Commit | Line | Data |
---|---|---|
3bd94003 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
991d9fa0 | 2 | /* |
e49e5829 | 3 | * Copyright (C) 2011-2012 Red Hat UK. |
991d9fa0 JT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm-thin-metadata.h" | |
742c8fdc | 9 | #include "dm-bio-prison-v1.h" |
1f4e0ff0 | 10 | #include "dm.h" |
991d9fa0 JT |
11 | |
12 | #include <linux/device-mapper.h> | |
13 | #include <linux/dm-io.h> | |
14 | #include <linux/dm-kcopyd.h> | |
0f30af98 | 15 | #include <linux/jiffies.h> |
604ea906 | 16 | #include <linux/log2.h> |
991d9fa0 | 17 | #include <linux/list.h> |
c140e1c4 | 18 | #include <linux/rculist.h> |
991d9fa0 JT |
19 | #include <linux/init.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/slab.h> | |
a822c83e | 22 | #include <linux/vmalloc.h> |
ac4c3f34 | 23 | #include <linux/sort.h> |
67324ea1 | 24 | #include <linux/rbtree.h> |
991d9fa0 JT |
25 | |
26 | #define DM_MSG_PREFIX "thin" | |
27 | ||
28 | /* | |
29 | * Tunable constants | |
30 | */ | |
7768ed33 | 31 | #define ENDIO_HOOK_POOL_SIZE 1024 |
991d9fa0 | 32 | #define MAPPING_POOL_SIZE 1024 |
905e51b3 | 33 | #define COMMIT_PERIOD HZ |
80c57893 MS |
34 | #define NO_SPACE_TIMEOUT_SECS 60 |
35 | ||
86a3238c | 36 | static unsigned int no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS; |
991d9fa0 | 37 | |
df5d2e90 MP |
38 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
39 | "A percentage of time allocated for copy on write"); | |
40 | ||
991d9fa0 JT |
41 | /* |
42 | * The block size of the device holding pool data must be | |
43 | * between 64KB and 1GB. | |
44 | */ | |
45 | #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT) | |
46 | #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT) | |
47 | ||
991d9fa0 JT |
48 | /* |
49 | * Device id is restricted to 24 bits. | |
50 | */ | |
51 | #define MAX_DEV_ID ((1 << 24) - 1) | |
52 | ||
53 | /* | |
54 | * How do we handle breaking sharing of data blocks? | |
55 | * ================================================= | |
56 | * | |
57 | * We use a standard copy-on-write btree to store the mappings for the | |
58 | * devices (note I'm talking about copy-on-write of the metadata here, not | |
59 | * the data). When you take an internal snapshot you clone the root node | |
60 | * of the origin btree. After this there is no concept of an origin or a | |
61 | * snapshot. They are just two device trees that happen to point to the | |
62 | * same data blocks. | |
63 | * | |
64 | * When we get a write in we decide if it's to a shared data block using | |
65 | * some timestamp magic. If it is, we have to break sharing. | |
66 | * | |
67 | * Let's say we write to a shared block in what was the origin. The | |
68 | * steps are: | |
69 | * | |
70 | * i) plug io further to this physical block. (see bio_prison code). | |
71 | * | |
72 | * ii) quiesce any read io to that shared data block. Obviously | |
44feb387 | 73 | * including all devices that share this block. (see dm_deferred_set code) |
991d9fa0 JT |
74 | * |
75 | * iii) copy the data block to a newly allocate block. This step can be | |
76 | * missed out if the io covers the block. (schedule_copy). | |
77 | * | |
78 | * iv) insert the new mapping into the origin's btree | |
fe878f34 | 79 | * (process_prepared_mapping). This act of inserting breaks some |
991d9fa0 JT |
80 | * sharing of btree nodes between the two devices. Breaking sharing only |
81 | * effects the btree of that specific device. Btrees for the other | |
82 | * devices that share the block never change. The btree for the origin | |
83 | * device as it was after the last commit is untouched, ie. we're using | |
84 | * persistent data structures in the functional programming sense. | |
85 | * | |
86 | * v) unplug io to this physical block, including the io that triggered | |
87 | * the breaking of sharing. | |
88 | * | |
89 | * Steps (ii) and (iii) occur in parallel. | |
90 | * | |
91 | * The metadata _doesn't_ need to be committed before the io continues. We | |
92 | * get away with this because the io is always written to a _new_ block. | |
93 | * If there's a crash, then: | |
94 | * | |
95 | * - The origin mapping will point to the old origin block (the shared | |
96 | * one). This will contain the data as it was before the io that triggered | |
97 | * the breaking of sharing came in. | |
98 | * | |
99 | * - The snap mapping still points to the old block. As it would after | |
100 | * the commit. | |
101 | * | |
102 | * The downside of this scheme is the timestamp magic isn't perfect, and | |
103 | * will continue to think that data block in the snapshot device is shared | |
104 | * even after the write to the origin has broken sharing. I suspect data | |
105 | * blocks will typically be shared by many different devices, so we're | |
106 | * breaking sharing n + 1 times, rather than n, where n is the number of | |
107 | * devices that reference this data block. At the moment I think the | |
108 | * benefits far, far outweigh the disadvantages. | |
109 | */ | |
110 | ||
111 | /*----------------------------------------------------------------*/ | |
112 | ||
991d9fa0 JT |
113 | /* |
114 | * Key building. | |
115 | */ | |
34fbcf62 JT |
116 | enum lock_space { |
117 | VIRTUAL, | |
118 | PHYSICAL | |
119 | }; | |
120 | ||
3f8d3f54 | 121 | static bool build_key(struct dm_thin_device *td, enum lock_space ls, |
34fbcf62 | 122 | dm_block_t b, dm_block_t e, struct dm_cell_key *key) |
991d9fa0 | 123 | { |
34fbcf62 | 124 | key->virtual = (ls == VIRTUAL); |
991d9fa0 | 125 | key->dev = dm_thin_dev_id(td); |
5f274d88 | 126 | key->block_begin = b; |
34fbcf62 | 127 | key->block_end = e; |
3f8d3f54 MS |
128 | |
129 | return dm_cell_key_has_valid_range(key); | |
34fbcf62 JT |
130 | } |
131 | ||
132 | static void build_data_key(struct dm_thin_device *td, dm_block_t b, | |
133 | struct dm_cell_key *key) | |
134 | { | |
3f8d3f54 | 135 | (void) build_key(td, PHYSICAL, b, b + 1llu, key); |
991d9fa0 JT |
136 | } |
137 | ||
138 | static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, | |
44feb387 | 139 | struct dm_cell_key *key) |
991d9fa0 | 140 | { |
3f8d3f54 | 141 | (void) build_key(td, VIRTUAL, b, b + 1llu, key); |
991d9fa0 JT |
142 | } |
143 | ||
144 | /*----------------------------------------------------------------*/ | |
145 | ||
7d327fe0 JT |
146 | #define THROTTLE_THRESHOLD (1 * HZ) |
147 | ||
148 | struct throttle { | |
149 | struct rw_semaphore lock; | |
150 | unsigned long threshold; | |
151 | bool throttle_applied; | |
152 | }; | |
153 | ||
154 | static void throttle_init(struct throttle *t) | |
155 | { | |
156 | init_rwsem(&t->lock); | |
157 | t->throttle_applied = false; | |
158 | } | |
159 | ||
160 | static void throttle_work_start(struct throttle *t) | |
161 | { | |
162 | t->threshold = jiffies + THROTTLE_THRESHOLD; | |
163 | } | |
164 | ||
165 | static void throttle_work_update(struct throttle *t) | |
166 | { | |
8ca8b1e1 | 167 | if (!t->throttle_applied && time_is_before_jiffies(t->threshold)) { |
7d327fe0 JT |
168 | down_write(&t->lock); |
169 | t->throttle_applied = true; | |
170 | } | |
171 | } | |
172 | ||
173 | static void throttle_work_complete(struct throttle *t) | |
174 | { | |
175 | if (t->throttle_applied) { | |
176 | t->throttle_applied = false; | |
177 | up_write(&t->lock); | |
178 | } | |
179 | } | |
180 | ||
181 | static void throttle_lock(struct throttle *t) | |
182 | { | |
183 | down_read(&t->lock); | |
184 | } | |
185 | ||
186 | static void throttle_unlock(struct throttle *t) | |
187 | { | |
188 | up_read(&t->lock); | |
189 | } | |
190 | ||
191 | /*----------------------------------------------------------------*/ | |
192 | ||
991d9fa0 JT |
193 | /* |
194 | * A pool device ties together a metadata device and a data device. It | |
195 | * also provides the interface for creating and destroying internal | |
196 | * devices. | |
197 | */ | |
a24c2569 | 198 | struct dm_thin_new_mapping; |
67e2e2b2 | 199 | |
e49e5829 | 200 | /* |
f6c36758 | 201 | * The pool runs in various modes. Ordered in degraded order for comparisons. |
e49e5829 JT |
202 | */ |
203 | enum pool_mode { | |
204 | PM_WRITE, /* metadata may be changed */ | |
3e1a0699 | 205 | PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */ |
3ab91828 JT |
206 | |
207 | /* | |
208 | * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY. | |
209 | */ | |
210 | PM_OUT_OF_METADATA_SPACE, | |
e49e5829 | 211 | PM_READ_ONLY, /* metadata may not be changed */ |
3ab91828 | 212 | |
e49e5829 JT |
213 | PM_FAIL, /* all I/O fails */ |
214 | }; | |
215 | ||
67e2e2b2 | 216 | struct pool_features { |
e49e5829 JT |
217 | enum pool_mode mode; |
218 | ||
9bc142dd MS |
219 | bool zero_new_blocks:1; |
220 | bool discard_enabled:1; | |
221 | bool discard_passdown:1; | |
787a996c | 222 | bool error_if_no_space:1; |
67e2e2b2 JT |
223 | }; |
224 | ||
e49e5829 JT |
225 | struct thin_c; |
226 | typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio); | |
a374bb21 | 227 | typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell); |
e49e5829 JT |
228 | typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m); |
229 | ||
ac4c3f34 JT |
230 | #define CELL_SORT_ARRAY_SIZE 8192 |
231 | ||
991d9fa0 JT |
232 | struct pool { |
233 | struct list_head list; | |
234 | struct dm_target *ti; /* Only set if a pool target is bound */ | |
235 | ||
236 | struct mapped_device *pool_md; | |
873937e7 | 237 | struct block_device *data_dev; |
991d9fa0 JT |
238 | struct block_device *md_dev; |
239 | struct dm_pool_metadata *pmd; | |
240 | ||
991d9fa0 | 241 | dm_block_t low_water_blocks; |
55f2b8bd | 242 | uint32_t sectors_per_block; |
f9a8e0cd | 243 | int sectors_per_block_shift; |
991d9fa0 | 244 | |
67e2e2b2 | 245 | struct pool_features pf; |
88a6621b | 246 | bool low_water_triggered:1; /* A dm event has been sent */ |
80e96c54 | 247 | bool suspended:1; |
c3667cc6 | 248 | bool out_of_data_space:1; |
991d9fa0 | 249 | |
44feb387 | 250 | struct dm_bio_prison *prison; |
991d9fa0 JT |
251 | struct dm_kcopyd_client *copier; |
252 | ||
72d711c8 | 253 | struct work_struct worker; |
991d9fa0 | 254 | struct workqueue_struct *wq; |
7d327fe0 | 255 | struct throttle throttle; |
905e51b3 | 256 | struct delayed_work waker; |
85ad643b | 257 | struct delayed_work no_space_timeout; |
991d9fa0 | 258 | |
905e51b3 | 259 | unsigned long last_commit_jiffies; |
86a3238c | 260 | unsigned int ref_count; |
991d9fa0 JT |
261 | |
262 | spinlock_t lock; | |
991d9fa0 | 263 | struct bio_list deferred_flush_bios; |
4ae280b4 | 264 | struct bio_list deferred_flush_completions; |
991d9fa0 | 265 | struct list_head prepared_mappings; |
104655fd | 266 | struct list_head prepared_discards; |
2a0fbffb | 267 | struct list_head prepared_discards_pt2; |
c140e1c4 | 268 | struct list_head active_thins; |
991d9fa0 | 269 | |
44feb387 MS |
270 | struct dm_deferred_set *shared_read_ds; |
271 | struct dm_deferred_set *all_io_ds; | |
991d9fa0 | 272 | |
a24c2569 | 273 | struct dm_thin_new_mapping *next_mapping; |
e49e5829 JT |
274 | |
275 | process_bio_fn process_bio; | |
276 | process_bio_fn process_discard; | |
277 | ||
a374bb21 JT |
278 | process_cell_fn process_cell; |
279 | process_cell_fn process_discard_cell; | |
280 | ||
e49e5829 JT |
281 | process_mapping_fn process_prepared_mapping; |
282 | process_mapping_fn process_prepared_discard; | |
2a0fbffb | 283 | process_mapping_fn process_prepared_discard_pt2; |
ac4c3f34 | 284 | |
a822c83e | 285 | struct dm_bio_prison_cell **cell_sort_array; |
72d711c8 MS |
286 | |
287 | mempool_t mapping_pool; | |
991d9fa0 JT |
288 | }; |
289 | ||
b5330655 | 290 | static void metadata_operation_failed(struct pool *pool, const char *op, int r); |
e49e5829 | 291 | |
f6c36758 MS |
292 | static enum pool_mode get_pool_mode(struct pool *pool) |
293 | { | |
294 | return pool->pf.mode; | |
295 | } | |
296 | ||
297 | static void notify_of_pool_mode_change(struct pool *pool) | |
298 | { | |
774f13ac | 299 | static const char *descs[] = { |
f6c36758 MS |
300 | "write", |
301 | "out-of-data-space", | |
302 | "read-only", | |
303 | "read-only", | |
304 | "fail" | |
305 | }; | |
306 | const char *extra_desc = NULL; | |
307 | enum pool_mode mode = get_pool_mode(pool); | |
308 | ||
309 | if (mode == PM_OUT_OF_DATA_SPACE) { | |
310 | if (!pool->pf.error_if_no_space) | |
311 | extra_desc = " (queue IO)"; | |
312 | else | |
313 | extra_desc = " (error IO)"; | |
314 | } | |
315 | ||
316 | dm_table_event(pool->ti->table); | |
317 | DMINFO("%s: switching pool to %s%s mode", | |
318 | dm_device_name(pool->pool_md), | |
319 | descs[(int)mode], extra_desc ? : ""); | |
320 | } | |
321 | ||
991d9fa0 JT |
322 | /* |
323 | * Target context for a pool. | |
324 | */ | |
325 | struct pool_c { | |
326 | struct dm_target *ti; | |
327 | struct pool *pool; | |
328 | struct dm_dev *data_dev; | |
329 | struct dm_dev *metadata_dev; | |
991d9fa0 JT |
330 | |
331 | dm_block_t low_water_blocks; | |
0424caa1 MS |
332 | struct pool_features requested_pf; /* Features requested during table load */ |
333 | struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */ | |
991d9fa0 JT |
334 | }; |
335 | ||
336 | /* | |
337 | * Target context for a thin. | |
338 | */ | |
339 | struct thin_c { | |
c140e1c4 | 340 | struct list_head list; |
991d9fa0 | 341 | struct dm_dev *pool_dev; |
2dd9c257 | 342 | struct dm_dev *origin_dev; |
e5aea7b4 | 343 | sector_t origin_size; |
991d9fa0 JT |
344 | dm_thin_id dev_id; |
345 | ||
346 | struct pool *pool; | |
347 | struct dm_thin_device *td; | |
583024d2 MS |
348 | struct mapped_device *thin_md; |
349 | ||
738211f7 | 350 | bool requeue_mode:1; |
c140e1c4 | 351 | spinlock_t lock; |
a374bb21 | 352 | struct list_head deferred_cells; |
c140e1c4 MS |
353 | struct bio_list deferred_bio_list; |
354 | struct bio_list retry_on_resume_list; | |
67324ea1 | 355 | struct rb_root sort_bio_list; /* sorted list of deferred bios */ |
b10ebd34 JT |
356 | |
357 | /* | |
358 | * Ensures the thin is not destroyed until the worker has finished | |
359 | * iterating the active_thins list. | |
360 | */ | |
22d4c291 | 361 | refcount_t refcount; |
b10ebd34 | 362 | struct completion can_destroy; |
991d9fa0 JT |
363 | }; |
364 | ||
365 | /*----------------------------------------------------------------*/ | |
366 | ||
34fbcf62 JT |
367 | static bool block_size_is_power_of_two(struct pool *pool) |
368 | { | |
369 | return pool->sectors_per_block_shift >= 0; | |
370 | } | |
371 | ||
372 | static sector_t block_to_sectors(struct pool *pool, dm_block_t b) | |
373 | { | |
374 | return block_size_is_power_of_two(pool) ? | |
375 | (b << pool->sectors_per_block_shift) : | |
376 | (b * pool->sectors_per_block); | |
377 | } | |
378 | ||
202bae52 JT |
379 | /*----------------------------------------------------------------*/ |
380 | ||
381 | struct discard_op { | |
382 | struct thin_c *tc; | |
383 | struct blk_plug plug; | |
384 | struct bio *parent_bio; | |
385 | struct bio *bio; | |
386 | }; | |
387 | ||
388 | static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent) | |
389 | { | |
390 | BUG_ON(!parent); | |
391 | ||
392 | op->tc = tc; | |
393 | blk_start_plug(&op->plug); | |
394 | op->parent_bio = parent; | |
395 | op->bio = NULL; | |
396 | } | |
397 | ||
398 | static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e) | |
34fbcf62 | 399 | { |
202bae52 | 400 | struct thin_c *tc = op->tc; |
34fbcf62 JT |
401 | sector_t s = block_to_sectors(tc->pool, data_b); |
402 | sector_t len = block_to_sectors(tc->pool, data_e - data_b); | |
3dba53a9 | 403 | |
44abff2c CH |
404 | return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, GFP_NOWAIT, |
405 | &op->bio); | |
202bae52 JT |
406 | } |
407 | ||
408 | static void end_discard(struct discard_op *op, int r) | |
409 | { | |
410 | if (op->bio) { | |
411 | /* | |
412 | * Even if one of the calls to issue_discard failed, we | |
413 | * need to wait for the chain to complete. | |
414 | */ | |
415 | bio_chain(op->bio, op->parent_bio); | |
c34b7ac6 | 416 | op->bio->bi_opf = REQ_OP_DISCARD; |
4e49ea4a | 417 | submit_bio(op->bio); |
3dba53a9 | 418 | } |
34fbcf62 | 419 | |
202bae52 JT |
420 | blk_finish_plug(&op->plug); |
421 | ||
422 | /* | |
423 | * Even if r is set, there could be sub discards in flight that we | |
424 | * need to wait for. | |
425 | */ | |
4e4cbee9 CH |
426 | if (r && !op->parent_bio->bi_status) |
427 | op->parent_bio->bi_status = errno_to_blk_status(r); | |
202bae52 | 428 | bio_endio(op->parent_bio); |
34fbcf62 JT |
429 | } |
430 | ||
431 | /*----------------------------------------------------------------*/ | |
432 | ||
025b9685 JT |
433 | /* |
434 | * wake_worker() is used when new work is queued and when pool_resume is | |
435 | * ready to continue deferred IO processing. | |
436 | */ | |
437 | static void wake_worker(struct pool *pool) | |
438 | { | |
439 | queue_work(pool->wq, &pool->worker); | |
440 | } | |
441 | ||
442 | /*----------------------------------------------------------------*/ | |
443 | ||
6beca5eb JT |
444 | static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio, |
445 | struct dm_bio_prison_cell **cell_result) | |
446 | { | |
447 | int r; | |
448 | struct dm_bio_prison_cell *cell_prealloc; | |
449 | ||
450 | /* | |
451 | * Allocate a cell from the prison's mempool. | |
452 | * This might block but it can't fail. | |
453 | */ | |
454 | cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO); | |
455 | ||
456 | r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result); | |
457 | if (r) | |
458 | /* | |
459 | * We reused an old cell; we can get rid of | |
460 | * the new one. | |
461 | */ | |
462 | dm_bio_prison_free_cell(pool->prison, cell_prealloc); | |
463 | ||
464 | return r; | |
465 | } | |
466 | ||
467 | static void cell_release(struct pool *pool, | |
468 | struct dm_bio_prison_cell *cell, | |
469 | struct bio_list *bios) | |
470 | { | |
471 | dm_cell_release(pool->prison, cell, bios); | |
472 | dm_bio_prison_free_cell(pool->prison, cell); | |
473 | } | |
474 | ||
2d759a46 JT |
475 | static void cell_visit_release(struct pool *pool, |
476 | void (*fn)(void *, struct dm_bio_prison_cell *), | |
477 | void *context, | |
478 | struct dm_bio_prison_cell *cell) | |
479 | { | |
480 | dm_cell_visit_release(pool->prison, fn, context, cell); | |
481 | dm_bio_prison_free_cell(pool->prison, cell); | |
482 | } | |
483 | ||
6beca5eb JT |
484 | static void cell_release_no_holder(struct pool *pool, |
485 | struct dm_bio_prison_cell *cell, | |
486 | struct bio_list *bios) | |
487 | { | |
488 | dm_cell_release_no_holder(pool->prison, cell, bios); | |
489 | dm_bio_prison_free_cell(pool->prison, cell); | |
490 | } | |
491 | ||
af91805a | 492 | static void cell_error_with_code(struct pool *pool, |
4e4cbee9 | 493 | struct dm_bio_prison_cell *cell, blk_status_t error_code) |
6beca5eb | 494 | { |
af91805a | 495 | dm_cell_error(pool->prison, cell, error_code); |
6beca5eb JT |
496 | dm_bio_prison_free_cell(pool->prison, cell); |
497 | } | |
498 | ||
4e4cbee9 | 499 | static blk_status_t get_pool_io_error_code(struct pool *pool) |
c3667cc6 | 500 | { |
4e4cbee9 | 501 | return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR; |
c3667cc6 MS |
502 | } |
503 | ||
af91805a MS |
504 | static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell) |
505 | { | |
4e4cbee9 | 506 | cell_error_with_code(pool, cell, get_pool_io_error_code(pool)); |
af91805a MS |
507 | } |
508 | ||
a374bb21 JT |
509 | static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell) |
510 | { | |
511 | cell_error_with_code(pool, cell, 0); | |
512 | } | |
513 | ||
514 | static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell) | |
515 | { | |
4e4cbee9 | 516 | cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE); |
a374bb21 JT |
517 | } |
518 | ||
6beca5eb JT |
519 | /*----------------------------------------------------------------*/ |
520 | ||
991d9fa0 JT |
521 | /* |
522 | * A global list of pools that uses a struct mapped_device as a key. | |
523 | */ | |
524 | static struct dm_thin_pool_table { | |
525 | struct mutex mutex; | |
526 | struct list_head pools; | |
527 | } dm_thin_pool_table; | |
528 | ||
529 | static void pool_table_init(void) | |
530 | { | |
531 | mutex_init(&dm_thin_pool_table.mutex); | |
532 | INIT_LIST_HEAD(&dm_thin_pool_table.pools); | |
533 | } | |
534 | ||
d5ffebdd MS |
535 | static void pool_table_exit(void) |
536 | { | |
537 | mutex_destroy(&dm_thin_pool_table.mutex); | |
538 | } | |
539 | ||
991d9fa0 JT |
540 | static void __pool_table_insert(struct pool *pool) |
541 | { | |
542 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
543 | list_add(&pool->list, &dm_thin_pool_table.pools); | |
544 | } | |
545 | ||
546 | static void __pool_table_remove(struct pool *pool) | |
547 | { | |
548 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
549 | list_del(&pool->list); | |
550 | } | |
551 | ||
552 | static struct pool *__pool_table_lookup(struct mapped_device *md) | |
553 | { | |
554 | struct pool *pool = NULL, *tmp; | |
555 | ||
556 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
557 | ||
558 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
559 | if (tmp->pool_md == md) { | |
560 | pool = tmp; | |
561 | break; | |
562 | } | |
563 | } | |
564 | ||
565 | return pool; | |
566 | } | |
567 | ||
568 | static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev) | |
569 | { | |
570 | struct pool *pool = NULL, *tmp; | |
571 | ||
572 | BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex)); | |
573 | ||
574 | list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) { | |
575 | if (tmp->md_dev == md_dev) { | |
576 | pool = tmp; | |
577 | break; | |
578 | } | |
579 | } | |
580 | ||
581 | return pool; | |
582 | } | |
583 | ||
584 | /*----------------------------------------------------------------*/ | |
585 | ||
a24c2569 | 586 | struct dm_thin_endio_hook { |
eb2aa48d | 587 | struct thin_c *tc; |
44feb387 MS |
588 | struct dm_deferred_entry *shared_read_entry; |
589 | struct dm_deferred_entry *all_io_entry; | |
a24c2569 | 590 | struct dm_thin_new_mapping *overwrite_mapping; |
67324ea1 | 591 | struct rb_node rb_node; |
34fbcf62 | 592 | struct dm_bio_prison_cell *cell; |
eb2aa48d JT |
593 | }; |
594 | ||
42d6a8ce MS |
595 | static void __merge_bio_list(struct bio_list *bios, struct bio_list *master) |
596 | { | |
597 | bio_list_merge(bios, master); | |
598 | bio_list_init(master); | |
599 | } | |
600 | ||
4e4cbee9 | 601 | static void error_bio_list(struct bio_list *bios, blk_status_t error) |
991d9fa0 JT |
602 | { |
603 | struct bio *bio; | |
42d6a8ce | 604 | |
4246a0b6 | 605 | while ((bio = bio_list_pop(bios))) { |
4e4cbee9 | 606 | bio->bi_status = error; |
4246a0b6 CH |
607 | bio_endio(bio); |
608 | } | |
42d6a8ce MS |
609 | } |
610 | ||
4e4cbee9 CH |
611 | static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, |
612 | blk_status_t error) | |
42d6a8ce | 613 | { |
991d9fa0 JT |
614 | struct bio_list bios; |
615 | ||
616 | bio_list_init(&bios); | |
18adc577 | 617 | |
8e0c9dac | 618 | spin_lock_irq(&tc->lock); |
42d6a8ce | 619 | __merge_bio_list(&bios, master); |
8e0c9dac | 620 | spin_unlock_irq(&tc->lock); |
991d9fa0 | 621 | |
42d6a8ce | 622 | error_bio_list(&bios, error); |
991d9fa0 JT |
623 | } |
624 | ||
a374bb21 JT |
625 | static void requeue_deferred_cells(struct thin_c *tc) |
626 | { | |
627 | struct pool *pool = tc->pool; | |
a374bb21 JT |
628 | struct list_head cells; |
629 | struct dm_bio_prison_cell *cell, *tmp; | |
630 | ||
631 | INIT_LIST_HEAD(&cells); | |
632 | ||
8e0c9dac | 633 | spin_lock_irq(&tc->lock); |
a374bb21 | 634 | list_splice_init(&tc->deferred_cells, &cells); |
8e0c9dac | 635 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
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 JT |
643 | struct bio_list bios; |
644 | ||
645 | bio_list_init(&bios); | |
646 | ||
8e0c9dac | 647 | spin_lock_irq(&tc->lock); |
42d6a8ce MS |
648 | __merge_bio_list(&bios, &tc->deferred_bio_list); |
649 | __merge_bio_list(&bios, &tc->retry_on_resume_list); | |
8e0c9dac | 650 | spin_unlock_irq(&tc->lock); |
3e1a0699 | 651 | |
4e4cbee9 | 652 | error_bio_list(&bios, BLK_STS_DM_REQUEUE); |
42d6a8ce | 653 | requeue_deferred_cells(tc); |
3e1a0699 JT |
654 | } |
655 | ||
4e4cbee9 | 656 | static void error_retry_list_with_code(struct pool *pool, blk_status_t error) |
c140e1c4 MS |
657 | { |
658 | struct thin_c *tc; | |
659 | ||
660 | rcu_read_lock(); | |
661 | list_for_each_entry_rcu(tc, &pool->active_thins, list) | |
0a927c2f | 662 | error_thin_bio_list(tc, &tc->retry_on_resume_list, error); |
c140e1c4 MS |
663 | rcu_read_unlock(); |
664 | } | |
665 | ||
0a927c2f MS |
666 | static void error_retry_list(struct pool *pool) |
667 | { | |
4e4cbee9 | 668 | error_retry_list_with_code(pool, get_pool_io_error_code(pool)); |
0a927c2f MS |
669 | } |
670 | ||
991d9fa0 JT |
671 | /* |
672 | * This section of code contains the logic for processing a thin device's IO. | |
673 | * Much of the code depends on pool object resources (lists, workqueues, etc) | |
674 | * but most is exclusively called from the thin target rather than the thin-pool | |
675 | * target. | |
676 | */ | |
677 | ||
678 | static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio) | |
679 | { | |
58f77a21 | 680 | struct pool *pool = tc->pool; |
4f024f37 | 681 | sector_t block_nr = bio->bi_iter.bi_sector; |
55f2b8bd | 682 | |
58f77a21 MS |
683 | if (block_size_is_power_of_two(pool)) |
684 | block_nr >>= pool->sectors_per_block_shift; | |
f9a8e0cd | 685 | else |
58f77a21 | 686 | (void) sector_div(block_nr, pool->sectors_per_block); |
55f2b8bd MS |
687 | |
688 | return block_nr; | |
991d9fa0 JT |
689 | } |
690 | ||
34fbcf62 JT |
691 | /* |
692 | * Returns the _complete_ blocks that this bio covers. | |
693 | */ | |
694 | static void get_bio_block_range(struct thin_c *tc, struct bio *bio, | |
695 | dm_block_t *begin, dm_block_t *end) | |
696 | { | |
697 | struct pool *pool = tc->pool; | |
698 | sector_t b = bio->bi_iter.bi_sector; | |
699 | sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
700 | ||
701 | b += pool->sectors_per_block - 1ull; /* so we round up */ | |
702 | ||
703 | if (block_size_is_power_of_two(pool)) { | |
704 | b >>= pool->sectors_per_block_shift; | |
705 | e >>= pool->sectors_per_block_shift; | |
706 | } else { | |
707 | (void) sector_div(b, pool->sectors_per_block); | |
708 | (void) sector_div(e, pool->sectors_per_block); | |
709 | } | |
710 | ||
711 | if (e < b) | |
712 | /* Can happen if the bio is within a single block. */ | |
713 | e = b; | |
714 | ||
715 | *begin = b; | |
716 | *end = e; | |
717 | } | |
718 | ||
991d9fa0 JT |
719 | static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block) |
720 | { | |
721 | struct pool *pool = tc->pool; | |
4f024f37 | 722 | sector_t bi_sector = bio->bi_iter.bi_sector; |
991d9fa0 | 723 | |
74d46992 | 724 | bio_set_dev(bio, tc->pool_dev->bdev); |
58f77a21 | 725 | if (block_size_is_power_of_two(pool)) |
4f024f37 KO |
726 | bio->bi_iter.bi_sector = |
727 | (block << pool->sectors_per_block_shift) | | |
728 | (bi_sector & (pool->sectors_per_block - 1)); | |
58f77a21 | 729 | else |
4f024f37 | 730 | bio->bi_iter.bi_sector = (block * pool->sectors_per_block) + |
58f77a21 | 731 | sector_div(bi_sector, pool->sectors_per_block); |
991d9fa0 JT |
732 | } |
733 | ||
2dd9c257 JT |
734 | static void remap_to_origin(struct thin_c *tc, struct bio *bio) |
735 | { | |
74d46992 | 736 | bio_set_dev(bio, tc->origin_dev->bdev); |
2dd9c257 JT |
737 | } |
738 | ||
4afdd680 JT |
739 | static int bio_triggers_commit(struct thin_c *tc, struct bio *bio) |
740 | { | |
f73f44eb | 741 | return op_is_flush(bio->bi_opf) && |
4afdd680 JT |
742 | dm_thin_changed_this_transaction(tc->td); |
743 | } | |
744 | ||
e8088073 JT |
745 | static void inc_all_io_entry(struct pool *pool, struct bio *bio) |
746 | { | |
747 | struct dm_thin_endio_hook *h; | |
748 | ||
e6047149 | 749 | if (bio_op(bio) == REQ_OP_DISCARD) |
e8088073 JT |
750 | return; |
751 | ||
59c3d2c6 | 752 | h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
e8088073 JT |
753 | h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds); |
754 | } | |
755 | ||
2dd9c257 | 756 | static void issue(struct thin_c *tc, struct bio *bio) |
991d9fa0 JT |
757 | { |
758 | struct pool *pool = tc->pool; | |
991d9fa0 | 759 | |
e49e5829 | 760 | if (!bio_triggers_commit(tc, bio)) { |
b7f8dff0 | 761 | dm_submit_bio_remap(bio, NULL); |
e49e5829 JT |
762 | return; |
763 | } | |
764 | ||
991d9fa0 | 765 | /* |
e49e5829 JT |
766 | * Complete bio with an error if earlier I/O caused changes to |
767 | * the metadata that can't be committed e.g, due to I/O errors | |
768 | * on the metadata device. | |
991d9fa0 | 769 | */ |
e49e5829 JT |
770 | if (dm_thin_aborted_changes(tc->td)) { |
771 | bio_io_error(bio); | |
772 | return; | |
773 | } | |
774 | ||
775 | /* | |
776 | * Batch together any bios that trigger commits and then issue a | |
777 | * single commit for them in process_deferred_bios(). | |
778 | */ | |
8e0c9dac | 779 | spin_lock_irq(&pool->lock); |
e49e5829 | 780 | bio_list_add(&pool->deferred_flush_bios, bio); |
8e0c9dac | 781 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
782 | } |
783 | ||
2dd9c257 JT |
784 | static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio) |
785 | { | |
786 | remap_to_origin(tc, bio); | |
787 | issue(tc, bio); | |
788 | } | |
789 | ||
790 | static void remap_and_issue(struct thin_c *tc, struct bio *bio, | |
791 | dm_block_t block) | |
792 | { | |
793 | remap(tc, bio, block); | |
794 | issue(tc, bio); | |
795 | } | |
796 | ||
991d9fa0 JT |
797 | /*----------------------------------------------------------------*/ |
798 | ||
799 | /* | |
800 | * Bio endio functions. | |
801 | */ | |
a24c2569 | 802 | struct dm_thin_new_mapping { |
991d9fa0 JT |
803 | struct list_head list; |
804 | ||
7f214665 | 805 | bool pass_discard:1; |
34fbcf62 | 806 | bool maybe_shared:1; |
991d9fa0 | 807 | |
50f3c3ef JT |
808 | /* |
809 | * Track quiescing, copying and zeroing preparation actions. When this | |
810 | * counter hits zero the block is prepared and can be inserted into the | |
811 | * btree. | |
812 | */ | |
813 | atomic_t prepare_actions; | |
814 | ||
4e4cbee9 | 815 | blk_status_t status; |
991d9fa0 | 816 | struct thin_c *tc; |
34fbcf62 | 817 | dm_block_t virt_begin, virt_end; |
991d9fa0 | 818 | dm_block_t data_block; |
34fbcf62 | 819 | struct dm_bio_prison_cell *cell; |
991d9fa0 JT |
820 | |
821 | /* | |
822 | * If the bio covers the whole area of a block then we can avoid | |
823 | * zeroing or copying. Instead this bio is hooked. The bio will | |
824 | * still be in the cell, so care has to be taken to avoid issuing | |
825 | * the bio twice. | |
826 | */ | |
827 | struct bio *bio; | |
828 | bio_end_io_t *saved_bi_end_io; | |
829 | }; | |
830 | ||
50f3c3ef | 831 | static void __complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
832 | { |
833 | struct pool *pool = m->tc->pool; | |
834 | ||
50f3c3ef | 835 | if (atomic_dec_and_test(&m->prepare_actions)) { |
daec338b | 836 | list_add_tail(&m->list, &pool->prepared_mappings); |
991d9fa0 JT |
837 | wake_worker(pool); |
838 | } | |
839 | } | |
840 | ||
e5aea7b4 | 841 | static void complete_mapping_preparation(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
842 | { |
843 | unsigned long flags; | |
991d9fa0 JT |
844 | struct pool *pool = m->tc->pool; |
845 | ||
991d9fa0 | 846 | spin_lock_irqsave(&pool->lock, flags); |
50f3c3ef | 847 | __complete_mapping_preparation(m); |
991d9fa0 JT |
848 | spin_unlock_irqrestore(&pool->lock, flags); |
849 | } | |
850 | ||
e5aea7b4 JT |
851 | static void copy_complete(int read_err, unsigned long write_err, void *context) |
852 | { | |
853 | struct dm_thin_new_mapping *m = context; | |
854 | ||
4e4cbee9 | 855 | m->status = read_err || write_err ? BLK_STS_IOERR : 0; |
e5aea7b4 JT |
856 | complete_mapping_preparation(m); |
857 | } | |
858 | ||
4246a0b6 | 859 | static void overwrite_endio(struct bio *bio) |
991d9fa0 | 860 | { |
59c3d2c6 | 861 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
a24c2569 | 862 | struct dm_thin_new_mapping *m = h->overwrite_mapping; |
991d9fa0 | 863 | |
8b908f8e MS |
864 | bio->bi_end_io = m->saved_bi_end_io; |
865 | ||
4e4cbee9 | 866 | m->status = bio->bi_status; |
e5aea7b4 | 867 | complete_mapping_preparation(m); |
991d9fa0 JT |
868 | } |
869 | ||
991d9fa0 JT |
870 | /*----------------------------------------------------------------*/ |
871 | ||
872 | /* | |
873 | * Workqueue. | |
874 | */ | |
875 | ||
876 | /* | |
877 | * Prepared mapping jobs. | |
878 | */ | |
879 | ||
880 | /* | |
2d759a46 JT |
881 | * This sends the bios in the cell, except the original holder, back |
882 | * to the deferred_bios list. | |
991d9fa0 | 883 | */ |
f286ba0e | 884 | static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 | 885 | { |
991d9fa0 JT |
886 | struct pool *pool = tc->pool; |
887 | unsigned long flags; | |
bb46c561 | 888 | struct bio_list bios; |
991d9fa0 | 889 | |
bb46c561 JT |
890 | bio_list_init(&bios); |
891 | cell_release_no_holder(pool, cell, &bios); | |
991d9fa0 | 892 | |
bb46c561 JT |
893 | if (!bio_list_empty(&bios)) { |
894 | spin_lock_irqsave(&tc->lock, flags); | |
895 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
896 | spin_unlock_irqrestore(&tc->lock, flags); | |
d256d796 | 897 | wake_worker(pool); |
bb46c561 | 898 | } |
991d9fa0 JT |
899 | } |
900 | ||
a374bb21 JT |
901 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio); |
902 | ||
2d759a46 JT |
903 | struct remap_info { |
904 | struct thin_c *tc; | |
905 | struct bio_list defer_bios; | |
906 | struct bio_list issue_bios; | |
907 | }; | |
908 | ||
909 | static void __inc_remap_and_issue_cell(void *context, | |
910 | struct dm_bio_prison_cell *cell) | |
a374bb21 | 911 | { |
2d759a46 | 912 | struct remap_info *info = context; |
a374bb21 | 913 | struct bio *bio; |
a374bb21 | 914 | |
2d759a46 | 915 | while ((bio = bio_list_pop(&cell->bios))) { |
f73f44eb | 916 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) |
2d759a46 | 917 | bio_list_add(&info->defer_bios, bio); |
a374bb21 | 918 | else { |
2d759a46 JT |
919 | inc_all_io_entry(info->tc->pool, bio); |
920 | ||
921 | /* | |
922 | * We can't issue the bios with the bio prison lock | |
923 | * held, so we add them to a list to issue on | |
924 | * return from this function. | |
925 | */ | |
926 | bio_list_add(&info->issue_bios, bio); | |
a374bb21 JT |
927 | } |
928 | } | |
929 | } | |
930 | ||
2d759a46 JT |
931 | static void inc_remap_and_issue_cell(struct thin_c *tc, |
932 | struct dm_bio_prison_cell *cell, | |
933 | dm_block_t block) | |
934 | { | |
935 | struct bio *bio; | |
936 | struct remap_info info; | |
937 | ||
938 | info.tc = tc; | |
939 | bio_list_init(&info.defer_bios); | |
940 | bio_list_init(&info.issue_bios); | |
941 | ||
942 | /* | |
943 | * We have to be careful to inc any bios we're about to issue | |
944 | * before the cell is released, and avoid a race with new bios | |
945 | * being added to the cell. | |
946 | */ | |
947 | cell_visit_release(tc->pool, __inc_remap_and_issue_cell, | |
948 | &info, cell); | |
949 | ||
950 | while ((bio = bio_list_pop(&info.defer_bios))) | |
951 | thin_defer_bio(tc, bio); | |
952 | ||
953 | while ((bio = bio_list_pop(&info.issue_bios))) | |
954 | remap_and_issue(info.tc, bio, block); | |
955 | } | |
956 | ||
e49e5829 JT |
957 | static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m) |
958 | { | |
6beca5eb | 959 | cell_error(m->tc->pool, m->cell); |
e49e5829 | 960 | list_del(&m->list); |
6f1c819c | 961 | mempool_free(m, &m->tc->pool->mapping_pool); |
e49e5829 | 962 | } |
025b9685 | 963 | |
4ae280b4 NT |
964 | static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio) |
965 | { | |
966 | struct pool *pool = tc->pool; | |
4ae280b4 NT |
967 | |
968 | /* | |
969 | * If the bio has the REQ_FUA flag set we must commit the metadata | |
970 | * before signaling its completion. | |
971 | */ | |
972 | if (!bio_triggers_commit(tc, bio)) { | |
973 | bio_endio(bio); | |
974 | return; | |
975 | } | |
976 | ||
977 | /* | |
978 | * Complete bio with an error if earlier I/O caused changes to the | |
979 | * metadata that can't be committed, e.g, due to I/O errors on the | |
980 | * metadata device. | |
981 | */ | |
982 | if (dm_thin_aborted_changes(tc->td)) { | |
983 | bio_io_error(bio); | |
984 | return; | |
985 | } | |
986 | ||
987 | /* | |
988 | * Batch together any bios that trigger commits and then issue a | |
989 | * single commit for them in process_deferred_bios(). | |
990 | */ | |
8e0c9dac | 991 | spin_lock_irq(&pool->lock); |
4ae280b4 | 992 | bio_list_add(&pool->deferred_flush_completions, bio); |
8e0c9dac | 993 | spin_unlock_irq(&pool->lock); |
4ae280b4 NT |
994 | } |
995 | ||
a24c2569 | 996 | static void process_prepared_mapping(struct dm_thin_new_mapping *m) |
991d9fa0 JT |
997 | { |
998 | struct thin_c *tc = m->tc; | |
6beca5eb | 999 | struct pool *pool = tc->pool; |
8b908f8e | 1000 | struct bio *bio = m->bio; |
991d9fa0 JT |
1001 | int r; |
1002 | ||
4e4cbee9 | 1003 | if (m->status) { |
6beca5eb | 1004 | cell_error(pool, m->cell); |
905386f8 | 1005 | goto out; |
991d9fa0 JT |
1006 | } |
1007 | ||
1008 | /* | |
1009 | * Commit the prepared block into the mapping btree. | |
1010 | * Any I/O for this block arriving after this point will get | |
1011 | * remapped to it directly. | |
1012 | */ | |
34fbcf62 | 1013 | r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block); |
991d9fa0 | 1014 | if (r) { |
b5330655 | 1015 | metadata_operation_failed(pool, "dm_thin_insert_block", r); |
6beca5eb | 1016 | cell_error(pool, m->cell); |
905386f8 | 1017 | goto out; |
991d9fa0 JT |
1018 | } |
1019 | ||
1020 | /* | |
1021 | * Release any bios held while the block was being provisioned. | |
1022 | * If we are processing a write bio that completely covers the block, | |
1023 | * we already processed it so can ignore it now when processing | |
1024 | * the bios in the cell. | |
1025 | */ | |
1026 | if (bio) { | |
2d759a46 | 1027 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); |
4ae280b4 | 1028 | complete_overwrite_bio(tc, bio); |
2d759a46 JT |
1029 | } else { |
1030 | inc_all_io_entry(tc->pool, m->cell->holder); | |
1031 | remap_and_issue(tc, m->cell->holder, m->data_block); | |
1032 | inc_remap_and_issue_cell(tc, m->cell, m->data_block); | |
1033 | } | |
991d9fa0 | 1034 | |
905386f8 | 1035 | out: |
991d9fa0 | 1036 | list_del(&m->list); |
6f1c819c | 1037 | mempool_free(m, &pool->mapping_pool); |
991d9fa0 JT |
1038 | } |
1039 | ||
34fbcf62 JT |
1040 | /*----------------------------------------------------------------*/ |
1041 | ||
1042 | static void free_discard_mapping(struct dm_thin_new_mapping *m) | |
104655fd | 1043 | { |
104655fd | 1044 | struct thin_c *tc = m->tc; |
0ef0b471 | 1045 | |
34fbcf62 JT |
1046 | if (m->cell) |
1047 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1048 | mempool_free(m, &tc->pool->mapping_pool); |
34fbcf62 | 1049 | } |
104655fd | 1050 | |
34fbcf62 JT |
1051 | static void process_prepared_discard_fail(struct dm_thin_new_mapping *m) |
1052 | { | |
e49e5829 | 1053 | bio_io_error(m->bio); |
34fbcf62 JT |
1054 | free_discard_mapping(m); |
1055 | } | |
1056 | ||
1057 | static void process_prepared_discard_success(struct dm_thin_new_mapping *m) | |
1058 | { | |
4246a0b6 | 1059 | bio_endio(m->bio); |
34fbcf62 JT |
1060 | free_discard_mapping(m); |
1061 | } | |
1062 | ||
1063 | static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m) | |
1064 | { | |
1065 | int r; | |
1066 | struct thin_c *tc = m->tc; | |
1067 | ||
1068 | r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end); | |
1069 | if (r) { | |
1070 | metadata_operation_failed(tc->pool, "dm_thin_remove_range", r); | |
1071 | bio_io_error(m->bio); | |
1072 | } else | |
4246a0b6 | 1073 | bio_endio(m->bio); |
34fbcf62 | 1074 | |
f286ba0e | 1075 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1076 | mempool_free(m, &tc->pool->mapping_pool); |
e49e5829 JT |
1077 | } |
1078 | ||
202bae52 JT |
1079 | /*----------------------------------------------------------------*/ |
1080 | ||
2a0fbffb JT |
1081 | static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m, |
1082 | struct bio *discard_parent) | |
e49e5829 | 1083 | { |
34fbcf62 JT |
1084 | /* |
1085 | * We've already unmapped this range of blocks, but before we | |
1086 | * passdown we have to check that these blocks are now unused. | |
1087 | */ | |
202bae52 | 1088 | int r = 0; |
d445bd9c | 1089 | bool shared = true; |
e49e5829 | 1090 | struct thin_c *tc = m->tc; |
34fbcf62 JT |
1091 | struct pool *pool = tc->pool; |
1092 | dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin; | |
202bae52 | 1093 | struct discard_op op; |
104655fd | 1094 | |
2a0fbffb | 1095 | begin_discard(&op, tc, discard_parent); |
34fbcf62 JT |
1096 | while (b != end) { |
1097 | /* find start of unmapped run */ | |
1098 | for (; b < end; b++) { | |
d445bd9c | 1099 | r = dm_pool_block_is_shared(pool->pmd, b, &shared); |
34fbcf62 | 1100 | if (r) |
202bae52 | 1101 | goto out; |
e8088073 | 1102 | |
d445bd9c | 1103 | if (!shared) |
34fbcf62 | 1104 | break; |
19fa1a67 | 1105 | } |
104655fd | 1106 | |
34fbcf62 JT |
1107 | if (b == end) |
1108 | break; | |
1109 | ||
1110 | /* find end of run */ | |
1111 | for (e = b + 1; e != end; e++) { | |
d445bd9c | 1112 | r = dm_pool_block_is_shared(pool->pmd, e, &shared); |
34fbcf62 | 1113 | if (r) |
202bae52 | 1114 | goto out; |
34fbcf62 | 1115 | |
d445bd9c | 1116 | if (shared) |
34fbcf62 JT |
1117 | break; |
1118 | } | |
1119 | ||
202bae52 | 1120 | r = issue_discard(&op, b, e); |
34fbcf62 | 1121 | if (r) |
202bae52 | 1122 | goto out; |
34fbcf62 JT |
1123 | |
1124 | b = e; | |
1125 | } | |
202bae52 JT |
1126 | out: |
1127 | end_discard(&op, r); | |
104655fd JT |
1128 | } |
1129 | ||
2a0fbffb JT |
1130 | static void queue_passdown_pt2(struct dm_thin_new_mapping *m) |
1131 | { | |
1132 | unsigned long flags; | |
1133 | struct pool *pool = m->tc->pool; | |
1134 | ||
1135 | spin_lock_irqsave(&pool->lock, flags); | |
1136 | list_add_tail(&m->list, &pool->prepared_discards_pt2); | |
1137 | spin_unlock_irqrestore(&pool->lock, flags); | |
1138 | wake_worker(pool); | |
1139 | } | |
1140 | ||
1141 | static void passdown_endio(struct bio *bio) | |
1142 | { | |
1143 | /* | |
1144 | * It doesn't matter if the passdown discard failed, we still want | |
1145 | * to unmap (we ignore err). | |
1146 | */ | |
1147 | queue_passdown_pt2(bio->bi_private); | |
948f581a | 1148 | bio_put(bio); |
2a0fbffb JT |
1149 | } |
1150 | ||
1151 | static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m) | |
e49e5829 JT |
1152 | { |
1153 | int r; | |
1154 | struct thin_c *tc = m->tc; | |
34fbcf62 | 1155 | struct pool *pool = tc->pool; |
2a0fbffb JT |
1156 | struct bio *discard_parent; |
1157 | dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin); | |
e49e5829 | 1158 | |
2a0fbffb JT |
1159 | /* |
1160 | * Only this thread allocates blocks, so we can be sure that the | |
1161 | * newly unmapped blocks will not be allocated before the end of | |
1162 | * the function. | |
1163 | */ | |
34fbcf62 | 1164 | r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end); |
202bae52 | 1165 | if (r) { |
34fbcf62 | 1166 | metadata_operation_failed(pool, "dm_thin_remove_range", r); |
202bae52 | 1167 | bio_io_error(m->bio); |
2a0fbffb | 1168 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1169 | mempool_free(m, &pool->mapping_pool); |
2a0fbffb JT |
1170 | return; |
1171 | } | |
34fbcf62 | 1172 | |
00a0ea33 VV |
1173 | /* |
1174 | * Increment the unmapped blocks. This prevents a race between the | |
1175 | * passdown io and reallocation of freed blocks. | |
1176 | */ | |
1177 | r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end); | |
1178 | if (r) { | |
1179 | metadata_operation_failed(pool, "dm_pool_inc_data_range", r); | |
1180 | bio_io_error(m->bio); | |
1181 | cell_defer_no_holder(tc, m->cell); | |
6f1c819c | 1182 | mempool_free(m, &pool->mapping_pool); |
00a0ea33 VV |
1183 | return; |
1184 | } | |
1185 | ||
07888c66 | 1186 | discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO); |
53db984e CH |
1187 | discard_parent->bi_end_io = passdown_endio; |
1188 | discard_parent->bi_private = m; | |
255e2646 HM |
1189 | if (m->maybe_shared) |
1190 | passdown_double_checking_shared_status(m, discard_parent); | |
1191 | else { | |
53db984e | 1192 | struct discard_op op; |
2a0fbffb | 1193 | |
53db984e CH |
1194 | begin_discard(&op, tc, discard_parent); |
1195 | r = issue_discard(&op, m->data_block, data_end); | |
1196 | end_discard(&op, r); | |
202bae52 | 1197 | } |
2a0fbffb JT |
1198 | } |
1199 | ||
1200 | static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m) | |
1201 | { | |
1202 | int r; | |
1203 | struct thin_c *tc = m->tc; | |
1204 | struct pool *pool = tc->pool; | |
1205 | ||
1206 | /* | |
1207 | * The passdown has completed, so now we can decrement all those | |
1208 | * unmapped blocks. | |
1209 | */ | |
1210 | r = dm_pool_dec_data_range(pool->pmd, m->data_block, | |
1211 | m->data_block + (m->virt_end - m->virt_begin)); | |
1212 | if (r) { | |
1213 | metadata_operation_failed(pool, "dm_pool_dec_data_range", r); | |
1214 | bio_io_error(m->bio); | |
1215 | } else | |
1216 | bio_endio(m->bio); | |
1217 | ||
34fbcf62 | 1218 | cell_defer_no_holder(tc, m->cell); |
6f1c819c | 1219 | mempool_free(m, &pool->mapping_pool); |
e49e5829 JT |
1220 | } |
1221 | ||
104655fd | 1222 | static void process_prepared(struct pool *pool, struct list_head *head, |
e49e5829 | 1223 | process_mapping_fn *fn) |
991d9fa0 | 1224 | { |
991d9fa0 | 1225 | struct list_head maps; |
a24c2569 | 1226 | struct dm_thin_new_mapping *m, *tmp; |
991d9fa0 JT |
1227 | |
1228 | INIT_LIST_HEAD(&maps); | |
8e0c9dac | 1229 | spin_lock_irq(&pool->lock); |
104655fd | 1230 | list_splice_init(head, &maps); |
8e0c9dac | 1231 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
1232 | |
1233 | list_for_each_entry_safe(m, tmp, &maps, list) | |
e49e5829 | 1234 | (*fn)(m); |
991d9fa0 JT |
1235 | } |
1236 | ||
1237 | /* | |
1238 | * Deferred bio jobs. | |
1239 | */ | |
104655fd | 1240 | static int io_overlaps_block(struct pool *pool, struct bio *bio) |
991d9fa0 | 1241 | { |
4f024f37 KO |
1242 | return bio->bi_iter.bi_size == |
1243 | (pool->sectors_per_block << SECTOR_SHIFT); | |
104655fd JT |
1244 | } |
1245 | ||
1246 | static int io_overwrites_block(struct pool *pool, struct bio *bio) | |
1247 | { | |
1248 | return (bio_data_dir(bio) == WRITE) && | |
1249 | io_overlaps_block(pool, bio); | |
991d9fa0 JT |
1250 | } |
1251 | ||
1252 | static void save_and_set_endio(struct bio *bio, bio_end_io_t **save, | |
1253 | bio_end_io_t *fn) | |
1254 | { | |
1255 | *save = bio->bi_end_io; | |
1256 | bio->bi_end_io = fn; | |
1257 | } | |
1258 | ||
1259 | static int ensure_next_mapping(struct pool *pool) | |
1260 | { | |
1261 | if (pool->next_mapping) | |
1262 | return 0; | |
1263 | ||
6f1c819c | 1264 | pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC); |
991d9fa0 JT |
1265 | |
1266 | return pool->next_mapping ? 0 : -ENOMEM; | |
1267 | } | |
1268 | ||
a24c2569 | 1269 | static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool) |
991d9fa0 | 1270 | { |
16961b04 | 1271 | struct dm_thin_new_mapping *m = pool->next_mapping; |
991d9fa0 JT |
1272 | |
1273 | BUG_ON(!pool->next_mapping); | |
1274 | ||
16961b04 MS |
1275 | memset(m, 0, sizeof(struct dm_thin_new_mapping)); |
1276 | INIT_LIST_HEAD(&m->list); | |
1277 | m->bio = NULL; | |
1278 | ||
991d9fa0 JT |
1279 | pool->next_mapping = NULL; |
1280 | ||
16961b04 | 1281 | return m; |
991d9fa0 JT |
1282 | } |
1283 | ||
e5aea7b4 JT |
1284 | static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m, |
1285 | sector_t begin, sector_t end) | |
1286 | { | |
e5aea7b4 JT |
1287 | struct dm_io_region to; |
1288 | ||
1289 | to.bdev = tc->pool_dev->bdev; | |
1290 | to.sector = begin; | |
1291 | to.count = end - begin; | |
1292 | ||
7209049d | 1293 | dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m); |
e5aea7b4 JT |
1294 | } |
1295 | ||
452d7a62 | 1296 | static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio, |
34fbcf62 | 1297 | dm_block_t data_begin, |
452d7a62 MS |
1298 | struct dm_thin_new_mapping *m) |
1299 | { | |
1300 | struct pool *pool = tc->pool; | |
1301 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1302 | ||
1303 | h->overwrite_mapping = m; | |
1304 | m->bio = bio; | |
1305 | save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio); | |
1306 | inc_all_io_entry(pool, bio); | |
34fbcf62 | 1307 | remap_and_issue(tc, bio, data_begin); |
452d7a62 MS |
1308 | } |
1309 | ||
e5aea7b4 JT |
1310 | /* |
1311 | * A partial copy also needs to zero the uncopied region. | |
1312 | */ | |
991d9fa0 | 1313 | static void schedule_copy(struct thin_c *tc, dm_block_t virt_block, |
2dd9c257 JT |
1314 | struct dm_dev *origin, dm_block_t data_origin, |
1315 | dm_block_t data_dest, | |
e5aea7b4 JT |
1316 | struct dm_bio_prison_cell *cell, struct bio *bio, |
1317 | sector_t len) | |
991d9fa0 | 1318 | { |
991d9fa0 | 1319 | struct pool *pool = tc->pool; |
a24c2569 | 1320 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1321 | |
991d9fa0 | 1322 | m->tc = tc; |
34fbcf62 JT |
1323 | m->virt_begin = virt_block; |
1324 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1325 | m->data_block = data_dest; |
1326 | m->cell = cell; | |
991d9fa0 | 1327 | |
e5aea7b4 JT |
1328 | /* |
1329 | * quiesce action + copy action + an extra reference held for the | |
1330 | * duration of this function (we may need to inc later for a | |
1331 | * partial zero). | |
1332 | */ | |
1333 | atomic_set(&m->prepare_actions, 3); | |
1334 | ||
44feb387 | 1335 | if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list)) |
e5aea7b4 | 1336 | complete_mapping_preparation(m); /* already quiesced */ |
991d9fa0 JT |
1337 | |
1338 | /* | |
1339 | * IO to pool_dev remaps to the pool target's data_dev. | |
1340 | * | |
1341 | * If the whole block of data is being overwritten, we can issue the | |
1342 | * bio immediately. Otherwise we use kcopyd to clone the data first. | |
1343 | */ | |
452d7a62 MS |
1344 | if (io_overwrites_block(pool, bio)) |
1345 | remap_and_issue_overwrite(tc, bio, data_dest, m); | |
1346 | else { | |
991d9fa0 JT |
1347 | struct dm_io_region from, to; |
1348 | ||
2dd9c257 | 1349 | from.bdev = origin->bdev; |
991d9fa0 | 1350 | from.sector = data_origin * pool->sectors_per_block; |
e5aea7b4 | 1351 | from.count = len; |
991d9fa0 JT |
1352 | |
1353 | to.bdev = tc->pool_dev->bdev; | |
1354 | to.sector = data_dest * pool->sectors_per_block; | |
e5aea7b4 | 1355 | to.count = len; |
991d9fa0 | 1356 | |
7209049d MS |
1357 | dm_kcopyd_copy(pool->copier, &from, 1, &to, |
1358 | 0, copy_complete, m); | |
e5aea7b4 JT |
1359 | |
1360 | /* | |
1361 | * Do we need to zero a tail region? | |
1362 | */ | |
1363 | if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) { | |
1364 | atomic_inc(&m->prepare_actions); | |
1365 | ll_zero(tc, m, | |
1366 | data_dest * pool->sectors_per_block + len, | |
1367 | (data_dest + 1) * pool->sectors_per_block); | |
991d9fa0 JT |
1368 | } |
1369 | } | |
e5aea7b4 JT |
1370 | |
1371 | complete_mapping_preparation(m); /* drop our ref */ | |
991d9fa0 JT |
1372 | } |
1373 | ||
2dd9c257 JT |
1374 | static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block, |
1375 | dm_block_t data_origin, dm_block_t data_dest, | |
a24c2569 | 1376 | struct dm_bio_prison_cell *cell, struct bio *bio) |
2dd9c257 JT |
1377 | { |
1378 | schedule_copy(tc, virt_block, tc->pool_dev, | |
e5aea7b4 JT |
1379 | data_origin, data_dest, cell, bio, |
1380 | tc->pool->sectors_per_block); | |
2dd9c257 JT |
1381 | } |
1382 | ||
991d9fa0 | 1383 | static void schedule_zero(struct thin_c *tc, dm_block_t virt_block, |
a24c2569 | 1384 | dm_block_t data_block, struct dm_bio_prison_cell *cell, |
991d9fa0 JT |
1385 | struct bio *bio) |
1386 | { | |
1387 | struct pool *pool = tc->pool; | |
a24c2569 | 1388 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
991d9fa0 | 1389 | |
50f3c3ef | 1390 | atomic_set(&m->prepare_actions, 1); /* no need to quiesce */ |
991d9fa0 | 1391 | m->tc = tc; |
34fbcf62 JT |
1392 | m->virt_begin = virt_block; |
1393 | m->virt_end = virt_block + 1u; | |
991d9fa0 JT |
1394 | m->data_block = data_block; |
1395 | m->cell = cell; | |
991d9fa0 JT |
1396 | |
1397 | /* | |
1398 | * If the whole block of data is being overwritten or we are not | |
1399 | * zeroing pre-existing data, we can issue the bio immediately. | |
1400 | * Otherwise we use kcopyd to zero the data first. | |
1401 | */ | |
f8ae7525 MS |
1402 | if (pool->pf.zero_new_blocks) { |
1403 | if (io_overwrites_block(pool, bio)) | |
1404 | remap_and_issue_overwrite(tc, bio, data_block, m); | |
1405 | else | |
1406 | ll_zero(tc, m, data_block * pool->sectors_per_block, | |
1407 | (data_block + 1) * pool->sectors_per_block); | |
1408 | } else | |
991d9fa0 | 1409 | process_prepared_mapping(m); |
e5aea7b4 | 1410 | } |
991d9fa0 | 1411 | |
e5aea7b4 JT |
1412 | static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block, |
1413 | dm_block_t data_dest, | |
1414 | struct dm_bio_prison_cell *cell, struct bio *bio) | |
1415 | { | |
1416 | struct pool *pool = tc->pool; | |
1417 | sector_t virt_block_begin = virt_block * pool->sectors_per_block; | |
1418 | sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block; | |
1419 | ||
1420 | if (virt_block_end <= tc->origin_size) | |
1421 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1422 | virt_block, data_dest, cell, bio, | |
1423 | pool->sectors_per_block); | |
1424 | ||
1425 | else if (virt_block_begin < tc->origin_size) | |
1426 | schedule_copy(tc, virt_block, tc->origin_dev, | |
1427 | virt_block, data_dest, cell, bio, | |
1428 | tc->origin_size - virt_block_begin); | |
1429 | ||
1430 | else | |
1431 | schedule_zero(tc, virt_block, data_dest, cell, bio); | |
991d9fa0 JT |
1432 | } |
1433 | ||
2c43fd26 JT |
1434 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode); |
1435 | ||
a685557f MS |
1436 | static void requeue_bios(struct pool *pool); |
1437 | ||
3ab91828 JT |
1438 | static bool is_read_only_pool_mode(enum pool_mode mode) |
1439 | { | |
1440 | return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY); | |
1441 | } | |
1442 | ||
1443 | static bool is_read_only(struct pool *pool) | |
1444 | { | |
1445 | return is_read_only_pool_mode(get_pool_mode(pool)); | |
1446 | } | |
1447 | ||
1448 | static void check_for_metadata_space(struct pool *pool) | |
1449 | { | |
1450 | int r; | |
1451 | const char *ooms_reason = NULL; | |
1452 | dm_block_t nr_free; | |
1453 | ||
1454 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free); | |
1455 | if (r) | |
1456 | ooms_reason = "Could not get free metadata blocks"; | |
1457 | else if (!nr_free) | |
1458 | ooms_reason = "No free metadata blocks"; | |
1459 | ||
1460 | if (ooms_reason && !is_read_only(pool)) { | |
1461 | DMERR("%s", ooms_reason); | |
1462 | set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE); | |
1463 | } | |
1464 | } | |
1465 | ||
1466 | static void check_for_data_space(struct pool *pool) | |
2c43fd26 JT |
1467 | { |
1468 | int r; | |
1469 | dm_block_t nr_free; | |
1470 | ||
1471 | if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE) | |
1472 | return; | |
1473 | ||
1474 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free); | |
1475 | if (r) | |
1476 | return; | |
1477 | ||
a685557f | 1478 | if (nr_free) { |
2c43fd26 | 1479 | set_pool_mode(pool, PM_WRITE); |
a685557f MS |
1480 | requeue_bios(pool); |
1481 | } | |
2c43fd26 JT |
1482 | } |
1483 | ||
e49e5829 JT |
1484 | /* |
1485 | * A non-zero return indicates read_only or fail_io mode. | |
1486 | * Many callers don't care about the return value. | |
1487 | */ | |
020cc3b5 | 1488 | static int commit(struct pool *pool) |
e49e5829 JT |
1489 | { |
1490 | int r; | |
1491 | ||
3ab91828 | 1492 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) |
e49e5829 JT |
1493 | return -EINVAL; |
1494 | ||
020cc3b5 | 1495 | r = dm_pool_commit_metadata(pool->pmd); |
b5330655 JT |
1496 | if (r) |
1497 | metadata_operation_failed(pool, "dm_pool_commit_metadata", r); | |
3ab91828 JT |
1498 | else { |
1499 | check_for_metadata_space(pool); | |
1500 | check_for_data_space(pool); | |
1501 | } | |
e49e5829 JT |
1502 | |
1503 | return r; | |
1504 | } | |
1505 | ||
88a6621b JT |
1506 | static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks) |
1507 | { | |
88a6621b JT |
1508 | if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) { |
1509 | DMWARN("%s: reached low water mark for data device: sending event.", | |
1510 | dm_device_name(pool->pool_md)); | |
8e0c9dac | 1511 | spin_lock_irq(&pool->lock); |
88a6621b | 1512 | pool->low_water_triggered = true; |
8e0c9dac | 1513 | spin_unlock_irq(&pool->lock); |
88a6621b JT |
1514 | dm_table_event(pool->ti->table); |
1515 | } | |
1516 | } | |
1517 | ||
991d9fa0 JT |
1518 | static int alloc_data_block(struct thin_c *tc, dm_block_t *result) |
1519 | { | |
1520 | int r; | |
1521 | dm_block_t free_blocks; | |
991d9fa0 JT |
1522 | struct pool *pool = tc->pool; |
1523 | ||
3e1a0699 | 1524 | if (WARN_ON(get_pool_mode(pool) != PM_WRITE)) |
8d30abff JT |
1525 | return -EINVAL; |
1526 | ||
991d9fa0 | 1527 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1528 | if (r) { |
1529 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
991d9fa0 | 1530 | return r; |
b5330655 | 1531 | } |
991d9fa0 | 1532 | |
88a6621b | 1533 | check_low_water_mark(pool, free_blocks); |
991d9fa0 JT |
1534 | |
1535 | if (!free_blocks) { | |
94563bad MS |
1536 | /* |
1537 | * Try to commit to see if that will free up some | |
1538 | * more space. | |
1539 | */ | |
020cc3b5 JT |
1540 | r = commit(pool); |
1541 | if (r) | |
1542 | return r; | |
991d9fa0 | 1543 | |
94563bad | 1544 | r = dm_pool_get_free_block_count(pool->pmd, &free_blocks); |
b5330655 JT |
1545 | if (r) { |
1546 | metadata_operation_failed(pool, "dm_pool_get_free_block_count", r); | |
94563bad | 1547 | return r; |
b5330655 | 1548 | } |
991d9fa0 | 1549 | |
94563bad | 1550 | if (!free_blocks) { |
3e1a0699 | 1551 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); |
94563bad | 1552 | return -ENOSPC; |
991d9fa0 JT |
1553 | } |
1554 | } | |
1555 | ||
1556 | r = dm_pool_alloc_data_block(pool->pmd, result); | |
4a02b34e | 1557 | if (r) { |
a685557f MS |
1558 | if (r == -ENOSPC) |
1559 | set_pool_mode(pool, PM_OUT_OF_DATA_SPACE); | |
1560 | else | |
1561 | metadata_operation_failed(pool, "dm_pool_alloc_data_block", r); | |
991d9fa0 | 1562 | return r; |
4a02b34e | 1563 | } |
991d9fa0 | 1564 | |
3ab91828 JT |
1565 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks); |
1566 | if (r) { | |
1567 | metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r); | |
1568 | return r; | |
1569 | } | |
1570 | ||
1571 | if (!free_blocks) { | |
1572 | /* Let's commit before we use up the metadata reserve. */ | |
1573 | r = commit(pool); | |
1574 | if (r) | |
1575 | return r; | |
1576 | } | |
1577 | ||
991d9fa0 JT |
1578 | return 0; |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * If we have run out of space, queue bios until the device is | |
1583 | * resumed, presumably after having been reloaded with more space. | |
1584 | */ | |
1585 | static void retry_on_resume(struct bio *bio) | |
1586 | { | |
59c3d2c6 | 1587 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 1588 | struct thin_c *tc = h->tc; |
991d9fa0 | 1589 | |
8e0c9dac | 1590 | spin_lock_irq(&tc->lock); |
c140e1c4 | 1591 | bio_list_add(&tc->retry_on_resume_list, bio); |
8e0c9dac | 1592 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
1593 | } |
1594 | ||
4e4cbee9 | 1595 | static blk_status_t should_error_unserviceable_bio(struct pool *pool) |
8c0f0e8c | 1596 | { |
3e1a0699 JT |
1597 | enum pool_mode m = get_pool_mode(pool); |
1598 | ||
1599 | switch (m) { | |
1600 | case PM_WRITE: | |
1601 | /* Shouldn't get here */ | |
1602 | DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode"); | |
4e4cbee9 | 1603 | return BLK_STS_IOERR; |
3e1a0699 JT |
1604 | |
1605 | case PM_OUT_OF_DATA_SPACE: | |
4e4cbee9 | 1606 | return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0; |
3e1a0699 | 1607 | |
3ab91828 | 1608 | case PM_OUT_OF_METADATA_SPACE: |
3e1a0699 JT |
1609 | case PM_READ_ONLY: |
1610 | case PM_FAIL: | |
4e4cbee9 | 1611 | return BLK_STS_IOERR; |
3e1a0699 JT |
1612 | default: |
1613 | /* Shouldn't get here */ | |
1614 | DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode"); | |
4e4cbee9 | 1615 | return BLK_STS_IOERR; |
3e1a0699 JT |
1616 | } |
1617 | } | |
8c0f0e8c | 1618 | |
3e1a0699 JT |
1619 | static void handle_unserviceable_bio(struct pool *pool, struct bio *bio) |
1620 | { | |
4e4cbee9 | 1621 | blk_status_t error = should_error_unserviceable_bio(pool); |
af91805a | 1622 | |
4246a0b6 | 1623 | if (error) { |
4e4cbee9 | 1624 | bio->bi_status = error; |
4246a0b6 CH |
1625 | bio_endio(bio); |
1626 | } else | |
6d16202b | 1627 | retry_on_resume(bio); |
8c0f0e8c MS |
1628 | } |
1629 | ||
399caddf | 1630 | static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1631 | { |
1632 | struct bio *bio; | |
1633 | struct bio_list bios; | |
4e4cbee9 | 1634 | blk_status_t error; |
991d9fa0 | 1635 | |
af91805a MS |
1636 | error = should_error_unserviceable_bio(pool); |
1637 | if (error) { | |
1638 | cell_error_with_code(pool, cell, error); | |
3e1a0699 JT |
1639 | return; |
1640 | } | |
1641 | ||
991d9fa0 | 1642 | bio_list_init(&bios); |
6beca5eb | 1643 | cell_release(pool, cell, &bios); |
991d9fa0 | 1644 | |
9d094eeb MS |
1645 | while ((bio = bio_list_pop(&bios))) |
1646 | retry_on_resume(bio); | |
991d9fa0 JT |
1647 | } |
1648 | ||
34fbcf62 JT |
1649 | static void process_discard_cell_no_passdown(struct thin_c *tc, |
1650 | struct dm_bio_prison_cell *virt_cell) | |
104655fd | 1651 | { |
104655fd | 1652 | struct pool *pool = tc->pool; |
34fbcf62 | 1653 | struct dm_thin_new_mapping *m = get_next_mapping(pool); |
104655fd | 1654 | |
34fbcf62 JT |
1655 | /* |
1656 | * We don't need to lock the data blocks, since there's no | |
1657 | * passdown. We only lock data blocks for allocation and breaking sharing. | |
1658 | */ | |
1659 | m->tc = tc; | |
1660 | m->virt_begin = virt_cell->key.block_begin; | |
1661 | m->virt_end = virt_cell->key.block_end; | |
1662 | m->cell = virt_cell; | |
1663 | m->bio = virt_cell->holder; | |
104655fd | 1664 | |
34fbcf62 JT |
1665 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) |
1666 | pool->process_prepared_discard(m); | |
1667 | } | |
104655fd | 1668 | |
34fbcf62 JT |
1669 | static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end, |
1670 | struct bio *bio) | |
1671 | { | |
1672 | struct pool *pool = tc->pool; | |
1673 | ||
1674 | int r; | |
1675 | bool maybe_shared; | |
1676 | struct dm_cell_key data_key; | |
1677 | struct dm_bio_prison_cell *data_cell; | |
1678 | struct dm_thin_new_mapping *m; | |
e2dd8aca JT |
1679 | dm_block_t virt_begin, virt_end, data_begin, data_end; |
1680 | dm_block_t len, next_boundary; | |
34fbcf62 JT |
1681 | |
1682 | while (begin != end) { | |
34fbcf62 JT |
1683 | r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, |
1684 | &data_begin, &maybe_shared); | |
e2dd8aca | 1685 | if (r) { |
104655fd | 1686 | /* |
34fbcf62 JT |
1687 | * Silently fail, letting any mappings we've |
1688 | * created complete. | |
104655fd | 1689 | */ |
34fbcf62 | 1690 | break; |
104655fd | 1691 | } |
104655fd | 1692 | |
e2dd8aca | 1693 | data_end = data_begin + (virt_end - virt_begin); |
104655fd | 1694 | |
34fbcf62 | 1695 | /* |
e2dd8aca | 1696 | * Make sure the data region obeys the bio prison restrictions. |
34fbcf62 | 1697 | */ |
e2dd8aca JT |
1698 | while (data_begin < data_end) { |
1699 | r = ensure_next_mapping(pool); | |
1700 | if (r) | |
1701 | return; /* we did our best */ | |
1702 | ||
1703 | next_boundary = ((data_begin >> BIO_PRISON_MAX_RANGE_SHIFT) + 1) | |
1704 | << BIO_PRISON_MAX_RANGE_SHIFT; | |
1705 | len = min_t(sector_t, data_end - data_begin, next_boundary - data_begin); | |
1706 | ||
3f8d3f54 MS |
1707 | /* This key is certainly within range given the above splitting */ |
1708 | (void) build_key(tc->td, PHYSICAL, data_begin, data_begin + len, &data_key); | |
e2dd8aca JT |
1709 | if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { |
1710 | /* contention, we'll give up with this range */ | |
1711 | data_begin += len; | |
1712 | continue; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * IO may still be going to the destination block. We must | |
1717 | * quiesce before we can do the removal. | |
1718 | */ | |
1719 | m = get_next_mapping(pool); | |
1720 | m->tc = tc; | |
1721 | m->maybe_shared = maybe_shared; | |
1722 | m->virt_begin = virt_begin; | |
1723 | m->virt_end = virt_begin + len; | |
1724 | m->data_block = data_begin; | |
1725 | m->cell = data_cell; | |
1726 | m->bio = bio; | |
1727 | ||
1728 | /* | |
1729 | * The parent bio must not complete before sub discard bios are | |
1730 | * chained to it (see end_discard's bio_chain)! | |
1731 | * | |
1732 | * This per-mapping bi_remaining increment is paired with | |
1733 | * the implicit decrement that occurs via bio_endio() in | |
1734 | * end_discard(). | |
1735 | */ | |
1736 | bio_inc_remaining(bio); | |
1737 | if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) | |
1738 | pool->process_prepared_discard(m); | |
1739 | ||
1740 | virt_begin += len; | |
1741 | data_begin += len; | |
1742 | } | |
34fbcf62 JT |
1743 | |
1744 | begin = virt_end; | |
104655fd JT |
1745 | } |
1746 | } | |
1747 | ||
34fbcf62 JT |
1748 | static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell) |
1749 | { | |
1750 | struct bio *bio = virt_cell->holder; | |
1751 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
1752 | ||
1753 | /* | |
1754 | * The virt_cell will only get freed once the origin bio completes. | |
1755 | * This means it will remain locked while all the individual | |
1756 | * passdown bios are in flight. | |
1757 | */ | |
1758 | h->cell = virt_cell; | |
1759 | break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio); | |
1760 | ||
1761 | /* | |
1762 | * We complete the bio now, knowing that the bi_remaining field | |
1763 | * will prevent completion until the sub range discards have | |
1764 | * completed. | |
1765 | */ | |
4246a0b6 | 1766 | bio_endio(bio); |
34fbcf62 JT |
1767 | } |
1768 | ||
a374bb21 JT |
1769 | static void process_discard_bio(struct thin_c *tc, struct bio *bio) |
1770 | { | |
34fbcf62 JT |
1771 | dm_block_t begin, end; |
1772 | struct dm_cell_key virt_key; | |
1773 | struct dm_bio_prison_cell *virt_cell; | |
a374bb21 | 1774 | |
34fbcf62 JT |
1775 | get_bio_block_range(tc, bio, &begin, &end); |
1776 | if (begin == end) { | |
1777 | /* | |
1778 | * The discard covers less than a block. | |
1779 | */ | |
4246a0b6 | 1780 | bio_endio(bio); |
a374bb21 | 1781 | return; |
34fbcf62 | 1782 | } |
a374bb21 | 1783 | |
3f8d3f54 MS |
1784 | if (unlikely(!build_key(tc->td, VIRTUAL, begin, end, &virt_key))) { |
1785 | DMERR_LIMIT("Discard doesn't respect bio prison limits"); | |
1786 | bio_endio(bio); | |
1787 | return; | |
1788 | } | |
1789 | ||
1790 | if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) { | |
34fbcf62 JT |
1791 | /* |
1792 | * Potential starvation issue: We're relying on the | |
1793 | * fs/application being well behaved, and not trying to | |
1794 | * send IO to a region at the same time as discarding it. | |
1795 | * If they do this persistently then it's possible this | |
1796 | * cell will never be granted. | |
1797 | */ | |
1798 | return; | |
3f8d3f54 | 1799 | } |
34fbcf62 JT |
1800 | |
1801 | tc->pool->process_discard_cell(tc, virt_cell); | |
a374bb21 JT |
1802 | } |
1803 | ||
991d9fa0 | 1804 | static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block, |
44feb387 | 1805 | struct dm_cell_key *key, |
991d9fa0 | 1806 | struct dm_thin_lookup_result *lookup_result, |
a24c2569 | 1807 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1808 | { |
1809 | int r; | |
1810 | dm_block_t data_block; | |
d6fc2042 | 1811 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1812 | |
1813 | r = alloc_data_block(tc, &data_block); | |
1814 | switch (r) { | |
1815 | case 0: | |
2dd9c257 JT |
1816 | schedule_internal_copy(tc, block, lookup_result->block, |
1817 | data_block, cell, bio); | |
991d9fa0 JT |
1818 | break; |
1819 | ||
1820 | case -ENOSPC: | |
399caddf | 1821 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1822 | break; |
1823 | ||
1824 | default: | |
c397741c MS |
1825 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1826 | __func__, r); | |
d6fc2042 | 1827 | cell_error(pool, cell); |
991d9fa0 JT |
1828 | break; |
1829 | } | |
1830 | } | |
1831 | ||
23ca2bb6 JT |
1832 | static void __remap_and_issue_shared_cell(void *context, |
1833 | struct dm_bio_prison_cell *cell) | |
1834 | { | |
1835 | struct remap_info *info = context; | |
1836 | struct bio *bio; | |
1837 | ||
1838 | while ((bio = bio_list_pop(&cell->bios))) { | |
f73f44eb CH |
1839 | if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) || |
1840 | bio_op(bio) == REQ_OP_DISCARD) | |
23ca2bb6 JT |
1841 | bio_list_add(&info->defer_bios, bio); |
1842 | else { | |
bd6d1e0a | 1843 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
23ca2bb6 JT |
1844 | |
1845 | h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds); | |
1846 | inc_all_io_entry(info->tc->pool, bio); | |
1847 | bio_list_add(&info->issue_bios, bio); | |
1848 | } | |
1849 | } | |
1850 | } | |
1851 | ||
1852 | static void remap_and_issue_shared_cell(struct thin_c *tc, | |
1853 | struct dm_bio_prison_cell *cell, | |
1854 | dm_block_t block) | |
1855 | { | |
1856 | struct bio *bio; | |
1857 | struct remap_info info; | |
1858 | ||
1859 | info.tc = tc; | |
1860 | bio_list_init(&info.defer_bios); | |
1861 | bio_list_init(&info.issue_bios); | |
1862 | ||
1863 | cell_visit_release(tc->pool, __remap_and_issue_shared_cell, | |
1864 | &info, cell); | |
1865 | ||
1866 | while ((bio = bio_list_pop(&info.defer_bios))) | |
1867 | thin_defer_bio(tc, bio); | |
1868 | ||
1869 | while ((bio = bio_list_pop(&info.issue_bios))) | |
1870 | remap_and_issue(tc, bio, block); | |
1871 | } | |
1872 | ||
991d9fa0 JT |
1873 | static void process_shared_bio(struct thin_c *tc, struct bio *bio, |
1874 | dm_block_t block, | |
23ca2bb6 JT |
1875 | struct dm_thin_lookup_result *lookup_result, |
1876 | struct dm_bio_prison_cell *virt_cell) | |
991d9fa0 | 1877 | { |
23ca2bb6 | 1878 | struct dm_bio_prison_cell *data_cell; |
991d9fa0 | 1879 | struct pool *pool = tc->pool; |
44feb387 | 1880 | struct dm_cell_key key; |
991d9fa0 JT |
1881 | |
1882 | /* | |
1883 | * If cell is already occupied, then sharing is already in the process | |
1884 | * of being broken so we have nothing further to do here. | |
1885 | */ | |
1886 | build_data_key(tc->td, lookup_result->block, &key); | |
23ca2bb6 JT |
1887 | if (bio_detain(pool, &key, bio, &data_cell)) { |
1888 | cell_defer_no_holder(tc, virt_cell); | |
991d9fa0 | 1889 | return; |
23ca2bb6 | 1890 | } |
991d9fa0 | 1891 | |
23ca2bb6 JT |
1892 | if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) { |
1893 | break_sharing(tc, bio, block, &key, lookup_result, data_cell); | |
1894 | cell_defer_no_holder(tc, virt_cell); | |
1895 | } else { | |
59c3d2c6 | 1896 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
991d9fa0 | 1897 | |
44feb387 | 1898 | h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds); |
e8088073 | 1899 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1900 | remap_and_issue(tc, bio, lookup_result->block); |
23ca2bb6 JT |
1901 | |
1902 | remap_and_issue_shared_cell(tc, data_cell, lookup_result->block); | |
1903 | remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block); | |
991d9fa0 JT |
1904 | } |
1905 | } | |
1906 | ||
1907 | static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block, | |
a24c2569 | 1908 | struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1909 | { |
1910 | int r; | |
1911 | dm_block_t data_block; | |
6beca5eb | 1912 | struct pool *pool = tc->pool; |
991d9fa0 JT |
1913 | |
1914 | /* | |
1915 | * Remap empty bios (flushes) immediately, without provisioning. | |
1916 | */ | |
4f024f37 | 1917 | if (!bio->bi_iter.bi_size) { |
6beca5eb | 1918 | inc_all_io_entry(pool, bio); |
f286ba0e | 1919 | cell_defer_no_holder(tc, cell); |
e8088073 | 1920 | |
991d9fa0 JT |
1921 | remap_and_issue(tc, bio, 0); |
1922 | return; | |
1923 | } | |
1924 | ||
1925 | /* | |
1926 | * Fill read bios with zeroes and complete them immediately. | |
1927 | */ | |
1928 | if (bio_data_dir(bio) == READ) { | |
1929 | zero_fill_bio(bio); | |
f286ba0e | 1930 | cell_defer_no_holder(tc, cell); |
4246a0b6 | 1931 | bio_endio(bio); |
991d9fa0 JT |
1932 | return; |
1933 | } | |
1934 | ||
1935 | r = alloc_data_block(tc, &data_block); | |
1936 | switch (r) { | |
1937 | case 0: | |
2dd9c257 JT |
1938 | if (tc->origin_dev) |
1939 | schedule_external_copy(tc, block, data_block, cell, bio); | |
1940 | else | |
1941 | schedule_zero(tc, block, data_block, cell, bio); | |
991d9fa0 JT |
1942 | break; |
1943 | ||
1944 | case -ENOSPC: | |
399caddf | 1945 | retry_bios_on_resume(pool, cell); |
991d9fa0 JT |
1946 | break; |
1947 | ||
1948 | default: | |
c397741c MS |
1949 | DMERR_LIMIT("%s: alloc_data_block() failed: error = %d", |
1950 | __func__, r); | |
6beca5eb | 1951 | cell_error(pool, cell); |
991d9fa0 JT |
1952 | break; |
1953 | } | |
1954 | } | |
1955 | ||
a374bb21 | 1956 | static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
991d9fa0 JT |
1957 | { |
1958 | int r; | |
6beca5eb | 1959 | struct pool *pool = tc->pool; |
a374bb21 | 1960 | struct bio *bio = cell->holder; |
991d9fa0 | 1961 | dm_block_t block = get_bio_block(tc, bio); |
991d9fa0 JT |
1962 | struct dm_thin_lookup_result lookup_result; |
1963 | ||
a374bb21 JT |
1964 | if (tc->requeue_mode) { |
1965 | cell_requeue(pool, cell); | |
991d9fa0 | 1966 | return; |
a374bb21 | 1967 | } |
991d9fa0 JT |
1968 | |
1969 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
1970 | switch (r) { | |
1971 | case 0: | |
23ca2bb6 JT |
1972 | if (lookup_result.shared) |
1973 | process_shared_bio(tc, bio, block, &lookup_result, cell); | |
1974 | else { | |
6beca5eb | 1975 | inc_all_io_entry(pool, bio); |
991d9fa0 | 1976 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 | 1977 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); |
e8088073 | 1978 | } |
991d9fa0 JT |
1979 | break; |
1980 | ||
1981 | case -ENODATA: | |
2dd9c257 | 1982 | if (bio_data_dir(bio) == READ && tc->origin_dev) { |
6beca5eb | 1983 | inc_all_io_entry(pool, bio); |
f286ba0e | 1984 | cell_defer_no_holder(tc, cell); |
e8088073 | 1985 | |
e5aea7b4 JT |
1986 | if (bio_end_sector(bio) <= tc->origin_size) |
1987 | remap_to_origin_and_issue(tc, bio); | |
1988 | ||
1989 | else if (bio->bi_iter.bi_sector < tc->origin_size) { | |
1990 | zero_fill_bio(bio); | |
1991 | bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT; | |
1992 | remap_to_origin_and_issue(tc, bio); | |
1993 | ||
1994 | } else { | |
1995 | zero_fill_bio(bio); | |
4246a0b6 | 1996 | bio_endio(bio); |
e5aea7b4 | 1997 | } |
2dd9c257 JT |
1998 | } else |
1999 | provision_block(tc, bio, block, cell); | |
991d9fa0 JT |
2000 | break; |
2001 | ||
2002 | default: | |
c397741c MS |
2003 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
2004 | __func__, r); | |
f286ba0e | 2005 | cell_defer_no_holder(tc, cell); |
991d9fa0 JT |
2006 | bio_io_error(bio); |
2007 | break; | |
2008 | } | |
2009 | } | |
2010 | ||
a374bb21 JT |
2011 | static void process_bio(struct thin_c *tc, struct bio *bio) |
2012 | { | |
2013 | struct pool *pool = tc->pool; | |
2014 | dm_block_t block = get_bio_block(tc, bio); | |
2015 | struct dm_bio_prison_cell *cell; | |
2016 | struct dm_cell_key key; | |
2017 | ||
2018 | /* | |
2019 | * If cell is already occupied, then the block is already | |
2020 | * being provisioned so we have nothing further to do here. | |
2021 | */ | |
2022 | build_virtual_key(tc->td, block, &key); | |
2023 | if (bio_detain(pool, &key, bio, &cell)) | |
2024 | return; | |
2025 | ||
2026 | process_cell(tc, cell); | |
2027 | } | |
2028 | ||
2029 | static void __process_bio_read_only(struct thin_c *tc, struct bio *bio, | |
2030 | struct dm_bio_prison_cell *cell) | |
e49e5829 JT |
2031 | { |
2032 | int r; | |
2033 | int rw = bio_data_dir(bio); | |
2034 | dm_block_t block = get_bio_block(tc, bio); | |
2035 | struct dm_thin_lookup_result lookup_result; | |
2036 | ||
2037 | r = dm_thin_find_block(tc->td, block, 1, &lookup_result); | |
2038 | switch (r) { | |
2039 | case 0: | |
a374bb21 | 2040 | if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) { |
8c0f0e8c | 2041 | handle_unserviceable_bio(tc->pool, bio); |
a374bb21 JT |
2042 | if (cell) |
2043 | cell_defer_no_holder(tc, cell); | |
2044 | } else { | |
e8088073 | 2045 | inc_all_io_entry(tc->pool, bio); |
e49e5829 | 2046 | remap_and_issue(tc, bio, lookup_result.block); |
a374bb21 JT |
2047 | if (cell) |
2048 | inc_remap_and_issue_cell(tc, cell, lookup_result.block); | |
e8088073 | 2049 | } |
e49e5829 JT |
2050 | break; |
2051 | ||
2052 | case -ENODATA: | |
a374bb21 JT |
2053 | if (cell) |
2054 | cell_defer_no_holder(tc, cell); | |
e49e5829 | 2055 | if (rw != READ) { |
8c0f0e8c | 2056 | handle_unserviceable_bio(tc->pool, bio); |
e49e5829 JT |
2057 | break; |
2058 | } | |
2059 | ||
2060 | if (tc->origin_dev) { | |
e8088073 | 2061 | inc_all_io_entry(tc->pool, bio); |
e49e5829 JT |
2062 | remap_to_origin_and_issue(tc, bio); |
2063 | break; | |
2064 | } | |
2065 | ||
2066 | zero_fill_bio(bio); | |
4246a0b6 | 2067 | bio_endio(bio); |
e49e5829 JT |
2068 | break; |
2069 | ||
2070 | default: | |
c397741c MS |
2071 | DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d", |
2072 | __func__, r); | |
a374bb21 JT |
2073 | if (cell) |
2074 | cell_defer_no_holder(tc, cell); | |
e49e5829 JT |
2075 | bio_io_error(bio); |
2076 | break; | |
2077 | } | |
2078 | } | |
2079 | ||
a374bb21 JT |
2080 | static void process_bio_read_only(struct thin_c *tc, struct bio *bio) |
2081 | { | |
2082 | __process_bio_read_only(tc, bio, NULL); | |
2083 | } | |
2084 | ||
2085 | static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2086 | { | |
2087 | __process_bio_read_only(tc, cell->holder, cell); | |
2088 | } | |
2089 | ||
3e1a0699 JT |
2090 | static void process_bio_success(struct thin_c *tc, struct bio *bio) |
2091 | { | |
4246a0b6 | 2092 | bio_endio(bio); |
3e1a0699 JT |
2093 | } |
2094 | ||
e49e5829 JT |
2095 | static void process_bio_fail(struct thin_c *tc, struct bio *bio) |
2096 | { | |
2097 | bio_io_error(bio); | |
2098 | } | |
2099 | ||
a374bb21 JT |
2100 | static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2101 | { | |
2102 | cell_success(tc->pool, cell); | |
2103 | } | |
2104 | ||
2105 | static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell) | |
2106 | { | |
2107 | cell_error(tc->pool, cell); | |
2108 | } | |
2109 | ||
ac8c3f3d JT |
2110 | /* |
2111 | * FIXME: should we also commit due to size of transaction, measured in | |
2112 | * metadata blocks? | |
2113 | */ | |
905e51b3 JT |
2114 | static int need_commit_due_to_time(struct pool *pool) |
2115 | { | |
0f30af98 MS |
2116 | return !time_in_range(jiffies, pool->last_commit_jiffies, |
2117 | pool->last_commit_jiffies + COMMIT_PERIOD); | |
905e51b3 JT |
2118 | } |
2119 | ||
67324ea1 MS |
2120 | #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node) |
2121 | #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook)) | |
2122 | ||
2123 | static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio) | |
2124 | { | |
2125 | struct rb_node **rbp, *parent; | |
2126 | struct dm_thin_endio_hook *pbd; | |
2127 | sector_t bi_sector = bio->bi_iter.bi_sector; | |
2128 | ||
2129 | rbp = &tc->sort_bio_list.rb_node; | |
2130 | parent = NULL; | |
2131 | while (*rbp) { | |
2132 | parent = *rbp; | |
2133 | pbd = thin_pbd(parent); | |
2134 | ||
2135 | if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector) | |
2136 | rbp = &(*rbp)->rb_left; | |
2137 | else | |
2138 | rbp = &(*rbp)->rb_right; | |
2139 | } | |
2140 | ||
2141 | pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); | |
2142 | rb_link_node(&pbd->rb_node, parent, rbp); | |
2143 | rb_insert_color(&pbd->rb_node, &tc->sort_bio_list); | |
2144 | } | |
2145 | ||
2146 | static void __extract_sorted_bios(struct thin_c *tc) | |
2147 | { | |
2148 | struct rb_node *node; | |
2149 | struct dm_thin_endio_hook *pbd; | |
2150 | struct bio *bio; | |
2151 | ||
2152 | for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) { | |
2153 | pbd = thin_pbd(node); | |
2154 | bio = thin_bio(pbd); | |
2155 | ||
2156 | bio_list_add(&tc->deferred_bio_list, bio); | |
2157 | rb_erase(&pbd->rb_node, &tc->sort_bio_list); | |
2158 | } | |
2159 | ||
2160 | WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list)); | |
2161 | } | |
2162 | ||
2163 | static void __sort_thin_deferred_bios(struct thin_c *tc) | |
2164 | { | |
2165 | struct bio *bio; | |
2166 | struct bio_list bios; | |
2167 | ||
2168 | bio_list_init(&bios); | |
2169 | bio_list_merge(&bios, &tc->deferred_bio_list); | |
2170 | bio_list_init(&tc->deferred_bio_list); | |
2171 | ||
2172 | /* Sort deferred_bio_list using rb-tree */ | |
2173 | while ((bio = bio_list_pop(&bios))) | |
2174 | __thin_bio_rb_add(tc, bio); | |
2175 | ||
2176 | /* | |
2177 | * Transfer the sorted bios in sort_bio_list back to | |
2178 | * deferred_bio_list to allow lockless submission of | |
2179 | * all bios. | |
2180 | */ | |
2181 | __extract_sorted_bios(tc); | |
2182 | } | |
2183 | ||
c140e1c4 | 2184 | static void process_thin_deferred_bios(struct thin_c *tc) |
991d9fa0 | 2185 | { |
c140e1c4 | 2186 | struct pool *pool = tc->pool; |
991d9fa0 JT |
2187 | struct bio *bio; |
2188 | struct bio_list bios; | |
67324ea1 | 2189 | struct blk_plug plug; |
86a3238c | 2190 | unsigned int count = 0; |
991d9fa0 | 2191 | |
c140e1c4 | 2192 | if (tc->requeue_mode) { |
4e4cbee9 CH |
2193 | error_thin_bio_list(tc, &tc->deferred_bio_list, |
2194 | BLK_STS_DM_REQUEUE); | |
c140e1c4 MS |
2195 | return; |
2196 | } | |
2197 | ||
991d9fa0 JT |
2198 | bio_list_init(&bios); |
2199 | ||
8e0c9dac | 2200 | spin_lock_irq(&tc->lock); |
67324ea1 MS |
2201 | |
2202 | if (bio_list_empty(&tc->deferred_bio_list)) { | |
8e0c9dac | 2203 | spin_unlock_irq(&tc->lock); |
67324ea1 MS |
2204 | return; |
2205 | } | |
2206 | ||
2207 | __sort_thin_deferred_bios(tc); | |
2208 | ||
c140e1c4 MS |
2209 | bio_list_merge(&bios, &tc->deferred_bio_list); |
2210 | bio_list_init(&tc->deferred_bio_list); | |
67324ea1 | 2211 | |
8e0c9dac | 2212 | spin_unlock_irq(&tc->lock); |
991d9fa0 | 2213 | |
67324ea1 | 2214 | blk_start_plug(&plug); |
991d9fa0 | 2215 | while ((bio = bio_list_pop(&bios))) { |
991d9fa0 JT |
2216 | /* |
2217 | * If we've got no free new_mapping structs, and processing | |
2218 | * this bio might require one, we pause until there are some | |
2219 | * prepared mappings to process. | |
2220 | */ | |
2221 | if (ensure_next_mapping(pool)) { | |
8e0c9dac | 2222 | spin_lock_irq(&tc->lock); |
c140e1c4 MS |
2223 | bio_list_add(&tc->deferred_bio_list, bio); |
2224 | bio_list_merge(&tc->deferred_bio_list, &bios); | |
8e0c9dac | 2225 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
2226 | break; |
2227 | } | |
104655fd | 2228 | |
e6047149 | 2229 | if (bio_op(bio) == REQ_OP_DISCARD) |
e49e5829 | 2230 | pool->process_discard(tc, bio); |
104655fd | 2231 | else |
e49e5829 | 2232 | pool->process_bio(tc, bio); |
8a01a6af JT |
2233 | |
2234 | if ((count++ & 127) == 0) { | |
7d327fe0 | 2235 | throttle_work_update(&pool->throttle); |
8a01a6af JT |
2236 | dm_pool_issue_prefetches(pool->pmd); |
2237 | } | |
e4f80303 | 2238 | cond_resched(); |
991d9fa0 | 2239 | } |
67324ea1 | 2240 | blk_finish_plug(&plug); |
c140e1c4 MS |
2241 | } |
2242 | ||
ac4c3f34 JT |
2243 | static int cmp_cells(const void *lhs, const void *rhs) |
2244 | { | |
2245 | struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs); | |
2246 | struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs); | |
2247 | ||
2248 | BUG_ON(!lhs_cell->holder); | |
2249 | BUG_ON(!rhs_cell->holder); | |
2250 | ||
2251 | if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector) | |
2252 | return -1; | |
2253 | ||
2254 | if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector) | |
2255 | return 1; | |
2256 | ||
2257 | return 0; | |
2258 | } | |
2259 | ||
86a3238c | 2260 | static unsigned int sort_cells(struct pool *pool, struct list_head *cells) |
ac4c3f34 | 2261 | { |
86a3238c | 2262 | unsigned int count = 0; |
ac4c3f34 JT |
2263 | struct dm_bio_prison_cell *cell, *tmp; |
2264 | ||
2265 | list_for_each_entry_safe(cell, tmp, cells, user_list) { | |
2266 | if (count >= CELL_SORT_ARRAY_SIZE) | |
2267 | break; | |
2268 | ||
2269 | pool->cell_sort_array[count++] = cell; | |
2270 | list_del(&cell->user_list); | |
2271 | } | |
2272 | ||
2273 | sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL); | |
2274 | ||
2275 | return count; | |
2276 | } | |
2277 | ||
a374bb21 JT |
2278 | static void process_thin_deferred_cells(struct thin_c *tc) |
2279 | { | |
2280 | struct pool *pool = tc->pool; | |
a374bb21 | 2281 | struct list_head cells; |
ac4c3f34 | 2282 | struct dm_bio_prison_cell *cell; |
86a3238c | 2283 | unsigned int i, j, count; |
a374bb21 JT |
2284 | |
2285 | INIT_LIST_HEAD(&cells); | |
2286 | ||
8e0c9dac | 2287 | spin_lock_irq(&tc->lock); |
a374bb21 | 2288 | list_splice_init(&tc->deferred_cells, &cells); |
8e0c9dac | 2289 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
2290 | |
2291 | if (list_empty(&cells)) | |
2292 | return; | |
2293 | ||
ac4c3f34 JT |
2294 | do { |
2295 | count = sort_cells(tc->pool, &cells); | |
a374bb21 | 2296 | |
ac4c3f34 JT |
2297 | for (i = 0; i < count; i++) { |
2298 | cell = pool->cell_sort_array[i]; | |
2299 | BUG_ON(!cell->holder); | |
a374bb21 | 2300 | |
ac4c3f34 JT |
2301 | /* |
2302 | * If we've got no free new_mapping structs, and processing | |
2303 | * this bio might require one, we pause until there are some | |
2304 | * prepared mappings to process. | |
2305 | */ | |
2306 | if (ensure_next_mapping(pool)) { | |
2307 | for (j = i; j < count; j++) | |
2308 | list_add(&pool->cell_sort_array[j]->user_list, &cells); | |
2309 | ||
8e0c9dac | 2310 | spin_lock_irq(&tc->lock); |
ac4c3f34 | 2311 | list_splice(&cells, &tc->deferred_cells); |
8e0c9dac | 2312 | spin_unlock_irq(&tc->lock); |
ac4c3f34 JT |
2313 | return; |
2314 | } | |
2315 | ||
e6047149 | 2316 | if (bio_op(cell->holder) == REQ_OP_DISCARD) |
ac4c3f34 JT |
2317 | pool->process_discard_cell(tc, cell); |
2318 | else | |
2319 | pool->process_cell(tc, cell); | |
2320 | } | |
e4f80303 | 2321 | cond_resched(); |
ac4c3f34 | 2322 | } while (!list_empty(&cells)); |
a374bb21 JT |
2323 | } |
2324 | ||
b10ebd34 JT |
2325 | static void thin_get(struct thin_c *tc); |
2326 | static void thin_put(struct thin_c *tc); | |
2327 | ||
2328 | /* | |
2329 | * We can't hold rcu_read_lock() around code that can block. So we | |
2330 | * find a thin with the rcu lock held; bump a refcount; then drop | |
2331 | * the lock. | |
2332 | */ | |
2333 | static struct thin_c *get_first_thin(struct pool *pool) | |
2334 | { | |
2335 | struct thin_c *tc = NULL; | |
2336 | ||
2337 | rcu_read_lock(); | |
2338 | if (!list_empty(&pool->active_thins)) { | |
2339 | tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list); | |
2340 | thin_get(tc); | |
2341 | } | |
2342 | rcu_read_unlock(); | |
2343 | ||
2344 | return tc; | |
2345 | } | |
2346 | ||
2347 | static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc) | |
2348 | { | |
2349 | struct thin_c *old_tc = tc; | |
2350 | ||
2351 | rcu_read_lock(); | |
2352 | list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) { | |
2353 | thin_get(tc); | |
2354 | thin_put(old_tc); | |
2355 | rcu_read_unlock(); | |
2356 | return tc; | |
2357 | } | |
2358 | thin_put(old_tc); | |
2359 | rcu_read_unlock(); | |
2360 | ||
2361 | return NULL; | |
2362 | } | |
2363 | ||
c140e1c4 MS |
2364 | static void process_deferred_bios(struct pool *pool) |
2365 | { | |
c140e1c4 | 2366 | struct bio *bio; |
4ae280b4 | 2367 | struct bio_list bios, bio_completions; |
c140e1c4 MS |
2368 | struct thin_c *tc; |
2369 | ||
b10ebd34 JT |
2370 | tc = get_first_thin(pool); |
2371 | while (tc) { | |
a374bb21 | 2372 | process_thin_deferred_cells(tc); |
c140e1c4 | 2373 | process_thin_deferred_bios(tc); |
b10ebd34 JT |
2374 | tc = get_next_thin(pool, tc); |
2375 | } | |
991d9fa0 JT |
2376 | |
2377 | /* | |
4ae280b4 NT |
2378 | * If there are any deferred flush bios, we must commit the metadata |
2379 | * before issuing them or signaling their completion. | |
991d9fa0 JT |
2380 | */ |
2381 | bio_list_init(&bios); | |
4ae280b4 NT |
2382 | bio_list_init(&bio_completions); |
2383 | ||
8e0c9dac | 2384 | spin_lock_irq(&pool->lock); |
991d9fa0 JT |
2385 | bio_list_merge(&bios, &pool->deferred_flush_bios); |
2386 | bio_list_init(&pool->deferred_flush_bios); | |
4ae280b4 NT |
2387 | |
2388 | bio_list_merge(&bio_completions, &pool->deferred_flush_completions); | |
2389 | bio_list_init(&pool->deferred_flush_completions); | |
8e0c9dac | 2390 | spin_unlock_irq(&pool->lock); |
991d9fa0 | 2391 | |
4ae280b4 | 2392 | if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) && |
4d1662a3 | 2393 | !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool))) |
991d9fa0 JT |
2394 | return; |
2395 | ||
020cc3b5 | 2396 | if (commit(pool)) { |
4ae280b4 NT |
2397 | bio_list_merge(&bios, &bio_completions); |
2398 | ||
991d9fa0 JT |
2399 | while ((bio = bio_list_pop(&bios))) |
2400 | bio_io_error(bio); | |
2401 | return; | |
2402 | } | |
905e51b3 | 2403 | pool->last_commit_jiffies = jiffies; |
991d9fa0 | 2404 | |
4ae280b4 NT |
2405 | while ((bio = bio_list_pop(&bio_completions))) |
2406 | bio_endio(bio); | |
2407 | ||
694cfe7f NT |
2408 | while ((bio = bio_list_pop(&bios))) { |
2409 | /* | |
2410 | * The data device was flushed as part of metadata commit, | |
2411 | * so complete redundant flushes immediately. | |
2412 | */ | |
2413 | if (bio->bi_opf & REQ_PREFLUSH) | |
2414 | bio_endio(bio); | |
2415 | else | |
b7f8dff0 | 2416 | dm_submit_bio_remap(bio, NULL); |
694cfe7f | 2417 | } |
991d9fa0 JT |
2418 | } |
2419 | ||
2420 | static void do_worker(struct work_struct *ws) | |
2421 | { | |
2422 | struct pool *pool = container_of(ws, struct pool, worker); | |
2423 | ||
7d327fe0 | 2424 | throttle_work_start(&pool->throttle); |
8a01a6af | 2425 | dm_pool_issue_prefetches(pool->pmd); |
7d327fe0 | 2426 | throttle_work_update(&pool->throttle); |
e49e5829 | 2427 | process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping); |
7d327fe0 | 2428 | throttle_work_update(&pool->throttle); |
e49e5829 | 2429 | process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard); |
7d327fe0 | 2430 | throttle_work_update(&pool->throttle); |
2a0fbffb JT |
2431 | process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2); |
2432 | throttle_work_update(&pool->throttle); | |
991d9fa0 | 2433 | process_deferred_bios(pool); |
7d327fe0 | 2434 | throttle_work_complete(&pool->throttle); |
991d9fa0 JT |
2435 | } |
2436 | ||
905e51b3 JT |
2437 | /* |
2438 | * We want to commit periodically so that not too much | |
2439 | * unwritten data builds up. | |
2440 | */ | |
2441 | static void do_waker(struct work_struct *ws) | |
2442 | { | |
2443 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker); | |
0ef0b471 | 2444 | |
905e51b3 JT |
2445 | wake_worker(pool); |
2446 | queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD); | |
2447 | } | |
2448 | ||
85ad643b JT |
2449 | /* |
2450 | * We're holding onto IO to allow userland time to react. After the | |
2451 | * timeout either the pool will have been resized (and thus back in | |
bcc696fa | 2452 | * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space. |
85ad643b JT |
2453 | */ |
2454 | static void do_no_space_timeout(struct work_struct *ws) | |
2455 | { | |
2456 | struct pool *pool = container_of(to_delayed_work(ws), struct pool, | |
2457 | no_space_timeout); | |
2458 | ||
bcc696fa MS |
2459 | if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) { |
2460 | pool->pf.error_if_no_space = true; | |
f6c36758 | 2461 | notify_of_pool_mode_change(pool); |
4e4cbee9 | 2462 | error_retry_list_with_code(pool, BLK_STS_NOSPC); |
bcc696fa | 2463 | } |
85ad643b JT |
2464 | } |
2465 | ||
991d9fa0 JT |
2466 | /*----------------------------------------------------------------*/ |
2467 | ||
e7a3e871 | 2468 | struct pool_work { |
738211f7 | 2469 | struct work_struct worker; |
e7a3e871 JT |
2470 | struct completion complete; |
2471 | }; | |
2472 | ||
2473 | static struct pool_work *to_pool_work(struct work_struct *ws) | |
2474 | { | |
2475 | return container_of(ws, struct pool_work, worker); | |
2476 | } | |
2477 | ||
2478 | static void pool_work_complete(struct pool_work *pw) | |
2479 | { | |
2480 | complete(&pw->complete); | |
2481 | } | |
738211f7 | 2482 | |
e7a3e871 JT |
2483 | static void pool_work_wait(struct pool_work *pw, struct pool *pool, |
2484 | void (*fn)(struct work_struct *)) | |
2485 | { | |
2486 | INIT_WORK_ONSTACK(&pw->worker, fn); | |
2487 | init_completion(&pw->complete); | |
2488 | queue_work(pool->wq, &pw->worker); | |
2489 | wait_for_completion(&pw->complete); | |
2490 | } | |
2491 | ||
2492 | /*----------------------------------------------------------------*/ | |
2493 | ||
2494 | struct noflush_work { | |
2495 | struct pool_work pw; | |
2496 | struct thin_c *tc; | |
738211f7 JT |
2497 | }; |
2498 | ||
e7a3e871 | 2499 | static struct noflush_work *to_noflush(struct work_struct *ws) |
738211f7 | 2500 | { |
e7a3e871 | 2501 | return container_of(to_pool_work(ws), struct noflush_work, pw); |
738211f7 JT |
2502 | } |
2503 | ||
2504 | static void do_noflush_start(struct work_struct *ws) | |
2505 | { | |
e7a3e871 | 2506 | struct noflush_work *w = to_noflush(ws); |
0ef0b471 | 2507 | |
738211f7 JT |
2508 | w->tc->requeue_mode = true; |
2509 | requeue_io(w->tc); | |
e7a3e871 | 2510 | pool_work_complete(&w->pw); |
738211f7 JT |
2511 | } |
2512 | ||
2513 | static void do_noflush_stop(struct work_struct *ws) | |
2514 | { | |
e7a3e871 | 2515 | struct noflush_work *w = to_noflush(ws); |
0ef0b471 | 2516 | |
738211f7 | 2517 | w->tc->requeue_mode = false; |
e7a3e871 | 2518 | pool_work_complete(&w->pw); |
738211f7 JT |
2519 | } |
2520 | ||
2521 | static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *)) | |
2522 | { | |
2523 | struct noflush_work w; | |
2524 | ||
738211f7 | 2525 | w.tc = tc; |
e7a3e871 | 2526 | pool_work_wait(&w.pw, tc->pool, fn); |
738211f7 JT |
2527 | } |
2528 | ||
2529 | /*----------------------------------------------------------------*/ | |
2530 | ||
34fbcf62 JT |
2531 | static bool passdown_enabled(struct pool_c *pt) |
2532 | { | |
2533 | return pt->adjusted_pf.discard_passdown; | |
2534 | } | |
2535 | ||
2536 | static void set_discard_callbacks(struct pool *pool) | |
2537 | { | |
2538 | struct pool_c *pt = pool->ti->private; | |
2539 | ||
2540 | if (passdown_enabled(pt)) { | |
2541 | pool->process_discard_cell = process_discard_cell_passdown; | |
2a0fbffb JT |
2542 | pool->process_prepared_discard = process_prepared_discard_passdown_pt1; |
2543 | pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2; | |
34fbcf62 JT |
2544 | } else { |
2545 | pool->process_discard_cell = process_discard_cell_no_passdown; | |
2546 | pool->process_prepared_discard = process_prepared_discard_no_passdown; | |
2547 | } | |
2548 | } | |
2549 | ||
8b64e881 | 2550 | static void set_pool_mode(struct pool *pool, enum pool_mode new_mode) |
e49e5829 | 2551 | { |
cdc2b415 | 2552 | struct pool_c *pt = pool->ti->private; |
07f2b6e0 MS |
2553 | bool needs_check = dm_pool_metadata_needs_check(pool->pmd); |
2554 | enum pool_mode old_mode = get_pool_mode(pool); | |
6aa7de05 | 2555 | unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ; |
07f2b6e0 MS |
2556 | |
2557 | /* | |
2558 | * Never allow the pool to transition to PM_WRITE mode if user | |
2559 | * intervention is required to verify metadata and data consistency. | |
2560 | */ | |
2561 | if (new_mode == PM_WRITE && needs_check) { | |
2562 | DMERR("%s: unable to switch pool to write mode until repaired.", | |
2563 | dm_device_name(pool->pool_md)); | |
2564 | if (old_mode != new_mode) | |
2565 | new_mode = old_mode; | |
2566 | else | |
2567 | new_mode = PM_READ_ONLY; | |
2568 | } | |
2569 | /* | |
2570 | * If we were in PM_FAIL mode, rollback of metadata failed. We're | |
2571 | * not going to recover without a thin_repair. So we never let the | |
2572 | * pool move out of the old mode. | |
2573 | */ | |
2574 | if (old_mode == PM_FAIL) | |
2575 | new_mode = old_mode; | |
e49e5829 | 2576 | |
8b64e881 | 2577 | switch (new_mode) { |
e49e5829 | 2578 | case PM_FAIL: |
5383ef3a | 2579 | dm_pool_metadata_read_only(pool->pmd); |
e49e5829 JT |
2580 | pool->process_bio = process_bio_fail; |
2581 | pool->process_discard = process_bio_fail; | |
a374bb21 JT |
2582 | pool->process_cell = process_cell_fail; |
2583 | pool->process_discard_cell = process_cell_fail; | |
e49e5829 JT |
2584 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
2585 | pool->process_prepared_discard = process_prepared_discard_fail; | |
3e1a0699 JT |
2586 | |
2587 | error_retry_list(pool); | |
e49e5829 JT |
2588 | break; |
2589 | ||
3ab91828 | 2590 | case PM_OUT_OF_METADATA_SPACE: |
e49e5829 | 2591 | case PM_READ_ONLY: |
3e1a0699 JT |
2592 | dm_pool_metadata_read_only(pool->pmd); |
2593 | pool->process_bio = process_bio_read_only; | |
2594 | pool->process_discard = process_bio_success; | |
a374bb21 JT |
2595 | pool->process_cell = process_cell_read_only; |
2596 | pool->process_discard_cell = process_cell_success; | |
3e1a0699 | 2597 | pool->process_prepared_mapping = process_prepared_mapping_fail; |
34fbcf62 | 2598 | pool->process_prepared_discard = process_prepared_discard_success; |
3e1a0699 JT |
2599 | |
2600 | error_retry_list(pool); | |
2601 | break; | |
2602 | ||
2603 | case PM_OUT_OF_DATA_SPACE: | |
2604 | /* | |
2605 | * Ideally we'd never hit this state; the low water mark | |
2606 | * would trigger userland to extend the pool before we | |
2607 | * completely run out of data space. However, many small | |
2608 | * IOs to unprovisioned space can consume data space at an | |
2609 | * alarming rate. Adjust your low water mark if you're | |
2610 | * frequently seeing this mode. | |
2611 | */ | |
c3667cc6 | 2612 | pool->out_of_data_space = true; |
3e1a0699 | 2613 | pool->process_bio = process_bio_read_only; |
a374bb21 JT |
2614 | pool->process_discard = process_discard_bio; |
2615 | pool->process_cell = process_cell_read_only; | |
3e1a0699 | 2616 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2617 | set_discard_callbacks(pool); |
85ad643b | 2618 | |
80c57893 MS |
2619 | if (!pool->pf.error_if_no_space && no_space_timeout) |
2620 | queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout); | |
e49e5829 JT |
2621 | break; |
2622 | ||
2623 | case PM_WRITE: | |
75294442 HT |
2624 | if (old_mode == PM_OUT_OF_DATA_SPACE) |
2625 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
c3667cc6 | 2626 | pool->out_of_data_space = false; |
172c2386 | 2627 | pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space; |
9b7aaa64 | 2628 | dm_pool_metadata_read_write(pool->pmd); |
e49e5829 | 2629 | pool->process_bio = process_bio; |
a374bb21 JT |
2630 | pool->process_discard = process_discard_bio; |
2631 | pool->process_cell = process_cell; | |
e49e5829 | 2632 | pool->process_prepared_mapping = process_prepared_mapping; |
34fbcf62 | 2633 | set_discard_callbacks(pool); |
e49e5829 JT |
2634 | break; |
2635 | } | |
8b64e881 MS |
2636 | |
2637 | pool->pf.mode = new_mode; | |
cdc2b415 MS |
2638 | /* |
2639 | * The pool mode may have changed, sync it so bind_control_target() | |
2640 | * doesn't cause an unexpected mode transition on resume. | |
2641 | */ | |
2642 | pt->adjusted_pf.mode = new_mode; | |
f6c36758 MS |
2643 | |
2644 | if (old_mode != new_mode) | |
2645 | notify_of_pool_mode_change(pool); | |
e49e5829 JT |
2646 | } |
2647 | ||
07f2b6e0 | 2648 | static void abort_transaction(struct pool *pool) |
b5330655 | 2649 | { |
07f2b6e0 MS |
2650 | const char *dev_name = dm_device_name(pool->pool_md); |
2651 | ||
2652 | DMERR_LIMIT("%s: aborting current metadata transaction", dev_name); | |
2653 | if (dm_pool_abort_metadata(pool->pmd)) { | |
2654 | DMERR("%s: failed to abort metadata transaction", dev_name); | |
2655 | set_pool_mode(pool, PM_FAIL); | |
2656 | } | |
2657 | ||
2658 | if (dm_pool_metadata_set_needs_check(pool->pmd)) { | |
2659 | DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name); | |
2660 | set_pool_mode(pool, PM_FAIL); | |
2661 | } | |
2662 | } | |
399caddf | 2663 | |
07f2b6e0 MS |
2664 | static void metadata_operation_failed(struct pool *pool, const char *op, int r) |
2665 | { | |
b5330655 JT |
2666 | DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d", |
2667 | dm_device_name(pool->pool_md), op, r); | |
2668 | ||
07f2b6e0 | 2669 | abort_transaction(pool); |
b5330655 JT |
2670 | set_pool_mode(pool, PM_READ_ONLY); |
2671 | } | |
2672 | ||
e49e5829 JT |
2673 | /*----------------------------------------------------------------*/ |
2674 | ||
991d9fa0 JT |
2675 | /* |
2676 | * Mapping functions. | |
2677 | */ | |
2678 | ||
2679 | /* | |
2680 | * Called only while mapping a thin bio to hand it over to the workqueue. | |
2681 | */ | |
2682 | static void thin_defer_bio(struct thin_c *tc, struct bio *bio) | |
2683 | { | |
991d9fa0 JT |
2684 | struct pool *pool = tc->pool; |
2685 | ||
8e0c9dac | 2686 | spin_lock_irq(&tc->lock); |
c140e1c4 | 2687 | bio_list_add(&tc->deferred_bio_list, bio); |
8e0c9dac | 2688 | spin_unlock_irq(&tc->lock); |
991d9fa0 JT |
2689 | |
2690 | wake_worker(pool); | |
2691 | } | |
2692 | ||
7d327fe0 JT |
2693 | static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio) |
2694 | { | |
2695 | struct pool *pool = tc->pool; | |
2696 | ||
2697 | throttle_lock(&pool->throttle); | |
2698 | thin_defer_bio(tc, bio); | |
2699 | throttle_unlock(&pool->throttle); | |
2700 | } | |
2701 | ||
a374bb21 JT |
2702 | static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell) |
2703 | { | |
a374bb21 JT |
2704 | struct pool *pool = tc->pool; |
2705 | ||
2706 | throttle_lock(&pool->throttle); | |
8e0c9dac | 2707 | spin_lock_irq(&tc->lock); |
a374bb21 | 2708 | list_add_tail(&cell->user_list, &tc->deferred_cells); |
8e0c9dac | 2709 | spin_unlock_irq(&tc->lock); |
a374bb21 JT |
2710 | throttle_unlock(&pool->throttle); |
2711 | ||
2712 | wake_worker(pool); | |
2713 | } | |
2714 | ||
59c3d2c6 | 2715 | static void thin_hook_bio(struct thin_c *tc, struct bio *bio) |
eb2aa48d | 2716 | { |
59c3d2c6 | 2717 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d JT |
2718 | |
2719 | h->tc = tc; | |
2720 | h->shared_read_entry = NULL; | |
e8088073 | 2721 | h->all_io_entry = NULL; |
eb2aa48d | 2722 | h->overwrite_mapping = NULL; |
34fbcf62 | 2723 | h->cell = NULL; |
eb2aa48d JT |
2724 | } |
2725 | ||
991d9fa0 JT |
2726 | /* |
2727 | * Non-blocking function called from the thin target's map function. | |
2728 | */ | |
7de3ee57 | 2729 | static int thin_bio_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 JT |
2730 | { |
2731 | int r; | |
2732 | struct thin_c *tc = ti->private; | |
2733 | dm_block_t block = get_bio_block(tc, bio); | |
2734 | struct dm_thin_device *td = tc->td; | |
2735 | struct dm_thin_lookup_result result; | |
a374bb21 | 2736 | struct dm_bio_prison_cell *virt_cell, *data_cell; |
e8088073 | 2737 | struct dm_cell_key key; |
991d9fa0 | 2738 | |
59c3d2c6 | 2739 | thin_hook_bio(tc, bio); |
e49e5829 | 2740 | |
738211f7 | 2741 | if (tc->requeue_mode) { |
4e4cbee9 | 2742 | bio->bi_status = BLK_STS_DM_REQUEUE; |
4246a0b6 | 2743 | bio_endio(bio); |
738211f7 JT |
2744 | return DM_MAPIO_SUBMITTED; |
2745 | } | |
2746 | ||
e49e5829 JT |
2747 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
2748 | bio_io_error(bio); | |
2749 | return DM_MAPIO_SUBMITTED; | |
2750 | } | |
2751 | ||
f73f44eb | 2752 | if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) { |
7d327fe0 | 2753 | thin_defer_bio_with_throttle(tc, bio); |
991d9fa0 JT |
2754 | return DM_MAPIO_SUBMITTED; |
2755 | } | |
2756 | ||
c822ed96 JT |
2757 | /* |
2758 | * We must hold the virtual cell before doing the lookup, otherwise | |
2759 | * there's a race with discard. | |
2760 | */ | |
2761 | build_virtual_key(tc->td, block, &key); | |
a374bb21 | 2762 | if (bio_detain(tc->pool, &key, bio, &virt_cell)) |
c822ed96 JT |
2763 | return DM_MAPIO_SUBMITTED; |
2764 | ||
991d9fa0 JT |
2765 | r = dm_thin_find_block(td, block, 0, &result); |
2766 | ||
2767 | /* | |
2768 | * Note that we defer readahead too. | |
2769 | */ | |
2770 | switch (r) { | |
2771 | case 0: | |
2772 | if (unlikely(result.shared)) { | |
2773 | /* | |
2774 | * We have a race condition here between the | |
2775 | * result.shared value returned by the lookup and | |
2776 | * snapshot creation, which may cause new | |
2777 | * sharing. | |
2778 | * | |
2779 | * To avoid this always quiesce the origin before | |
2780 | * taking the snap. You want to do this anyway to | |
2781 | * ensure a consistent application view | |
2782 | * (i.e. lockfs). | |
2783 | * | |
2784 | * More distant ancestors are irrelevant. The | |
2785 | * shared flag will be set in their case. | |
2786 | */ | |
a374bb21 | 2787 | thin_defer_cell(tc, virt_cell); |
e8088073 | 2788 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2789 | } |
e8088073 | 2790 | |
e8088073 | 2791 | build_data_key(tc->td, result.block, &key); |
a374bb21 JT |
2792 | if (bio_detain(tc->pool, &key, bio, &data_cell)) { |
2793 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2794 | return DM_MAPIO_SUBMITTED; |
2795 | } | |
2796 | ||
2797 | inc_all_io_entry(tc->pool, bio); | |
a374bb21 JT |
2798 | cell_defer_no_holder(tc, data_cell); |
2799 | cell_defer_no_holder(tc, virt_cell); | |
e8088073 JT |
2800 | |
2801 | remap(tc, bio, result.block); | |
2802 | return DM_MAPIO_REMAPPED; | |
991d9fa0 JT |
2803 | |
2804 | case -ENODATA: | |
e49e5829 | 2805 | case -EWOULDBLOCK: |
a374bb21 | 2806 | thin_defer_cell(tc, virt_cell); |
2aab3850 | 2807 | return DM_MAPIO_SUBMITTED; |
e49e5829 JT |
2808 | |
2809 | default: | |
2810 | /* | |
2811 | * Must always call bio_io_error on failure. | |
2812 | * dm_thin_find_block can fail with -EINVAL if the | |
2813 | * pool is switched to fail-io mode. | |
2814 | */ | |
2815 | bio_io_error(bio); | |
a374bb21 | 2816 | cell_defer_no_holder(tc, virt_cell); |
2aab3850 | 2817 | return DM_MAPIO_SUBMITTED; |
991d9fa0 | 2818 | } |
991d9fa0 JT |
2819 | } |
2820 | ||
c140e1c4 | 2821 | static void requeue_bios(struct pool *pool) |
991d9fa0 | 2822 | { |
c140e1c4 MS |
2823 | struct thin_c *tc; |
2824 | ||
2825 | rcu_read_lock(); | |
2826 | list_for_each_entry_rcu(tc, &pool->active_thins, list) { | |
8e0c9dac | 2827 | spin_lock_irq(&tc->lock); |
c140e1c4 MS |
2828 | bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list); |
2829 | bio_list_init(&tc->retry_on_resume_list); | |
8e0c9dac | 2830 | spin_unlock_irq(&tc->lock); |
c140e1c4 MS |
2831 | } |
2832 | rcu_read_unlock(); | |
991d9fa0 JT |
2833 | } |
2834 | ||
a4a82ce3 HM |
2835 | /* |
2836 | *-------------------------------------------------------------- | |
991d9fa0 | 2837 | * Binding of control targets to a pool object |
a4a82ce3 HM |
2838 | *-------------------------------------------------------------- |
2839 | */ | |
58051b94 JT |
2840 | static bool is_factor(sector_t block_size, uint32_t n) |
2841 | { | |
2842 | return !sector_div(block_size, n); | |
2843 | } | |
2844 | ||
9bc142dd MS |
2845 | /* |
2846 | * If discard_passdown was enabled verify that the data device | |
0424caa1 | 2847 | * supports discards. Disable discard_passdown if not. |
9bc142dd | 2848 | */ |
0424caa1 | 2849 | static void disable_passdown_if_not_supported(struct pool_c *pt) |
9bc142dd | 2850 | { |
0424caa1 MS |
2851 | struct pool *pool = pt->pool; |
2852 | struct block_device *data_bdev = pt->data_dev->bdev; | |
2853 | struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits; | |
0424caa1 | 2854 | const char *reason = NULL; |
9bc142dd | 2855 | |
0424caa1 | 2856 | if (!pt->adjusted_pf.discard_passdown) |
9bc142dd MS |
2857 | return; |
2858 | ||
70200574 | 2859 | if (!bdev_max_discard_sectors(pt->data_dev->bdev)) |
0424caa1 MS |
2860 | reason = "discard unsupported"; |
2861 | ||
2862 | else if (data_limits->max_discard_sectors < pool->sectors_per_block) | |
2863 | reason = "max discard sectors smaller than a block"; | |
9bc142dd | 2864 | |
0424caa1 | 2865 | if (reason) { |
385411ff | 2866 | DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason); |
0424caa1 MS |
2867 | pt->adjusted_pf.discard_passdown = false; |
2868 | } | |
9bc142dd MS |
2869 | } |
2870 | ||
991d9fa0 JT |
2871 | static int bind_control_target(struct pool *pool, struct dm_target *ti) |
2872 | { | |
2873 | struct pool_c *pt = ti->private; | |
2874 | ||
e49e5829 | 2875 | /* |
9b7aaa64 | 2876 | * We want to make sure that a pool in PM_FAIL mode is never upgraded. |
e49e5829 | 2877 | */ |
07f2b6e0 | 2878 | enum pool_mode old_mode = get_pool_mode(pool); |
0424caa1 | 2879 | enum pool_mode new_mode = pt->adjusted_pf.mode; |
e49e5829 | 2880 | |
8b64e881 MS |
2881 | /* |
2882 | * Don't change the pool's mode until set_pool_mode() below. | |
2883 | * Otherwise the pool's process_* function pointers may | |
2884 | * not match the desired pool mode. | |
2885 | */ | |
2886 | pt->adjusted_pf.mode = old_mode; | |
2887 | ||
2888 | pool->ti = ti; | |
2889 | pool->pf = pt->adjusted_pf; | |
2890 | pool->low_water_blocks = pt->low_water_blocks; | |
2891 | ||
9bc142dd | 2892 | set_pool_mode(pool, new_mode); |
f402693d | 2893 | |
991d9fa0 JT |
2894 | return 0; |
2895 | } | |
2896 | ||
2897 | static void unbind_control_target(struct pool *pool, struct dm_target *ti) | |
2898 | { | |
2899 | if (pool->ti == ti) | |
2900 | pool->ti = NULL; | |
2901 | } | |
2902 | ||
a4a82ce3 HM |
2903 | /* |
2904 | *-------------------------------------------------------------- | |
991d9fa0 | 2905 | * Pool creation |
a4a82ce3 HM |
2906 | *-------------------------------------------------------------- |
2907 | */ | |
67e2e2b2 JT |
2908 | /* Initialize pool features. */ |
2909 | static void pool_features_init(struct pool_features *pf) | |
2910 | { | |
e49e5829 | 2911 | pf->mode = PM_WRITE; |
9bc142dd MS |
2912 | pf->zero_new_blocks = true; |
2913 | pf->discard_enabled = true; | |
2914 | pf->discard_passdown = true; | |
787a996c | 2915 | pf->error_if_no_space = false; |
67e2e2b2 JT |
2916 | } |
2917 | ||
991d9fa0 JT |
2918 | static void __pool_destroy(struct pool *pool) |
2919 | { | |
2920 | __pool_table_remove(pool); | |
2921 | ||
a822c83e | 2922 | vfree(pool->cell_sort_array); |
991d9fa0 JT |
2923 | if (dm_pool_metadata_close(pool->pmd) < 0) |
2924 | DMWARN("%s: dm_pool_metadata_close() failed.", __func__); | |
2925 | ||
44feb387 | 2926 | dm_bio_prison_destroy(pool->prison); |
991d9fa0 JT |
2927 | dm_kcopyd_client_destroy(pool->copier); |
2928 | ||
88430ebc LM |
2929 | cancel_delayed_work_sync(&pool->waker); |
2930 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
991d9fa0 JT |
2931 | if (pool->wq) |
2932 | destroy_workqueue(pool->wq); | |
2933 | ||
2934 | if (pool->next_mapping) | |
6f1c819c KO |
2935 | mempool_free(pool->next_mapping, &pool->mapping_pool); |
2936 | mempool_exit(&pool->mapping_pool); | |
44feb387 MS |
2937 | dm_deferred_set_destroy(pool->shared_read_ds); |
2938 | dm_deferred_set_destroy(pool->all_io_ds); | |
991d9fa0 JT |
2939 | kfree(pool); |
2940 | } | |
2941 | ||
a24c2569 | 2942 | static struct kmem_cache *_new_mapping_cache; |
a24c2569 | 2943 | |
991d9fa0 JT |
2944 | static struct pool *pool_create(struct mapped_device *pool_md, |
2945 | struct block_device *metadata_dev, | |
873937e7 | 2946 | struct block_device *data_dev, |
e49e5829 JT |
2947 | unsigned long block_size, |
2948 | int read_only, char **error) | |
991d9fa0 JT |
2949 | { |
2950 | int r; | |
2951 | void *err_p; | |
2952 | struct pool *pool; | |
2953 | struct dm_pool_metadata *pmd; | |
e49e5829 | 2954 | bool format_device = read_only ? false : true; |
991d9fa0 | 2955 | |
e49e5829 | 2956 | pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device); |
991d9fa0 JT |
2957 | if (IS_ERR(pmd)) { |
2958 | *error = "Error creating metadata object"; | |
2959 | return (struct pool *)pmd; | |
2960 | } | |
2961 | ||
d3775354 | 2962 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
991d9fa0 JT |
2963 | if (!pool) { |
2964 | *error = "Error allocating memory for pool"; | |
2965 | err_p = ERR_PTR(-ENOMEM); | |
2966 | goto bad_pool; | |
2967 | } | |
2968 | ||
2969 | pool->pmd = pmd; | |
2970 | pool->sectors_per_block = block_size; | |
f9a8e0cd MP |
2971 | if (block_size & (block_size - 1)) |
2972 | pool->sectors_per_block_shift = -1; | |
2973 | else | |
2974 | pool->sectors_per_block_shift = __ffs(block_size); | |
991d9fa0 | 2975 | pool->low_water_blocks = 0; |
67e2e2b2 | 2976 | pool_features_init(&pool->pf); |
a195db2d | 2977 | pool->prison = dm_bio_prison_create(); |
991d9fa0 JT |
2978 | if (!pool->prison) { |
2979 | *error = "Error creating pool's bio prison"; | |
2980 | err_p = ERR_PTR(-ENOMEM); | |
2981 | goto bad_prison; | |
2982 | } | |
2983 | ||
df5d2e90 | 2984 | pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
991d9fa0 JT |
2985 | if (IS_ERR(pool->copier)) { |
2986 | r = PTR_ERR(pool->copier); | |
2987 | *error = "Error creating pool's kcopyd client"; | |
2988 | err_p = ERR_PTR(r); | |
2989 | goto bad_kcopyd_client; | |
2990 | } | |
2991 | ||
2992 | /* | |
2993 | * Create singlethreaded workqueue that will service all devices | |
2994 | * that use this metadata. | |
2995 | */ | |
2996 | pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
2997 | if (!pool->wq) { | |
2998 | *error = "Error creating pool's workqueue"; | |
2999 | err_p = ERR_PTR(-ENOMEM); | |
3000 | goto bad_wq; | |
3001 | } | |
3002 | ||
7d327fe0 | 3003 | throttle_init(&pool->throttle); |
991d9fa0 | 3004 | INIT_WORK(&pool->worker, do_worker); |
905e51b3 | 3005 | INIT_DELAYED_WORK(&pool->waker, do_waker); |
85ad643b | 3006 | INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout); |
991d9fa0 | 3007 | spin_lock_init(&pool->lock); |
991d9fa0 | 3008 | bio_list_init(&pool->deferred_flush_bios); |
4ae280b4 | 3009 | bio_list_init(&pool->deferred_flush_completions); |
991d9fa0 | 3010 | INIT_LIST_HEAD(&pool->prepared_mappings); |
104655fd | 3011 | INIT_LIST_HEAD(&pool->prepared_discards); |
2a0fbffb | 3012 | INIT_LIST_HEAD(&pool->prepared_discards_pt2); |
c140e1c4 | 3013 | INIT_LIST_HEAD(&pool->active_thins); |
88a6621b | 3014 | pool->low_water_triggered = false; |
80e96c54 | 3015 | pool->suspended = true; |
c3667cc6 | 3016 | pool->out_of_data_space = false; |
44feb387 MS |
3017 | |
3018 | pool->shared_read_ds = dm_deferred_set_create(); | |
3019 | if (!pool->shared_read_ds) { | |
3020 | *error = "Error creating pool's shared read deferred set"; | |
3021 | err_p = ERR_PTR(-ENOMEM); | |
3022 | goto bad_shared_read_ds; | |
3023 | } | |
3024 | ||
3025 | pool->all_io_ds = dm_deferred_set_create(); | |
3026 | if (!pool->all_io_ds) { | |
3027 | *error = "Error creating pool's all io deferred set"; | |
3028 | err_p = ERR_PTR(-ENOMEM); | |
3029 | goto bad_all_io_ds; | |
3030 | } | |
991d9fa0 JT |
3031 | |
3032 | pool->next_mapping = NULL; | |
6f1c819c KO |
3033 | r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE, |
3034 | _new_mapping_cache); | |
3035 | if (r) { | |
991d9fa0 | 3036 | *error = "Error creating pool's mapping mempool"; |
6f1c819c | 3037 | err_p = ERR_PTR(r); |
991d9fa0 JT |
3038 | goto bad_mapping_pool; |
3039 | } | |
3040 | ||
42bc47b3 KC |
3041 | pool->cell_sort_array = |
3042 | vmalloc(array_size(CELL_SORT_ARRAY_SIZE, | |
3043 | sizeof(*pool->cell_sort_array))); | |
a822c83e JT |
3044 | if (!pool->cell_sort_array) { |
3045 | *error = "Error allocating cell sort array"; | |
3046 | err_p = ERR_PTR(-ENOMEM); | |
3047 | goto bad_sort_array; | |
3048 | } | |
3049 | ||
991d9fa0 | 3050 | pool->ref_count = 1; |
905e51b3 | 3051 | pool->last_commit_jiffies = jiffies; |
991d9fa0 JT |
3052 | pool->pool_md = pool_md; |
3053 | pool->md_dev = metadata_dev; | |
873937e7 | 3054 | pool->data_dev = data_dev; |
991d9fa0 JT |
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, | |
873937e7 | 3096 | struct block_device *data_dev, |
e49e5829 JT |
3097 | unsigned long block_size, int read_only, |
3098 | char **error, int *created) | |
991d9fa0 JT |
3099 | { |
3100 | struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev); | |
3101 | ||
3102 | if (pool) { | |
f09996c9 MS |
3103 | if (pool->pool_md != pool_md) { |
3104 | *error = "metadata device already in use by a pool"; | |
991d9fa0 | 3105 | return ERR_PTR(-EBUSY); |
f09996c9 | 3106 | } |
873937e7 MP |
3107 | if (pool->data_dev != data_dev) { |
3108 | *error = "data device already in use by a pool"; | |
3109 | return ERR_PTR(-EBUSY); | |
3110 | } | |
991d9fa0 JT |
3111 | __pool_inc(pool); |
3112 | ||
3113 | } else { | |
3114 | pool = __pool_table_lookup(pool_md); | |
3115 | if (pool) { | |
873937e7 | 3116 | if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) { |
f09996c9 | 3117 | *error = "different pool cannot replace a pool"; |
991d9fa0 | 3118 | return ERR_PTR(-EINVAL); |
f09996c9 | 3119 | } |
991d9fa0 JT |
3120 | __pool_inc(pool); |
3121 | ||
67e2e2b2 | 3122 | } else { |
873937e7 | 3123 | pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error); |
67e2e2b2 JT |
3124 | *created = 1; |
3125 | } | |
991d9fa0 JT |
3126 | } |
3127 | ||
3128 | return pool; | |
3129 | } | |
3130 | ||
a4a82ce3 HM |
3131 | /* |
3132 | *-------------------------------------------------------------- | |
991d9fa0 | 3133 | * Pool target methods |
a4a82ce3 HM |
3134 | *-------------------------------------------------------------- |
3135 | */ | |
991d9fa0 JT |
3136 | static void pool_dtr(struct dm_target *ti) |
3137 | { | |
3138 | struct pool_c *pt = ti->private; | |
3139 | ||
3140 | mutex_lock(&dm_thin_pool_table.mutex); | |
3141 | ||
3142 | unbind_control_target(pt->pool, ti); | |
3143 | __pool_dec(pt->pool); | |
3144 | dm_put_device(ti, pt->metadata_dev); | |
3145 | dm_put_device(ti, pt->data_dev); | |
3146 | kfree(pt); | |
3147 | ||
3148 | mutex_unlock(&dm_thin_pool_table.mutex); | |
3149 | } | |
3150 | ||
991d9fa0 JT |
3151 | static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf, |
3152 | struct dm_target *ti) | |
3153 | { | |
3154 | int r; | |
86a3238c | 3155 | unsigned int argc; |
991d9fa0 JT |
3156 | const char *arg_name; |
3157 | ||
5916a22b | 3158 | static const struct dm_arg _args[] = { |
74aa45c3 | 3159 | {0, 4, "Invalid number of pool feature arguments"}, |
991d9fa0 JT |
3160 | }; |
3161 | ||
3162 | /* | |
3163 | * No feature arguments supplied. | |
3164 | */ | |
3165 | if (!as->argc) | |
3166 | return 0; | |
3167 | ||
3168 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
3169 | if (r) | |
3170 | return -EINVAL; | |
3171 | ||
3172 | while (argc && !r) { | |
3173 | arg_name = dm_shift_arg(as); | |
3174 | argc--; | |
3175 | ||
e49e5829 | 3176 | if (!strcasecmp(arg_name, "skip_block_zeroing")) |
9bc142dd | 3177 | pf->zero_new_blocks = false; |
e49e5829 JT |
3178 | |
3179 | else if (!strcasecmp(arg_name, "ignore_discard")) | |
9bc142dd | 3180 | pf->discard_enabled = false; |
e49e5829 JT |
3181 | |
3182 | else if (!strcasecmp(arg_name, "no_discard_passdown")) | |
9bc142dd | 3183 | pf->discard_passdown = false; |
991d9fa0 | 3184 | |
e49e5829 JT |
3185 | else if (!strcasecmp(arg_name, "read_only")) |
3186 | pf->mode = PM_READ_ONLY; | |
3187 | ||
787a996c MS |
3188 | else if (!strcasecmp(arg_name, "error_if_no_space")) |
3189 | pf->error_if_no_space = true; | |
3190 | ||
e49e5829 JT |
3191 | else { |
3192 | ti->error = "Unrecognised pool feature requested"; | |
3193 | r = -EINVAL; | |
3194 | break; | |
3195 | } | |
991d9fa0 JT |
3196 | } |
3197 | ||
3198 | return r; | |
3199 | } | |
3200 | ||
ac8c3f3d JT |
3201 | static void metadata_low_callback(void *context) |
3202 | { | |
3203 | struct pool *pool = context; | |
3204 | ||
3205 | DMWARN("%s: reached low water mark for metadata device: sending event.", | |
3206 | dm_device_name(pool->pool_md)); | |
3207 | ||
3208 | dm_table_event(pool->ti->table); | |
3209 | } | |
3210 | ||
694cfe7f NT |
3211 | /* |
3212 | * We need to flush the data device **before** committing the metadata. | |
3213 | * | |
3214 | * This ensures that the data blocks of any newly inserted mappings are | |
3215 | * properly written to non-volatile storage and won't be lost in case of a | |
3216 | * crash. | |
3217 | * | |
3218 | * Failure to do so can result in data corruption in the case of internal or | |
3219 | * external snapshots and in the case of newly provisioned blocks, when block | |
3220 | * zeroing is enabled. | |
3221 | */ | |
3222 | static int metadata_pre_commit_callback(void *context) | |
3223 | { | |
f06c03d1 | 3224 | struct pool *pool = context; |
694cfe7f | 3225 | |
28d7d128 | 3226 | return blkdev_issue_flush(pool->data_dev); |
694cfe7f NT |
3227 | } |
3228 | ||
7d48935e MS |
3229 | static sector_t get_dev_size(struct block_device *bdev) |
3230 | { | |
6dcbb52c | 3231 | return bdev_nr_sectors(bdev); |
7d48935e MS |
3232 | } |
3233 | ||
3234 | static void warn_if_metadata_device_too_big(struct block_device *bdev) | |
b17446df | 3235 | { |
7d48935e | 3236 | sector_t metadata_dev_size = get_dev_size(bdev); |
b17446df | 3237 | |
7d48935e | 3238 | if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) |
385411ff CH |
3239 | DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.", |
3240 | bdev, 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; | |
0ef0b471 | 3276 | |
ac8c3f3d JT |
3277 | return min((dm_block_t)1024ULL /* 4M */, quarter); |
3278 | } | |
3279 | ||
991d9fa0 JT |
3280 | /* |
3281 | * thin-pool <metadata dev> <data dev> | |
3282 | * <data block size (sectors)> | |
3283 | * <low water mark (blocks)> | |
3284 | * [<#feature args> [<arg>]*] | |
3285 | * | |
3286 | * Optional feature arguments are: | |
3287 | * skip_block_zeroing: skips the zeroing of newly-provisioned blocks. | |
67e2e2b2 JT |
3288 | * ignore_discard: disable discard |
3289 | * no_discard_passdown: don't pass discards down to the data device | |
787a996c MS |
3290 | * read_only: Don't allow any changes to be made to the pool metadata. |
3291 | * error_if_no_space: error IOs, instead of queueing, if no space. | |
991d9fa0 | 3292 | */ |
86a3238c | 3293 | static int pool_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
991d9fa0 | 3294 | { |
67e2e2b2 | 3295 | int r, pool_created = 0; |
991d9fa0 JT |
3296 | struct pool_c *pt; |
3297 | struct pool *pool; | |
3298 | struct pool_features pf; | |
3299 | struct dm_arg_set as; | |
3300 | struct dm_dev *data_dev; | |
3301 | unsigned long block_size; | |
3302 | dm_block_t low_water_blocks; | |
3303 | struct dm_dev *metadata_dev; | |
05bdb996 | 3304 | blk_mode_t metadata_mode; |
991d9fa0 JT |
3305 | |
3306 | /* | |
3307 | * FIXME Remove validation from scope of lock. | |
3308 | */ | |
3309 | mutex_lock(&dm_thin_pool_table.mutex); | |
3310 | ||
3311 | if (argc < 4) { | |
3312 | ti->error = "Invalid argument count"; | |
3313 | r = -EINVAL; | |
3314 | goto out_unlock; | |
3315 | } | |
5d0db96d | 3316 | |
991d9fa0 JT |
3317 | as.argc = argc; |
3318 | as.argv = argv; | |
3319 | ||
70de2cbd JCXF |
3320 | /* make sure metadata and data are different devices */ |
3321 | if (!strcmp(argv[0], argv[1])) { | |
3322 | ti->error = "Error setting metadata or data device"; | |
3323 | r = -EINVAL; | |
3324 | goto out_unlock; | |
3325 | } | |
3326 | ||
5d0db96d JT |
3327 | /* |
3328 | * Set default pool features. | |
3329 | */ | |
3330 | pool_features_init(&pf); | |
3331 | ||
3332 | dm_consume_args(&as, 4); | |
3333 | r = parse_pool_features(&as, &pf, ti); | |
3334 | if (r) | |
3335 | goto out_unlock; | |
3336 | ||
05bdb996 CH |
3337 | metadata_mode = BLK_OPEN_READ | |
3338 | ((pf.mode == PM_READ_ONLY) ? 0 : BLK_OPEN_WRITE); | |
5d0db96d | 3339 | r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev); |
991d9fa0 JT |
3340 | if (r) { |
3341 | ti->error = "Error opening metadata block device"; | |
3342 | goto out_unlock; | |
3343 | } | |
7d48935e | 3344 | warn_if_metadata_device_too_big(metadata_dev->bdev); |
991d9fa0 | 3345 | |
05bdb996 | 3346 | r = dm_get_device(ti, argv[1], BLK_OPEN_READ | BLK_OPEN_WRITE, &data_dev); |
991d9fa0 JT |
3347 | if (r) { |
3348 | ti->error = "Error getting data device"; | |
3349 | goto out_metadata; | |
3350 | } | |
3351 | ||
3352 | if (kstrtoul(argv[2], 10, &block_size) || !block_size || | |
3353 | block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS || | |
3354 | block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS || | |
55f2b8bd | 3355 | block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) { |
991d9fa0 JT |
3356 | ti->error = "Invalid block size"; |
3357 | r = -EINVAL; | |
3358 | goto out; | |
3359 | } | |
3360 | ||
3361 | if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) { | |
3362 | ti->error = "Invalid low water mark"; | |
3363 | r = -EINVAL; | |
3364 | goto out; | |
3365 | } | |
3366 | ||
991d9fa0 JT |
3367 | pt = kzalloc(sizeof(*pt), GFP_KERNEL); |
3368 | if (!pt) { | |
3369 | r = -ENOMEM; | |
3370 | goto out; | |
3371 | } | |
3372 | ||
873937e7 | 3373 | pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev, |
e49e5829 | 3374 | block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created); |
991d9fa0 JT |
3375 | if (IS_ERR(pool)) { |
3376 | r = PTR_ERR(pool); | |
3377 | goto out_free_pt; | |
3378 | } | |
3379 | ||
67e2e2b2 JT |
3380 | /* |
3381 | * 'pool_created' reflects whether this is the first table load. | |
3382 | * Top level discard support is not allowed to be changed after | |
3383 | * initial load. This would require a pool reload to trigger thin | |
3384 | * device changes. | |
3385 | */ | |
3386 | if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) { | |
3387 | ti->error = "Discard support cannot be disabled once enabled"; | |
3388 | r = -EINVAL; | |
3389 | goto out_flags_changed; | |
3390 | } | |
3391 | ||
991d9fa0 JT |
3392 | pt->pool = pool; |
3393 | pt->ti = ti; | |
3394 | pt->metadata_dev = metadata_dev; | |
3395 | pt->data_dev = data_dev; | |
3396 | pt->low_water_blocks = low_water_blocks; | |
0424caa1 | 3397 | pt->adjusted_pf = pt->requested_pf = pf; |
55a62eef | 3398 | ti->num_flush_bios = 1; |
9bbf5fee | 3399 | ti->limit_swap_bios = true; |
9bc142dd | 3400 | |
67e2e2b2 JT |
3401 | /* |
3402 | * Only need to enable discards if the pool should pass | |
3403 | * them down to the data device. The thin device's discard | |
3404 | * processing will cause mappings to be removed from the btree. | |
3405 | */ | |
3406 | if (pf.discard_enabled && pf.discard_passdown) { | |
55a62eef | 3407 | ti->num_discard_bios = 1; |
67e2e2b2 JT |
3408 | /* |
3409 | * Setting 'discards_supported' circumvents the normal | |
3410 | * stacking of discard limits (this keeps the pool and | |
3411 | * thin devices' discard limits consistent). | |
3412 | */ | |
0ac55489 | 3413 | ti->discards_supported = true; |
e2dd8aca | 3414 | ti->max_discard_granularity = true; |
67e2e2b2 | 3415 | } |
991d9fa0 JT |
3416 | ti->private = pt; |
3417 | ||
ac8c3f3d JT |
3418 | r = dm_pool_register_metadata_threshold(pt->pool->pmd, |
3419 | calc_metadata_threshold(pt), | |
3420 | metadata_low_callback, | |
3421 | pool); | |
3534e5a5 LM |
3422 | if (r) { |
3423 | ti->error = "Error registering metadata threshold"; | |
ba30670f | 3424 | goto out_flags_changed; |
3534e5a5 | 3425 | } |
ac8c3f3d | 3426 | |
f06c03d1 MP |
3427 | dm_pool_register_pre_commit_callback(pool->pmd, |
3428 | metadata_pre_commit_callback, pool); | |
3429 | ||
991d9fa0 JT |
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; | |
991d9fa0 JT |
3453 | |
3454 | /* | |
3455 | * As this is a singleton target, ti->begin is always zero. | |
3456 | */ | |
8e0c9dac | 3457 | spin_lock_irq(&pool->lock); |
74d46992 | 3458 | bio_set_dev(bio, pt->data_dev->bdev); |
991d9fa0 | 3459 | r = DM_MAPIO_REMAPPED; |
8e0c9dac | 3460 | spin_unlock_irq(&pool->lock); |
991d9fa0 JT |
3461 | |
3462 | return r; | |
3463 | } | |
3464 | ||
b17446df | 3465 | static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit) |
991d9fa0 JT |
3466 | { |
3467 | int r; | |
3468 | struct pool_c *pt = ti->private; | |
3469 | struct pool *pool = pt->pool; | |
55f2b8bd MS |
3470 | sector_t data_size = ti->len; |
3471 | dm_block_t sb_data_size; | |
991d9fa0 | 3472 | |
b17446df | 3473 | *need_commit = false; |
991d9fa0 | 3474 | |
55f2b8bd MS |
3475 | (void) sector_div(data_size, pool->sectors_per_block); |
3476 | ||
991d9fa0 JT |
3477 | r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size); |
3478 | if (r) { | |
4fa5971a MS |
3479 | DMERR("%s: failed to retrieve data device size", |
3480 | dm_device_name(pool->pool_md)); | |
991d9fa0 JT |
3481 | return r; |
3482 | } | |
3483 | ||
3484 | if (data_size < sb_data_size) { | |
4fa5971a MS |
3485 | DMERR("%s: pool target (%llu blocks) too small: expected %llu", |
3486 | dm_device_name(pool->pool_md), | |
55f2b8bd | 3487 | (unsigned long long)data_size, sb_data_size); |
991d9fa0 JT |
3488 | return -EINVAL; |
3489 | ||
3490 | } else if (data_size > sb_data_size) { | |
07f2b6e0 MS |
3491 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3492 | DMERR("%s: unable to grow the data device until repaired.", | |
3493 | dm_device_name(pool->pool_md)); | |
3494 | return 0; | |
3495 | } | |
3496 | ||
6f7f51d4 MS |
3497 | if (sb_data_size) |
3498 | DMINFO("%s: growing the data device from %llu to %llu blocks", | |
3499 | dm_device_name(pool->pool_md), | |
3500 | sb_data_size, (unsigned long long)data_size); | |
991d9fa0 JT |
3501 | r = dm_pool_resize_data_dev(pool->pmd, data_size); |
3502 | if (r) { | |
b5330655 | 3503 | metadata_operation_failed(pool, "dm_pool_resize_data_dev", r); |
991d9fa0 JT |
3504 | return r; |
3505 | } | |
3506 | ||
b17446df | 3507 | *need_commit = true; |
991d9fa0 JT |
3508 | } |
3509 | ||
3510 | return 0; | |
3511 | } | |
3512 | ||
24347e95 JT |
3513 | static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit) |
3514 | { | |
3515 | int r; | |
3516 | struct pool_c *pt = ti->private; | |
3517 | struct pool *pool = pt->pool; | |
3518 | dm_block_t metadata_dev_size, sb_metadata_dev_size; | |
3519 | ||
3520 | *need_commit = false; | |
3521 | ||
610bba8b | 3522 | metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev); |
24347e95 JT |
3523 | |
3524 | r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size); | |
3525 | if (r) { | |
4fa5971a MS |
3526 | DMERR("%s: failed to retrieve metadata device size", |
3527 | dm_device_name(pool->pool_md)); | |
24347e95 JT |
3528 | return r; |
3529 | } | |
3530 | ||
3531 | if (metadata_dev_size < sb_metadata_dev_size) { | |
4fa5971a MS |
3532 | DMERR("%s: metadata device (%llu blocks) too small: expected %llu", |
3533 | dm_device_name(pool->pool_md), | |
24347e95 JT |
3534 | metadata_dev_size, sb_metadata_dev_size); |
3535 | return -EINVAL; | |
3536 | ||
3537 | } else if (metadata_dev_size > sb_metadata_dev_size) { | |
07f2b6e0 MS |
3538 | if (dm_pool_metadata_needs_check(pool->pmd)) { |
3539 | DMERR("%s: unable to grow the metadata device until repaired.", | |
3540 | dm_device_name(pool->pool_md)); | |
3541 | return 0; | |
3542 | } | |
3543 | ||
7d48935e | 3544 | warn_if_metadata_device_too_big(pool->md_dev); |
6f7f51d4 MS |
3545 | DMINFO("%s: growing the metadata device from %llu to %llu blocks", |
3546 | dm_device_name(pool->pool_md), | |
3547 | sb_metadata_dev_size, metadata_dev_size); | |
3ab91828 JT |
3548 | |
3549 | if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE) | |
3550 | set_pool_mode(pool, PM_WRITE); | |
3551 | ||
24347e95 JT |
3552 | r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size); |
3553 | if (r) { | |
b5330655 | 3554 | metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r); |
24347e95 JT |
3555 | return r; |
3556 | } | |
3557 | ||
3558 | *need_commit = true; | |
3559 | } | |
3560 | ||
3561 | return 0; | |
3562 | } | |
3563 | ||
b17446df JT |
3564 | /* |
3565 | * Retrieves the number of blocks of the data device from | |
3566 | * the superblock and compares it to the actual device size, | |
3567 | * thus resizing the data device in case it has grown. | |
3568 | * | |
3569 | * This both copes with opening preallocated data devices in the ctr | |
3570 | * being followed by a resume | |
3571 | * -and- | |
3572 | * calling the resume method individually after userspace has | |
3573 | * grown the data device in reaction to a table event. | |
3574 | */ | |
3575 | static int pool_preresume(struct dm_target *ti) | |
3576 | { | |
3577 | int r; | |
24347e95 | 3578 | bool need_commit1, need_commit2; |
b17446df JT |
3579 | struct pool_c *pt = ti->private; |
3580 | struct pool *pool = pt->pool; | |
3581 | ||
3582 | /* | |
3583 | * Take control of the pool object. | |
3584 | */ | |
3585 | r = bind_control_target(pool, ti); | |
3586 | if (r) | |
19eb1650 | 3587 | goto out; |
b17446df JT |
3588 | |
3589 | r = maybe_resize_data_dev(ti, &need_commit1); | |
3590 | if (r) | |
19eb1650 | 3591 | goto out; |
b17446df | 3592 | |
24347e95 JT |
3593 | r = maybe_resize_metadata_dev(ti, &need_commit2); |
3594 | if (r) | |
19eb1650 | 3595 | goto out; |
24347e95 JT |
3596 | |
3597 | if (need_commit1 || need_commit2) | |
020cc3b5 | 3598 | (void) commit(pool); |
19eb1650 LM |
3599 | out: |
3600 | /* | |
3601 | * When a thin-pool is PM_FAIL, it cannot be rebuilt if | |
3602 | * bio is in deferred list. Therefore need to return 0 | |
3603 | * to allow pool_resume() to flush IO. | |
3604 | */ | |
3605 | if (r && get_pool_mode(pool) == PM_FAIL) | |
3606 | r = 0; | |
b17446df | 3607 | |
19eb1650 | 3608 | return r; |
b17446df JT |
3609 | } |
3610 | ||
583024d2 MS |
3611 | static void pool_suspend_active_thins(struct pool *pool) |
3612 | { | |
3613 | struct thin_c *tc; | |
3614 | ||
3615 | /* Suspend all active thin devices */ | |
3616 | tc = get_first_thin(pool); | |
3617 | while (tc) { | |
3618 | dm_internal_suspend_noflush(tc->thin_md); | |
3619 | tc = get_next_thin(pool, tc); | |
3620 | } | |
3621 | } | |
3622 | ||
3623 | static void pool_resume_active_thins(struct pool *pool) | |
3624 | { | |
3625 | struct thin_c *tc; | |
3626 | ||
3627 | /* Resume all active thin devices */ | |
3628 | tc = get_first_thin(pool); | |
3629 | while (tc) { | |
3630 | dm_internal_resume(tc->thin_md); | |
3631 | tc = get_next_thin(pool, tc); | |
3632 | } | |
3633 | } | |
3634 | ||
991d9fa0 JT |
3635 | static void pool_resume(struct dm_target *ti) |
3636 | { | |
3637 | struct pool_c *pt = ti->private; | |
3638 | struct pool *pool = pt->pool; | |
991d9fa0 | 3639 | |
583024d2 MS |
3640 | /* |
3641 | * Must requeue active_thins' bios and then resume | |
3642 | * active_thins _before_ clearing 'suspend' flag. | |
3643 | */ | |
3644 | requeue_bios(pool); | |
3645 | pool_resume_active_thins(pool); | |
3646 | ||
8e0c9dac | 3647 | spin_lock_irq(&pool->lock); |
88a6621b | 3648 | pool->low_water_triggered = false; |
80e96c54 | 3649 | pool->suspended = false; |
8e0c9dac | 3650 | spin_unlock_irq(&pool->lock); |
80e96c54 | 3651 | |
905e51b3 | 3652 | do_waker(&pool->waker.work); |
991d9fa0 JT |
3653 | } |
3654 | ||
80e96c54 MS |
3655 | static void pool_presuspend(struct dm_target *ti) |
3656 | { | |
3657 | struct pool_c *pt = ti->private; | |
3658 | struct pool *pool = pt->pool; | |
80e96c54 | 3659 | |
8e0c9dac | 3660 | spin_lock_irq(&pool->lock); |
80e96c54 | 3661 | pool->suspended = true; |
8e0c9dac | 3662 | spin_unlock_irq(&pool->lock); |
583024d2 MS |
3663 | |
3664 | pool_suspend_active_thins(pool); | |
80e96c54 MS |
3665 | } |
3666 | ||
3667 | static void pool_presuspend_undo(struct dm_target *ti) | |
3668 | { | |
3669 | struct pool_c *pt = ti->private; | |
3670 | struct pool *pool = pt->pool; | |
80e96c54 | 3671 | |
583024d2 MS |
3672 | pool_resume_active_thins(pool); |
3673 | ||
8e0c9dac | 3674 | spin_lock_irq(&pool->lock); |
80e96c54 | 3675 | pool->suspended = false; |
8e0c9dac | 3676 | spin_unlock_irq(&pool->lock); |
80e96c54 MS |
3677 | } |
3678 | ||
991d9fa0 JT |
3679 | static void pool_postsuspend(struct dm_target *ti) |
3680 | { | |
991d9fa0 JT |
3681 | struct pool_c *pt = ti->private; |
3682 | struct pool *pool = pt->pool; | |
3683 | ||
18d03e8c NB |
3684 | cancel_delayed_work_sync(&pool->waker); |
3685 | cancel_delayed_work_sync(&pool->no_space_timeout); | |
991d9fa0 | 3686 | flush_workqueue(pool->wq); |
020cc3b5 | 3687 | (void) commit(pool); |
991d9fa0 JT |
3688 | } |
3689 | ||
86a3238c | 3690 | static int check_arg_count(unsigned int argc, unsigned int args_required) |
991d9fa0 JT |
3691 | { |
3692 | if (argc != args_required) { | |
3693 | DMWARN("Message received with %u arguments instead of %u.", | |
3694 | argc, args_required); | |
3695 | return -EINVAL; | |
3696 | } | |
3697 | ||
3698 | return 0; | |
3699 | } | |
3700 | ||
3701 | static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning) | |
3702 | { | |
3703 | if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) && | |
3704 | *dev_id <= MAX_DEV_ID) | |
3705 | return 0; | |
3706 | ||
3707 | if (warning) | |
3708 | DMWARN("Message received with invalid device id: %s", arg); | |
3709 | ||
3710 | return -EINVAL; | |
3711 | } | |
3712 | ||
86a3238c | 3713 | static int process_create_thin_mesg(unsigned int argc, char **argv, struct pool *pool) |
991d9fa0 JT |
3714 | { |
3715 | dm_thin_id dev_id; | |
3716 | int r; | |
3717 | ||
3718 | r = check_arg_count(argc, 2); | |
3719 | if (r) | |
3720 | return r; | |
3721 | ||
3722 | r = read_dev_id(argv[1], &dev_id, 1); | |
3723 | if (r) | |
3724 | return r; | |
3725 | ||
3726 | r = dm_pool_create_thin(pool->pmd, dev_id); | |
3727 | if (r) { | |
3728 | DMWARN("Creation of new thinly-provisioned device with id %s failed.", | |
3729 | argv[1]); | |
3730 | return r; | |
3731 | } | |
3732 | ||
3733 | return 0; | |
3734 | } | |
3735 | ||
86a3238c | 3736 | static int process_create_snap_mesg(unsigned int argc, char **argv, struct pool *pool) |
991d9fa0 JT |
3737 | { |
3738 | dm_thin_id dev_id; | |
3739 | dm_thin_id origin_dev_id; | |
3740 | int r; | |
3741 | ||
3742 | r = check_arg_count(argc, 3); | |
3743 | if (r) | |
3744 | return r; | |
3745 | ||
3746 | r = read_dev_id(argv[1], &dev_id, 1); | |
3747 | if (r) | |
3748 | return r; | |
3749 | ||
3750 | r = read_dev_id(argv[2], &origin_dev_id, 1); | |
3751 | if (r) | |
3752 | return r; | |
3753 | ||
3754 | r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id); | |
3755 | if (r) { | |
3756 | DMWARN("Creation of new snapshot %s of device %s failed.", | |
3757 | argv[1], argv[2]); | |
3758 | return r; | |
3759 | } | |
3760 | ||
3761 | return 0; | |
3762 | } | |
3763 | ||
86a3238c | 3764 | static int process_delete_mesg(unsigned int argc, char **argv, struct pool *pool) |
991d9fa0 JT |
3765 | { |
3766 | dm_thin_id dev_id; | |
3767 | int r; | |
3768 | ||
3769 | r = check_arg_count(argc, 2); | |
3770 | if (r) | |
3771 | return r; | |
3772 | ||
3773 | r = read_dev_id(argv[1], &dev_id, 1); | |
3774 | if (r) | |
3775 | return r; | |
3776 | ||
3777 | r = dm_pool_delete_thin_device(pool->pmd, dev_id); | |
3778 | if (r) | |
3779 | DMWARN("Deletion of thin device %s failed.", argv[1]); | |
3780 | ||
3781 | return r; | |
3782 | } | |
3783 | ||
86a3238c | 3784 | static int process_set_transaction_id_mesg(unsigned int argc, char **argv, struct pool *pool) |
991d9fa0 JT |
3785 | { |
3786 | dm_thin_id old_id, new_id; | |
3787 | int r; | |
3788 | ||
3789 | r = check_arg_count(argc, 3); | |
3790 | if (r) | |
3791 | return r; | |
3792 | ||
3793 | if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) { | |
3794 | DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]); | |
3795 | return -EINVAL; | |
3796 | } | |
3797 | ||
3798 | if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) { | |
3799 | DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]); | |
3800 | return -EINVAL; | |
3801 | } | |
3802 | ||
3803 | r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id); | |
3804 | if (r) { | |
3805 | DMWARN("Failed to change transaction id from %s to %s.", | |
3806 | argv[1], argv[2]); | |
3807 | return r; | |
3808 | } | |
3809 | ||
3810 | return 0; | |
3811 | } | |
3812 | ||
86a3238c | 3813 | static int process_reserve_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool) |
cc8394d8 JT |
3814 | { |
3815 | int r; | |
3816 | ||
3817 | r = check_arg_count(argc, 1); | |
3818 | if (r) | |
3819 | return r; | |
3820 | ||
020cc3b5 | 3821 | (void) commit(pool); |
0d200aef | 3822 | |
cc8394d8 JT |
3823 | r = dm_pool_reserve_metadata_snap(pool->pmd); |
3824 | if (r) | |
3825 | DMWARN("reserve_metadata_snap message failed."); | |
3826 | ||
3827 | return r; | |
3828 | } | |
3829 | ||
86a3238c | 3830 | static int process_release_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool) |
cc8394d8 JT |
3831 | { |
3832 | int r; | |
3833 | ||
3834 | r = check_arg_count(argc, 1); | |
3835 | if (r) | |
3836 | return r; | |
3837 | ||
3838 | r = dm_pool_release_metadata_snap(pool->pmd); | |
3839 | if (r) | |
3840 | DMWARN("release_metadata_snap message failed."); | |
3841 | ||
3842 | return r; | |
3843 | } | |
3844 | ||
991d9fa0 JT |
3845 | /* |
3846 | * Messages supported: | |
3847 | * create_thin <dev_id> | |
3848 | * create_snap <dev_id> <origin_id> | |
3849 | * delete <dev_id> | |
991d9fa0 | 3850 | * set_transaction_id <current_trans_id> <new_trans_id> |
cc8394d8 JT |
3851 | * reserve_metadata_snap |
3852 | * release_metadata_snap | |
991d9fa0 | 3853 | */ |
86a3238c HM |
3854 | static int pool_message(struct dm_target *ti, unsigned int argc, char **argv, |
3855 | char *result, unsigned int maxlen) | |
991d9fa0 JT |
3856 | { |
3857 | int r = -EINVAL; | |
3858 | struct pool_c *pt = ti->private; | |
3859 | struct pool *pool = pt->pool; | |
3860 | ||
3ab91828 | 3861 | if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) { |
2a7eaea0 JT |
3862 | DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode", |
3863 | dm_device_name(pool->pool_md)); | |
fd467696 | 3864 | return -EOPNOTSUPP; |
2a7eaea0 JT |
3865 | } |
3866 | ||
991d9fa0 JT |
3867 | if (!strcasecmp(argv[0], "create_thin")) |
3868 | r = process_create_thin_mesg(argc, argv, pool); | |
3869 | ||
3870 | else if (!strcasecmp(argv[0], "create_snap")) | |
3871 | r = process_create_snap_mesg(argc, argv, pool); | |
3872 | ||
3873 | else if (!strcasecmp(argv[0], "delete")) | |
3874 | r = process_delete_mesg(argc, argv, pool); | |
3875 | ||
3876 | else if (!strcasecmp(argv[0], "set_transaction_id")) | |
3877 | r = process_set_transaction_id_mesg(argc, argv, pool); | |
3878 | ||
cc8394d8 JT |
3879 | else if (!strcasecmp(argv[0], "reserve_metadata_snap")) |
3880 | r = process_reserve_metadata_snap_mesg(argc, argv, pool); | |
3881 | ||
3882 | else if (!strcasecmp(argv[0], "release_metadata_snap")) | |
3883 | r = process_release_metadata_snap_mesg(argc, argv, pool); | |
3884 | ||
991d9fa0 JT |
3885 | else |
3886 | DMWARN("Unrecognised thin pool target message received: %s", argv[0]); | |
3887 | ||
e49e5829 | 3888 | if (!r) |
020cc3b5 | 3889 | (void) commit(pool); |
991d9fa0 JT |
3890 | |
3891 | return r; | |
3892 | } | |
3893 | ||
e49e5829 | 3894 | static void emit_flags(struct pool_features *pf, char *result, |
86a3238c | 3895 | unsigned int sz, unsigned int maxlen) |
e49e5829 | 3896 | { |
86a3238c | 3897 | unsigned int count = !pf->zero_new_blocks + !pf->discard_enabled + |
787a996c MS |
3898 | !pf->discard_passdown + (pf->mode == PM_READ_ONLY) + |
3899 | pf->error_if_no_space; | |
e49e5829 JT |
3900 | DMEMIT("%u ", count); |
3901 | ||
3902 | if (!pf->zero_new_blocks) | |
3903 | DMEMIT("skip_block_zeroing "); | |
3904 | ||
3905 | if (!pf->discard_enabled) | |
3906 | DMEMIT("ignore_discard "); | |
3907 | ||
3908 | if (!pf->discard_passdown) | |
3909 | DMEMIT("no_discard_passdown "); | |
3910 | ||
3911 | if (pf->mode == PM_READ_ONLY) | |
3912 | DMEMIT("read_only "); | |
787a996c MS |
3913 | |
3914 | if (pf->error_if_no_space) | |
3915 | DMEMIT("error_if_no_space "); | |
e49e5829 JT |
3916 | } |
3917 | ||
991d9fa0 JT |
3918 | /* |
3919 | * Status line is: | |
3920 | * <transaction id> <used metadata sectors>/<total metadata sectors> | |
3921 | * <used data sectors>/<total data sectors> <held metadata root> | |
e4c78e21 | 3922 | * <pool mode> <discard config> <no space config> <needs_check> |
991d9fa0 | 3923 | */ |
fd7c092e | 3924 | static void pool_status(struct dm_target *ti, status_type_t type, |
86a3238c | 3925 | unsigned int status_flags, char *result, unsigned int maxlen) |
991d9fa0 | 3926 | { |
e49e5829 | 3927 | int r; |
86a3238c | 3928 | unsigned int sz = 0; |
991d9fa0 JT |
3929 | uint64_t transaction_id; |
3930 | dm_block_t nr_free_blocks_data; | |
3931 | dm_block_t nr_free_blocks_metadata; | |
3932 | dm_block_t nr_blocks_data; | |
3933 | dm_block_t nr_blocks_metadata; | |
3934 | dm_block_t held_root; | |
3ab91828 | 3935 | enum pool_mode mode; |
991d9fa0 JT |
3936 | char buf[BDEVNAME_SIZE]; |
3937 | char buf2[BDEVNAME_SIZE]; | |
3938 | struct pool_c *pt = ti->private; | |
3939 | struct pool *pool = pt->pool; | |
3940 | ||
3941 | switch (type) { | |
3942 | case STATUSTYPE_INFO: | |
e49e5829 JT |
3943 | if (get_pool_mode(pool) == PM_FAIL) { |
3944 | DMEMIT("Fail"); | |
3945 | break; | |
3946 | } | |
3947 | ||
1f4e0ff0 AK |
3948 | /* Commit to ensure statistics aren't out-of-date */ |
3949 | if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti)) | |
020cc3b5 | 3950 | (void) commit(pool); |
1f4e0ff0 | 3951 | |
fd7c092e MP |
3952 | r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id); |
3953 | if (r) { | |
4fa5971a MS |
3954 | DMERR("%s: dm_pool_get_metadata_transaction_id returned %d", |
3955 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3956 | goto err; |
3957 | } | |
991d9fa0 | 3958 | |
fd7c092e MP |
3959 | r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata); |
3960 | if (r) { | |
4fa5971a MS |
3961 | DMERR("%s: dm_pool_get_free_metadata_block_count returned %d", |
3962 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3963 | goto err; |
3964 | } | |
991d9fa0 JT |
3965 | |
3966 | r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata); | |
fd7c092e | 3967 | if (r) { |
4fa5971a MS |
3968 | DMERR("%s: dm_pool_get_metadata_dev_size returned %d", |
3969 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3970 | goto err; |
3971 | } | |
991d9fa0 | 3972 | |
fd7c092e MP |
3973 | r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data); |
3974 | if (r) { | |
4fa5971a MS |
3975 | DMERR("%s: dm_pool_get_free_block_count returned %d", |
3976 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3977 | goto err; |
3978 | } | |
991d9fa0 JT |
3979 | |
3980 | r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data); | |
fd7c092e | 3981 | if (r) { |
4fa5971a MS |
3982 | DMERR("%s: dm_pool_get_data_dev_size returned %d", |
3983 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3984 | goto err; |
3985 | } | |
991d9fa0 | 3986 | |
cc8394d8 | 3987 | r = dm_pool_get_metadata_snap(pool->pmd, &held_root); |
fd7c092e | 3988 | if (r) { |
4fa5971a MS |
3989 | DMERR("%s: dm_pool_get_metadata_snap returned %d", |
3990 | dm_device_name(pool->pool_md), r); | |
fd7c092e MP |
3991 | goto err; |
3992 | } | |
991d9fa0 JT |
3993 | |
3994 | DMEMIT("%llu %llu/%llu %llu/%llu ", | |
3995 | (unsigned long long)transaction_id, | |
3996 | (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata), | |
3997 | (unsigned long long)nr_blocks_metadata, | |
3998 | (unsigned long long)(nr_blocks_data - nr_free_blocks_data), | |
3999 | (unsigned long long)nr_blocks_data); | |
4000 | ||
4001 | if (held_root) | |
e49e5829 JT |
4002 | DMEMIT("%llu ", held_root); |
4003 | else | |
4004 | DMEMIT("- "); | |
4005 | ||
3ab91828 JT |
4006 | mode = get_pool_mode(pool); |
4007 | if (mode == PM_OUT_OF_DATA_SPACE) | |
3e1a0699 | 4008 | DMEMIT("out_of_data_space "); |
3ab91828 | 4009 | else if (is_read_only_pool_mode(mode)) |
e49e5829 | 4010 | DMEMIT("ro "); |
991d9fa0 | 4011 | else |
e49e5829 JT |
4012 | DMEMIT("rw "); |
4013 | ||
018debea | 4014 | if (!pool->pf.discard_enabled) |
787a996c | 4015 | DMEMIT("ignore_discard "); |
018debea | 4016 | else if (pool->pf.discard_passdown) |
787a996c MS |
4017 | DMEMIT("discard_passdown "); |
4018 | else | |
4019 | DMEMIT("no_discard_passdown "); | |
4020 | ||
4021 | if (pool->pf.error_if_no_space) | |
4022 | DMEMIT("error_if_no_space "); | |
e49e5829 | 4023 | else |
787a996c | 4024 | DMEMIT("queue_if_no_space "); |
991d9fa0 | 4025 | |
e4c78e21 MS |
4026 | if (dm_pool_metadata_needs_check(pool->pmd)) |
4027 | DMEMIT("needs_check "); | |
4028 | else | |
4029 | DMEMIT("- "); | |
4030 | ||
63c8ecb6 AG |
4031 | DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt)); |
4032 | ||
991d9fa0 JT |
4033 | break; |
4034 | ||
4035 | case STATUSTYPE_TABLE: | |
4036 | DMEMIT("%s %s %lu %llu ", | |
4037 | format_dev_t(buf, pt->metadata_dev->bdev->bd_dev), | |
4038 | format_dev_t(buf2, pt->data_dev->bdev->bd_dev), | |
4039 | (unsigned long)pool->sectors_per_block, | |
4040 | (unsigned long long)pt->low_water_blocks); | |
0424caa1 | 4041 | emit_flags(&pt->requested_pf, result, sz, maxlen); |
991d9fa0 | 4042 | break; |
8ec45662 TS |
4043 | |
4044 | case STATUSTYPE_IMA: | |
4045 | *result = '\0'; | |
4046 | break; | |
991d9fa0 | 4047 | } |
fd7c092e | 4048 | return; |
991d9fa0 | 4049 | |
fd7c092e MP |
4050 | err: |
4051 | DMEMIT("Error"); | |
991d9fa0 JT |
4052 | } |
4053 | ||
4054 | static int pool_iterate_devices(struct dm_target *ti, | |
4055 | iterate_devices_callout_fn fn, void *data) | |
4056 | { | |
4057 | struct pool_c *pt = ti->private; | |
4058 | ||
4059 | return fn(ti, pt->data_dev, 0, ti->len, data); | |
4060 | } | |
4061 | ||
991d9fa0 JT |
4062 | static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4063 | { | |
4064 | struct pool_c *pt = ti->private; | |
4065 | struct pool *pool = pt->pool; | |
604ea906 MS |
4066 | sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; |
4067 | ||
4068 | /* | |
d200c30e MS |
4069 | * If max_sectors is smaller than pool->sectors_per_block adjust it |
4070 | * to the highest possible power-of-2 factor of pool->sectors_per_block. | |
4071 | * This is especially beneficial when the pool's data device is a RAID | |
4072 | * device that has a full stripe width that matches pool->sectors_per_block | |
4073 | * -- because even though partial RAID stripe-sized IOs will be issued to a | |
4074 | * single RAID stripe; when aggregated they will end on a full RAID stripe | |
4075 | * boundary.. which avoids additional partial RAID stripe writes cascading | |
604ea906 | 4076 | */ |
604ea906 MS |
4077 | if (limits->max_sectors < pool->sectors_per_block) { |
4078 | while (!is_factor(pool->sectors_per_block, limits->max_sectors)) { | |
4079 | if ((limits->max_sectors & (limits->max_sectors - 1)) == 0) | |
4080 | limits->max_sectors--; | |
4081 | limits->max_sectors = rounddown_pow_of_two(limits->max_sectors); | |
4082 | } | |
604ea906 | 4083 | } |
991d9fa0 | 4084 | |
0cc67cd9 MS |
4085 | /* |
4086 | * If the system-determined stacked limits are compatible with the | |
4087 | * pool's blocksize (io_opt is a factor) do not override them. | |
4088 | */ | |
4089 | if (io_opt_sectors < pool->sectors_per_block || | |
604ea906 MS |
4090 | !is_factor(io_opt_sectors, pool->sectors_per_block)) { |
4091 | if (is_factor(pool->sectors_per_block, limits->max_sectors)) | |
4092 | blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT); | |
4093 | else | |
4094 | blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT); | |
0cc67cd9 MS |
4095 | blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT); |
4096 | } | |
0424caa1 MS |
4097 | |
4098 | /* | |
4099 | * pt->adjusted_pf is a staging area for the actual features to use. | |
4100 | * They get transferred to the live pool in bind_control_target() | |
4101 | * called from pool_preresume(). | |
4102 | */ | |
b60ab990 MS |
4103 | if (!pt->adjusted_pf.discard_enabled) { |
4104 | /* | |
4105 | * Must explicitly disallow stacking discard limits otherwise the | |
4106 | * block layer will stack them if pool's data device has support. | |
b60ab990 MS |
4107 | */ |
4108 | limits->discard_granularity = 0; | |
0424caa1 | 4109 | return; |
b60ab990 | 4110 | } |
0424caa1 MS |
4111 | |
4112 | disable_passdown_if_not_supported(pt); | |
4113 | ||
34fbcf62 JT |
4114 | /* |
4115 | * The pool uses the same discard limits as the underlying data | |
4116 | * device. DM core has already set this up. | |
4117 | */ | |
991d9fa0 JT |
4118 | } |
4119 | ||
4120 | static struct target_type pool_target = { | |
4121 | .name = "thin-pool", | |
4122 | .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | | |
4123 | DM_TARGET_IMMUTABLE, | |
e2dd8aca | 4124 | .version = {1, 23, 0}, |
991d9fa0 JT |
4125 | .module = THIS_MODULE, |
4126 | .ctr = pool_ctr, | |
4127 | .dtr = pool_dtr, | |
4128 | .map = pool_map, | |
80e96c54 MS |
4129 | .presuspend = pool_presuspend, |
4130 | .presuspend_undo = pool_presuspend_undo, | |
991d9fa0 JT |
4131 | .postsuspend = pool_postsuspend, |
4132 | .preresume = pool_preresume, | |
4133 | .resume = pool_resume, | |
4134 | .message = pool_message, | |
4135 | .status = pool_status, | |
991d9fa0 JT |
4136 | .iterate_devices = pool_iterate_devices, |
4137 | .io_hints = pool_io_hints, | |
4138 | }; | |
4139 | ||
a4a82ce3 HM |
4140 | /* |
4141 | *-------------------------------------------------------------- | |
991d9fa0 | 4142 | * Thin target methods |
a4a82ce3 HM |
4143 | *-------------------------------------------------------------- |
4144 | */ | |
b10ebd34 JT |
4145 | static void thin_get(struct thin_c *tc) |
4146 | { | |
22d4c291 | 4147 | refcount_inc(&tc->refcount); |
b10ebd34 JT |
4148 | } |
4149 | ||
4150 | static void thin_put(struct thin_c *tc) | |
4151 | { | |
22d4c291 | 4152 | if (refcount_dec_and_test(&tc->refcount)) |
b10ebd34 JT |
4153 | complete(&tc->can_destroy); |
4154 | } | |
4155 | ||
991d9fa0 JT |
4156 | static void thin_dtr(struct dm_target *ti) |
4157 | { | |
4158 | struct thin_c *tc = ti->private; | |
c140e1c4 | 4159 | |
8e0c9dac | 4160 | spin_lock_irq(&tc->pool->lock); |
c140e1c4 | 4161 | list_del_rcu(&tc->list); |
8e0c9dac | 4162 | spin_unlock_irq(&tc->pool->lock); |
c140e1c4 | 4163 | synchronize_rcu(); |
991d9fa0 | 4164 | |
17181fb7 MP |
4165 | thin_put(tc); |
4166 | wait_for_completion(&tc->can_destroy); | |
4167 | ||
991d9fa0 JT |
4168 | mutex_lock(&dm_thin_pool_table.mutex); |
4169 | ||
4170 | __pool_dec(tc->pool); | |
4171 | dm_pool_close_thin_device(tc->td); | |
4172 | dm_put_device(ti, tc->pool_dev); | |
2dd9c257 JT |
4173 | if (tc->origin_dev) |
4174 | dm_put_device(ti, tc->origin_dev); | |
991d9fa0 JT |
4175 | kfree(tc); |
4176 | ||
4177 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4178 | } | |
4179 | ||
4180 | /* | |
4181 | * Thin target parameters: | |
4182 | * | |
2dd9c257 | 4183 | * <pool_dev> <dev_id> [origin_dev] |
991d9fa0 JT |
4184 | * |
4185 | * pool_dev: the path to the pool (eg, /dev/mapper/my_pool) | |
4186 | * dev_id: the internal device identifier | |
2dd9c257 | 4187 | * origin_dev: a device external to the pool that should act as the origin |
67e2e2b2 JT |
4188 | * |
4189 | * If the pool device has discards disabled, they get disabled for the thin | |
4190 | * device as well. | |
991d9fa0 | 4191 | */ |
86a3238c | 4192 | static int thin_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
991d9fa0 JT |
4193 | { |
4194 | int r; | |
4195 | struct thin_c *tc; | |
2dd9c257 | 4196 | struct dm_dev *pool_dev, *origin_dev; |
991d9fa0 JT |
4197 | struct mapped_device *pool_md; |
4198 | ||
4199 | mutex_lock(&dm_thin_pool_table.mutex); | |
4200 | ||
2dd9c257 | 4201 | if (argc != 2 && argc != 3) { |
991d9fa0 JT |
4202 | ti->error = "Invalid argument count"; |
4203 | r = -EINVAL; | |
4204 | goto out_unlock; | |
4205 | } | |
4206 | ||
4207 | tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL); | |
4208 | if (!tc) { | |
4209 | ti->error = "Out of memory"; | |
4210 | r = -ENOMEM; | |
4211 | goto out_unlock; | |
4212 | } | |
583024d2 | 4213 | tc->thin_md = dm_table_get_md(ti->table); |
c140e1c4 | 4214 | spin_lock_init(&tc->lock); |
a374bb21 | 4215 | INIT_LIST_HEAD(&tc->deferred_cells); |
c140e1c4 MS |
4216 | bio_list_init(&tc->deferred_bio_list); |
4217 | bio_list_init(&tc->retry_on_resume_list); | |
67324ea1 | 4218 | tc->sort_bio_list = RB_ROOT; |
991d9fa0 | 4219 | |
2dd9c257 | 4220 | if (argc == 3) { |
70de2cbd JCXF |
4221 | if (!strcmp(argv[0], argv[2])) { |
4222 | ti->error = "Error setting origin device"; | |
4223 | r = -EINVAL; | |
4224 | goto bad_origin_dev; | |
4225 | } | |
4226 | ||
05bdb996 | 4227 | r = dm_get_device(ti, argv[2], BLK_OPEN_READ, &origin_dev); |
2dd9c257 JT |
4228 | if (r) { |
4229 | ti->error = "Error opening origin device"; | |
4230 | goto bad_origin_dev; | |
4231 | } | |
4232 | tc->origin_dev = origin_dev; | |
4233 | } | |
4234 | ||
991d9fa0 JT |
4235 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev); |
4236 | if (r) { | |
4237 | ti->error = "Error opening pool device"; | |
4238 | goto bad_pool_dev; | |
4239 | } | |
4240 | tc->pool_dev = pool_dev; | |
4241 | ||
4242 | if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) { | |
4243 | ti->error = "Invalid device id"; | |
4244 | r = -EINVAL; | |
4245 | goto bad_common; | |
4246 | } | |
4247 | ||
4248 | pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev); | |
4249 | if (!pool_md) { | |
4250 | ti->error = "Couldn't get pool mapped device"; | |
4251 | r = -EINVAL; | |
4252 | goto bad_common; | |
4253 | } | |
4254 | ||
4255 | tc->pool = __pool_table_lookup(pool_md); | |
4256 | if (!tc->pool) { | |
4257 | ti->error = "Couldn't find pool object"; | |
4258 | r = -EINVAL; | |
4259 | goto bad_pool_lookup; | |
4260 | } | |
4261 | __pool_inc(tc->pool); | |
4262 | ||
e49e5829 JT |
4263 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4264 | ti->error = "Couldn't open thin device, Pool is in fail mode"; | |
1acacc07 | 4265 | r = -EINVAL; |
80e96c54 | 4266 | goto bad_pool; |
e49e5829 JT |
4267 | } |
4268 | ||
991d9fa0 JT |
4269 | r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td); |
4270 | if (r) { | |
4271 | ti->error = "Couldn't open thin internal device"; | |
80e96c54 | 4272 | goto bad_pool; |
991d9fa0 JT |
4273 | } |
4274 | ||
542f9038 MS |
4275 | r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block); |
4276 | if (r) | |
80e96c54 | 4277 | goto bad; |
542f9038 | 4278 | |
55a62eef | 4279 | ti->num_flush_bios = 1; |
9bbf5fee | 4280 | ti->limit_swap_bios = true; |
16ad3d10 | 4281 | ti->flush_supported = true; |
a9251281 | 4282 | ti->accounts_remapped_io = true; |
30187e1d | 4283 | ti->per_io_data_size = sizeof(struct dm_thin_endio_hook); |
67e2e2b2 JT |
4284 | |
4285 | /* In case the pool supports discards, pass them on. */ | |
4286 | if (tc->pool->pf.discard_enabled) { | |
0ac55489 | 4287 | ti->discards_supported = true; |
55a62eef | 4288 | ti->num_discard_bios = 1; |
e2dd8aca | 4289 | ti->max_discard_granularity = true; |
67e2e2b2 | 4290 | } |
991d9fa0 | 4291 | |
991d9fa0 JT |
4292 | mutex_unlock(&dm_thin_pool_table.mutex); |
4293 | ||
8e0c9dac | 4294 | spin_lock_irq(&tc->pool->lock); |
80e96c54 | 4295 | if (tc->pool->suspended) { |
8e0c9dac | 4296 | spin_unlock_irq(&tc->pool->lock); |
80e96c54 MS |
4297 | mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */ |
4298 | ti->error = "Unable to activate thin device while pool is suspended"; | |
4299 | r = -EINVAL; | |
4300 | goto bad; | |
4301 | } | |
22d4c291 | 4302 | refcount_set(&tc->refcount, 1); |
2b94e896 | 4303 | init_completion(&tc->can_destroy); |
c140e1c4 | 4304 | list_add_tail_rcu(&tc->list, &tc->pool->active_thins); |
8e0c9dac | 4305 | spin_unlock_irq(&tc->pool->lock); |
c140e1c4 MS |
4306 | /* |
4307 | * This synchronize_rcu() call is needed here otherwise we risk a | |
4308 | * wake_worker() call finding no bios to process (because the newly | |
4309 | * added tc isn't yet visible). So this reduces latency since we | |
4310 | * aren't then dependent on the periodic commit to wake_worker(). | |
4311 | */ | |
4312 | synchronize_rcu(); | |
4313 | ||
80e96c54 MS |
4314 | dm_put(pool_md); |
4315 | ||
991d9fa0 JT |
4316 | return 0; |
4317 | ||
80e96c54 | 4318 | bad: |
1acacc07 | 4319 | dm_pool_close_thin_device(tc->td); |
80e96c54 | 4320 | bad_pool: |
991d9fa0 JT |
4321 | __pool_dec(tc->pool); |
4322 | bad_pool_lookup: | |
4323 | dm_put(pool_md); | |
4324 | bad_common: | |
4325 | dm_put_device(ti, tc->pool_dev); | |
4326 | bad_pool_dev: | |
2dd9c257 JT |
4327 | if (tc->origin_dev) |
4328 | dm_put_device(ti, tc->origin_dev); | |
4329 | bad_origin_dev: | |
991d9fa0 JT |
4330 | kfree(tc); |
4331 | out_unlock: | |
4332 | mutex_unlock(&dm_thin_pool_table.mutex); | |
4333 | ||
4334 | return r; | |
4335 | } | |
4336 | ||
7de3ee57 | 4337 | static int thin_map(struct dm_target *ti, struct bio *bio) |
991d9fa0 | 4338 | { |
4f024f37 | 4339 | bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); |
991d9fa0 | 4340 | |
7de3ee57 | 4341 | return thin_bio_map(ti, bio); |
991d9fa0 JT |
4342 | } |
4343 | ||
4e4cbee9 CH |
4344 | static int thin_endio(struct dm_target *ti, struct bio *bio, |
4345 | blk_status_t *err) | |
eb2aa48d JT |
4346 | { |
4347 | unsigned long flags; | |
59c3d2c6 | 4348 | struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook)); |
eb2aa48d | 4349 | struct list_head work; |
a24c2569 | 4350 | struct dm_thin_new_mapping *m, *tmp; |
eb2aa48d JT |
4351 | struct pool *pool = h->tc->pool; |
4352 | ||
4353 | if (h->shared_read_entry) { | |
4354 | INIT_LIST_HEAD(&work); | |
44feb387 | 4355 | dm_deferred_entry_dec(h->shared_read_entry, &work); |
eb2aa48d JT |
4356 | |
4357 | spin_lock_irqsave(&pool->lock, flags); | |
4358 | list_for_each_entry_safe(m, tmp, &work, list) { | |
4359 | list_del(&m->list); | |
50f3c3ef | 4360 | __complete_mapping_preparation(m); |
eb2aa48d JT |
4361 | } |
4362 | spin_unlock_irqrestore(&pool->lock, flags); | |
4363 | } | |
4364 | ||
104655fd JT |
4365 | if (h->all_io_entry) { |
4366 | INIT_LIST_HEAD(&work); | |
44feb387 | 4367 | dm_deferred_entry_dec(h->all_io_entry, &work); |
563af186 JT |
4368 | if (!list_empty(&work)) { |
4369 | spin_lock_irqsave(&pool->lock, flags); | |
4370 | list_for_each_entry_safe(m, tmp, &work, list) | |
daec338b | 4371 | list_add_tail(&m->list, &pool->prepared_discards); |
563af186 JT |
4372 | spin_unlock_irqrestore(&pool->lock, flags); |
4373 | wake_worker(pool); | |
4374 | } | |
104655fd JT |
4375 | } |
4376 | ||
34fbcf62 JT |
4377 | if (h->cell) |
4378 | cell_defer_no_holder(h->tc, h->cell); | |
4379 | ||
1be56909 | 4380 | return DM_ENDIO_DONE; |
eb2aa48d JT |
4381 | } |
4382 | ||
738211f7 | 4383 | static void thin_presuspend(struct dm_target *ti) |
991d9fa0 | 4384 | { |
738211f7 JT |
4385 | struct thin_c *tc = ti->private; |
4386 | ||
991d9fa0 | 4387 | if (dm_noflush_suspending(ti)) |
738211f7 JT |
4388 | noflush_work(tc, do_noflush_start); |
4389 | } | |
4390 | ||
4391 | static void thin_postsuspend(struct dm_target *ti) | |
4392 | { | |
4393 | struct thin_c *tc = ti->private; | |
4394 | ||
4395 | /* | |
4396 | * The dm_noflush_suspending flag has been cleared by now, so | |
4397 | * unfortunately we must always run this. | |
4398 | */ | |
4399 | noflush_work(tc, do_noflush_stop); | |
991d9fa0 JT |
4400 | } |
4401 | ||
e5aea7b4 JT |
4402 | static int thin_preresume(struct dm_target *ti) |
4403 | { | |
4404 | struct thin_c *tc = ti->private; | |
4405 | ||
4406 | if (tc->origin_dev) | |
4407 | tc->origin_size = get_dev_size(tc->origin_dev->bdev); | |
4408 | ||
4409 | return 0; | |
4410 | } | |
4411 | ||
991d9fa0 JT |
4412 | /* |
4413 | * <nr mapped sectors> <highest mapped sector> | |
4414 | */ | |
fd7c092e | 4415 | static void thin_status(struct dm_target *ti, status_type_t type, |
86a3238c | 4416 | unsigned int status_flags, char *result, unsigned int maxlen) |
991d9fa0 JT |
4417 | { |
4418 | int r; | |
4419 | ssize_t sz = 0; | |
4420 | dm_block_t mapped, highest; | |
4421 | char buf[BDEVNAME_SIZE]; | |
4422 | struct thin_c *tc = ti->private; | |
4423 | ||
e49e5829 JT |
4424 | if (get_pool_mode(tc->pool) == PM_FAIL) { |
4425 | DMEMIT("Fail"); | |
fd7c092e | 4426 | return; |
e49e5829 JT |
4427 | } |
4428 | ||
991d9fa0 JT |
4429 | if (!tc->td) |
4430 | DMEMIT("-"); | |
4431 | else { | |
4432 | switch (type) { | |
4433 | case STATUSTYPE_INFO: | |
4434 | r = dm_thin_get_mapped_count(tc->td, &mapped); | |
fd7c092e MP |
4435 | if (r) { |
4436 | DMERR("dm_thin_get_mapped_count returned %d", r); | |
4437 | goto err; | |
4438 | } | |
991d9fa0 JT |
4439 | |
4440 | r = dm_thin_get_highest_mapped_block(tc->td, &highest); | |
fd7c092e MP |
4441 | if (r < 0) { |
4442 | DMERR("dm_thin_get_highest_mapped_block returned %d", r); | |
4443 | goto err; | |
4444 | } | |
991d9fa0 JT |
4445 | |
4446 | DMEMIT("%llu ", mapped * tc->pool->sectors_per_block); | |
4447 | if (r) | |
4448 | DMEMIT("%llu", ((highest + 1) * | |
4449 | tc->pool->sectors_per_block) - 1); | |
4450 | else | |
4451 | DMEMIT("-"); | |
4452 | break; | |
4453 | ||
4454 | case STATUSTYPE_TABLE: | |
4455 | DMEMIT("%s %lu", | |
4456 | format_dev_t(buf, tc->pool_dev->bdev->bd_dev), | |
4457 | (unsigned long) tc->dev_id); | |
2dd9c257 JT |
4458 | if (tc->origin_dev) |
4459 | DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev)); | |
991d9fa0 | 4460 | break; |
8ec45662 TS |
4461 | |
4462 | case STATUSTYPE_IMA: | |
4463 | *result = '\0'; | |
4464 | break; | |
991d9fa0 JT |
4465 | } |
4466 | } | |
4467 | ||
fd7c092e MP |
4468 | return; |
4469 | ||
4470 | err: | |
4471 | DMEMIT("Error"); | |
991d9fa0 JT |
4472 | } |
4473 | ||
4474 | static int thin_iterate_devices(struct dm_target *ti, | |
4475 | iterate_devices_callout_fn fn, void *data) | |
4476 | { | |
55f2b8bd | 4477 | sector_t blocks; |
991d9fa0 | 4478 | struct thin_c *tc = ti->private; |
55f2b8bd | 4479 | struct pool *pool = tc->pool; |
991d9fa0 JT |
4480 | |
4481 | /* | |
4482 | * We can't call dm_pool_get_data_dev_size() since that blocks. So | |
4483 | * we follow a more convoluted path through to the pool's target. | |
4484 | */ | |
55f2b8bd | 4485 | if (!pool->ti) |
991d9fa0 JT |
4486 | return 0; /* nothing is bound */ |
4487 | ||
55f2b8bd MS |
4488 | blocks = pool->ti->len; |
4489 | (void) sector_div(blocks, pool->sectors_per_block); | |
991d9fa0 | 4490 | if (blocks) |
55f2b8bd | 4491 | return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data); |
991d9fa0 JT |
4492 | |
4493 | return 0; | |
4494 | } | |
4495 | ||
34fbcf62 JT |
4496 | static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) |
4497 | { | |
4498 | struct thin_c *tc = ti->private; | |
4499 | struct pool *pool = tc->pool; | |
21607670 | 4500 | |
0fcb04d5 MS |
4501 | if (!pool->pf.discard_enabled) |
4502 | return; | |
34fbcf62 JT |
4503 | |
4504 | limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; | |
e2dd8aca | 4505 | limits->max_discard_sectors = pool->sectors_per_block * BIO_PRISON_MAX_RANGE; |
34fbcf62 JT |
4506 | } |
4507 | ||
991d9fa0 JT |
4508 | static struct target_type thin_target = { |
4509 | .name = "thin", | |
e2dd8aca | 4510 | .version = {1, 23, 0}, |
991d9fa0 JT |
4511 | .module = THIS_MODULE, |
4512 | .ctr = thin_ctr, | |
4513 | .dtr = thin_dtr, | |
4514 | .map = thin_map, | |
eb2aa48d | 4515 | .end_io = thin_endio, |
e5aea7b4 | 4516 | .preresume = thin_preresume, |
738211f7 | 4517 | .presuspend = thin_presuspend, |
991d9fa0 JT |
4518 | .postsuspend = thin_postsuspend, |
4519 | .status = thin_status, | |
4520 | .iterate_devices = thin_iterate_devices, | |
34fbcf62 | 4521 | .io_hints = thin_io_hints, |
991d9fa0 JT |
4522 | }; |
4523 | ||
4524 | /*----------------------------------------------------------------*/ | |
4525 | ||
4526 | static int __init dm_thin_init(void) | |
4527 | { | |
7e6358d2 | 4528 | int r = -ENOMEM; |
991d9fa0 JT |
4529 | |
4530 | pool_table_init(); | |
4531 | ||
7e6358d2 | 4532 | _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0); |
4533 | if (!_new_mapping_cache) | |
4534 | return r; | |
4535 | ||
991d9fa0 JT |
4536 | r = dm_register_target(&thin_target); |
4537 | if (r) | |
7e6358d2 | 4538 | goto bad_new_mapping_cache; |
991d9fa0 JT |
4539 | |
4540 | r = dm_register_target(&pool_target); | |
4541 | if (r) | |
7e6358d2 | 4542 | goto bad_thin_target; |
a24c2569 | 4543 | |
a24c2569 MS |
4544 | return 0; |
4545 | ||
7e6358d2 | 4546 | bad_thin_target: |
a24c2569 | 4547 | dm_unregister_target(&thin_target); |
7e6358d2 | 4548 | bad_new_mapping_cache: |
4549 | kmem_cache_destroy(_new_mapping_cache); | |
991d9fa0 JT |
4550 | |
4551 | return r; | |
4552 | } | |
4553 | ||
4554 | static void dm_thin_exit(void) | |
4555 | { | |
4556 | dm_unregister_target(&thin_target); | |
4557 | dm_unregister_target(&pool_target); | |
a24c2569 | 4558 | |
a24c2569 | 4559 | kmem_cache_destroy(_new_mapping_cache); |
d5ffebdd MS |
4560 | |
4561 | pool_table_exit(); | |
991d9fa0 JT |
4562 | } |
4563 | ||
4564 | module_init(dm_thin_init); | |
4565 | module_exit(dm_thin_exit); | |
4566 | ||
6a808034 | 4567 | module_param_named(no_space_timeout, no_space_timeout_secs, uint, 0644); |
80c57893 MS |
4568 | MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds"); |
4569 | ||
7cab8bf1 | 4570 | MODULE_DESCRIPTION(DM_NAME " thin provisioning target"); |
991d9fa0 JT |
4571 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
4572 | MODULE_LICENSE("GPL"); |