202bae52 JT |
1123 | out: |
1124 | end_discard(&op, r); | |
104655fd JT |
1125 | } |
1126 | ||
2a0fbffb JT |
1127 | static void queue_passdown_pt2(struct dm_thin_new_mapping *m) |
1128 | { | |
1129 | unsigned long flags; | |
1130 | struct pool *pool = m->tc->pool; | |
1131 | ||
1132 | spin_lock_irqsave(&pool->lock, flags); | |
1133 | list_add_tail(&m->list, &pool->prepared_discards_pt2); | |
1134 | spin_unlock_irqrestore(&pool->lock, flags); | |
1135 | wake_worker(pool); | |
1136 | } | |
1137 | ||
1138 | static void passdown_endio(struct bio *bio) | |
1139 | { | |
1140 | /* | |
1141 | * It doesn't matter if the passdown discard failed, we still want | |
1142 | * to unmap (we ignore err). | |
1143 | */ | |
1144 | queue_passdown_pt2(bio->bi_private); | |
948f581a | 1145 | bio_put(bio); |
2a0fbffb JT |
1146 | } |
1147 | ||
1148 | static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m) | |
e49e5829 JT |
1149 | { |
1150 | int r; | |
1151 | struct thin_c *tc = m->tc; | |
34fbcf62 | 1152 | struct pool *pool = tc->pool; |
2a0fbffb JT |
1153 | struct bio *discard_parent; |
1154 | dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin); | |
e49e5829 | 1155 | |
2a0fbffb JT |
1156 | /* |
1157 | * Only this thread allocates blocks, so we can be sure that the | |
1158 | * newly unmapped blocks will not be allocated before the end of | |
1159 | * the function. | |
1160 | */ | |
34fbcf62 | 1161 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
202bae52 | 1162 | if (r) { |
34fbcf62 | 1163 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
202bae52 | 1164 | bio_io_error(m->bio); |
2a0fbffb | 1165 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1166 | mempool_free(m, &pool->mapping_pool); |
2a0fbffb JT |
1167 | return; |
1168 | } | |
34fbcf62 | 1169 | |
00a0ea33 VV |
1170 | /* |
1171 | * Increment the unmapped blocks. This prevents a race between the | |
1172 | * passdown io and reallocation of freed blocks. | |
1173 | */ | |
1174 | r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end); | |
1175 | if (r) { | |
1176 | metadata_operation_failed(pool, "dm_pool_inc_data_range", r); | |
1177 | bio_io_error(m->bio); | |
1178 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1179 | mempool_free(m, &pool->mapping_pool); |
00a0ea33 VV |
1180 | return; |
1181 | } | |
1182 | ||
2a0fbffb JT |
1183 | discard_parent = bio_alloc(GFP_NOIO, 1); |
1184 | if (!discard_parent) { | |
1185 | DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.", | |
1186 | dm_device_name(tc->pool->pool_md)); | |
1187 | queue_passdown_pt2(m); | |
202bae52 JT |
1188 | |
1189 | } else { | |
2a0fbffb JT |
1190 | discard_parent->bi_end_io = passdown_endio; |
1191 | discard_parent->bi_private = m; | |
1192 | ||
1193 | if (m->maybe_shared) | |
1194 | passdown_double_checking_shared_status(m, discard_parent); | |
1195 | else { | |
1196 | struct discard_op op; | |
1197 | ||
1198 | begin_discard(&op, tc, discard_parent); | |
1199 | r = issue_discard(&op, m->data_block, data_end); | |
1200 | end_discard(&op, r); | |
1201 | } | |
202bae52 | 1202 | } |
2a0fbffb JT |
1203 | } |
1204 | ||
1205 | static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m) | |
1206 | { | |
1207 | int r; | |
1208 | struct thin_c *tc = m->tc; | |
1209 | struct pool *pool = tc->pool; | |
1210 | ||
1211 | /* | |
1212 | * The passdown has completed, so now we can decrement all those | |
1213 | * unmapped blocks. | |
1214 | */ | |
1215 | r = dm_pool_dec_data_range(pool->pmd, m->data_block, | |
1216 | m->data_block + (m->virt_end - m->virt_begin)); | |
1217 | if (r) { | |
1218 | metadata_operation_failed(pool, "dm_pool_dec_data_range", r); | |
1219 | bio_io_error(m->bio); | |
1220 | } else | |
1221 | bio_endio(m->bio); | |
1222 | ||
34fbcf62 | 1223 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1224 | mempool_free(m, &pool->mapping_pool); |
e49e5829 JT |
1225 | } |
1226 | ||
104655fd | 1227 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 1228 | process_mapping_fn *fn) |
991d9fa0 JT |
1229 | { |
1230 | unsigned long flags; | |
1231 | struct list_head maps; | |
a24c2569 | 1232 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
1233 | |
1234 | INIT_LIST_HEAD(&maps); | |
1235 | spin_lock_irqsave(&pool->lock, flags); | |
104655fd | 1236 | list_splice_init(head, &maps); |
991d9fa0 JT |
1237 | spin_unlock_irqrestore(&pool->lock, flags); |
1238 | ||
1239 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1240 | (*fn)(m); |
991d9fa0 JT |
1241 | } |
1242 | ||
1243 | /* | |
1244 | * Deferred bio jobs. | |
1245 | */ | |
104655fd | 1246 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1247 | { |
4f024f37 KO |
1248 | return bio->bi_iter.bi_size == |
1249 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
1250 | } |
1251 | ||
1252 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1253 | { | |
1254 | return (bio_data_dir(bio) == WRITE) && | |
1255 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1256 | } |
1257 | ||
1258 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1259 | bio_end_io_t *fn) | |
1260 | { | |
1261 | *save = bio->bi_end_io; | |
1262 | bio->bi_end_io = fn; | |
1263 | } | |
1264 | ||
1265 | static int ensure_next_mapping(struct pool *pool) | |
1266 | { | |
1267 | if (pool->next_mapping) | |
1268 | return 0; | |
1269 | ||
6f1c819c | 1270 | pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC); |
991d9fa0 JT |
1271 | |
1272 | return pool->next_mapping ? 0 : -ENOMEM; | |
1273 | } | |
1274 | ||
a24c2569 | 1275 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1276 | { |
16961b04 | 1277 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
1278 | |
1279 | BUG_ON(!pool->next_mapping); | |
1280 | ||
16961b04 MS |
1281 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
1282 | INIT_LIST_HEAD(&m->list); | |
1283 | m->bio = NULL; | |
1284 | ||
991d9fa0 JT |
1285 | pool->next_mapping = NULL; |
1286 | ||
16961b04 | 1287 | return m; |
991d9fa0 JT |
1288 | } |
1289 | ||
e5aea7b4 JT |
1290 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
1291 | sector_t begin, sector_t end) | |
1292 | { | |
e5aea7b4 JT |
1293 | struct dm_io_region to; |
1294 | ||
1295 | to.bdev = tc->pool_dev->bdev; | |
1296 | to.sector = begin; | |
1297 | to.count = end - begin; | |
1298 | ||
7209049d | 1299 | dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); |
e5aea7b4 JT |
1300 | } |
1301 | ||
452d7a62 | 1302 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
34fbcf62 | 1303 | dm_block_t data_begin, |
452d7a62 MS |
1304 | struct dm_thin_new_mapping *m) |
1305 | { | |
1306 | struct pool *pool = tc->pool; | |
1307 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1308 | ||
1309 | h->overwrite_mapping = m; | |
1310 | m->bio = bio; | |
1311 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
1312 | inc_all_io_entry(pool, bio); | |
34fbcf62 | 1313 | remap_and_issue(tc, bio, data_begin); |
452d7a62 MS |
1314 | } |
1315 | ||
e5aea7b4 JT |
1316 | /* |
1317 | * A partial copy also needs to zero the uncopied region. | |
1318 | */ | |
991d9fa0 | 1319 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
1320 | struct dm_dev *origin, dm_block_t data_origin, |
1321 | dm_block_t data_dest, | |
e5aea7b4 JT |
1322 | struct dm_bio_prison_cell *cell, struct bio *bio, |
1323 | sector_t len) | |
991d9fa0 | 1324 | { |
991d9fa0 | 1325 | struct pool *pool = tc->pool; |
a24c2569 | 1326 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1327 | |
991d9fa0 | 1328 | m->tc = tc; |
34fbcf62 JT |
1329 | m->virt_begin = virt_block; |
1330 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1331 | m->data_block = data_dest; |
1332 | m->cell = cell; | |
991d9fa0 | 1333 | |
e5aea7b4 JT |
1334 | /* |
1335 | * quiesce action + copy action + an extra reference held for the | |
1336 | * duration of this function (we may need to inc later for a | |
1337 | * partial zero). | |
1338 | */ | |
1339 | atomic_set(&m->prepare_actions, 3); | |
1340 | ||
44feb387 | 1341 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1342 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1343 | |
1344 | /* | |
1345 | * IO to pool_dev remaps to the pool target's data_dev. | |
1346 | * | |
1347 | * If the whole block of data is being overwritten, we can issue the | |
1348 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1349 | */ | |
452d7a62 MS |
1350 | if (io_overwrites_block(pool, bio)) |
1351 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1352 | else { | |
991d9fa0 JT |
1353 | struct dm_io_region from, to; |
1354 | ||
2dd9c257 | 1355 | from.bdev = origin->bdev; |
991d9fa0 | 1356 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1357 | from.count = len; |
991d9fa0 JT |
1358 | |
1359 | to.bdev = tc->pool_dev->bdev; | |
1360 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1361 | to.count = len; |
991d9fa0 | 1362 | |
7209049d MS |
1363 | dm_kcopyd_copy(pool->copier, &from, 1, &to, |
1364 | 0, copy_complete, m); | |
e5aea7b4 JT |
1365 | |
1366 | /* | |
1367 | * Do we need to zero a tail region? | |
1368 | */ | |
1369 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1370 | atomic_inc(&m->prepare_actions); | |
1371 | ll_zero(tc, m, | |
1372 | data_dest * pool->sectors_per_block + len, | |
1373 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1374 | } |
1375 | } | |
e5aea7b4 JT |
1376 | |
1377 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1378 | } |
1379 | ||
2dd9c257 JT |
1380 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1381 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1382 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1383 | { |
1384 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1385 | data_origin, data_dest, cell, bio, |
1386 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1387 | } |
1388 | ||
991d9fa0 | 1389 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1390 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1391 | struct bio *bio) |
1392 | { | |
1393 | struct pool *pool = tc->pool; | |
a24c2569 | 1394 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1395 | |
50f3c3ef | 1396 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 | 1397 | m->tc = tc; |
34fbcf62 JT |
1398 | m->virt_begin = virt_block; |
1399 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1400 | m->data_block = data_block; |
1401 | m->cell = cell; | |
991d9fa0 JT |
1402 | |
1403 | /* | |
1404 | * If the whole block of data is being overwritten or we are not | |
1405 | * zeroing pre-existing data, we can issue the bio immediately. | |
1406 | * Otherwise we use kcopyd to zero the data first. | |
1407 | */ | |
f8ae7525 MS |
1408 | if (pool->pf.zero_new_blocks) { |
1409 | if (io_overwrites_block(pool, bio)) | |
1410 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
1411 | else | |
1412 | ll_zero(tc, m, data_block * pool->sectors_per_block, | |
1413 | (data_block + 1) * pool->sectors_per_block); | |
1414 | } else | |
991d9fa0 | 1415 | process_prepared_mapping(m); |
e5aea7b4 | 1416 | } |
991d9fa0 | 1417 | |
e5aea7b4 JT |
1418 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1419 | dm_block_t data_dest, | |
1420 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1421 | { | |
1422 | struct pool *pool = tc->pool; | |
1423 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1424 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1425 | ||
1426 | if (virt_block_end <= tc->origin_size) | |
1427 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1428 | virt_block, data_dest, cell, bio, | |
1429 | pool->sectors_per_block); | |
1430 | ||
1431 | else if (virt_block_begin < tc->origin_size) | |
1432 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1433 | virt_block, data_dest, cell, bio, | |
1434 | tc->origin_size - virt_block_begin); | |
1435 | ||
1436 | else | |
1437 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1438 | } |
1439 | ||
2c43fd26 JT |
1440 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1441 | ||
a685557f MS |
1442 | static void requeue_bios(struct pool *pool); |
1443 | ||
3ab91828 JT |
1444 | static bool is_read_only_pool_mode(enum pool_mode mode) |
1445 | { | |
1446 | return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY); | |
1447 | } | |
1448 | ||
1449 | static bool is_read_only(struct pool *pool) | |
1450 | { | |
1451 | return is_read_only_pool_mode(get_pool_mode(pool)); | |
1452 | } | |
1453 | ||
1454 | static void check_for_metadata_space(struct pool *pool) | |
1455 | { | |
1456 | int r; | |
1457 | const char *ooms_reason = NULL; | |
1458 | dm_block_t nr_free; | |
1459 | ||
1460 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free); | |
1461 | if (r) | |
1462 | ooms_reason = "Could not get free metadata blocks"; | |
1463 | else if (!nr_free) | |
1464 | ooms_reason = "No free metadata blocks"; | |
1465 | ||
1466 | if (ooms_reason && !is_read_only(pool)) { | |
1467 | DMERR("%s", ooms_reason); | |
1468 | set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE); | |
1469 | } | |
1470 | } | |
1471 | ||
1472 | static void check_for_data_space(struct pool *pool) | |
2c43fd26 JT |
1473 | { |
1474 | int r; | |
1475 | dm_block_t nr_free; | |
1476 | ||
1477 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) | |
1478 | return; | |
1479 | ||
1480 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); | |
1481 | if (r) | |
1482 | return; | |
1483 | ||
a685557f | 1484 | if (nr_free) { |
2c43fd26 | 1485 | set_pool_mode(pool, PM_WRITE); |
a685557f MS |
1486 | requeue_bios(pool); |
1487 | } | |
2c43fd26 JT |
1488 | } |
1489 | ||
e49e5829 JT |
1490 | /* |
1491 | * A non-zero return indicates read_only or fail_io mode. | |
1492 | * Many callers don't care about the return value. | |
1493 | */ | |
020cc3b5 | 1494 | static int commit(struct pool *pool) |
e49e5829 JT |
1495 | { |
1496 | int r; | |
1497 | ||
3ab91828 | 1498 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) |
e49e5829 JT |
1499 | return -EINVAL; |
1500 | ||
020cc3b5 | 1501 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1502 | if (r) |
1503 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
3ab91828 JT |
1504 | else { |
1505 | check_for_metadata_space(pool); | |
1506 | check_for_data_space(pool); | |
1507 | } | |
e49e5829 JT |
1508 | |
1509 | return r; | |
1510 | } | |
1511 | ||
88a6621b JT |
1512 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1513 | { | |
1514 | unsigned long flags; | |
1515 | ||
1516 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { | |
1517 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1518 | dm_device_name(pool->pool_md)); | |
1519 | spin_lock_irqsave(&pool->lock, flags); | |
1520 | pool->low_water_triggered = true; | |
1521 | spin_unlock_irqrestore(&pool->lock, flags); | |
1522 | dm_table_event(pool->ti->table); | |
1523 | } | |
1524 | } | |
1525 | ||
991d9fa0 JT |
1526 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1527 | { | |
1528 | int r; | |
1529 | dm_block_t free_blocks; | |
991d9fa0 JT |
1530 | struct pool *pool = tc->pool; |
1531 | ||
3e1a0699 | 1532 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1533 | return -EINVAL; |
1534 | ||
991d9fa0 | 1535 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1536 | if (r) { |
1537 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1538 | return r; |
b5330655 | 1539 | } |
991d9fa0 | 1540 | |
88a6621b | 1541 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1542 | |
1543 | if (!free_blocks) { | |
94563bad MS |
1544 | /* |
1545 | * Try to commit to see if that will free up some | |
1546 | * more space. | |
1547 | */ | |
020cc3b5 JT |
1548 | r = commit(pool); |
1549 | if (r) | |
1550 | return r; | |
991d9fa0 | 1551 | |
94563bad | 1552 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1553 | if (r) { |
1554 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1555 | return r; |
b5330655 | 1556 | } |
991d9fa0 | 1557 | |
94563bad | 1558 | if (!free_blocks) { |
3e1a0699 | 1559 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1560 | return -ENOSPC; |
991d9fa0 JT |
1561 | } |
1562 | } | |
1563 | ||
1564 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1565 | if (r) { |
a685557f MS |
1566 | if (r == -ENOSPC) |
1567 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); | |
1568 | else | |
1569 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); | |
991d9fa0 | 1570 | return r; |
4a02b34e | 1571 | } |
991d9fa0 | 1572 | |
3ab91828 JT |
1573 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks); |
1574 | if (r) { | |
1575 | metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r); | |
1576 | return r; | |
1577 | } | |
1578 | ||
1579 | if (!free_blocks) { | |
1580 | /* Let's commit before we use up the metadata reserve. */ | |
1581 | r = commit(pool); | |
1582 | if (r) | |
1583 | return r; | |
1584 | } | |
1585 | ||
991d9fa0 JT |
1586 | return 0; |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * If we have run out of space, queue bios until the device is | |
1591 | * resumed, presumably after having been reloaded with more space. | |
1592 | */ | |
1593 | static void retry_on_resume(struct bio *bio) | |
1594 | { | |
59c3d2c6 | 1595 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1596 | struct thin_c *tc = h->tc; |
991d9fa0 JT |
1597 | unsigned long flags; |
1598 | ||
c140e1c4 MS |
1599 | spin_lock_irqsave(&tc->lock, flags); |
1600 | bio_list_add(&tc->retry_on_resume_list, bio); | |
1601 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
1602 | } |
1603 | ||
4e4cbee9 | 1604 | static blk_status_t should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1605 | { |
3e1a0699 JT |
1606 | enum pool_mode m = get_pool_mode(pool); |
1607 | ||
1608 | switch (m) { | |
1609 | case PM_WRITE: | |
1610 | /* Shouldn't get here */ | |
1611 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
4e4cbee9 | 1612 | return BLK_STS_IOERR; |
3e1a0699 JT |
1613 | |
1614 | case PM_OUT_OF_DATA_SPACE: | |
4e4cbee9 | 1615 | return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0; |
3e1a0699 | 1616 | |
3ab91828 | 1617 | case PM_OUT_OF_METADATA_SPACE: |
3e1a0699 JT |
1618 | case PM_READ_ONLY: |
1619 | case PM_FAIL: | |
4e4cbee9 | 1620 | return BLK_STS_IOERR; |
3e1a0699 JT |
1621 | default: |
1622 | /* Shouldn't get here */ | |
1623 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
4e4cbee9 | 1624 | return BLK_STS_IOERR; |
3e1a0699 JT |
1625 | } |
1626 | } | |
8c0f0e8c | 1627 | |
3e1a0699 JT |
1628 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1629 | { | |
4e4cbee9 | 1630 | blk_status_t error = should_error_unserviceable_bio(pool); |
af91805a | 1631 | |
4246a0b6 | 1632 | if (error) { |
4e4cbee9 | 1633 | bio->bi_status = error; |
4246a0b6 CH |
1634 | bio_endio(bio); |
1635 | } else | |
6d16202b | 1636 | retry_on_resume(bio); |
8c0f0e8c MS |
1637 | } |
1638 | ||
399caddf | 1639 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1640 | { |
1641 | struct bio *bio; | |
1642 | struct bio_list bios; | |
4e4cbee9 | 1643 | blk_status_t error; |
991d9fa0 | 1644 | |
af91805a MS |
1645 | error = should_error_unserviceable_bio(pool); |
1646 | if (error) { | |
1647 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1648 | return; |
1649 | } | |
1650 | ||
991d9fa0 | 1651 | bio_list_init(&bios); |
6beca5eb | 1652 | cell_release(pool, cell, &bios); |
991d9fa0 | 1653 | |
9d094eeb MS |
1654 | while ((bio = bio_list_pop(&bios))) |
1655 | retry_on_resume(bio); | |
991d9fa0 JT |
1656 | } |
1657 | ||
34fbcf62 JT |
1658 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
1659 | struct dm_bio_prison_cell *virt_cell) | |
104655fd | 1660 | { |
104655fd | 1661 | struct pool *pool = tc->pool; |
34fbcf62 | 1662 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
104655fd | 1663 | |
34fbcf62 JT |
1664 | /* |
1665 | * We don't need to lock the data blocks, since there's no | |
1666 | * passdown. We only lock data blocks for allocation and breaking sharing. | |
1667 | */ | |
1668 | m->tc = tc; | |
1669 | m->virt_begin = virt_cell->key.block_begin; | |
1670 | m->virt_end = virt_cell->key.block_end; | |
1671 | m->cell = virt_cell; | |
1672 | m->bio = virt_cell->holder; | |
104655fd | 1673 | |
34fbcf62 JT |
1674 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1675 | pool->process_prepared_discard(m); | |
1676 | } | |
104655fd | 1677 | |
34fbcf62 JT |
1678 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
1679 | struct bio *bio) | |
1680 | { | |
1681 | struct pool *pool = tc->pool; | |
1682 | ||
1683 | int r; | |
1684 | bool maybe_shared; | |
1685 | struct dm_cell_key data_key; | |
1686 | struct dm_bio_prison_cell *data_cell; | |
1687 | struct dm_thin_new_mapping *m; | |
1688 | dm_block_t virt_begin, virt_end, data_begin; | |
1689 | ||
1690 | while (begin != end) { | |
1691 | r = ensure_next_mapping(pool); | |
1692 | if (r) | |
1693 | /* we did our best */ | |
1694 | return; | |
e8088073 | 1695 | |
34fbcf62 JT |
1696 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
1697 | &data_begin, &maybe_shared); | |
1698 | if (r) | |
104655fd | 1699 | /* |
34fbcf62 JT |
1700 | * Silently fail, letting any mappings we've |
1701 | * created complete. | |
104655fd | 1702 | */ |
34fbcf62 JT |
1703 | break; |
1704 | ||
1705 | build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); | |
1706 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { | |
1707 | /* contention, we'll give up with this range */ | |
1708 | begin = virt_end; | |
1709 | continue; | |
104655fd | 1710 | } |
104655fd | 1711 | |
104655fd | 1712 | /* |
34fbcf62 JT |
1713 | * IO may still be going to the destination block. We must |
1714 | * quiesce before we can do the removal. | |
104655fd | 1715 | */ |
34fbcf62 JT |
1716 | m = get_next_mapping(pool); |
1717 | m->tc = tc; | |
1718 | m->maybe_shared = maybe_shared; | |
1719 | m->virt_begin = virt_begin; | |
1720 | m->virt_end = virt_end; | |
1721 | m->data_block = data_begin; | |
1722 | m->cell = data_cell; | |
1723 | m->bio = bio; | |
104655fd | 1724 | |
34fbcf62 JT |
1725 | /* |
1726 | * The parent bio must not complete before sub discard bios are | |
202bae52 | 1727 | * chained to it (see end_discard's bio_chain)! |
34fbcf62 JT |
1728 | * |
1729 | * This per-mapping bi_remaining increment is paired with | |
1730 | * the implicit decrement that occurs via bio_endio() in | |
202bae52 | 1731 | * end_discard(). |
34fbcf62 | 1732 | */ |
13e4f8a6 | 1733 | bio_inc_remaining(bio); |
34fbcf62 JT |
1734 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1735 | pool->process_prepared_discard(m); | |
1736 | ||
1737 | begin = virt_end; | |
104655fd JT |
1738 | } |
1739 | } | |
1740 | ||
34fbcf62 JT |
1741 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
1742 | { | |
1743 | struct bio *bio = virt_cell->holder; | |
1744 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1745 | ||
1746 | /* | |
1747 | * The virt_cell will only get freed once the origin bio completes. | |
1748 | * This means it will remain locked while all the individual | |
1749 | * passdown bios are in flight. | |
1750 | */ | |
1751 | h->cell = virt_cell; | |
1752 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); | |
1753 | ||
1754 | /* | |
1755 | * We complete the bio now, knowing that the bi_remaining field | |
1756 | * will prevent completion until the sub range discards have | |
1757 | * completed. | |
1758 | */ | |
4246a0b6 | 1759 | bio_endio(bio); |
34fbcf62 JT |
1760 | } |
1761 | ||
a374bb21 JT |
1762 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1763 | { | |
34fbcf62 JT |
1764 | dm_block_t begin, end; |
1765 | struct dm_cell_key virt_key; | |
1766 | struct dm_bio_prison_cell *virt_cell; | |
a374bb21 | 1767 | |
34fbcf62 JT |
1768 | get_bio_block_range(tc, bio, &begin, &end); |
1769 | if (begin == end) { | |
1770 | /* | |
1771 | * The discard covers less than a block. | |
1772 | */ | |
4246a0b6 | 1773 | bio_endio(bio); |
a374bb21 | 1774 | return; |
34fbcf62 | 1775 | } |
a374bb21 | 1776 | |
34fbcf62 JT |
1777 | build_key(tc->td, VIRTUAL, begin, end, &virt_key); |
1778 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) | |
1779 | /* | |
1780 | * Potential starvation issue: We're relying on the | |
1781 | * fs/application being well behaved, and not trying to | |
1782 | * send IO to a region at the same time as discarding it. | |
1783 | * If they do this persistently then it's possible this | |
1784 | * cell will never be granted. | |
1785 | */ | |
1786 | return; | |
1787 | ||
1788 | tc->pool->process_discard_cell(tc, virt_cell); | |
a374bb21 JT |
1789 | } |
1790 | ||
991d9fa0 | 1791 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1792 | struct dm_cell_key *key, |
991d9fa0 | 1793 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1794 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1795 | { |
1796 | int r; | |
1797 | dm_block_t data_block; | |
d6fc2042 | 1798 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1799 | |
1800 | r = alloc_data_block(tc, &data_block); | |
1801 | switch (r) { | |
1802 | case 0: | |
2dd9c257 JT |
1803 | schedule_internal_copy(tc, block, lookup_result->block, |
1804 | data_block, cell, bio); | |
991d9fa0 JT |
1805 | break; |
1806 | ||
1807 | case -ENOSPC: | |
399caddf | 1808 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1809 | break; |
1810 | ||
1811 | default: | |
c397741c MS |
1812 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1813 | __func__, r); | |
d6fc2042 | 1814 | cell_error(pool, cell); |
991d9fa0 JT |
1815 | break; |
1816 | } | |
1817 | } | |
1818 | ||
23ca2bb6 JT |
1819 | static void __remap_and_issue_shared_cell(void *context, |
1820 | struct dm_bio_prison_cell *cell) | |
1821 | { | |
1822 | struct remap_info *info = context; | |
1823 | struct bio *bio; | |
1824 | ||
1825 | while ((bio = bio_list_pop(&cell->bios))) { | |
f73f44eb CH |
1826 | if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) || |
1827 | bio_op(bio) == REQ_OP_DISCARD) | |
23ca2bb6 JT |
1828 | bio_list_add(&info->defer_bios, bio); |
1829 | else { | |
bd6d1e0a | 1830 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
23ca2bb6 JT |
1831 | |
1832 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1833 | inc_all_io_entry(info->tc->pool, bio); | |
1834 | bio_list_add(&info->issue_bios, bio); | |
1835 | } | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1840 | struct dm_bio_prison_cell *cell, | |
1841 | dm_block_t block) | |
1842 | { | |
1843 | struct bio *bio; | |
1844 | struct remap_info info; | |
1845 | ||
1846 | info.tc = tc; | |
1847 | bio_list_init(&info.defer_bios); | |
1848 | bio_list_init(&info.issue_bios); | |
1849 | ||
1850 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1851 | &info, cell); | |
1852 | ||
1853 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1854 | thin_defer_bio(tc, bio); | |
1855 | ||
1856 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1857 | remap_and_issue(tc, bio, block); | |
1858 | } | |
1859 | ||
991d9fa0 JT |
1860 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1861 | dm_block_t block, | |
23ca2bb6 JT |
1862 | struct dm_thin_lookup_result *lookup_result, |
1863 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1864 | { |
23ca2bb6 | 1865 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1866 | struct pool *pool = tc->pool; |
44feb387 | 1867 | struct dm_cell_key key; |
991d9fa0 JT |
1868 | |
1869 | /* | |
1870 | * If cell is already occupied, then sharing is already in the process | |
1871 | * of being broken so we have nothing further to do here. | |
1872 | */ | |
1873 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1874 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1875 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1876 | return; |
23ca2bb6 | 1877 | } |
991d9fa0 | 1878 | |
23ca2bb6 JT |
1879 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1880 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1881 | cell_defer_no_holder(tc, virt_cell); | |
1882 | } else { | |
59c3d2c6 | 1883 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1884 | |
44feb387 | 1885 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1886 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1887 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1888 | |
1889 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1890 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1891 | } |
1892 | } | |
1893 | ||
1894 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1895 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1896 | { |
1897 | int r; | |
1898 | dm_block_t data_block; | |
6beca5eb | 1899 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1900 | |
1901 | /* | |
1902 | * Remap empty bios (flushes) immediately, without provisioning. | |
1903 | */ | |
4f024f37 | 1904 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1905 | inc_all_io_entry(pool, bio); |
f286ba0e | 1906 | cell_defer_no_holder(tc, cell); |
e8088073 | 1907 | |
991d9fa0 JT |
1908 | remap_and_issue(tc, bio, 0); |
1909 | return; | |
1910 | } | |
1911 | ||
1912 | /* | |
1913 | * Fill read bios with zeroes and complete them immediately. | |
1914 | */ | |
1915 | if (bio_data_dir(bio) == READ) { | |
1916 | zero_fill_bio(bio); | |
f286ba0e | 1917 | cell_defer_no_holder(tc, cell); |
4246a0b6 | 1918 | bio_endio(bio); |
991d9fa0 JT |
1919 | return; |
1920 | } | |
1921 | ||
1922 | r = alloc_data_block(tc, &data_block); | |
1923 | switch (r) { | |
1924 | case 0: | |
2dd9c257 JT |
1925 | if (tc->origin_dev) |
1926 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1927 | else | |
1928 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1929 | break; |
1930 | ||
1931 | case -ENOSPC: | |
399caddf | 1932 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1933 | break; |
1934 | ||
1935 | default: | |
c397741c MS |
1936 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1937 | __func__, r); | |
6beca5eb | 1938 | cell_error(pool, cell); |
991d9fa0 JT |
1939 | break; |
1940 | } | |
1941 | } | |
1942 | ||
a374bb21 | 1943 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1944 | { |
1945 | int r; | |
6beca5eb | 1946 | struct pool *pool = tc->pool; |
a374bb21 | 1947 | struct bio *bio = cell->holder; |
991d9fa0 | 1948 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1949 | struct dm_thin_lookup_result lookup_result; |
1950 | ||
a374bb21 JT |
1951 | if (tc->requeue_mode) { |
1952 | cell_requeue(pool, cell); | |
991d9fa0 | 1953 | return; |
a374bb21 | 1954 | } |
991d9fa0 JT |
1955 | |
1956 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1957 | switch (r) { | |
1958 | case 0: | |
23ca2bb6 JT |
1959 | if (lookup_result.shared) |
1960 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1961 | else { | |
6beca5eb | 1962 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1963 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1964 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1965 | } |
991d9fa0 JT |
1966 | break; |
1967 | ||
1968 | case -ENODATA: | |
2dd9c257 | 1969 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1970 | inc_all_io_entry(pool, bio); |
f286ba0e | 1971 | cell_defer_no_holder(tc, cell); |
e8088073 | 1972 | |
e5aea7b4 JT |
1973 | if (bio_end_sector(bio) <= tc->origin_size) |
1974 | remap_to_origin_and_issue(tc, bio); | |
1975 | ||
1976 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1977 | zero_fill_bio(bio); | |
1978 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1979 | remap_to_origin_and_issue(tc, bio); | |
1980 | ||
1981 | } else { | |
1982 | zero_fill_bio(bio); | |
4246a0b6 | 1983 | bio_endio(bio); |
e5aea7b4 | 1984 | } |
2dd9c257 JT |
1985 | } else |
1986 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
1987 | break; |
1988 | ||
1989 | default: | |
c397741c MS |
1990 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
1991 | __func__, r); | |
f286ba0e | 1992 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
1993 | bio_io_error(bio); |
1994 | break; | |
1995 | } | |
1996 | } | |
1997 | ||
a374bb21 JT |
1998 | static void process_bio(struct thin_c *tc, struct bio *bio) |
1999 | { | |
2000 | struct pool *pool = tc->pool; | |
2001 | dm_block_t block = get_bio_block(tc, bio); | |
2002 | struct dm_bio_prison_cell *cell; | |
2003 | struct dm_cell_key key; | |
2004 | ||
2005 | /* | |
2006 | * If cell is already occupied, then the block is already | |
2007 | * being provisioned so we have nothing further to do here. | |
2008 | */ | |
2009 | build_virtual_key(tc->td, block, &key); | |
2010 | if (bio_detain(pool, &key, bio, &cell)) | |
2011 | return; | |
2012 | ||
2013 | process_cell(tc, cell); | |
2014 | } | |
2015 | ||
2016 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
2017 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
2018 | { |
2019 | int r; | |
2020 | int rw = bio_data_dir(bio); | |
2021 | dm_block_t block = get_bio_block(tc, bio); | |
2022 | struct dm_thin_lookup_result lookup_result; | |
2023 | ||
2024 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
2025 | switch (r) { | |
2026 | case 0: | |
a374bb21 | 2027 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 2028 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
2029 | if (cell) |
2030 | cell_defer_no_holder(tc, cell); | |
2031 | } else { | |
e8088073 | 2032 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 2033 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
2034 | if (cell) |
2035 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 2036 | } |
e49e5829 JT |
2037 | break; |
2038 | ||
2039 | case -ENODATA: | |
a374bb21 JT |
2040 | if (cell) |
2041 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 2042 | if (rw != READ) { |
8c0f0e8c | 2043 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
2044 | break; |
2045 | } | |
2046 | ||
2047 | if (tc->origin_dev) { | |
e8088073 | 2048 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
2049 | remap_to_origin_and_issue(tc, bio); |
2050 | break; | |
2051 | } | |
2052 | ||
2053 | zero_fill_bio(bio); | |
4246a0b6 | 2054 | bio_endio(bio); |
e49e5829 JT |
2055 | break; |
2056 | ||
2057 | default: | |
c397741c MS |
2058 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
2059 | __func__, r); | |
a374bb21 JT |
2060 | if (cell) |
2061 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
2062 | bio_io_error(bio); |
2063 | break; | |
2064 | } | |
2065 | } | |
2066 | ||
a374bb21 JT |
2067 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
2068 | { | |
2069 | __process_bio_read_only(tc, bio, NULL); | |
2070 | } | |
2071 | ||
2072 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2073 | { | |
2074 | __process_bio_read_only(tc, cell->holder, cell); | |
2075 | } | |
2076 | ||
3e1a0699 JT |
2077 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
2078 | { | |
4246a0b6 | 2079 | bio_endio(bio); |
3e1a0699 JT |
2080 | } |
2081 | ||
e49e5829 JT |
2082 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
2083 | { | |
2084 | bio_io_error(bio); | |
2085 | } | |
2086 | ||
a374bb21 JT |
2087 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2088 | { | |
2089 | cell_success(tc->pool, cell); | |
2090 | } | |
2091 | ||
2092 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2093 | { | |
2094 | cell_error(tc->pool, cell); | |
2095 | } | |
2096 | ||
ac8c3f3d JT |
2097 | /* |
2098 | * FIXME: should we also commit due to size of transaction, measured in | |
2099 | * metadata blocks? | |
2100 | */ | |
905e51b3 JT |
2101 | static int need_commit_due_to_time(struct pool *pool) |
2102 | { | |
0f30af98 MS |
2103 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
2104 | pool->last_commit_jiffies + COMMIT_PERIOD); | |
905e51b3 JT |
2105 | } |
2106 | ||
67324ea1 MS |
2107 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
2108 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
2109 | ||
2110 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
2111 | { | |
2112 | struct rb_node **rbp, *parent; | |
2113 | struct dm_thin_endio_hook *pbd; | |
2114 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
2115 | ||
2116 | rbp = &tc->sort_bio_list.rb_node; | |
2117 | parent = NULL; | |
2118 | while (*rbp) { | |
2119 | parent = *rbp; | |
2120 | pbd = thin_pbd(parent); | |
2121 | ||
2122 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
2123 | rbp = &(*rbp)->rb_left; | |
2124 | else | |
2125 | rbp = &(*rbp)->rb_right; | |
2126 | } | |
2127 | ||
2128 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
2129 | rb_link_node(&pbd->rb_node, parent, rbp); | |
2130 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
2131 | } | |
2132 | ||
2133 | static void __extract_sorted_bios(struct thin_c *tc) | |
2134 | { | |
2135 | struct rb_node *node; | |
2136 | struct dm_thin_endio_hook *pbd; | |
2137 | struct bio *bio; | |
2138 | ||
2139 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
2140 | pbd = thin_pbd(node); | |
2141 | bio = thin_bio(pbd); | |
2142 | ||
2143 | bio_list_add(&tc->deferred_bio_list, bio); | |
2144 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
2145 | } | |
2146 | ||
2147 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
2148 | } | |
2149 | ||
2150 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
2151 | { | |
2152 | struct bio *bio; | |
2153 | struct bio_list bios; | |
2154 | ||
2155 | bio_list_init(&bios); | |
2156 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
2157 | bio_list_init(&tc->deferred_bio_list); | |
2158 | ||
2159 | /* Sort deferred_bio_list using rb-tree */ | |
2160 | while ((bio = bio_list_pop(&bios))) | |
2161 | __thin_bio_rb_add(tc, bio); | |
2162 | ||
2163 | /* | |
2164 | * Transfer the sorted bios in sort_bio_list back to | |
2165 | * deferred_bio_list to allow lockless submission of | |
2166 | * all bios. | |
2167 | */ | |
2168 | __extract_sorted_bios(tc); | |
2169 | } | |
2170 | ||
c140e1c4 | 2171 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 2172 | { |
c140e1c4 | 2173 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2174 | unsigned long flags; |
2175 | struct bio *bio; | |
2176 | struct bio_list bios; | |
67324ea1 | 2177 | struct blk_plug plug; |
8a01a6af | 2178 | unsigned count = 0; |
991d9fa0 | 2179 | |
c140e1c4 | 2180 | if (tc->requeue_mode) { |
4e4cbee9 CH |
2181 | error_thin_bio_list(tc, &tc->deferred_bio_list, |
2182 | BLK_STS_DM_REQUEUE); | |
c140e1c4 MS |
2183 | return; |
2184 | } | |
2185 | ||
991d9fa0 JT |
2186 | bio_list_init(&bios); |
2187 | ||
c140e1c4 | 2188 | spin_lock_irqsave(&tc->lock, flags); |
67324ea1 MS |
2189 | |
2190 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
2191 | spin_unlock_irqrestore(&tc->lock, flags); | |
2192 | return; | |
2193 | } | |
2194 | ||
2195 | __sort_thin_deferred_bios(tc); | |
2196 | ||
c140e1c4 MS |
2197 | bio_list_merge(&bios, &tc->deferred_bio_list); |
2198 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 2199 | |
c140e1c4 | 2200 | spin_unlock_irqrestore(&tc->lock, flags); |
991d9fa0 | 2201 | |
67324ea1 | 2202 | blk_start_plug(&plug); |
991d9fa0 | 2203 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
2204 | /* |
2205 | * If we've got no free new_mapping structs, and processing | |
2206 | * this bio might require one, we pause until there are some | |
2207 | * prepared mappings to process. | |
2208 | */ | |
2209 | if (ensure_next_mapping(pool)) { | |
c140e1c4 MS |
2210 | spin_lock_irqsave(&tc->lock, flags); |
2211 | bio_list_add(&tc->deferred_bio_list, bio); | |
2212 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
2213 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2214 | break; |
2215 | } | |
104655fd | 2216 | |
e6047149 | 2217 | if (bio_op(bio) == REQ_OP_DISCARD) |
e49e5829 | 2218 | pool->process_discard(tc, bio); |
104655fd | 2219 | else |
e49e5829 | 2220 | pool->process_bio(tc, bio); |
8a01a6af JT |
2221 | |
2222 | if ((count++ & 127) == 0) { | |
7d327fe0 | 2223 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
2224 | dm_pool_issue_prefetches(pool->pmd); |
2225 | } | |
991d9fa0 | 2226 | } |
67324ea1 | 2227 | blk_finish_plug(&plug); |
c140e1c4 MS |
2228 | } |
2229 | ||
ac4c3f34 JT |
2230 | static int cmp_cells(const void *lhs, const void *rhs) |
2231 | { | |
2232 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
2233 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
2234 | ||
2235 | BUG_ON(!lhs_cell->holder); | |
2236 | BUG_ON(!rhs_cell->holder); | |
2237 | ||
2238 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
2239 | return -1; | |
2240 | ||
2241 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
2242 | return 1; | |
2243 | ||
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | static unsigned sort_cells(struct pool *pool, struct list_head *cells) | |
2248 | { | |
2249 | unsigned count = 0; | |
2250 | struct dm_bio_prison_cell *cell, *tmp; | |
2251 | ||
2252 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
2253 | if (count >= CELL_SORT_ARRAY_SIZE) | |
2254 | break; | |
2255 | ||
2256 | pool->cell_sort_array[count++] = cell; | |
2257 | list_del(&cell->user_list); | |
2258 | } | |
2259 | ||
2260 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
2261 | ||
2262 | return count; | |
2263 | } | |
2264 | ||
a374bb21 JT |
2265 | static void process_thin_deferred_cells(struct thin_c *tc) |
2266 | { | |
2267 | struct pool *pool = tc->pool; | |
2268 | unsigned long flags; | |
2269 | struct list_head cells; | |
ac4c3f34 JT |
2270 | struct dm_bio_prison_cell *cell; |
2271 | unsigned i, j, count; | |
a374bb21 JT |
2272 | |
2273 | INIT_LIST_HEAD(&cells); | |
2274 | ||
2275 | spin_lock_irqsave(&tc->lock, flags); | |
2276 | list_splice_init(&tc->deferred_cells, &cells); | |
2277 | spin_unlock_irqrestore(&tc->lock, flags); | |
2278 | ||
2279 | if (list_empty(&cells)) | |
2280 | return; | |
2281 | ||
ac4c3f34 JT |
2282 | do { |
2283 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 2284 | |
ac4c3f34 JT |
2285 | for (i = 0; i < count; i++) { |
2286 | cell = pool->cell_sort_array[i]; | |
2287 | BUG_ON(!cell->holder); | |
a374bb21 | 2288 | |
ac4c3f34 JT |
2289 | /* |
2290 | * If we've got no free new_mapping structs, and processing | |
2291 | * this bio might require one, we pause until there are some | |
2292 | * prepared mappings to process. | |
2293 | */ | |
2294 | if (ensure_next_mapping(pool)) { | |
2295 | for (j = i; j < count; j++) | |
2296 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
2297 | ||
2298 | spin_lock_irqsave(&tc->lock, flags); | |
2299 | list_splice(&cells, &tc->deferred_cells); | |
2300 | spin_unlock_irqrestore(&tc->lock, flags); | |
2301 | return; | |
2302 | } | |
2303 | ||
e6047149 | 2304 | if (bio_op(cell->holder) == REQ_OP_DISCARD) |
ac4c3f34 JT |
2305 | pool->process_discard_cell(tc, cell); |
2306 | else | |
2307 | pool->process_cell(tc, cell); | |
2308 | } | |
2309 | } while (!list_empty(&cells)); | |
a374bb21 JT |
2310 | } |
2311 | ||
b10ebd34 JT |
2312 | static void thin_get(struct thin_c *tc); |
2313 | static void thin_put(struct thin_c *tc); | |
2314 | ||
2315 | /* | |
2316 | * We can't hold rcu_read_lock() around code that can block. So we | |
2317 | * find a thin with the rcu lock held; bump a refcount; then drop | |
2318 | * the lock. | |
2319 | */ | |
2320 | static struct thin_c *get_first_thin(struct pool *pool) | |
2321 | { | |
2322 | struct thin_c *tc = NULL; | |
2323 | ||
2324 | rcu_read_lock(); | |
2325 | if (!list_empty(&pool->active_thins)) { | |
2326 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
2327 | thin_get(tc); | |
2328 | } | |
2329 | rcu_read_unlock(); | |
2330 | ||
2331 | return tc; | |
2332 | } | |
2333 | ||
2334 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
2335 | { | |
2336 | struct thin_c *old_tc = tc; | |
2337 | ||
2338 | rcu_read_lock(); | |
2339 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
2340 | thin_get(tc); | |
2341 | thin_put(old_tc); | |
2342 | rcu_read_unlock(); | |
2343 | return tc; | |
2344 | } | |
2345 | thin_put(old_tc); | |
2346 | rcu_read_unlock(); | |
2347 | ||
2348 | return NULL; | |
2349 | } | |
2350 | ||
c140e1c4 MS |
2351 | static void process_deferred_bios(struct pool *pool) |
2352 | { | |
2353 | unsigned long flags; | |
2354 | struct bio *bio; | |
4ae280b4 | 2355 | struct bio_list bios, bio_completions; |
c140e1c4 MS |
2356 | struct thin_c *tc; |
2357 | ||
b10ebd34 JT |
2358 | tc = get_first_thin(pool); |
2359 | while (tc) { | |
a374bb21 | 2360 | process_thin_deferred_cells(tc); |
c140e1c4 | 2361 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
2362 | tc = get_next_thin(pool, tc); |
2363 | } | |
991d9fa0 JT |
2364 | |
2365 | /* | |
4ae280b4 NT |
2366 | * If there are any deferred flush bios, we must commit the metadata |
2367 | * before issuing them or signaling their completion. | |
991d9fa0 JT |
2368 | */ |
2369 | bio_list_init(&bios); | |
4ae280b4 NT |
2370 | bio_list_init(&bio_completions); |
2371 | ||
991d9fa0 JT |
2372 | spin_lock_irqsave(&pool->lock, flags); |
2373 | bio_list_merge(&bios, &pool->deferred_flush_bios); | |
2374 | bio_list_init(&pool->deferred_flush_bios); | |
4ae280b4 NT |
2375 | |
2376 | bio_list_merge(&bio_completions, &pool->deferred_flush_completions); | |
2377 | bio_list_init(&pool->deferred_flush_completions); | |
991d9fa0 JT |
2378 | spin_unlock_irqrestore(&pool->lock, flags); |
2379 | ||
4ae280b4 | 2380 | if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) && |
4d1662a3 | 2381 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) |
991d9fa0 JT |
2382 | return; |
2383 | ||
020cc3b5 | 2384 | if (commit(pool)) { |
4ae280b4 NT |
2385 | bio_list_merge(&bios, &bio_completions); |
2386 | ||
991d9fa0 JT |
2387 | while ((bio = bio_list_pop(&bios))) |
2388 | bio_io_error(bio); | |
2389 | return; | |
2390 | } | |
905e51b3 | 2391 | pool->last_commit_jiffies = jiffies; |
991d9fa0 | 2392 | |
4ae280b4 NT |
2393 | while ((bio = bio_list_pop(&bio_completions))) |
2394 | bio_endio(bio); | |
2395 | ||
ddd2185e NT |
2396 | while ((bio = bio_list_pop(&bios))) { |
2397 | /* | |
2398 | * The data device was flushed as part of metadata commit, | |
2399 | * so complete redundant flushes immediately. | |
2400 | */ | |
2401 | if (bio->bi_opf & REQ_PREFLUSH) | |
2402 | bio_endio(bio); | |
2403 | else | |
2404 | generic_make_request(bio); | |
2405 | } | |
991d9fa0 JT |
2406 | } |
2407 | ||
2408 | static void do_worker(struct work_struct *ws) | |
2409 | { | |
2410 | struct pool *pool = container_of(ws, struct pool, worker); | |
2411 | ||
7d327fe0 | 2412 | throttle_work_start(&pool->throttle); |
8a01a6af | 2413 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 2414 | throttle_work_update(&pool->throttle); |
e49e5829 | 2415 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 2416 | throttle_work_update(&pool->throttle); |
e49e5829 | 2417 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 2418 | throttle_work_update(&pool->throttle); |
2a0fbffb JT |
2419 | process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2); |
2420 | throttle_work_update(&pool->throttle); | |
991d9fa0 | 2421 | process_deferred_bios(pool); |
7d327fe0 | 2422 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
2423 | } |
2424 | ||
905e51b3 JT |
2425 | /* |
2426 | * We want to commit periodically so that not too much | |
2427 | * unwritten data builds up. | |
2428 | */ | |
2429 | static void do_waker(struct work_struct *ws) | |
2430 | { | |
2431 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
2432 | wake_worker(pool); | |
2433 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
2434 | } | |
2435 | ||
85ad643b JT |
2436 | /* |
2437 | * We're holding onto IO to allow userland time to react. After the | |
2438 | * timeout either the pool will have been resized (and thus back in | |
bcc696fa | 2439 | * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space. |
85ad643b JT |
2440 | */ |
2441 | static void do_no_space_timeout(struct work_struct *ws) | |
2442 | { | |
2443 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2444 | no_space_timeout); | |
2445 | ||
bcc696fa MS |
2446 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) { |
2447 | pool->pf.error_if_no_space = true; | |
f6c36758 | 2448 | notify_of_pool_mode_change(pool); |
4e4cbee9 | 2449 | error_retry_list_with_code(pool, BLK_STS_NOSPC); |
bcc696fa | 2450 | } |
85ad643b JT |
2451 | } |
2452 | ||
991d9fa0 JT |
2453 | /*----------------------------------------------------------------*/ |
2454 | ||
e7a3e871 | 2455 | struct pool_work { |
738211f7 | 2456 | struct work_struct worker; |
e7a3e871 JT |
2457 | struct completion complete; |
2458 | }; | |
2459 | ||
2460 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2461 | { | |
2462 | return container_of(ws, struct pool_work, worker); | |
2463 | } | |
2464 | ||
2465 | static void pool_work_complete(struct pool_work *pw) | |
2466 | { | |
2467 | complete(&pw->complete); | |
2468 | } | |
738211f7 | 2469 | |
e7a3e871 JT |
2470 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2471 | void (*fn)(struct work_struct *)) | |
2472 | { | |
2473 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2474 | init_completion(&pw->complete); | |
2475 | queue_work(pool->wq, &pw->worker); | |
2476 | wait_for_completion(&pw->complete); | |
2477 | } | |
2478 | ||
2479 | /*----------------------------------------------------------------*/ | |
2480 | ||
2481 | struct noflush_work { | |
2482 | struct pool_work pw; | |
2483 | struct thin_c *tc; | |
738211f7 JT |
2484 | }; |
2485 | ||
e7a3e871 | 2486 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2487 | { |
e7a3e871 | 2488 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2489 | } |
2490 | ||
2491 | static void do_noflush_start(struct work_struct *ws) | |
2492 | { | |
e7a3e871 | 2493 | struct noflush_work *w = to_noflush(ws); |
738211f7 JT |
2494 | w->tc->requeue_mode = true; |
2495 | requeue_io(w->tc); | |
e7a3e871 | 2496 | pool_work_complete(&w->pw); |
738211f7 JT |
2497 | } |
2498 | ||
2499 | static void do_noflush_stop(struct work_struct *ws) | |
2500 | { | |
e7a3e871 | 2501 | struct noflush_work *w = to_noflush(ws); |
738211f7 | 2502 | w->tc->requeue_mode = false; |
e7a3e871 | 2503 | pool_work_complete(&w->pw); |
738211f7 JT |
2504 | } |
2505 | ||
2506 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2507 | { | |
2508 | struct noflush_work w; | |
2509 | ||
738211f7 | 2510 | w.tc = tc; |
e7a3e871 | 2511 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2512 | } |
2513 | ||
2514 | /*----------------------------------------------------------------*/ | |
2515 | ||
34fbcf62 JT |
2516 | static bool passdown_enabled(struct pool_c *pt) |
2517 | { | |
2518 | return pt->adjusted_pf.discard_passdown; | |
2519 | } | |
2520 | ||
2521 | static void set_discard_callbacks(struct pool *pool) | |
2522 | { | |
2523 | struct pool_c *pt = pool->ti->private; | |
2524 | ||
2525 | if (passdown_enabled(pt)) { | |
2526 | pool->process_discard_cell = process_discard_cell_passdown; | |
2a0fbffb JT |
2527 | pool->process_prepared_discard = process_prepared_discard_passdown_pt1; |
2528 | pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2; | |
34fbcf62 JT |
2529 | } else { |
2530 | pool->process_discard_cell = process_discard_cell_no_passdown; | |
2531 | pool->process_prepared_discard = process_prepared_discard_no_passdown; | |
2532 | } | |
2533 | } | |
2534 | ||
8b64e881 | 2535 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2536 | { |
cdc2b415 | 2537 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2538 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2539 | enum pool_mode old_mode = get_pool_mode(pool); | |
6aa7de05 | 2540 | unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2541 | |
2542 | /* | |
2543 | * Never allow the pool to transition to PM_WRITE mode if user | |
2544 | * intervention is required to verify metadata and data consistency. | |
2545 | */ | |
2546 | if (new_mode == PM_WRITE && needs_check) { | |
2547 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2548 | dm_device_name(pool->pool_md)); | |
2549 | if (old_mode != new_mode) | |
2550 | new_mode = old_mode; | |
2551 | else | |
2552 | new_mode = PM_READ_ONLY; | |
2553 | } | |
2554 | /* | |
2555 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2556 | * not going to recover without a thin_repair. So we never let the | |
2557 | * pool move out of the old mode. | |
2558 | */ | |
2559 | if (old_mode == PM_FAIL) | |
2560 | new_mode = old_mode; | |
e49e5829 | 2561 | |
8b64e881 | 2562 | switch (new_mode) { |
e49e5829 | 2563 | case PM_FAIL: |
5383ef3a | 2564 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2565 | pool->process_bio = process_bio_fail; |
2566 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2567 | pool->process_cell = process_cell_fail; |
2568 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2569 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2570 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2571 | |
2572 | error_retry_list(pool); | |
e49e5829 JT |
2573 | break; |
2574 | ||
3ab91828 | 2575 | case PM_OUT_OF_METADATA_SPACE: |
e49e5829 | 2576 | case PM_READ_ONLY: |
3e1a0699 JT |
2577 | dm_pool_metadata_read_only(pool->pmd); |
2578 | pool->process_bio = process_bio_read_only; | |
2579 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2580 | pool->process_cell = process_cell_read_only; |
2581 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 | 2582 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
34fbcf62 | 2583 | pool->process_prepared_discard = process_prepared_discard_success; |
3e1a0699 JT |
2584 | |
2585 | error_retry_list(pool); | |
2586 | break; | |
2587 | ||
2588 | case PM_OUT_OF_DATA_SPACE: | |
2589 | /* | |
2590 | * Ideally we'd never hit this state; the low water mark | |
2591 | * would trigger userland to extend the pool before we | |
2592 | * completely run out of data space. However, many small | |
2593 | * IOs to unprovisioned space can consume data space at an | |
2594 | * alarming rate. Adjust your low water mark if you're | |
2595 | * frequently seeing this mode. | |
2596 | */ | |
c3667cc6 | 2597 | pool->out_of_data_space = true; |
3e1a0699 | 2598 | pool->process_bio = process_bio_read_only; |
a374bb21 JT |
2599 | pool->process_discard = process_discard_bio; |
2600 | pool->process_cell = process_cell_read_only; | |
3e1a0699 | 2601 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2602 | set_discard_callbacks(pool); |
85ad643b | 2603 | |
80c57893 MS |
2604 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2605 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2606 | break; |
2607 | ||
2608 | case PM_WRITE: | |
75294442 HT |
2609 | if (old_mode == PM_OUT_OF_DATA_SPACE) |
2610 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
c3667cc6 | 2611 | pool->out_of_data_space = false; |
172c2386 | 2612 | pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space; |
9b7aaa64 | 2613 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2614 | pool->process_bio = process_bio; |
a374bb21 JT |
2615 | pool->process_discard = process_discard_bio; |
2616 | pool->process_cell = process_cell; | |
e49e5829 | 2617 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2618 | set_discard_callbacks(pool); |
e49e5829 JT |
2619 | break; |
2620 | } | |
8b64e881 MS |
2621 | |
2622 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2623 | /* |
2624 | * The pool mode may have changed, sync it so bind_control_target() | |
2625 | * doesn't cause an unexpected mode transition on resume. | |
2626 | */ | |
2627 | pt->adjusted_pf.mode = new_mode; | |
f6c36758 MS |
2628 | |
2629 | if (old_mode != new_mode) | |
2630 | notify_of_pool_mode_change(pool); | |
e49e5829 JT |
2631 | } |
2632 | ||
07f2b6e0 | 2633 | static void abort_transaction(struct pool *pool) |
b5330655 | 2634 | { |
07f2b6e0 MS |
2635 | const char *dev_name = dm_device_name(pool->pool_md); |
2636 | ||
2637 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2638 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2639 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2640 | set_pool_mode(pool, PM_FAIL); | |
2641 | } | |
2642 | ||
2643 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2644 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2645 | set_pool_mode(pool, PM_FAIL); | |
2646 | } | |
2647 | } | |
399caddf | 2648 | |
07f2b6e0 MS |
2649 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2650 | { | |
b5330655 JT |
2651 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2652 | dm_device_name(pool->pool_md), op, r); | |
2653 | ||
07f2b6e0 | 2654 | abort_transaction(pool); |
b5330655 JT |
2655 | set_pool_mode(pool, PM_READ_ONLY); |
2656 | } | |
2657 | ||
e49e5829 JT |
2658 | /*----------------------------------------------------------------*/ |
2659 | ||
991d9fa0 JT |
2660 | /* |
2661 | * Mapping functions. | |
2662 | */ | |
2663 | ||
2664 | /* | |
2665 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2666 | */ | |
2667 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2668 | { | |
2669 | unsigned long flags; | |
2670 | struct pool *pool = tc->pool; | |
2671 | ||
c140e1c4 MS |
2672 | spin_lock_irqsave(&tc->lock, flags); |
2673 | bio_list_add(&tc->deferred_bio_list, bio); | |
2674 | spin_unlock_irqrestore(&tc->lock, flags); | |
991d9fa0 JT |
2675 | |
2676 | wake_worker(pool); | |
2677 | } | |
2678 | ||
7d327fe0 JT |
2679 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2680 | { | |
2681 | struct pool *pool = tc->pool; | |
2682 | ||
2683 | throttle_lock(&pool->throttle); | |
2684 | thin_defer_bio(tc, bio); | |
2685 | throttle_unlock(&pool->throttle); | |
2686 | } | |
2687 | ||
a374bb21 JT |
2688 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2689 | { | |
2690 | unsigned long flags; | |
2691 | struct pool *pool = tc->pool; | |
2692 | ||
2693 | throttle_lock(&pool->throttle); | |
2694 | spin_lock_irqsave(&tc->lock, flags); | |
2695 | list_add_tail(&cell->user_list, &tc->deferred_cells); | |
2696 | spin_unlock_irqrestore(&tc->lock, flags); | |
2697 | throttle_unlock(&pool->throttle); | |
2698 | ||
2699 | wake_worker(pool); | |
2700 | } | |
2701 | ||
59c3d2c6 | 2702 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2703 | { |
59c3d2c6 | 2704 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2705 | |
2706 | h->tc = tc; | |
2707 | h->shared_read_entry = NULL; | |
e8088073 | 2708 | h->all_io_entry = NULL; |
eb2aa48d | 2709 | h->overwrite_mapping = NULL; |
34fbcf62 | 2710 | h->cell = NULL; |
eb2aa48d JT |
2711 | } |
2712 | ||
991d9fa0 JT |
2713 | /* |
2714 | * Non-blocking function called from the thin target's map function. | |
2715 | */ | |
7de3ee57 | 2716 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2717 | { |
2718 | int r; | |
2719 | struct thin_c *tc = ti->private; | |
2720 | dm_block_t block = get_bio_block(tc, bio); | |
2721 | struct dm_thin_device *td = tc->td; | |
2722 | struct dm_thin_lookup_result result; | |
a374bb21 | 2723 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2724 | struct dm_cell_key key; |
991d9fa0 | 2725 | |
59c3d2c6 | 2726 | thin_hook_bio(tc, bio); |
e49e5829 | 2727 | |
738211f7 | 2728 | if (tc->requeue_mode) { |
4e4cbee9 | 2729 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 | 2730 | bio_endio(bio); |
738211f7 JT |
2731 | return DM_MAPIO_SUBMITTED; |
2732 | } | |
2733 | ||
e49e5829 JT |
2734 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2735 | bio_io_error(bio); | |
2736 | return DM_MAPIO_SUBMITTED; | |
2737 | } | |
2738 | ||
f73f44eb | 2739 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) { |
7d327fe0 | 2740 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2741 | return DM_MAPIO_SUBMITTED; |
2742 | } | |
2743 | ||
c822ed96 JT |
2744 | /* |
2745 | * We must hold the virtual cell before doing the lookup, otherwise | |
2746 | * there's a race with discard. | |
2747 | */ | |
2748 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2749 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2750 | return DM_MAPIO_SUBMITTED; |
2751 | ||
991d9fa0 JT |
2752 | r = dm_thin_find_block(td, block, 0, &result); |
2753 | ||
2754 | /* | |
2755 | * Note that we defer readahead too. | |
2756 | */ | |
2757 | switch (r) { | |
2758 | case 0: | |
2759 | if (unlikely(result.shared)) { | |
2760 | /* | |
2761 | * We have a race condition here between the | |
2762 | * result.shared value returned by the lookup and | |
2763 | * snapshot creation, which may cause new | |
2764 | * sharing. | |
2765 | * | |
2766 | * To avoid this always quiesce the origin before | |
2767 | * taking the snap. You want to do this anyway to | |
2768 | * ensure a consistent application view | |
2769 | * (i.e. lockfs). | |
2770 | * | |
2771 | * More distant ancestors are irrelevant. The | |
2772 | * shared flag will be set in their case. | |
2773 | */ | |
a374bb21 | 2774 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2775 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2776 | } |
e8088073 | 2777 | |
e8088073 | 2778 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2779 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2780 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2781 | return DM_MAPIO_SUBMITTED; |
2782 | } | |
2783 | ||
2784 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2785 | cell_defer_no_holder(tc, data_cell); |
2786 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2787 | |
2788 | remap(tc, bio, result.block); | |
2789 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2790 | |
2791 | case -ENODATA: | |
e49e5829 | 2792 | case -EWOULDBLOCK: |
a374bb21 | 2793 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2794 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2795 | |
2796 | default: | |
2797 | /* | |
2798 | * Must always call bio_io_error on failure. | |
2799 | * dm_thin_find_block can fail with -EINVAL if the | |
2800 | * pool is switched to fail-io mode. | |
2801 | */ | |
2802 | bio_io_error(bio); | |
a374bb21 | 2803 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2804 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2805 | } |
991d9fa0 JT |
2806 | } |
2807 | ||
2808 | static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
2809 | { | |
991d9fa0 | 2810 | struct pool_c *pt = container_of(cb, struct pool_c, callbacks); |
760fe67e | 2811 | struct request_queue *q; |
991d9fa0 | 2812 | |
760fe67e MS |
2813 | if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE) |
2814 | return 1; | |
991d9fa0 | 2815 | |
760fe67e | 2816 | q = bdev_get_queue(pt->data_dev->bdev); |
dc3b17cc | 2817 | return bdi_congested(q->backing_dev_info, bdi_bits); |
991d9fa0 JT |
2818 | } |
2819 | ||
c140e1c4 | 2820 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2821 | { |
c140e1c4 MS |
2822 | unsigned long flags; |
2823 | struct thin_c *tc; | |
2824 | ||
2825 | rcu_read_lock(); | |
2826 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
2827 | spin_lock_irqsave(&tc->lock, flags); | |
2828 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); | |
2829 | bio_list_init(&tc->retry_on_resume_list); | |
2830 | spin_unlock_irqrestore(&tc->lock, flags); | |
2831 | } | |
2832 | rcu_read_unlock(); | |
991d9fa0 JT |
2833 | } |
2834 | ||
2835 | /*---------------------------------------------------------------- | |
2836 | * Binding of control targets to a pool object | |
2837 | *--------------------------------------------------------------*/ | |
9bc142dd MS |
2838 | static bool data_dev_supports_discard(struct pool_c *pt) |
2839 | { | |
2840 | struct request_queue *q = bdev_get_queue(pt->data_dev->bdev); | |
2841 | ||
2842 | return q && blk_queue_discard(q); | |
2843 | } | |
2844 | ||
58051b94 JT |
2845 | static bool is_factor(sector_t block_size, uint32_t n) |
2846 | { | |
2847 | return !sector_div(block_size, n); | |
2848 | } | |
2849 | ||
9bc142dd MS |
2850 | /* |
2851 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2852 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2853 | */ |
0424caa1 | 2854 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2855 | { |
0424caa1 MS |
2856 | struct pool *pool = pt->pool; |
2857 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2858 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
0424caa1 | 2859 | const char *reason = NULL; |
9bc142dd MS |
2860 | char buf[BDEVNAME_SIZE]; |
2861 | ||
0424caa1 | 2862 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2863 | return; |
2864 | ||
0424caa1 MS |
2865 | if (!data_dev_supports_discard(pt)) |
2866 | reason = "discard unsupported"; | |
2867 | ||
2868 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2869 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2870 | |
0424caa1 MS |
2871 | if (reason) { |
2872 | DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason); | |
2873 | pt->adjusted_pf.discard_passdown = false; | |
2874 | } | |
9bc142dd MS |
2875 | } |
2876 | ||
991d9fa0 JT |
2877 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2878 | { | |
2879 | struct pool_c *pt = ti->private; | |
2880 | ||
e49e5829 | 2881 | /* |
9b7aaa64 | 2882 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2883 | */ |
07f2b6e0 | 2884 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2885 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2886 | |
8b64e881 MS |
2887 | /* |
2888 | * Don't change the pool's mode until set_pool_mode() below. | |
2889 | * Otherwise the pool's process_* function pointers may | |
2890 | * not match the desired pool mode. | |
2891 | */ | |
2892 | pt->adjusted_pf.mode = old_mode; | |
2893 | ||
2894 | pool->ti = ti; | |
2895 | pool->pf = pt->adjusted_pf; | |
2896 | pool->low_water_blocks = pt->low_water_blocks; | |
2897 | ||
9bc142dd | 2898 | set_pool_mode(pool, new_mode); |
f402693d | 2899 | |
991d9fa0 JT |
2900 | return 0; |
2901 | } | |
2902 | ||
2903 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2904 | { | |
2905 | if (pool->ti == ti) | |
2906 | pool->ti = NULL; | |
2907 | } | |
2908 | ||
2909 | /*---------------------------------------------------------------- | |
2910 | * Pool creation | |
2911 | *--------------------------------------------------------------*/ | |
67e2e2b2 JT |
2912 | /* Initialize pool features. */ |
2913 | static void pool_features_init(struct pool_features *pf) | |
2914 | { | |
e49e5829 | 2915 | pf->mode = PM_WRITE; |
9bc142dd MS |
2916 | pf->zero_new_blocks = true; |
2917 | pf->discard_enabled = true; | |
2918 | pf->discard_passdown = true; | |
787a996c | 2919 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2920 | } |
2921 | ||
991d9fa0 JT |
2922 | static void __pool_destroy(struct pool *pool) |
2923 | { | |
2924 | __pool_table_remove(pool); | |
2925 | ||
a822c83e | 2926 | vfree(pool->cell_sort_array); |
991d9fa0 JT |
2927 | if (dm_pool_metadata_close(pool->pmd) < 0) |
2928 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2929 | ||
44feb387 | 2930 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2931 | dm_kcopyd_client_destroy(pool->copier); |
2932 | ||
2933 | if (pool->wq) | |
2934 | destroy_workqueue(pool->wq); | |
2935 | ||
2936 | if (pool->next_mapping) | |
6f1c819c KO |
2937 | mempool_free(pool->next_mapping, &pool->mapping_pool); |
2938 | mempool_exit(&pool->mapping_pool); | |
44feb387 MS |
2939 | dm_deferred_set_destroy(pool->shared_read_ds); |
2940 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2941 | kfree(pool); |
2942 | } | |
2943 | ||
a24c2569 | 2944 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2945 | |
991d9fa0 JT |
2946 | static struct pool *pool_create(struct mapped_device *pool_md, |
2947 | struct block_device *metadata_dev, | |
e49e5829 JT |
2948 | unsigned long block_size, |
2949 | int read_only, char **error) | |
991d9fa0 JT |
2950 | { |
2951 | int r; | |
2952 | void *err_p; | |
2953 | struct pool *pool; | |
2954 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2955 | bool format_device = read_only ? false : true; |
991d9fa0 | 2956 | |
e49e5829 | 2957 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2958 | if (IS_ERR(pmd)) { |
2959 | *error = "Error creating metadata object"; | |
2960 | return (struct pool *)pmd; | |
2961 | } | |
2962 | ||
d3775354 | 2963 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
991d9fa0 JT |
2964 | if (!pool) { |
2965 | *error = "Error allocating memory for pool"; | |
2966 | err_p = ERR_PTR(-ENOMEM); | |
2967 | goto bad_pool; | |
2968 | } | |
2969 | ||
2970 | pool->pmd = pmd; | |
2971 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2972 | if (block_size & (block_size - 1)) |
2973 | pool->sectors_per_block_shift = -1; | |
2974 | else | |
2975 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2976 | pool->low_water_blocks = 0; |
67e2e2b2 | 2977 | pool_features_init(&pool->pf); |
a195db2d | 2978 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2979 | if (!pool->prison) { |
2980 | *error = "Error creating pool's bio prison"; | |
2981 | err_p = ERR_PTR(-ENOMEM); | |
2982 | goto bad_prison; | |
2983 | } | |
2984 | ||
df5d2e90 | 2985 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2986 | if (IS_ERR(pool->copier)) { |
2987 | r = PTR_ERR(pool->copier); | |
2988 | *error = "Error creating pool's kcopyd client"; | |
2989 | err_p = ERR_PTR(r); | |
2990 | goto bad_kcopyd_client; | |
2991 | } | |
2992 | ||
2993 | /* | |
2994 | * Create singlethreaded workqueue that will service all devices | |
2995 | * that use this metadata. | |
2996 | */ | |
2997 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2998 | if (!pool->wq) { | |
2999 | *error = "Error creating pool's workqueue"; | |
3000 | err_p = ERR_PTR(-ENOMEM); | |
3001 | goto bad_wq; | |
3002 | } | |
3003 | ||
7d327fe0 | 3004 | throttle_init(&pool->throttle); |
991d9fa0 | 3005 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 3006 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 3007 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 3008 | spin_lock_init(&pool->lock); |
991d9fa0 | 3009 | bio_list_init(&pool->deferred_flush_bios); |
4ae280b4 | 3010 | bio_list_init(&pool->deferred_flush_completions); |
991d9fa0 | 3011 | INIT_LIST_HEAD(&pool->prepared_mappings); |
104655fd | 3012 | INIT_LIST_HEAD(&pool->prepared_discards); |
2a0fbffb | 3013 | INIT_LIST_HEAD(&pool->prepared_discards_pt2); |
c140e1c4 | 3014 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 3015 | pool->low_water_triggered = false; |
80e96c54 | 3016 | pool->suspended = true; |
c3667cc6 | 3017 | pool->out_of_data_space = false; |
44feb387 MS |
3018 | |
3019 | pool->shared_read_ds = dm_deferred_set_create(); | |
3020 | if (!pool->shared_read_ds) { | |
3021 | *error = "Error creating pool's shared read deferred set"; | |
3022 | err_p = ERR_PTR(-ENOMEM); | |
3023 | goto bad_shared_read_ds; | |
3024 | } | |
3025 | ||
3026 | pool->all_io_ds = dm_deferred_set_create(); | |
3027 | if (!pool->all_io_ds) { | |
3028 | *error = "Error creating pool's all io deferred set"; | |
3029 | err_p = ERR_PTR(-ENOMEM); | |
3030 | goto bad_all_io_ds; | |
3031 | } | |
991d9fa0 JT |
3032 | |
3033 | pool->next_mapping = NULL; | |
6f1c819c KO |
3034 | r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE, |
3035 | _new_mapping_cache); | |
3036 | if (r) { | |
991d9fa0 | 3037 | *error = "Error creating pool's mapping mempool"; |
6f1c819c | 3038 | err_p = ERR_PTR(r); |
991d9fa0 JT |
3039 | goto bad_mapping_pool; |
3040 | } | |
3041 | ||
42bc47b3 KC |
3042 | pool->cell_sort_array = |
3043 | vmalloc(array_size(CELL_SORT_ARRAY_SIZE, | |
3044 | sizeof(*pool->cell_sort_array))); | |
a822c83e JT |
3045 | if (!pool->cell_sort_array) { |
3046 | *error = "Error allocating cell sort array"; | |
3047 | err_p = ERR_PTR(-ENOMEM); | |
3048 | goto bad_sort_array; | |
3049 | } | |
3050 | ||
991d9fa0 | 3051 | pool->ref_count = 1; |
905e51b3 | 3052 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
3053 | pool->pool_md = pool_md; |
3054 | pool->md_dev = metadata_dev; | |
3055 | __pool_table_insert(pool); | |
3056 | ||
3057 | return pool; | |
3058 | ||
a822c83e | 3059 | bad_sort_array: |
6f1c819c | 3060 | mempool_exit(&pool->mapping_pool); |
991d9fa0 | 3061 | bad_mapping_pool: |
44feb387 MS |
3062 | dm_deferred_set_destroy(pool->all_io_ds); |
3063 | bad_all_io_ds: | |
3064 | dm_deferred_set_destroy(pool->shared_read_ds); | |
3065 | bad_shared_read_ds: | |
991d9fa0 JT |
3066 | destroy_workqueue(pool->wq); |
3067 | bad_wq: | |
3068 | dm_kcopyd_client_destroy(pool->copier); | |
3069 | bad_kcopyd_client: | |
44feb387 | 3070 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
3071 | bad_prison: |
3072 | kfree(pool); | |
3073 | bad_pool: | |
3074 | if (dm_pool_metadata_close(pmd)) | |
3075 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
3076 | ||
3077 | return err_p; | |
3078 | } | |
3079 | ||
3080 | static void __pool_inc(struct pool *pool) | |
3081 | { | |
3082 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
3083 | pool->ref_count++; | |
3084 | } | |
3085 | ||
3086 | static void __pool_dec(struct pool *pool) | |
3087 | { | |
3088 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
3089 | BUG_ON(!pool->ref_count); | |
3090 | if (!--pool->ref_count) | |
3091 | __pool_destroy(pool); | |
3092 | } | |
3093 | ||
3094 | static struct pool *__pool_find(struct mapped_device *pool_md, | |
3095 | struct block_device *metadata_dev, | |
e49e5829 JT |
3096 | unsigned long block_size, int read_only, |
3097 | char **error, int *created) | |
991d9fa0 JT |
3098 | { |
3099 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
3100 | ||
3101 | if (pool) { | |
f09996c9 MS |
3102 | if (pool->pool_md != pool_md) { |
3103 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 3104 | return ERR_PTR(-EBUSY); |
f09996c9 | 3105 | } |
991d9fa0 JT |
3106 | __pool_inc(pool); |
3107 | ||
3108 | } else { | |
3109 | pool = __pool_table_lookup(pool_md); | |
3110 | if (pool) { | |
f09996c9 MS |
3111 | if (pool->md_dev != metadata_dev) { |
3112 | *error = "different pool cannot replace a pool"; | |
991d9fa0 | 3113 | return ERR_PTR(-EINVAL); |
f09996c9 | 3114 | } |
991d9fa0 JT |
3115 | __pool_inc(pool); |
3116 | ||
67e2e2b2 | 3117 | } else { |
e49e5829 | 3118 | pool = pool_create(pool_md, metadata_dev, block_size, read_only, error); |
67e2e2b2 JT |
3119 | *created = 1; |
3120 | } | |
991d9fa0 JT |
3121 | } |
3122 | ||
3123 | return pool; | |
3124 | } | |
3125 | ||
3126 | /*---------------------------------------------------------------- | |
3127 | * Pool target methods | |
3128 | *--------------------------------------------------------------*/ | |
3129 | static void pool_dtr(struct dm_target *ti) | |
3130 | { | |
3131 | struct pool_c *pt = ti->private; | |
3132 | ||
3133 | mutex_lock(&dm_thin_pool_table.mutex); | |
3134 | ||
3135 | unbind_control_target(pt->pool, ti); | |
3136 | __pool_dec(pt->pool); | |
3137 | dm_put_device(ti, pt->metadata_dev); | |
3138 | dm_put_device(ti, pt->data_dev); | |
ddd2185e | 3139 | bio_uninit(&pt->flush_bio); |
991d9fa0 JT |
3140 | kfree(pt); |
3141 | ||
3142 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3143 | } | |
3144 | ||
991d9fa0 JT |
3145 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
3146 | struct dm_target *ti) | |
3147 | { | |
3148 | int r; | |
3149 | unsigned argc; | |
3150 | const char *arg_name; | |
3151 | ||
5916a22b | 3152 | static const struct dm_arg _args[] = { |
74aa45c3 | 3153 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
3154 | }; |
3155 | ||
3156 | /* | |
3157 | * No feature arguments supplied. | |
3158 | */ | |
3159 | if (!as->argc) | |
3160 | return 0; | |
3161 | ||
3162 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
3163 | if (r) | |
3164 | return -EINVAL; | |
3165 | ||
3166 | while (argc && !r) { | |
3167 | arg_name = dm_shift_arg(as); | |
3168 | argc--; | |
3169 | ||
e49e5829 | 3170 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 3171 | pf->zero_new_blocks = false; |
e49e5829 JT |
3172 | |
3173 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 3174 | pf->discard_enabled = false; |
e49e5829 JT |
3175 | |
3176 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 3177 | pf->discard_passdown = false; |
991d9fa0 | 3178 | |
e49e5829 JT |
3179 | else if (!strcasecmp(arg_name, "read_only")) |
3180 | pf->mode = PM_READ_ONLY; | |
3181 | ||
787a996c MS |
3182 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
3183 | pf->error_if_no_space = true; | |
3184 | ||
e49e5829 JT |
3185 | else { |
3186 | ti->error = "Unrecognised pool feature requested"; | |
3187 | r = -EINVAL; | |
3188 | break; | |
3189 | } | |
991d9fa0 JT |
3190 | } |
3191 | ||
3192 | return r; | |
3193 | } | |
3194 | ||
ac8c3f3d JT |
3195 | static void metadata_low_callback(void *context) |
3196 | { | |
3197 | struct pool *pool = context; | |
3198 | ||
3199 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
3200 | dm_device_name(pool->pool_md)); | |
3201 | ||
3202 | dm_table_event(pool->ti->table); | |
3203 | } | |
3204 | ||
ddd2185e NT |
3205 | /* |
3206 | * We need to flush the data device **before** committing the metadata. | |
3207 | * | |
3208 | * This ensures that the data blocks of any newly inserted mappings are | |
3209 | * properly written to non-volatile storage and won't be lost in case of a | |
3210 | * crash. | |
3211 | * | |
3212 | * Failure to do so can result in data corruption in the case of internal or | |
3213 | * external snapshots and in the case of newly provisioned blocks, when block | |
3214 | * zeroing is enabled. | |
3215 | */ | |
3216 | static int metadata_pre_commit_callback(void *context) | |
3217 | { | |
3218 | struct pool_c *pt = context; | |
3219 | struct bio *flush_bio = &pt->flush_bio; | |
3220 | ||
3221 | bio_reset(flush_bio); | |
3222 | bio_set_dev(flush_bio, pt->data_dev->bdev); | |
3223 | flush_bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; | |
3224 | ||
3225 | return submit_bio_wait(flush_bio); | |
3226 | } | |
3227 | ||
7d48935e MS |
3228 | static sector_t get_dev_size(struct block_device *bdev) |
3229 | { | |
3230 | return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
3231 | } | |
3232 | ||
3233 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 3234 | { |
7d48935e | 3235 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df JT |
3236 | char buffer[BDEVNAME_SIZE]; |
3237 | ||
7d48935e | 3238 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
b17446df JT |
3239 | DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.", |
3240 | bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS); | |
7d48935e MS |
3241 | } |
3242 | ||
3243 | static sector_t get_metadata_dev_size(struct block_device *bdev) | |
3244 | { | |
3245 | sector_t metadata_dev_size = get_dev_size(bdev); | |
3246 | ||
3247 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS) | |
3248 | metadata_dev_size = THIN_METADATA_MAX_SECTORS; | |
b17446df JT |
3249 | |
3250 | return metadata_dev_size; | |
3251 | } | |
3252 | ||
24347e95 JT |
3253 | static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev) |
3254 | { | |
3255 | sector_t metadata_dev_size = get_metadata_dev_size(bdev); | |
3256 | ||
7d48935e | 3257 | sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE); |
24347e95 JT |
3258 | |
3259 | return metadata_dev_size; | |
3260 | } | |
3261 | ||
ac8c3f3d JT |
3262 | /* |
3263 | * When a metadata threshold is crossed a dm event is triggered, and | |
3264 | * userland should respond by growing the metadata device. We could let | |
3265 | * userland set the threshold, like we do with the data threshold, but I'm | |
3266 | * not sure they know enough to do this well. | |
3267 | */ | |
3268 | static dm_block_t calc_metadata_threshold(struct pool_c *pt) | |
3269 | { | |
3270 | /* | |
3271 | * 4M is ample for all ops with the possible exception of thin | |
3272 | * device deletion which is harmless if it fails (just retry the | |
3273 | * delete after you've grown the device). | |
3274 | */ | |
3275 | dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4; | |
3276 | return min((dm_block_t)1024ULL /* 4M */, quarter); | |
3277 | } | |
3278 | ||
991d9fa0 JT |
3279 | /* |
3280 | * thin-pool <metadata dev> <data dev> | |
3281 | * <data block size (sectors)> | |
3282 | * <low water mark (blocks)> | |
3283 | * [<#feature args> [<arg>]*] | |
3284 | * | |
3285 | * Optional feature arguments are: | |
3286 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
3287 | * ignore_discard: disable discard |
3288 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
3289 | * read_only: Don't allow any changes to be made to the pool metadata. |
3290 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 JT |
3291 | */ |
3292 | static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
3293 | { | |
67e2e2b2 | 3294 | int r, pool_created = 0; |
991d9fa0 JT |
3295 | struct pool_c *pt; |
3296 | struct pool *pool; | |
3297 | struct pool_features pf; | |
3298 | struct dm_arg_set as; | |
3299 | struct dm_dev *data_dev; | |
3300 | unsigned long block_size; | |
3301 | dm_block_t low_water_blocks; | |
3302 | struct dm_dev *metadata_dev; | |
5d0db96d | 3303 | fmode_t metadata_mode; |
991d9fa0 JT |
3304 | |
3305 | /* | |
3306 | * FIXME Remove validation from scope of lock. | |
3307 | */ | |
3308 | mutex_lock(&dm_thin_pool_table.mutex); | |
3309 | ||
3310 | if (argc < 4) { | |
3311 | ti->error = "Invalid argument count"; | |
3312 | r = -EINVAL; | |
3313 | goto out_unlock; | |
3314 | } | |
5d0db96d | 3315 | |
991d9fa0 JT |
3316 | as.argc = argc; |
3317 | as.argv = argv; | |
3318 | ||
70de2cbd JCXF |
3319 | /* make sure metadata and data are different devices */ |
3320 | if (!strcmp(argv[0], argv[1])) { | |
3321 | ti->error = "Error setting metadata or data device"; | |
3322 | r = -EINVAL; | |
3323 | goto out_unlock; | |
3324 | } | |
3325 | ||
5d0db96d JT |
3326 | /* |
3327 | * Set default pool features. | |
3328 | */ | |
3329 | pool_features_init(&pf); | |
3330 | ||
3331 | dm_consume_args(&as, 4); | |
3332 | r = parse_pool_features(&as, &pf, ti); | |
3333 | if (r) | |
3334 | goto out_unlock; | |
3335 | ||
3336 | metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE); | |
3337 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); | |
991d9fa0 JT |
3338 | if (r) { |
3339 | ti->error = "Error opening metadata block device"; | |
3340 | goto out_unlock; | |
3341 | } | |
7d48935e | 3342 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 JT |
3343 | |
3344 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev); | |
3345 | if (r) { | |
3346 | ti->error = "Error getting data device"; | |
3347 | goto out_metadata; | |
3348 | } | |
3349 | ||
3350 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
3351 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
3352 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 3353 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
3354 | ti->error = "Invalid block size"; |
3355 | r = -EINVAL; | |
3356 | goto out; | |
3357 | } | |
3358 | ||
3359 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
3360 | ti->error = "Invalid low water mark"; | |
3361 | r = -EINVAL; | |
3362 | goto out; | |
3363 | } | |
3364 | ||
991d9fa0 JT |
3365 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
3366 | if (!pt) { | |
3367 | r = -ENOMEM; | |
3368 | goto out; | |
3369 | } | |
3370 | ||
3371 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, | |
e49e5829 | 3372 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
3373 | if (IS_ERR(pool)) { |
3374 | r = PTR_ERR(pool); | |
3375 | goto out_free_pt; | |
3376 | } | |
3377 | ||
67e2e2b2 JT |
3378 | /* |
3379 | * 'pool_created' reflects whether this is the first table load. | |
3380 | * Top level discard support is not allowed to be changed after | |
3381 | * initial load. This would require a pool reload to trigger thin | |
3382 | * device changes. | |
3383 | */ | |
3384 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
3385 | ti->error = "Discard support cannot be disabled once enabled"; | |
3386 | r = -EINVAL; | |
3387 | goto out_flags_changed; | |
3388 | } | |
3389 | ||
991d9fa0 JT |
3390 | pt->pool = pool; |
3391 | pt->ti = ti; | |
3392 | pt->metadata_dev = metadata_dev; | |
3393 | pt->data_dev = data_dev; | |
3394 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 3395 | pt->adjusted_pf = pt->requested_pf = pf; |
ddd2185e | 3396 | bio_init(&pt->flush_bio, NULL, 0); |
55a62eef | 3397 | ti->num_flush_bios = 1; |
9bc142dd | 3398 | |
67e2e2b2 JT |
3399 | /* |
3400 | * Only need to enable discards if the pool should pass | |
3401 | * them down to the data device. The thin device's discard | |
3402 | * processing will cause mappings to be removed from the btree. | |
3403 | */ | |
3404 | if (pf.discard_enabled && pf.discard_passdown) { | |
55a62eef | 3405 | ti->num_discard_bios = 1; |
9bc142dd | 3406 | |
67e2e2b2 JT |
3407 | /* |
3408 | * Setting 'discards_supported' circumvents the normal | |
3409 | * stacking of discard limits (this keeps the pool and | |
3410 | * thin devices' discard limits consistent). | |
3411 | */ | |
0ac55489 | 3412 | ti->discards_supported = true; |
67e2e2b2 | 3413 | } |
991d9fa0 JT |
3414 | ti->private = pt; |
3415 | ||
ac8c3f3d JT |
3416 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
3417 | calc_metadata_threshold(pt), | |
3418 | metadata_low_callback, | |
3419 | pool); | |
3420 | if (r) | |
ba30670f | 3421 | goto out_flags_changed; |
ac8c3f3d | 3422 | |
ddd2185e NT |
3423 | dm_pool_register_pre_commit_callback(pt->pool->pmd, |
3424 | metadata_pre_commit_callback, | |
3425 | pt); | |
3426 | ||
991d9fa0 JT |
3427 | pt->callbacks.congested_fn = pool_is_congested; |
3428 | dm_table_add_target_callbacks(ti->table, &pt->callbacks); | |
3429 | ||
3430 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3431 | ||
3432 | return 0; | |
3433 | ||
67e2e2b2 JT |
3434 | out_flags_changed: |
3435 | __pool_dec(pool); | |
991d9fa0 JT |
3436 | out_free_pt: |
3437 | kfree(pt); | |
3438 | out: | |
3439 | dm_put_device(ti, data_dev); | |
3440 | out_metadata: | |
3441 | dm_put_device(ti, metadata_dev); | |
3442 | out_unlock: | |
3443 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3444 | ||
3445 | return r; | |
3446 | } | |
3447 | ||
7de3ee57 | 3448 | static int pool_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
3449 | { |
3450 | int r; | |
3451 | struct pool_c *pt = ti->private; | |
3452 | struct pool *pool = pt->pool; | |
3453 | unsigned long flags; | |
3454 | ||
3455 | /* | |
3456 | * As this is a singleton target, ti->begin is always zero. | |
3457 | */ | |
3458 | spin_lock_irqsave(&pool->lock, flags); | |
74d46992 | 3459 | bio_set_dev(bio, pt->data_dev->bdev); |
991d9fa0 JT |
3460 | r = DM_MAPIO_REMAPPED; |
3461 | spin_unlock_irqrestore(&pool->lock, flags); | |
3462 | ||
3463 | return r; | |
3464 | } | |
3465 | ||
b17446df | 3466 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
3467 | { |
3468 | int r; | |
3469 | struct pool_c *pt = ti->private; | |
3470 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
3471 | sector_t data_size = ti->len; |
3472 | dm_block_t sb_data_size; | |
991d9fa0 | 3473 | |
b17446df | 3474 | *need_commit = false; |
991d9fa0 | 3475 | |
55f2b8bd MS |
3476 | (void) sector_div(data_size, pool->sectors_per_block); |
3477 | ||
991d9fa0 JT |
3478 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
3479 | if (r) { | |
4fa5971a MS |
3480 | DMERR("%s: failed to retrieve data device size", |
3481 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
3482 | return r; |
3483 | } | |
3484 | ||
3485 | if (data_size < sb_data_size) { | |
4fa5971a MS |
3486 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
3487 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3488 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3489 | return -EINVAL; |
3490 | ||
3491 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3492 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3493 | DMERR("%s: unable to grow the data device until repaired.", | |
3494 | dm_device_name(pool->pool_md)); | |
3495 | return 0; | |
3496 | } | |
3497 | ||
6f7f51d4 MS |
3498 | if (sb_data_size) |
3499 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3500 | dm_device_name(pool->pool_md), | |
3501 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3502 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3503 | if (r) { | |
b5330655 | 3504 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3505 | return r; |
3506 | } | |
3507 | ||
b17446df | 3508 | *need_commit = true; |
991d9fa0 JT |
3509 | } |
3510 | ||
3511 | return 0; | |
3512 | } | |
3513 | ||
24347e95 JT |
3514 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3515 | { | |
3516 | int r; | |
3517 | struct pool_c *pt = ti->private; | |
3518 | struct pool *pool = pt->pool; | |
3519 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3520 | ||
3521 | *need_commit = false; | |
3522 | ||
610bba8b | 3523 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3524 | |
3525 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3526 | if (r) { | |
4fa5971a MS |
3527 | DMERR("%s: failed to retrieve metadata device size", |
3528 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3529 | return r; |
3530 | } | |
3531 | ||
3532 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3533 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3534 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3535 | metadata_dev_size, sb_metadata_dev_size); |
3536 | return -EINVAL; | |
3537 | ||
3538 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3539 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3540 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3541 | dm_device_name(pool->pool_md)); | |
3542 | return 0; | |
3543 | } | |
3544 | ||
7d48935e | 3545 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3546 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3547 | dm_device_name(pool->pool_md), | |
3548 | sb_metadata_dev_size, metadata_dev_size); | |
3ab91828 JT |
3549 | |
3550 | if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE) | |
3551 | set_pool_mode(pool, PM_WRITE); | |
3552 | ||
24347e95 JT |
3553 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3554 | if (r) { | |
b5330655 | 3555 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3556 | return r; |
3557 | } | |
3558 | ||
3559 | *need_commit = true; | |
3560 | } | |
3561 | ||
3562 | return 0; | |
3563 | } | |
3564 | ||
b17446df JT |
3565 | /* |
3566 | * Retrieves the number of blocks of the data device from | |
3567 | * the superblock and compares it to the actual device size, | |
3568 | * thus resizing the data device in case it has grown. | |
3569 | * | |
3570 | * This both copes with opening preallocated data devices in the ctr | |
3571 | * being followed by a resume | |
3572 | * -and- | |
3573 | * calling the resume method individually after userspace has | |
3574 | * grown the data device in reaction to a table event. | |
3575 | */ | |
3576 | static int pool_preresume(struct dm_target *ti) | |
3577 | { | |
3578 | int r; | |
24347e95 | 3579 | bool need_commit1, need_commit2; |
b17446df JT |
3580 | struct pool_c *pt = ti->private; |
3581 | struct pool *pool = pt->pool; | |
3582 | ||
3583 | /* | |
3584 | * Take control of the pool object. | |
3585 | */ | |
3586 | r = bind_control_target(pool, ti); | |
3587 | if (r) | |
3588 | return r; | |
3589 | ||
3590 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3591 | if (r) | |
3592 | return r; | |
3593 | ||
24347e95 JT |
3594 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3595 | if (r) | |
3596 | return r; | |
3597 | ||
3598 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3599 | (void) commit(pool); |
b17446df JT |
3600 | |
3601 | return 0; | |
3602 | } | |
3603 | ||
583024d2 MS |
3604 | static void pool_suspend_active_thins(struct pool *pool) |
3605 | { | |
3606 | struct thin_c *tc; | |
3607 | ||
3608 | /* Suspend all active thin devices */ | |
3609 | tc = get_first_thin(pool); | |
3610 | while (tc) { | |
3611 | dm_internal_suspend_noflush(tc->thin_md); | |
3612 | tc = get_next_thin(pool, tc); | |
3613 | } | |
3614 | } | |
3615 | ||
3616 | static void pool_resume_active_thins(struct pool *pool) | |
3617 | { | |
3618 | struct thin_c *tc; | |
3619 | ||
3620 | /* Resume all active thin devices */ | |
3621 | tc = get_first_thin(pool); | |
3622 | while (tc) { | |
3623 | dm_internal_resume(tc->thin_md); | |
3624 | tc = get_next_thin(pool, tc); | |
3625 | } | |
3626 | } | |
3627 | ||
991d9fa0 JT |
3628 | static void pool_resume(struct dm_target *ti) |
3629 | { | |
3630 | struct pool_c *pt = ti->private; | |
3631 | struct pool *pool = pt->pool; | |
3632 | unsigned long flags; | |
3633 | ||
583024d2 MS |
3634 | /* |
3635 | * Must requeue active_thins' bios and then resume | |
3636 | * active_thins _before_ clearing 'suspend' flag. | |
3637 | */ | |
3638 | requeue_bios(pool); | |
3639 | pool_resume_active_thins(pool); | |
3640 | ||
991d9fa0 | 3641 | spin_lock_irqsave(&pool->lock, flags); |
88a6621b | 3642 | pool->low_water_triggered = false; |
80e96c54 | 3643 | pool->suspended = false; |
991d9fa0 | 3644 | spin_unlock_irqrestore(&pool->lock, flags); |
80e96c54 | 3645 | |
905e51b3 | 3646 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3647 | } |
3648 | ||
80e96c54 MS |
3649 | static void pool_presuspend(struct dm_target *ti) |
3650 | { | |
3651 | struct pool_c *pt = ti->private; | |
3652 | struct pool *pool = pt->pool; | |
3653 | unsigned long flags; | |
3654 | ||
3655 | spin_lock_irqsave(&pool->lock, flags); | |
3656 | pool->suspended = true; | |
3657 | spin_unlock_irqrestore(&pool->lock, flags); | |
583024d2 MS |
3658 | |
3659 | pool_suspend_active_thins(pool); | |
80e96c54 MS |
3660 | } |
3661 | ||
3662 | static void pool_presuspend_undo(struct dm_target *ti) | |
3663 | { | |
3664 | struct pool_c *pt = ti->private; | |
3665 | struct pool *pool = pt->pool; | |
3666 | unsigned long flags; | |
3667 | ||
583024d2 MS |
3668 | pool_resume_active_thins(pool); |
3669 | ||
80e96c54 MS |
3670 | spin_lock_irqsave(&pool->lock, flags); |
3671 | pool->suspended = false; | |
3672 | spin_unlock_irqrestore(&pool->lock, flags); | |
3673 | } | |
3674 | ||
991d9fa0 JT |
3675 | static void pool_postsuspend(struct dm_target *ti) |
3676 | { | |
991d9fa0 JT |
3677 | struct pool_c *pt = ti->private; |
3678 | struct pool *pool = pt->pool; | |
3679 | ||
18d03e8c NB |
3680 | cancel_delayed_work_sync(&pool->waker); |
3681 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
991d9fa0 | 3682 | flush_workqueue(pool->wq); |
020cc3b5 | 3683 | (void) commit(pool); |
991d9fa0 JT |
3684 | } |
3685 | ||
3686 | static int check_arg_count(unsigned argc, unsigned args_required) | |
3687 | { | |
3688 | if (argc != args_required) { | |
3689 | DMWARN("Message received with %u arguments instead of %u.", | |
3690 | argc, args_required); | |
3691 | return -EINVAL; | |
3692 | } | |
3693 | ||
3694 | return 0; | |
3695 | } | |
3696 | ||
3697 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3698 | { | |
3699 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3700 | *dev_id <= MAX_DEV_ID) | |
3701 | return 0; | |
3702 | ||
3703 | if (warning) | |
3704 | DMWARN("Message received with invalid device id: %s", arg); | |
3705 | ||
3706 | return -EINVAL; | |
3707 | } | |
3708 | ||
3709 | static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool) | |
3710 | { | |
3711 | dm_thin_id dev_id; | |
3712 | int r; | |
3713 | ||
3714 | r = check_arg_count(argc, 2); | |
3715 | if (r) | |
3716 | return r; | |
3717 | ||
3718 | r = read_dev_id(argv[1], &dev_id, 1); | |
3719 | if (r) | |
3720 | return r; | |
3721 | ||
3722 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3723 | if (r) { | |
3724 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3725 | argv[1]); | |
3726 | return r; | |
3727 | } | |
3728 | ||
3729 | return 0; | |
3730 | } | |
3731 | ||
3732 | static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3733 | { | |
3734 | dm_thin_id dev_id; | |
3735 | dm_thin_id origin_dev_id; | |
3736 | int r; | |
3737 | ||
3738 | r = check_arg_count(argc, 3); | |
3739 | if (r) | |
3740 | return r; | |
3741 | ||
3742 | r = read_dev_id(argv[1], &dev_id, 1); | |
3743 | if (r) | |
3744 | return r; | |
3745 | ||
3746 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3747 | if (r) | |
3748 | return r; | |
3749 | ||
3750 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3751 | if (r) { | |
3752 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3753 | argv[1], argv[2]); | |
3754 | return r; | |
3755 | } | |
3756 | ||
3757 | return 0; | |
3758 | } | |
3759 | ||
3760 | static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool) | |
3761 | { | |
3762 | dm_thin_id dev_id; | |
3763 | int r; | |
3764 | ||
3765 | r = check_arg_count(argc, 2); | |
3766 | if (r) | |
3767 | return r; | |
3768 | ||
3769 | r = read_dev_id(argv[1], &dev_id, 1); | |
3770 | if (r) | |
3771 | return r; | |
3772 | ||
3773 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3774 | if (r) | |
3775 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3776 | ||
3777 | return r; | |
3778 | } | |
3779 | ||
3780 | static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool) | |
3781 | { | |
3782 | dm_thin_id old_id, new_id; | |
3783 | int r; | |
3784 | ||
3785 | r = check_arg_count(argc, 3); | |
3786 | if (r) | |
3787 | return r; | |
3788 | ||
3789 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3790 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3791 | return -EINVAL; | |
3792 | } | |
3793 | ||
3794 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3795 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3796 | return -EINVAL; | |
3797 | } | |
3798 | ||
3799 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3800 | if (r) { | |
3801 | DMWARN("Failed to change transaction id from %s to %s.", | |
3802 | argv[1], argv[2]); | |
3803 | return r; | |
3804 | } | |
3805 | ||
3806 | return 0; | |
3807 | } | |
3808 | ||
cc8394d8 JT |
3809 | static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) |
3810 | { | |
3811 | int r; | |
3812 | ||
3813 | r = check_arg_count(argc, 1); | |
3814 | if (r) | |
3815 | return r; | |
3816 | ||
020cc3b5 | 3817 | (void) commit(pool); |
0d200aef | 3818 | |
cc8394d8 JT |
3819 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3820 | if (r) | |
3821 | DMWARN("reserve_metadata_snap message failed."); | |
3822 | ||
3823 | return r; | |
3824 | } | |
3825 | ||
3826 | static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool) | |
3827 | { | |
3828 | int r; | |
3829 | ||
3830 | r = check_arg_count(argc, 1); | |
3831 | if (r) | |
3832 | return r; | |
3833 | ||
3834 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3835 | if (r) | |
3836 | DMWARN("release_metadata_snap message failed."); | |
3837 | ||
3838 | return r; | |
3839 | } | |
3840 | ||
991d9fa0 JT |
3841 | /* |
3842 | * Messages supported: | |
3843 | * create_thin <dev_id> | |
3844 | * create_snap <dev_id> <origin_id> | |
3845 | * delete <dev_id> | |
991d9fa0 | 3846 | * set_transaction_id <current_trans_id> <new_trans_id> |
cc8394d8 JT |
3847 | * reserve_metadata_snap |
3848 | * release_metadata_snap | |
991d9fa0 | 3849 | */ |
1eb5fa84 MS |
3850 | static int pool_message(struct dm_target *ti, unsigned argc, char **argv, |
3851 | char *result, unsigned maxlen) | |
991d9fa0 JT |
3852 | { |
3853 | int r = -EINVAL; | |
3854 | struct pool_c *pt = ti->private; | |
3855 | struct pool *pool = pt->pool; | |
3856 | ||
3ab91828 | 3857 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) { |
2a7eaea0 JT |
3858 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", |
3859 | dm_device_name(pool->pool_md)); | |
fd467696 | 3860 | return -EOPNOTSUPP; |
2a7eaea0 JT |
3861 | } |
3862 | ||
991d9fa0 JT |
3863 | if (!strcasecmp(argv[0], "create_thin")) |
3864 | r = process_create_thin_mesg(argc, argv, pool); | |
3865 | ||
3866 | else if (!strcasecmp(argv[0], "create_snap")) | |
3867 | r = process_create_snap_mesg(argc, argv, pool); | |
3868 | ||
3869 | else if (!strcasecmp(argv[0], "delete")) | |
3870 | r = process_delete_mesg(argc, argv, pool); | |
3871 | ||
3872 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3873 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3874 | ||
cc8394d8 JT |
3875 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3876 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3877 | ||
3878 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3879 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3880 | ||
991d9fa0 JT |
3881 | else |
3882 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3883 | ||
e49e5829 | 3884 | if (!r) |
020cc3b5 | 3885 | (void) commit(pool); |
991d9fa0 JT |
3886 | |
3887 | return r; | |
3888 | } | |
3889 | ||
e49e5829 JT |
3890 | static void emit_flags(struct pool_features *pf, char *result, |
3891 | unsigned sz, unsigned maxlen) | |
3892 | { | |
3893 | unsigned count = !pf->zero_new_blocks + !pf->discard_enabled + | |
787a996c MS |
3894 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3895 | pf->error_if_no_space; | |
e49e5829 JT |
3896 | DMEMIT("%u ", count); |
3897 | ||
3898 | if (!pf->zero_new_blocks) | |
3899 | DMEMIT("skip_block_zeroing "); | |
3900 | ||
3901 | if (!pf->discard_enabled) | |
3902 | DMEMIT("ignore_discard "); | |
3903 | ||
3904 | if (!pf->discard_passdown) | |
3905 | DMEMIT("no_discard_passdown "); | |
3906 | ||
3907 | if (pf->mode == PM_READ_ONLY) | |
3908 | DMEMIT("read_only "); | |
787a996c MS |
3909 | |
3910 | if (pf->error_if_no_space) | |
3911 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3912 | } |
3913 | ||
991d9fa0 JT |
3914 | /* |
3915 | * Status line is: | |
3916 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3917 | * <used data sectors>/<total data sectors> <held metadata root> | |
e4c78e21 | 3918 | * <pool mode> <discard config> <no space config> <needs_check> |
991d9fa0 | 3919 | */ |
fd7c092e MP |
3920 | static void pool_status(struct dm_target *ti, status_type_t type, |
3921 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 | 3922 | { |
e49e5829 | 3923 | int r; |
991d9fa0 JT |
3924 | unsigned sz = 0; |
3925 | uint64_t transaction_id; | |
3926 | dm_block_t nr_free_blocks_data; | |
3927 | dm_block_t nr_free_blocks_metadata; | |
3928 | dm_block_t nr_blocks_data; | |
3929 | dm_block_t nr_blocks_metadata; | |
3930 | dm_block_t held_root; | |
3ab91828 | 3931 | enum pool_mode mode; |
991d9fa0 JT |
3932 | char buf[BDEVNAME_SIZE]; |
3933 | char buf2[BDEVNAME_SIZE]; | |
3934 | struct pool_c *pt = ti->private; | |
3935 | struct pool *pool = pt->pool; | |
3936 | ||
3937 | switch (type) { | |
3938 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3939 | if (get_pool_mode(pool) == PM_FAIL) { |
3940 | DMEMIT("Fail"); | |
3941 | break; | |
3942 | } | |
3943 | ||
1f4e0ff0 AK |
3944 | /* Commit to ensure statistics aren't out-of-date */ |
3945 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3946 | (void) commit(pool); |
1f4e0ff0 | 3947 | |
fd7c092e MP |
3948 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3949 | if (r) { | |
4fa5971a MS |
3950 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3951 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3952 | goto err; |
3953 | } | |
991d9fa0 | 3954 | |
fd7c092e MP |
3955 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3956 | if (r) { | |
4fa5971a MS |
3957 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3958 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3959 | goto err; |
3960 | } | |
991d9fa0 JT |
3961 | |
3962 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3963 | if (r) { |
4fa5971a MS |
3964 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3965 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3966 | goto err; |
3967 | } | |
991d9fa0 | 3968 | |
fd7c092e MP |
3969 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3970 | if (r) { | |
4fa5971a MS |
3971 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3972 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3973 | goto err; |
3974 | } | |
991d9fa0 JT |
3975 | |
3976 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3977 | if (r) { |
4fa5971a MS |
3978 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3979 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3980 | goto err; |
3981 | } | |
991d9fa0 | 3982 | |
cc8394d8 | 3983 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3984 | if (r) { |
4fa5971a MS |
3985 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3986 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3987 | goto err; |
3988 | } | |
991d9fa0 JT |
3989 | |
3990 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3991 | (unsigned long long)transaction_id, | |
3992 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3993 | (unsigned long long)nr_blocks_metadata, | |
3994 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3995 | (unsigned long long)nr_blocks_data); | |
3996 | ||
3997 | if (held_root) | |
e49e5829 JT |
3998 | DMEMIT("%llu ", held_root); |
3999 | else | |
4000 | DMEMIT("- "); | |
4001 | ||
3ab91828 JT |
4002 | mode = get_pool_mode(pool); |
4003 | if (mode == PM_OUT_OF_DATA_SPACE) | |
3e1a0699 | 4004 | DMEMIT("out_of_data_space "); |
3ab91828 | 4005 | else if (is_read_only_pool_mode(mode)) |
e49e5829 | 4006 | DMEMIT("ro "); |
991d9fa0 | 4007 | else |
e49e5829 JT |
4008 | DMEMIT("rw "); |
4009 | ||
018debea | 4010 | if (!pool->pf.discard_enabled) |
787a996c | 4011 | DMEMIT("ignore_discard "); |
018debea | 4012 | else if (pool->pf.discard_passdown) |
787a996c MS |
4013 | DMEMIT("discard_passdown "); |
4014 | else | |
4015 | DMEMIT("no_discard_passdown "); | |
4016 | ||
4017 | if (pool->pf.error_if_no_space) | |
4018 | DMEMIT("error_if_no_space "); | |
e49e5829 | 4019 | else |
787a996c | 4020 | DMEMIT("queue_if_no_space "); |
991d9fa0 | 4021 | |
e4c78e21 MS |
4022 | if (dm_pool_metadata_needs_check(pool->pmd)) |
4023 | DMEMIT("needs_check "); | |
4024 | else | |
4025 | DMEMIT("- "); | |
4026 | ||
63c8ecb6 AG |
4027 | DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt)); |
4028 | ||
991d9fa0 JT |
4029 | break; |
4030 | ||
4031 | case STATUSTYPE_TABLE: | |
4032 | DMEMIT("%s %s %lu %llu ", | |
4033 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
4034 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
4035 | (unsigned long)pool->sectors_per_block, | |
4036 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 4037 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 JT |
4038 | break; |
4039 | } | |
fd7c092e | 4040 | return; |
991d9fa0 | 4041 | |
fd7c092e MP |
4042 | err: |
4043 | DMEMIT("Error"); | |
991d9fa0 JT |
4044 | } |
4045 | ||
4046 | static int pool_iterate_devices(struct dm_target *ti, | |
4047 | iterate_devices_callout_fn fn, void *data) | |
4048 | { | |
4049 | struct pool_c *pt = ti->private; | |
4050 | ||
4051 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
4052 | } | |
4053 | ||
991d9fa0 JT |
4054 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4055 | { | |
4056 | struct pool_c *pt = ti->private; | |
4057 | struct pool *pool = pt->pool; | |
604ea906 MS |
4058 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
4059 | ||
4060 | /* | |
d200c30e MS |
4061 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
4062 | * to the highest possible power-of-2 factor of pool->sectors_per_block. | |
4063 | * This is especially beneficial when the pool's data device is a RAID | |
4064 | * device that has a full stripe width that matches pool->sectors_per_block | |
4065 | * -- because even though partial RAID stripe-sized IOs will be issued to a | |
4066 | * single RAID stripe; when aggregated they will end on a full RAID stripe | |
4067 | * boundary.. which avoids additional partial RAID stripe writes cascading | |
604ea906 | 4068 | */ |
604ea906 MS |
4069 | if (limits->max_sectors < pool->sectors_per_block) { |
4070 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
4071 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
4072 | limits->max_sectors--; | |
4073 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
4074 | } | |
604ea906 | 4075 | } |
991d9fa0 | 4076 | |
0cc67cd9 MS |
4077 | /* |
4078 | * If the system-determined stacked limits are compatible with the | |
4079 | * pool's blocksize (io_opt is a factor) do not override them. | |
4080 | */ | |
4081 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
4082 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
4083 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
4084 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
4085 | else | |
4086 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
4087 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
4088 | } | |
0424caa1 MS |
4089 | |
4090 | /* | |
4091 | * pt->adjusted_pf is a staging area for the actual features to use. | |
4092 | * They get transferred to the live pool in bind_control_target() | |
4093 | * called from pool_preresume(). | |
4094 | */ | |
b60ab990 MS |
4095 | if (!pt->adjusted_pf.discard_enabled) { |
4096 | /* | |
4097 | * Must explicitly disallow stacking discard limits otherwise the | |
4098 | * block layer will stack them if pool's data device has support. | |
4099 | * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the | |
4100 | * user to see that, so make sure to set all discard limits to 0. | |
4101 | */ | |
4102 | limits->discard_granularity = 0; | |
0424caa1 | 4103 | return; |
b60ab990 | 4104 | } |
0424caa1 MS |
4105 | |
4106 | disable_passdown_if_not_supported(pt); | |
4107 | ||
34fbcf62 JT |
4108 | /* |
4109 | * The pool uses the same discard limits as the underlying data | |
4110 | * device. DM core has already set this up. | |
4111 | */ | |
991d9fa0 JT |
4112 | } |
4113 | ||
4114 | static struct target_type pool_target = { | |
4115 | .name = "thin-pool", | |
4116 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
4117 | DM_TARGET_IMMUTABLE, | |
2af6c070 | 4118 | .version = {1, 21, 0}, |
991d9fa0 JT |
4119 | .module = THIS_MODULE, |
4120 | .ctr = pool_ctr, | |
4121 | .dtr = pool_dtr, | |
4122 | .map = pool_map, | |
80e96c54 MS |
4123 | .presuspend = pool_presuspend, |
4124 | .presuspend_undo = pool_presuspend_undo, | |
991d9fa0 JT |
4125 | .postsuspend = pool_postsuspend, |
4126 | .preresume = pool_preresume, | |
4127 | .resume = pool_resume, | |
4128 | .message = pool_message, | |
4129 | .status = pool_status, | |
991d9fa0 JT |
4130 | .iterate_devices = pool_iterate_devices, |
4131 | .io_hints = pool_io_hints, | |
4132 | }; | |
4133 | ||
4134 | /*---------------------------------------------------------------- | |
4135 | * Thin target methods | |
4136 | *--------------------------------------------------------------*/ | |
b10ebd34 JT |
4137 | static void thin_get(struct thin_c *tc) |
4138 | { | |
22d4c291 | 4139 | refcount_inc(&tc->refcount); |
b10ebd34 JT |
4140 | } |
4141 | ||
4142 | static void thin_put(struct thin_c *tc) | |
4143 | { | |
22d4c291 | 4144 | if (refcount_dec_and_test(&tc->refcount)) |
b10ebd34 JT |
4145 | complete(&tc->can_destroy); |
4146 | } | |
4147 | ||
991d9fa0 JT |
4148 | static void thin_dtr(struct dm_target *ti) |
4149 | { | |
4150 | struct thin_c *tc = ti->private; | |
c140e1c4 MS |
4151 | unsigned long flags; |
4152 | ||
4153 | spin_lock_irqsave(&tc->pool->lock, flags); | |
4154 | list_del_rcu(&tc->list); | |
4155 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
4156 | synchronize_rcu(); | |
991d9fa0 | 4157 | |
17181fb7 MP |
4158 | thin_put(tc); |
4159 | wait_for_completion(&tc->can_destroy); | |
4160 | ||
991d9fa0 JT |
4161 | mutex_lock(&dm_thin_pool_table.mutex); |
4162 | ||
4163 | __pool_dec(tc->pool); | |
4164 | dm_pool_close_thin_device(tc->td); | |
4165 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
4166 | if (tc->origin_dev) |
4167 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
4168 | kfree(tc); |
4169 | ||
4170 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4171 | } | |
4172 | ||
4173 | /* | |
4174 | * Thin target parameters: | |
4175 | * | |
2dd9c257 | 4176 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
4177 | * |
4178 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
4179 | * dev_id: the internal device identifier | |
2dd9c257 | 4180 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
4181 | * |
4182 | * If the pool device has discards disabled, they get disabled for the thin | |
4183 | * device as well. | |
991d9fa0 JT |
4184 | */ |
4185 | static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
4186 | { | |
4187 | int r; | |
4188 | struct thin_c *tc; | |
2dd9c257 | 4189 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 | 4190 | struct mapped_device *pool_md; |
5e3283e2 | 4191 | unsigned long flags; |
991d9fa0 JT |
4192 | |
4193 | mutex_lock(&dm_thin_pool_table.mutex); | |
4194 | ||
2dd9c257 | 4195 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
4196 | ti->error = "Invalid argument count"; |
4197 | r = -EINVAL; | |
4198 | goto out_unlock; | |
4199 | } | |
4200 | ||
4201 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
4202 | if (!tc) { | |
4203 | ti->error = "Out of memory"; | |
4204 | r = -ENOMEM; | |
4205 | goto out_unlock; | |
4206 | } | |
583024d2 | 4207 | tc->thin_md = dm_table_get_md(ti->table); |
c140e1c4 | 4208 | spin_lock_init(&tc->lock); |
a374bb21 | 4209 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
4210 | bio_list_init(&tc->deferred_bio_list); |
4211 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 4212 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 4213 | |
2dd9c257 | 4214 | if (argc == 3) { |
70de2cbd JCXF |
4215 | if (!strcmp(argv[0], argv[2])) { |
4216 | ti->error = "Error setting origin device"; | |
4217 | r = -EINVAL; | |
4218 | goto bad_origin_dev; | |
4219 | } | |
4220 | ||
2dd9c257 JT |
4221 | r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev); |
4222 | if (r) { | |
4223 | ti->error = "Error opening origin device"; | |
4224 | goto bad_origin_dev; | |
4225 | } | |
4226 | tc->origin_dev = origin_dev; | |
4227 | } | |
4228 | ||
991d9fa0 JT |
4229 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
4230 | if (r) { | |
4231 | ti->error = "Error opening pool device"; | |
4232 | goto bad_pool_dev; | |
4233 | } | |
4234 | tc->pool_dev = pool_dev; | |
4235 | ||
4236 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
4237 | ti->error = "Invalid device id"; | |
4238 | r = -EINVAL; | |
4239 | goto bad_common; | |
4240 | } | |
4241 | ||
4242 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
4243 | if (!pool_md) { | |
4244 | ti->error = "Couldn't get pool mapped device"; | |
4245 | r = -EINVAL; | |
4246 | goto bad_common; | |
4247 | } | |
4248 | ||
4249 | tc->pool = __pool_table_lookup(pool_md); | |
4250 | if (!tc->pool) { | |
4251 | ti->error = "Couldn't find pool object"; | |
4252 | r = -EINVAL; | |
4253 | goto bad_pool_lookup; | |
4254 | } | |
4255 | __pool_inc(tc->pool); | |
4256 | ||
e49e5829 JT |
4257 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4258 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 4259 | r = -EINVAL; |
80e96c54 | 4260 | goto bad_pool; |
e49e5829 JT |
4261 | } |
4262 | ||
991d9fa0 JT |
4263 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
4264 | if (r) { | |
4265 | ti->error = "Couldn't open thin internal device"; | |
80e96c54 | 4266 | goto bad_pool; |
991d9fa0 JT |
4267 | } |
4268 | ||
542f9038 MS |
4269 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
4270 | if (r) | |
80e96c54 | 4271 | goto bad; |
542f9038 | 4272 | |
55a62eef | 4273 | ti->num_flush_bios = 1; |
16ad3d10 | 4274 | ti->flush_supported = true; |
30187e1d | 4275 | ti->per_io_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
4276 | |
4277 | /* In case the pool supports discards, pass them on. */ | |
4278 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 4279 | ti->discards_supported = true; |
55a62eef | 4280 | ti->num_discard_bios = 1; |
67e2e2b2 | 4281 | } |
991d9fa0 | 4282 | |
991d9fa0 JT |
4283 | mutex_unlock(&dm_thin_pool_table.mutex); |
4284 | ||
5e3283e2 | 4285 | spin_lock_irqsave(&tc->pool->lock, flags); |
80e96c54 MS |
4286 | if (tc->pool->suspended) { |
4287 | spin_unlock_irqrestore(&tc->pool->lock, flags); | |
4288 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ | |
4289 | ti->error = "Unable to activate thin device while pool is suspended"; | |
4290 | r = -EINVAL; | |
4291 | goto bad; | |
4292 | } | |
22d4c291 | 4293 | refcount_set(&tc->refcount, 1); |
2b94e896 | 4294 | init_completion(&tc->can_destroy); |
c140e1c4 | 4295 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
5e3283e2 | 4296 | spin_unlock_irqrestore(&tc->pool->lock, flags); |
c140e1c4 MS |
4297 | /* |
4298 | * This synchronize_rcu() call is needed here otherwise we risk a | |
4299 | * wake_worker() call finding no bios to process (because the newly | |
4300 | * added tc isn't yet visible). So this reduces latency since we | |
4301 | * aren't then dependent on the periodic commit to wake_worker(). | |
4302 | */ | |
4303 | synchronize_rcu(); | |
4304 | ||
80e96c54 MS |
4305 | dm_put(pool_md); |
4306 | ||
991d9fa0 JT |
4307 | return 0; |
4308 | ||
80e96c54 | 4309 | bad: |
1acacc07 | 4310 | dm_pool_close_thin_device(tc->td); |
80e96c54 | 4311 | bad_pool: |
991d9fa0 JT |
4312 | __pool_dec(tc->pool); |
4313 | bad_pool_lookup: | |
4314 | dm_put(pool_md); | |
4315 | bad_common: | |
4316 | dm_put_device(ti, tc->pool_dev); | |
4317 | bad_pool_dev: | |
2dd9c257 JT |
4318 | if (tc->origin_dev) |
4319 | dm_put_device(ti, tc->origin_dev); | |
4320 | bad_origin_dev: | |
991d9fa0 JT |
4321 | kfree(tc); |
4322 | out_unlock: | |
4323 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4324 | ||
4325 | return r; | |
4326 | } | |
4327 | ||
7de3ee57 | 4328 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 4329 | { |
4f024f37 | 4330 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 4331 | |
7de3ee57 | 4332 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
4333 | } |
4334 | ||
4e4cbee9 CH |
4335 | static int thin_endio(struct dm_target *ti, struct bio *bio, |
4336 | blk_status_t *err) | |
eb2aa48d JT |
4337 | { |
4338 | unsigned long flags; | |
59c3d2c6 | 4339 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 4340 | struct list_head work; |
a24c2569 | 4341 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
4342 | struct pool *pool = h->tc->pool; |
4343 | ||
4344 | if (h->shared_read_entry) { | |
4345 | INIT_LIST_HEAD(&work); | |
44feb387 | 4346 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
4347 | |
4348 | spin_lock_irqsave(&pool->lock, flags); | |
4349 | list_for_each_entry_safe(m, tmp, &work, list) { | |
4350 | list_del(&m->list); | |
50f3c3ef | 4351 | __complete_mapping_preparation(m); |
eb2aa48d JT |
4352 | } |
4353 | spin_unlock_irqrestore(&pool->lock, flags); | |
4354 | } | |
4355 | ||
104655fd JT |
4356 | if (h->all_io_entry) { |
4357 | INIT_LIST_HEAD(&work); | |
44feb387 | 4358 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
4359 | if (!list_empty(&work)) { |
4360 | spin_lock_irqsave(&pool->lock, flags); | |
4361 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 4362 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
4363 | spin_unlock_irqrestore(&pool->lock, flags); |
4364 | wake_worker(pool); | |
4365 | } | |
104655fd JT |
4366 | } |
4367 | ||
34fbcf62 JT |
4368 | if (h->cell) |
4369 | cell_defer_no_holder(h->tc, h->cell); | |
4370 | ||
1be56909 | 4371 | return DM_ENDIO_DONE; |
eb2aa48d JT |
4372 | } |
4373 | ||
738211f7 | 4374 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 4375 | { |
738211f7 JT |
4376 | struct thin_c *tc = ti->private; |
4377 | ||
991d9fa0 | 4378 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
4379 | noflush_work(tc, do_noflush_start); |
4380 | } | |
4381 | ||
4382 | static void thin_postsuspend(struct dm_target *ti) | |
4383 | { | |
4384 | struct thin_c *tc = ti->private; | |
4385 | ||
4386 | /* | |
4387 | * The dm_noflush_suspending flag has been cleared by now, so | |
4388 | * unfortunately we must always run this. | |
4389 | */ | |
4390 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
4391 | } |
4392 | ||
e5aea7b4 JT |
4393 | static int thin_preresume(struct dm_target *ti) |
4394 | { | |
4395 | struct thin_c *tc = ti->private; | |
4396 | ||
4397 | if (tc->origin_dev) | |
4398 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
4399 | ||
4400 | return 0; | |
4401 | } | |
4402 | ||
991d9fa0 JT |
4403 | /* |
4404 | * <nr mapped sectors> <highest mapped sector> | |
4405 | */ | |
fd7c092e MP |
4406 | static void thin_status(struct dm_target *ti, status_type_t type, |
4407 | unsigned status_flags, char *result, unsigned maxlen) | |
991d9fa0 JT |
4408 | { |
4409 | int r; | |
4410 | ssize_t sz = 0; | |
4411 | dm_block_t mapped, highest; | |
4412 | char buf[BDEVNAME_SIZE]; | |
4413 | struct thin_c *tc = ti->private; | |
4414 | ||
e49e5829 JT |
4415 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4416 | DMEMIT("Fail"); | |
fd7c092e | 4417 | return; |
e49e5829 JT |
4418 | } |
4419 | ||
991d9fa0 JT |
4420 | if (!tc->td) |
4421 | DMEMIT("-"); | |
4422 | else { | |
4423 | switch (type) { | |
4424 | case STATUSTYPE_INFO: | |
4425 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
4426 | if (r) { |
4427 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
4428 | goto err; | |
4429 | } | |
991d9fa0 JT |
4430 | |
4431 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
4432 | if (r < 0) { |
4433 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
4434 | goto err; | |
4435 | } | |
991d9fa0 JT |
4436 | |
4437 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
4438 | if (r) | |
4439 | DMEMIT("%llu", ((highest + 1) * | |
4440 | tc->pool->sectors_per_block) - 1); | |
4441 | else | |
4442 | DMEMIT("-"); | |
4443 | break; | |
4444 | ||
4445 | case STATUSTYPE_TABLE: | |
4446 | DMEMIT("%s %lu", | |
4447 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
4448 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
4449 | if (tc->origin_dev) |
4450 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 JT |
4451 | break; |
4452 | } | |
4453 | } | |
4454 | ||
fd7c092e MP |
4455 | return; |
4456 | ||
4457 | err: | |
4458 | DMEMIT("Error"); | |
991d9fa0 JT |
4459 | } |
4460 | ||
4461 | static int thin_iterate_devices(struct dm_target *ti, | |
4462 | iterate_devices_callout_fn fn, void *data) | |
4463 | { | |
55f2b8bd | 4464 | sector_t blocks; |
991d9fa0 | 4465 | struct thin_c *tc = ti->private; |
55f2b8bd | 4466 | struct pool *pool = tc->pool; |
991d9fa0 JT |
4467 | |
4468 | /* | |
4469 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
4470 | * we follow a more convoluted path through to the pool's target. | |
4471 | */ | |
55f2b8bd | 4472 | if (!pool->ti) |
991d9fa0 JT |
4473 | return 0; /* nothing is bound */ |
4474 | ||
55f2b8bd MS |
4475 | blocks = pool->ti->len; |
4476 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 4477 | if (blocks) |
55f2b8bd | 4478 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
4479 | |
4480 | return 0; | |
4481 | } | |
4482 | ||
34fbcf62 JT |
4483 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4484 | { | |
4485 | struct thin_c *tc = ti->private; | |
4486 | struct pool *pool = tc->pool; | |
21607670 | 4487 | |
0fcb04d5 MS |
4488 | if (!pool->pf.discard_enabled) |
4489 | return; | |
34fbcf62 JT |
4490 | |
4491 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
4492 | limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ | |
4493 | } | |
4494 | ||
991d9fa0 JT |
4495 | static struct target_type thin_target = { |
4496 | .name = "thin", | |
2af6c070 | 4497 | .version = {1, 21, 0}, |
991d9fa0 JT |
4498 | .module = THIS_MODULE, |
4499 | .ctr = thin_ctr, | |
4500 | .dtr = thin_dtr, | |
4501 | .map = thin_map, | |
eb2aa48d | 4502 | .end_io = thin_endio, |
e5aea7b4 | 4503 | .preresume = thin_preresume, |
738211f7 | 4504 | .presuspend = thin_presuspend, |
991d9fa0 JT |
4505 | .postsuspend = thin_postsuspend, |
4506 | .status = thin_status, | |
4507 | .iterate_devices = thin_iterate_devices, | |
34fbcf62 | 4508 | .io_hints = thin_io_hints, |
991d9fa0 JT |
4509 | }; |
4510 | ||
4511 | /*----------------------------------------------------------------*/ | |
4512 | ||
4513 | static int __init dm_thin_init(void) | |
4514 | { | |
7e6358d2 | 4515 | int r = -ENOMEM; |
991d9fa0 JT |
4516 | |
4517 | pool_table_init(); | |
4518 | ||
7e6358d2 | 4519 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
4520 | if (!_new_mapping_cache) | |
4521 | return r; | |
4522 | ||
991d9fa0 JT |
4523 | r = dm_register_target(&thin_target); |
4524 | if (r) | |
7e6358d2 | 4525 | goto bad_new_mapping_cache; |
991d9fa0 JT |
4526 | |
4527 | r = dm_register_target(&pool_target); | |
4528 | if (r) | |
7e6358d2 | 4529 | goto bad_thin_target; |
a24c2569 | 4530 | |
a24c2569 MS |
4531 | return 0; |
4532 | ||
7e6358d2 | 4533 | bad_thin_target: |
a24c2569 | 4534 | dm_unregister_target(&thin_target); |
7e6358d2 | 4535 | bad_new_mapping_cache: |
4536 | kmem_cache_destroy(_new_mapping_cache); | |
991d9fa0 JT |
4537 | |
4538 | return r; | |
4539 | } | |
4540 | ||
4541 | static void dm_thin_exit(void) | |
4542 | { | |
4543 | dm_unregister_target(&thin_target); | |
4544 | dm_unregister_target(&pool_target); | |
a24c2569 | 4545 | |
a24c2569 | 4546 | kmem_cache_destroy(_new_mapping_cache); |
d5ffebdd MS |
4547 | |
4548 | pool_table_exit(); | |
991d9fa0 JT |
4549 | } |
4550 | ||
4551 | module_init(dm_thin_init); | |
4552 | module_exit(dm_thin_exit); | |
4553 | ||
80c57893 MS |
4554 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR); |
4555 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); | |
4556 | ||
7cab8bf1 | 4557 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4558 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4559 | MODULE_LICENSE("GPL"); |