blob: ee29037ffc2e74633050b708718ccbe963bf20d1 [file] [log] [blame]
Joe Thornber991d9fa2011-10-31 20:21:18 +00001/*
Joe Thornbere49e5822012-07-27 15:08:16 +01002 * Copyright (C) 2011-2012 Red Hat UK.
Joe Thornber991d9fa2011-10-31 20:21:18 +00003 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-thin-metadata.h"
Mike Snitzer4f81a412012-10-12 21:02:13 +01008#include "dm-bio-prison.h"
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01009#include "dm.h"
Joe Thornber991d9fa2011-10-31 20:21:18 +000010
11#include <linux/device-mapper.h>
12#include <linux/dm-io.h>
13#include <linux/dm-kcopyd.h>
14#include <linux/list.h>
15#include <linux/init.h>
16#include <linux/module.h>
17#include <linux/slab.h>
18
19#define DM_MSG_PREFIX "thin"
20
21/*
22 * Tunable constants
23 */
Alasdair G Kergon7768ed32012-07-27 15:07:57 +010024#define ENDIO_HOOK_POOL_SIZE 1024
Joe Thornber991d9fa2011-10-31 20:21:18 +000025#define MAPPING_POOL_SIZE 1024
26#define PRISON_CELLS 1024
Joe Thornber905e51b2012-03-28 18:41:27 +010027#define COMMIT_PERIOD HZ
Joe Thornber991d9fa2011-10-31 20:21:18 +000028
Mikulas Patockadf5d2e92013-03-01 22:45:49 +000029DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
30 "A percentage of time allocated for copy on write");
31
Joe Thornber991d9fa2011-10-31 20:21:18 +000032/*
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
35 */
36#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
38
39/*
Joe Thornber991d9fa2011-10-31 20:21:18 +000040 * Device id is restricted to 24 bits.
41 */
42#define MAX_DEV_ID ((1 << 24) - 1)
43
44/*
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
47 *
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
53 * same data blocks.
54 *
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
57 *
58 * Let's say we write to a shared block in what was the origin. The
59 * steps are:
60 *
61 * i) plug io further to this physical block. (see bio_prison code).
62 *
63 * ii) quiesce any read io to that shared data block. Obviously
Mike Snitzer44feb382012-10-12 21:02:10 +010064 * including all devices that share this block. (see dm_deferred_set code)
Joe Thornber991d9fa2011-10-31 20:21:18 +000065 *
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
68 *
69 * iv) insert the new mapping into the origin's btree
Joe Thornberfe878f32012-03-28 18:41:24 +010070 * (process_prepared_mapping). This act of inserting breaks some
Joe Thornber991d9fa2011-10-31 20:21:18 +000071 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
76 *
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
79 *
80 * Steps (ii) and (iii) occur in parallel.
81 *
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
85 *
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
89 *
90 * - The snap mapping still points to the old block. As it would after
91 * the commit.
92 *
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
100 */
101
102/*----------------------------------------------------------------*/
103
104/*
Joe Thornber991d9fa2011-10-31 20:21:18 +0000105 * Key building.
106 */
107static void build_data_key(struct dm_thin_device *td,
Mike Snitzer44feb382012-10-12 21:02:10 +0100108 dm_block_t b, struct dm_cell_key *key)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000109{
110 key->virtual = 0;
111 key->dev = dm_thin_dev_id(td);
112 key->block = b;
113}
114
115static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
Mike Snitzer44feb382012-10-12 21:02:10 +0100116 struct dm_cell_key *key)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000117{
118 key->virtual = 1;
119 key->dev = dm_thin_dev_id(td);
120 key->block = b;
121}
122
123/*----------------------------------------------------------------*/
124
125/*
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
128 * devices.
129 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100130struct dm_thin_new_mapping;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100131
Joe Thornbere49e5822012-07-27 15:08:16 +0100132/*
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
134 */
135enum pool_mode {
136 PM_WRITE, /* metadata may be changed */
137 PM_READ_ONLY, /* metadata may not be changed */
138 PM_FAIL, /* all I/O fails */
139};
140
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100141struct pool_features {
Joe Thornbere49e5822012-07-27 15:08:16 +0100142 enum pool_mode mode;
143
Mike Snitzer9bc142d2012-09-26 23:45:46 +0100144 bool zero_new_blocks:1;
145 bool discard_enabled:1;
146 bool discard_passdown:1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100147};
148
Joe Thornbere49e5822012-07-27 15:08:16 +0100149struct thin_c;
150typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
151typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
152
Joe Thornber991d9fa2011-10-31 20:21:18 +0000153struct pool {
154 struct list_head list;
155 struct dm_target *ti; /* Only set if a pool target is bound */
156
157 struct mapped_device *pool_md;
158 struct block_device *md_dev;
159 struct dm_pool_metadata *pmd;
160
Joe Thornber991d9fa2011-10-31 20:21:18 +0000161 dm_block_t low_water_blocks;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100162 uint32_t sectors_per_block;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100163 int sectors_per_block_shift;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000164
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100165 struct pool_features pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000166 unsigned low_water_triggered:1; /* A dm event has been sent */
167 unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
168
Mike Snitzer44feb382012-10-12 21:02:10 +0100169 struct dm_bio_prison *prison;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000170 struct dm_kcopyd_client *copier;
171
172 struct workqueue_struct *wq;
173 struct work_struct worker;
Joe Thornber905e51b2012-03-28 18:41:27 +0100174 struct delayed_work waker;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000175
Joe Thornber905e51b2012-03-28 18:41:27 +0100176 unsigned long last_commit_jiffies;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100177 unsigned ref_count;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000178
179 spinlock_t lock;
180 struct bio_list deferred_bios;
181 struct bio_list deferred_flush_bios;
182 struct list_head prepared_mappings;
Joe Thornber104655f2012-03-28 18:41:28 +0100183 struct list_head prepared_discards;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000184
185 struct bio_list retry_on_resume_list;
186
Mike Snitzer44feb382012-10-12 21:02:10 +0100187 struct dm_deferred_set *shared_read_ds;
188 struct dm_deferred_set *all_io_ds;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000189
Mike Snitzera24c2562012-06-03 00:30:00 +0100190 struct dm_thin_new_mapping *next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000191 mempool_t *mapping_pool;
Joe Thornbere49e5822012-07-27 15:08:16 +0100192
193 process_bio_fn process_bio;
194 process_bio_fn process_discard;
195
196 process_mapping_fn process_prepared_mapping;
197 process_mapping_fn process_prepared_discard;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000198};
199
Joe Thornbere49e5822012-07-27 15:08:16 +0100200static enum pool_mode get_pool_mode(struct pool *pool);
201static void set_pool_mode(struct pool *pool, enum pool_mode mode);
202
Joe Thornber991d9fa2011-10-31 20:21:18 +0000203/*
204 * Target context for a pool.
205 */
206struct pool_c {
207 struct dm_target *ti;
208 struct pool *pool;
209 struct dm_dev *data_dev;
210 struct dm_dev *metadata_dev;
211 struct dm_target_callbacks callbacks;
212
213 dm_block_t low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +0100214 struct pool_features requested_pf; /* Features requested during table load */
215 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
Joe Thornber991d9fa2011-10-31 20:21:18 +0000216};
217
218/*
219 * Target context for a thin.
220 */
221struct thin_c {
222 struct dm_dev *pool_dev;
Joe Thornber2dd9c252012-03-28 18:41:28 +0100223 struct dm_dev *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000224 dm_thin_id dev_id;
225
226 struct pool *pool;
227 struct dm_thin_device *td;
228};
229
230/*----------------------------------------------------------------*/
231
Joe Thornber025b9682013-03-01 22:45:50 +0000232/*
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
235 */
236static void wake_worker(struct pool *pool)
237{
238 queue_work(pool->wq, &pool->worker);
239}
240
241/*----------------------------------------------------------------*/
242
Joe Thornber6beca5e2013-03-01 22:45:50 +0000243static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
244 struct dm_bio_prison_cell **cell_result)
245{
246 int r;
247 struct dm_bio_prison_cell *cell_prealloc;
248
249 /*
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
252 */
253 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
254
255 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
256 if (r)
257 /*
258 * We reused an old cell; we can get rid of
259 * the new one.
260 */
261 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
262
263 return r;
264}
265
266static void cell_release(struct pool *pool,
267 struct dm_bio_prison_cell *cell,
268 struct bio_list *bios)
269{
270 dm_cell_release(pool->prison, cell, bios);
271 dm_bio_prison_free_cell(pool->prison, cell);
272}
273
274static void cell_release_no_holder(struct pool *pool,
275 struct dm_bio_prison_cell *cell,
276 struct bio_list *bios)
277{
278 dm_cell_release_no_holder(pool->prison, cell, bios);
279 dm_bio_prison_free_cell(pool->prison, cell);
280}
281
Joe Thornber025b9682013-03-01 22:45:50 +0000282static void cell_defer_no_holder_no_free(struct thin_c *tc,
283 struct dm_bio_prison_cell *cell)
284{
285 struct pool *pool = tc->pool;
286 unsigned long flags;
287
288 spin_lock_irqsave(&pool->lock, flags);
289 dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
290 spin_unlock_irqrestore(&pool->lock, flags);
291
292 wake_worker(pool);
293}
294
Joe Thornber6beca5e2013-03-01 22:45:50 +0000295static void cell_error(struct pool *pool,
296 struct dm_bio_prison_cell *cell)
297{
298 dm_cell_error(pool->prison, cell);
299 dm_bio_prison_free_cell(pool->prison, cell);
300}
301
302/*----------------------------------------------------------------*/
303
Joe Thornber991d9fa2011-10-31 20:21:18 +0000304/*
305 * A global list of pools that uses a struct mapped_device as a key.
306 */
307static struct dm_thin_pool_table {
308 struct mutex mutex;
309 struct list_head pools;
310} dm_thin_pool_table;
311
312static void pool_table_init(void)
313{
314 mutex_init(&dm_thin_pool_table.mutex);
315 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
316}
317
318static void __pool_table_insert(struct pool *pool)
319{
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
321 list_add(&pool->list, &dm_thin_pool_table.pools);
322}
323
324static void __pool_table_remove(struct pool *pool)
325{
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
327 list_del(&pool->list);
328}
329
330static struct pool *__pool_table_lookup(struct mapped_device *md)
331{
332 struct pool *pool = NULL, *tmp;
333
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
335
336 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
337 if (tmp->pool_md == md) {
338 pool = tmp;
339 break;
340 }
341 }
342
343 return pool;
344}
345
346static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
347{
348 struct pool *pool = NULL, *tmp;
349
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
351
352 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
353 if (tmp->md_dev == md_dev) {
354 pool = tmp;
355 break;
356 }
357 }
358
359 return pool;
360}
361
362/*----------------------------------------------------------------*/
363
Mike Snitzera24c2562012-06-03 00:30:00 +0100364struct dm_thin_endio_hook {
Joe Thornbereb2aa482012-03-28 18:41:28 +0100365 struct thin_c *tc;
Mike Snitzer44feb382012-10-12 21:02:10 +0100366 struct dm_deferred_entry *shared_read_entry;
367 struct dm_deferred_entry *all_io_entry;
Mike Snitzera24c2562012-06-03 00:30:00 +0100368 struct dm_thin_new_mapping *overwrite_mapping;
Joe Thornbereb2aa482012-03-28 18:41:28 +0100369};
370
Joe Thornber991d9fa2011-10-31 20:21:18 +0000371static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
372{
373 struct bio *bio;
374 struct bio_list bios;
375
376 bio_list_init(&bios);
377 bio_list_merge(&bios, master);
378 bio_list_init(master);
379
380 while ((bio = bio_list_pop(&bios))) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000381 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100382
Joe Thornbereb2aa482012-03-28 18:41:28 +0100383 if (h->tc == tc)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000384 bio_endio(bio, DM_ENDIO_REQUEUE);
385 else
386 bio_list_add(master, bio);
387 }
388}
389
390static void requeue_io(struct thin_c *tc)
391{
392 struct pool *pool = tc->pool;
393 unsigned long flags;
394
395 spin_lock_irqsave(&pool->lock, flags);
396 __requeue_bio_list(tc, &pool->deferred_bios);
397 __requeue_bio_list(tc, &pool->retry_on_resume_list);
398 spin_unlock_irqrestore(&pool->lock, flags);
399}
400
401/*
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
405 * target.
406 */
407
Mike Snitzer58f77a22013-03-01 22:45:45 +0000408static bool block_size_is_power_of_two(struct pool *pool)
409{
410 return pool->sectors_per_block_shift >= 0;
411}
412
Joe Thornber991d9fa2011-10-31 20:21:18 +0000413static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
414{
Mike Snitzer58f77a22013-03-01 22:45:45 +0000415 struct pool *pool = tc->pool;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100416 sector_t block_nr = bio->bi_sector;
417
Mike Snitzer58f77a22013-03-01 22:45:45 +0000418 if (block_size_is_power_of_two(pool))
419 block_nr >>= pool->sectors_per_block_shift;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100420 else
Mike Snitzer58f77a22013-03-01 22:45:45 +0000421 (void) sector_div(block_nr, pool->sectors_per_block);
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100422
423 return block_nr;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000424}
425
426static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
427{
428 struct pool *pool = tc->pool;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100429 sector_t bi_sector = bio->bi_sector;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000430
431 bio->bi_bdev = tc->pool_dev->bdev;
Mike Snitzer58f77a22013-03-01 22:45:45 +0000432 if (block_size_is_power_of_two(pool))
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100433 bio->bi_sector = (block << pool->sectors_per_block_shift) |
434 (bi_sector & (pool->sectors_per_block - 1));
Mike Snitzer58f77a22013-03-01 22:45:45 +0000435 else
436 bio->bi_sector = (block * pool->sectors_per_block) +
437 sector_div(bi_sector, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000438}
439
Joe Thornber2dd9c252012-03-28 18:41:28 +0100440static void remap_to_origin(struct thin_c *tc, struct bio *bio)
441{
442 bio->bi_bdev = tc->origin_dev->bdev;
443}
444
Joe Thornber4afdd682012-07-27 15:08:14 +0100445static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
446{
447 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
448 dm_thin_changed_this_transaction(tc->td);
449}
450
Joe Thornbere8088072012-12-21 20:23:31 +0000451static void inc_all_io_entry(struct pool *pool, struct bio *bio)
452{
453 struct dm_thin_endio_hook *h;
454
455 if (bio->bi_rw & REQ_DISCARD)
456 return;
457
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000458 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbere8088072012-12-21 20:23:31 +0000459 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
460}
461
Joe Thornber2dd9c252012-03-28 18:41:28 +0100462static void issue(struct thin_c *tc, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000463{
464 struct pool *pool = tc->pool;
465 unsigned long flags;
466
Joe Thornbere49e5822012-07-27 15:08:16 +0100467 if (!bio_triggers_commit(tc, bio)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000468 generic_make_request(bio);
Joe Thornbere49e5822012-07-27 15:08:16 +0100469 return;
470 }
471
472 /*
473 * Complete bio with an error if earlier I/O caused changes to
474 * the metadata that can't be committed e.g, due to I/O errors
475 * on the metadata device.
476 */
477 if (dm_thin_aborted_changes(tc->td)) {
478 bio_io_error(bio);
479 return;
480 }
481
482 /*
483 * Batch together any bios that trigger commits and then issue a
484 * single commit for them in process_deferred_bios().
485 */
486 spin_lock_irqsave(&pool->lock, flags);
487 bio_list_add(&pool->deferred_flush_bios, bio);
488 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000489}
490
Joe Thornber2dd9c252012-03-28 18:41:28 +0100491static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
492{
493 remap_to_origin(tc, bio);
494 issue(tc, bio);
495}
496
497static void remap_and_issue(struct thin_c *tc, struct bio *bio,
498 dm_block_t block)
499{
500 remap(tc, bio, block);
501 issue(tc, bio);
502}
503
Joe Thornber991d9fa2011-10-31 20:21:18 +0000504/*----------------------------------------------------------------*/
505
506/*
507 * Bio endio functions.
508 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100509struct dm_thin_new_mapping {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000510 struct list_head list;
511
Joe Thornbereb2aa482012-03-28 18:41:28 +0100512 unsigned quiesced:1;
513 unsigned prepared:1;
Joe Thornber104655f2012-03-28 18:41:28 +0100514 unsigned pass_discard:1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000515
516 struct thin_c *tc;
517 dm_block_t virt_block;
518 dm_block_t data_block;
Mike Snitzera24c2562012-06-03 00:30:00 +0100519 struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000520 int err;
521
522 /*
523 * If the bio covers the whole area of a block then we can avoid
524 * zeroing or copying. Instead this bio is hooked. The bio will
525 * still be in the cell, so care has to be taken to avoid issuing
526 * the bio twice.
527 */
528 struct bio *bio;
529 bio_end_io_t *saved_bi_end_io;
530};
531
Mike Snitzera24c2562012-06-03 00:30:00 +0100532static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000533{
534 struct pool *pool = m->tc->pool;
535
Joe Thornbereb2aa482012-03-28 18:41:28 +0100536 if (m->quiesced && m->prepared) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000537 list_add(&m->list, &pool->prepared_mappings);
538 wake_worker(pool);
539 }
540}
541
542static void copy_complete(int read_err, unsigned long write_err, void *context)
543{
544 unsigned long flags;
Mike Snitzera24c2562012-06-03 00:30:00 +0100545 struct dm_thin_new_mapping *m = context;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000546 struct pool *pool = m->tc->pool;
547
548 m->err = read_err || write_err ? -EIO : 0;
549
550 spin_lock_irqsave(&pool->lock, flags);
551 m->prepared = 1;
552 __maybe_add_mapping(m);
553 spin_unlock_irqrestore(&pool->lock, flags);
554}
555
556static void overwrite_endio(struct bio *bio, int err)
557{
558 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000559 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100560 struct dm_thin_new_mapping *m = h->overwrite_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000561 struct pool *pool = m->tc->pool;
562
563 m->err = err;
564
565 spin_lock_irqsave(&pool->lock, flags);
566 m->prepared = 1;
567 __maybe_add_mapping(m);
568 spin_unlock_irqrestore(&pool->lock, flags);
569}
570
Joe Thornber991d9fa2011-10-31 20:21:18 +0000571/*----------------------------------------------------------------*/
572
573/*
574 * Workqueue.
575 */
576
577/*
578 * Prepared mapping jobs.
579 */
580
581/*
582 * This sends the bios in the cell back to the deferred_bios list.
583 */
Joe Thornber2aab3852012-12-21 20:23:33 +0000584static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000585{
586 struct pool *pool = tc->pool;
587 unsigned long flags;
588
589 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000590 cell_release(pool, cell, &pool->deferred_bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000591 spin_unlock_irqrestore(&tc->pool->lock, flags);
592
593 wake_worker(pool);
594}
595
596/*
Joe Thornber6beca5e2013-03-01 22:45:50 +0000597 * Same as cell_defer above, except it omits the original holder of the cell.
Joe Thornber991d9fa2011-10-31 20:21:18 +0000598 */
Joe Thornberf286ba02012-12-21 20:23:33 +0000599static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000600{
Joe Thornber991d9fa2011-10-31 20:21:18 +0000601 struct pool *pool = tc->pool;
602 unsigned long flags;
603
Joe Thornber991d9fa2011-10-31 20:21:18 +0000604 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000605 cell_release_no_holder(pool, cell, &pool->deferred_bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000606 spin_unlock_irqrestore(&pool->lock, flags);
607
608 wake_worker(pool);
609}
610
Joe Thornbere49e5822012-07-27 15:08:16 +0100611static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
612{
613 if (m->bio)
614 m->bio->bi_end_io = m->saved_bi_end_io;
Joe Thornber6beca5e2013-03-01 22:45:50 +0000615 cell_error(m->tc->pool, m->cell);
Joe Thornbere49e5822012-07-27 15:08:16 +0100616 list_del(&m->list);
617 mempool_free(m, m->tc->pool->mapping_pool);
618}
Joe Thornber025b9682013-03-01 22:45:50 +0000619
Mike Snitzera24c2562012-06-03 00:30:00 +0100620static void process_prepared_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000621{
622 struct thin_c *tc = m->tc;
Joe Thornber6beca5e2013-03-01 22:45:50 +0000623 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000624 struct bio *bio;
625 int r;
626
627 bio = m->bio;
628 if (bio)
629 bio->bi_end_io = m->saved_bi_end_io;
630
631 if (m->err) {
Joe Thornber6beca5e2013-03-01 22:45:50 +0000632 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100633 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000634 }
635
636 /*
637 * Commit the prepared block into the mapping btree.
638 * Any I/O for this block arriving after this point will get
639 * remapped to it directly.
640 */
641 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
642 if (r) {
Joe Thornberfafc7a82013-12-02 17:57:42 -0500643 DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
644 dm_device_name(pool->pool_md), r);
645 set_pool_mode(pool, PM_READ_ONLY);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000646 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100647 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000648 }
649
650 /*
651 * Release any bios held while the block was being provisioned.
652 * If we are processing a write bio that completely covers the block,
653 * we already processed it so can ignore it now when processing
654 * the bios in the cell.
655 */
656 if (bio) {
Joe Thornberf286ba02012-12-21 20:23:33 +0000657 cell_defer_no_holder(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000658 bio_endio(bio, 0);
659 } else
Joe Thornber2aab3852012-12-21 20:23:33 +0000660 cell_defer(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000661
Joe Thornber905386f2012-07-27 15:08:05 +0100662out:
Joe Thornber991d9fa2011-10-31 20:21:18 +0000663 list_del(&m->list);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000664 mempool_free(m, pool->mapping_pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000665}
666
Joe Thornbere49e5822012-07-27 15:08:16 +0100667static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber104655f2012-03-28 18:41:28 +0100668{
Joe Thornber104655f2012-03-28 18:41:28 +0100669 struct thin_c *tc = m->tc;
670
Joe Thornbere49e5822012-07-27 15:08:16 +0100671 bio_io_error(m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000672 cell_defer_no_holder(tc, m->cell);
673 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere49e5822012-07-27 15:08:16 +0100674 mempool_free(m, tc->pool->mapping_pool);
675}
Joe Thornber104655f2012-03-28 18:41:28 +0100676
Joe Thornbere49e5822012-07-27 15:08:16 +0100677static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
678{
679 struct thin_c *tc = m->tc;
680
Joe Thornbere8088072012-12-21 20:23:31 +0000681 inc_all_io_entry(tc->pool, m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000682 cell_defer_no_holder(tc, m->cell);
683 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere8088072012-12-21 20:23:31 +0000684
Joe Thornber104655f2012-03-28 18:41:28 +0100685 if (m->pass_discard)
686 remap_and_issue(tc, m->bio, m->data_block);
687 else
688 bio_endio(m->bio, 0);
689
Joe Thornber104655f2012-03-28 18:41:28 +0100690 mempool_free(m, tc->pool->mapping_pool);
691}
692
Joe Thornbere49e5822012-07-27 15:08:16 +0100693static void process_prepared_discard(struct dm_thin_new_mapping *m)
694{
695 int r;
696 struct thin_c *tc = m->tc;
697
698 r = dm_thin_remove_block(tc->td, m->virt_block);
699 if (r)
Mike Snitzerc3977412012-12-21 20:23:34 +0000700 DMERR_LIMIT("dm_thin_remove_block() failed");
Joe Thornbere49e5822012-07-27 15:08:16 +0100701
702 process_prepared_discard_passdown(m);
703}
704
Joe Thornber104655f2012-03-28 18:41:28 +0100705static void process_prepared(struct pool *pool, struct list_head *head,
Joe Thornbere49e5822012-07-27 15:08:16 +0100706 process_mapping_fn *fn)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000707{
708 unsigned long flags;
709 struct list_head maps;
Mike Snitzera24c2562012-06-03 00:30:00 +0100710 struct dm_thin_new_mapping *m, *tmp;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000711
712 INIT_LIST_HEAD(&maps);
713 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +0100714 list_splice_init(head, &maps);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000715 spin_unlock_irqrestore(&pool->lock, flags);
716
717 list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornbere49e5822012-07-27 15:08:16 +0100718 (*fn)(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000719}
720
721/*
722 * Deferred bio jobs.
723 */
Joe Thornber104655f2012-03-28 18:41:28 +0100724static int io_overlaps_block(struct pool *pool, struct bio *bio)
725{
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100726 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber104655f2012-03-28 18:41:28 +0100727}
728
Joe Thornber991d9fa2011-10-31 20:21:18 +0000729static int io_overwrites_block(struct pool *pool, struct bio *bio)
730{
Joe Thornber104655f2012-03-28 18:41:28 +0100731 return (bio_data_dir(bio) == WRITE) &&
732 io_overlaps_block(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000733}
734
735static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
736 bio_end_io_t *fn)
737{
738 *save = bio->bi_end_io;
739 bio->bi_end_io = fn;
740}
741
742static int ensure_next_mapping(struct pool *pool)
743{
744 if (pool->next_mapping)
745 return 0;
746
747 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
748
749 return pool->next_mapping ? 0 : -ENOMEM;
750}
751
Mike Snitzera24c2562012-06-03 00:30:00 +0100752static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000753{
Mike Snitzera24c2562012-06-03 00:30:00 +0100754 struct dm_thin_new_mapping *r = pool->next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000755
756 BUG_ON(!pool->next_mapping);
757
758 pool->next_mapping = NULL;
759
760 return r;
761}
762
763static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
Joe Thornber2dd9c252012-03-28 18:41:28 +0100764 struct dm_dev *origin, dm_block_t data_origin,
765 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100766 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000767{
768 int r;
769 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100770 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000771
772 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +0100773 m->quiesced = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000774 m->prepared = 0;
775 m->tc = tc;
776 m->virt_block = virt_block;
777 m->data_block = data_dest;
778 m->cell = cell;
779 m->err = 0;
780 m->bio = NULL;
781
Mike Snitzer44feb382012-10-12 21:02:10 +0100782 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
Joe Thornbereb2aa482012-03-28 18:41:28 +0100783 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000784
785 /*
786 * IO to pool_dev remaps to the pool target's data_dev.
787 *
788 * If the whole block of data is being overwritten, we can issue the
789 * bio immediately. Otherwise we use kcopyd to clone the data first.
790 */
791 if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000792 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100793
Joe Thornbereb2aa482012-03-28 18:41:28 +0100794 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000795 m->bio = bio;
796 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000797 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000798 remap_and_issue(tc, bio, data_dest);
799 } else {
800 struct dm_io_region from, to;
801
Joe Thornber2dd9c252012-03-28 18:41:28 +0100802 from.bdev = origin->bdev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000803 from.sector = data_origin * pool->sectors_per_block;
804 from.count = pool->sectors_per_block;
805
806 to.bdev = tc->pool_dev->bdev;
807 to.sector = data_dest * pool->sectors_per_block;
808 to.count = pool->sectors_per_block;
809
810 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
811 0, copy_complete, m);
812 if (r < 0) {
813 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000814 DMERR_LIMIT("dm_kcopyd_copy() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000815 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000816 }
817 }
818}
819
Joe Thornber2dd9c252012-03-28 18:41:28 +0100820static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
821 dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100822 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100823{
824 schedule_copy(tc, virt_block, tc->pool_dev,
825 data_origin, data_dest, cell, bio);
826}
827
828static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
829 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100830 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100831{
832 schedule_copy(tc, virt_block, tc->origin_dev,
833 virt_block, data_dest, cell, bio);
834}
835
Joe Thornber991d9fa2011-10-31 20:21:18 +0000836static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzera24c2562012-06-03 00:30:00 +0100837 dm_block_t data_block, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +0000838 struct bio *bio)
839{
840 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100841 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000842
843 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +0100844 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000845 m->prepared = 0;
846 m->tc = tc;
847 m->virt_block = virt_block;
848 m->data_block = data_block;
849 m->cell = cell;
850 m->err = 0;
851 m->bio = NULL;
852
853 /*
854 * If the whole block of data is being overwritten or we are not
855 * zeroing pre-existing data, we can issue the bio immediately.
856 * Otherwise we use kcopyd to zero the data first.
857 */
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100858 if (!pool->pf.zero_new_blocks)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000859 process_prepared_mapping(m);
860
861 else if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000862 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100863
Joe Thornbereb2aa482012-03-28 18:41:28 +0100864 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000865 m->bio = bio;
866 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000867 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000868 remap_and_issue(tc, bio, data_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000869 } else {
870 int r;
871 struct dm_io_region to;
872
873 to.bdev = tc->pool_dev->bdev;
874 to.sector = data_block * pool->sectors_per_block;
875 to.count = pool->sectors_per_block;
876
877 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
878 if (r < 0) {
879 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000880 DMERR_LIMIT("dm_kcopyd_zero() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000881 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000882 }
883 }
884}
885
Joe Thornbere49e5822012-07-27 15:08:16 +0100886/*
887 * A non-zero return indicates read_only or fail_io mode.
888 * Many callers don't care about the return value.
889 */
Joe Thornber020cc3b2013-12-04 15:05:36 -0500890static int commit(struct pool *pool)
Joe Thornbere49e5822012-07-27 15:08:16 +0100891{
892 int r;
893
894 if (get_pool_mode(pool) != PM_WRITE)
895 return -EINVAL;
896
Joe Thornber020cc3b2013-12-04 15:05:36 -0500897 r = dm_pool_commit_metadata(pool->pmd);
898 if (r) {
899 DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
900 dm_device_name(pool->pool_md), r);
Joe Thornbere49e5822012-07-27 15:08:16 +0100901 set_pool_mode(pool, PM_READ_ONLY);
Joe Thornber020cc3b2013-12-04 15:05:36 -0500902 }
Joe Thornbere49e5822012-07-27 15:08:16 +0100903
904 return r;
905}
906
Joe Thornber991d9fa2011-10-31 20:21:18 +0000907static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
908{
909 int r;
910 dm_block_t free_blocks;
911 unsigned long flags;
912 struct pool *pool = tc->pool;
913
Mike Snitzer94563ba2013-08-22 09:56:18 -0400914 /*
915 * Once no_free_space is set we must not allow allocation to succeed.
916 * Otherwise it is difficult to explain, debug, test and support.
917 */
918 if (pool->no_free_space)
919 return -ENOSPC;
920
Joe Thornber991d9fa2011-10-31 20:21:18 +0000921 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
922 if (r)
923 return r;
924
925 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
Joe Thornberb17446d2013-05-10 14:37:18 +0100926 DMWARN("%s: reached low water mark for data device: sending event.",
Joe Thornber991d9fa2011-10-31 20:21:18 +0000927 dm_device_name(pool->pool_md));
928 spin_lock_irqsave(&pool->lock, flags);
929 pool->low_water_triggered = 1;
930 spin_unlock_irqrestore(&pool->lock, flags);
931 dm_table_event(pool->ti->table);
932 }
933
934 if (!free_blocks) {
Mike Snitzer94563ba2013-08-22 09:56:18 -0400935 /*
936 * Try to commit to see if that will free up some
937 * more space.
938 */
Joe Thornber020cc3b2013-12-04 15:05:36 -0500939 r = commit(pool);
940 if (r)
941 return r;
Mike Snitzer94563ba2013-08-22 09:56:18 -0400942
943 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
944 if (r)
945 return r;
946
947 /*
948 * If we still have no space we set a flag to avoid
949 * doing all this checking and return -ENOSPC. This
950 * flag serves as a latch that disallows allocations from
951 * this pool until the admin takes action (e.g. resize or
952 * table reload).
953 */
954 if (!free_blocks) {
Mike Snitzer4a02b342013-12-03 12:20:57 -0500955 DMWARN("%s: no free data space available.",
Mike Snitzer94563ba2013-08-22 09:56:18 -0400956 dm_device_name(pool->pool_md));
957 spin_lock_irqsave(&pool->lock, flags);
958 pool->no_free_space = 1;
959 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000960 return -ENOSPC;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000961 }
962 }
963
964 r = dm_pool_alloc_data_block(pool->pmd, result);
Mike Snitzer4a02b342013-12-03 12:20:57 -0500965 if (r) {
966 if (r == -ENOSPC &&
967 !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
968 !free_blocks) {
969 DMWARN("%s: no free metadata space available.",
970 dm_device_name(pool->pool_md));
971 set_pool_mode(pool, PM_READ_ONLY);
972 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000973 return r;
Mike Snitzer4a02b342013-12-03 12:20:57 -0500974 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000975
976 return 0;
977}
978
979/*
980 * If we have run out of space, queue bios until the device is
981 * resumed, presumably after having been reloaded with more space.
982 */
983static void retry_on_resume(struct bio *bio)
984{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000985 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +0100986 struct thin_c *tc = h->tc;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000987 struct pool *pool = tc->pool;
988 unsigned long flags;
989
990 spin_lock_irqsave(&pool->lock, flags);
991 bio_list_add(&pool->retry_on_resume_list, bio);
992 spin_unlock_irqrestore(&pool->lock, flags);
993}
994
Joe Thornber6beca5e2013-03-01 22:45:50 +0000995static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000996{
997 struct bio *bio;
998 struct bio_list bios;
999
1000 bio_list_init(&bios);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001001 cell_release(pool, cell, &bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001002
1003 while ((bio = bio_list_pop(&bios)))
1004 retry_on_resume(bio);
1005}
1006
Joe Thornber104655f2012-03-28 18:41:28 +01001007static void process_discard(struct thin_c *tc, struct bio *bio)
1008{
1009 int r;
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001010 unsigned long flags;
Joe Thornber104655f2012-03-28 18:41:28 +01001011 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001012 struct dm_bio_prison_cell *cell, *cell2;
Mike Snitzer44feb382012-10-12 21:02:10 +01001013 struct dm_cell_key key, key2;
Joe Thornber104655f2012-03-28 18:41:28 +01001014 dm_block_t block = get_bio_block(tc, bio);
1015 struct dm_thin_lookup_result lookup_result;
Mike Snitzera24c2562012-06-03 00:30:00 +01001016 struct dm_thin_new_mapping *m;
Joe Thornber104655f2012-03-28 18:41:28 +01001017
1018 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001019 if (bio_detain(tc->pool, &key, bio, &cell))
Joe Thornber104655f2012-03-28 18:41:28 +01001020 return;
1021
1022 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1023 switch (r) {
1024 case 0:
1025 /*
1026 * Check nobody is fiddling with this pool block. This can
1027 * happen if someone's in the process of breaking sharing
1028 * on this block.
1029 */
1030 build_data_key(tc->td, lookup_result.block, &key2);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001031 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
Joe Thornberf286ba02012-12-21 20:23:33 +00001032 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001033 break;
1034 }
1035
1036 if (io_overlaps_block(pool, bio)) {
1037 /*
1038 * IO may still be going to the destination block. We must
1039 * quiesce before we can do the removal.
1040 */
1041 m = get_next_mapping(pool);
1042 m->tc = tc;
Mike Snitzer17b7d632012-07-27 15:07:57 +01001043 m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
Joe Thornber104655f2012-03-28 18:41:28 +01001044 m->virt_block = block;
1045 m->data_block = lookup_result.block;
1046 m->cell = cell;
1047 m->cell2 = cell2;
1048 m->err = 0;
1049 m->bio = bio;
1050
Mike Snitzer44feb382012-10-12 21:02:10 +01001051 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001052 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001053 list_add(&m->list, &pool->prepared_discards);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001054 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001055 wake_worker(pool);
1056 }
1057 } else {
Joe Thornbere8088072012-12-21 20:23:31 +00001058 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001059 cell_defer_no_holder(tc, cell);
1060 cell_defer_no_holder(tc, cell2);
Joe Thornbere8088072012-12-21 20:23:31 +00001061
Joe Thornber104655f2012-03-28 18:41:28 +01001062 /*
Mikulas Patocka49296302012-07-27 15:08:03 +01001063 * The DM core makes sure that the discard doesn't span
1064 * a block boundary. So we submit the discard of a
1065 * partial block appropriately.
Joe Thornber104655f2012-03-28 18:41:28 +01001066 */
Mikulas Patocka650d2a02012-07-20 14:25:05 +01001067 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1068 remap_and_issue(tc, bio, lookup_result.block);
1069 else
1070 bio_endio(bio, 0);
Joe Thornber104655f2012-03-28 18:41:28 +01001071 }
1072 break;
1073
1074 case -ENODATA:
1075 /*
1076 * It isn't provisioned, just forget it.
1077 */
Joe Thornberf286ba02012-12-21 20:23:33 +00001078 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001079 bio_endio(bio, 0);
1080 break;
1081
1082 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001083 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1084 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001085 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001086 bio_io_error(bio);
1087 break;
1088 }
1089}
1090
Joe Thornber991d9fa2011-10-31 20:21:18 +00001091static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzer44feb382012-10-12 21:02:10 +01001092 struct dm_cell_key *key,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001093 struct dm_thin_lookup_result *lookup_result,
Mike Snitzera24c2562012-06-03 00:30:00 +01001094 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001095{
1096 int r;
1097 dm_block_t data_block;
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001098 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001099
1100 r = alloc_data_block(tc, &data_block);
1101 switch (r) {
1102 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001103 schedule_internal_copy(tc, block, lookup_result->block,
1104 data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001105 break;
1106
1107 case -ENOSPC:
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001108 no_space(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001109 break;
1110
1111 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001112 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1113 __func__, r);
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001114 set_pool_mode(pool, PM_READ_ONLY);
1115 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001116 break;
1117 }
1118}
1119
1120static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1121 dm_block_t block,
1122 struct dm_thin_lookup_result *lookup_result)
1123{
Mike Snitzera24c2562012-06-03 00:30:00 +01001124 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001125 struct pool *pool = tc->pool;
Mike Snitzer44feb382012-10-12 21:02:10 +01001126 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001127
1128 /*
1129 * If cell is already occupied, then sharing is already in the process
1130 * of being broken so we have nothing further to do here.
1131 */
1132 build_data_key(tc->td, lookup_result->block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001133 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001134 return;
1135
Joe Thornber60049702012-07-27 15:08:06 +01001136 if (bio_data_dir(bio) == WRITE && bio->bi_size)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001137 break_sharing(tc, bio, block, &key, lookup_result, cell);
1138 else {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001139 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornber991d9fa2011-10-31 20:21:18 +00001140
Mike Snitzer44feb382012-10-12 21:02:10 +01001141 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
Joe Thornbere8088072012-12-21 20:23:31 +00001142 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001143 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001144
Joe Thornber991d9fa2011-10-31 20:21:18 +00001145 remap_and_issue(tc, bio, lookup_result->block);
1146 }
1147}
1148
1149static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001150 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001151{
1152 int r;
1153 dm_block_t data_block;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001154 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001155
1156 /*
1157 * Remap empty bios (flushes) immediately, without provisioning.
1158 */
1159 if (!bio->bi_size) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001160 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001161 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001162
Joe Thornber991d9fa2011-10-31 20:21:18 +00001163 remap_and_issue(tc, bio, 0);
1164 return;
1165 }
1166
1167 /*
1168 * Fill read bios with zeroes and complete them immediately.
1169 */
1170 if (bio_data_dir(bio) == READ) {
1171 zero_fill_bio(bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001172 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001173 bio_endio(bio, 0);
1174 return;
1175 }
1176
1177 r = alloc_data_block(tc, &data_block);
1178 switch (r) {
1179 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001180 if (tc->origin_dev)
1181 schedule_external_copy(tc, block, data_block, cell, bio);
1182 else
1183 schedule_zero(tc, block, data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001184 break;
1185
1186 case -ENOSPC:
Joe Thornber6beca5e2013-03-01 22:45:50 +00001187 no_space(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001188 break;
1189
1190 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001191 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1192 __func__, r);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001193 set_pool_mode(pool, PM_READ_ONLY);
1194 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001195 break;
1196 }
1197}
1198
1199static void process_bio(struct thin_c *tc, struct bio *bio)
1200{
1201 int r;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001202 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001203 dm_block_t block = get_bio_block(tc, bio);
Mike Snitzera24c2562012-06-03 00:30:00 +01001204 struct dm_bio_prison_cell *cell;
Mike Snitzer44feb382012-10-12 21:02:10 +01001205 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001206 struct dm_thin_lookup_result lookup_result;
1207
1208 /*
1209 * If cell is already occupied, then the block is already
1210 * being provisioned so we have nothing further to do here.
1211 */
1212 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001213 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001214 return;
1215
1216 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1217 switch (r) {
1218 case 0:
Joe Thornbere8088072012-12-21 20:23:31 +00001219 if (lookup_result.shared) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001220 process_shared_bio(tc, bio, block, &lookup_result);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001221 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
Joe Thornbere8088072012-12-21 20:23:31 +00001222 } else {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001223 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001224 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001225
Joe Thornber991d9fa2011-10-31 20:21:18 +00001226 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001227 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001228 break;
1229
1230 case -ENODATA:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001231 if (bio_data_dir(bio) == READ && tc->origin_dev) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001232 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001233 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001234
Joe Thornber2dd9c252012-03-28 18:41:28 +01001235 remap_to_origin_and_issue(tc, bio);
1236 } else
1237 provision_block(tc, bio, block, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001238 break;
1239
1240 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001241 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1242 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001243 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001244 bio_io_error(bio);
1245 break;
1246 }
1247}
1248
Joe Thornbere49e5822012-07-27 15:08:16 +01001249static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1250{
1251 int r;
1252 int rw = bio_data_dir(bio);
1253 dm_block_t block = get_bio_block(tc, bio);
1254 struct dm_thin_lookup_result lookup_result;
1255
1256 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1257 switch (r) {
1258 case 0:
1259 if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1260 bio_io_error(bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001261 else {
1262 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001263 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001264 }
Joe Thornbere49e5822012-07-27 15:08:16 +01001265 break;
1266
1267 case -ENODATA:
1268 if (rw != READ) {
1269 bio_io_error(bio);
1270 break;
1271 }
1272
1273 if (tc->origin_dev) {
Joe Thornbere8088072012-12-21 20:23:31 +00001274 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001275 remap_to_origin_and_issue(tc, bio);
1276 break;
1277 }
1278
1279 zero_fill_bio(bio);
1280 bio_endio(bio, 0);
1281 break;
1282
1283 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001284 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1285 __func__, r);
Joe Thornbere49e5822012-07-27 15:08:16 +01001286 bio_io_error(bio);
1287 break;
1288 }
1289}
1290
1291static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1292{
1293 bio_io_error(bio);
1294}
1295
Joe Thornberac8c3f32013-05-10 14:37:21 +01001296/*
1297 * FIXME: should we also commit due to size of transaction, measured in
1298 * metadata blocks?
1299 */
Joe Thornber905e51b2012-03-28 18:41:27 +01001300static int need_commit_due_to_time(struct pool *pool)
1301{
1302 return jiffies < pool->last_commit_jiffies ||
1303 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1304}
1305
Joe Thornber991d9fa2011-10-31 20:21:18 +00001306static void process_deferred_bios(struct pool *pool)
1307{
1308 unsigned long flags;
1309 struct bio *bio;
1310 struct bio_list bios;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001311
1312 bio_list_init(&bios);
1313
1314 spin_lock_irqsave(&pool->lock, flags);
1315 bio_list_merge(&bios, &pool->deferred_bios);
1316 bio_list_init(&pool->deferred_bios);
1317 spin_unlock_irqrestore(&pool->lock, flags);
1318
1319 while ((bio = bio_list_pop(&bios))) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001320 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001321 struct thin_c *tc = h->tc;
1322
Joe Thornber991d9fa2011-10-31 20:21:18 +00001323 /*
1324 * If we've got no free new_mapping structs, and processing
1325 * this bio might require one, we pause until there are some
1326 * prepared mappings to process.
1327 */
1328 if (ensure_next_mapping(pool)) {
1329 spin_lock_irqsave(&pool->lock, flags);
1330 bio_list_merge(&pool->deferred_bios, &bios);
1331 spin_unlock_irqrestore(&pool->lock, flags);
1332
1333 break;
1334 }
Joe Thornber104655f2012-03-28 18:41:28 +01001335
1336 if (bio->bi_rw & REQ_DISCARD)
Joe Thornbere49e5822012-07-27 15:08:16 +01001337 pool->process_discard(tc, bio);
Joe Thornber104655f2012-03-28 18:41:28 +01001338 else
Joe Thornbere49e5822012-07-27 15:08:16 +01001339 pool->process_bio(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001340 }
1341
1342 /*
1343 * If there are any deferred flush bios, we must commit
1344 * the metadata before issuing them.
1345 */
1346 bio_list_init(&bios);
1347 spin_lock_irqsave(&pool->lock, flags);
1348 bio_list_merge(&bios, &pool->deferred_flush_bios);
1349 bio_list_init(&pool->deferred_flush_bios);
1350 spin_unlock_irqrestore(&pool->lock, flags);
1351
Joe Thornber905e51b2012-03-28 18:41:27 +01001352 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001353 return;
1354
Joe Thornber020cc3b2013-12-04 15:05:36 -05001355 if (commit(pool)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001356 while ((bio = bio_list_pop(&bios)))
1357 bio_io_error(bio);
1358 return;
1359 }
Joe Thornber905e51b2012-03-28 18:41:27 +01001360 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001361
1362 while ((bio = bio_list_pop(&bios)))
1363 generic_make_request(bio);
1364}
1365
1366static void do_worker(struct work_struct *ws)
1367{
1368 struct pool *pool = container_of(ws, struct pool, worker);
1369
Joe Thornbere49e5822012-07-27 15:08:16 +01001370 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1371 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001372 process_deferred_bios(pool);
1373}
1374
Joe Thornber905e51b2012-03-28 18:41:27 +01001375/*
1376 * We want to commit periodically so that not too much
1377 * unwritten data builds up.
1378 */
1379static void do_waker(struct work_struct *ws)
1380{
1381 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1382 wake_worker(pool);
1383 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1384}
1385
Joe Thornber991d9fa2011-10-31 20:21:18 +00001386/*----------------------------------------------------------------*/
1387
Joe Thornbere49e5822012-07-27 15:08:16 +01001388static enum pool_mode get_pool_mode(struct pool *pool)
1389{
1390 return pool->pf.mode;
1391}
1392
1393static void set_pool_mode(struct pool *pool, enum pool_mode mode)
1394{
1395 int r;
1396
1397 pool->pf.mode = mode;
1398
1399 switch (mode) {
1400 case PM_FAIL:
Mike Snitzer4fa59712013-08-21 17:30:40 -04001401 DMERR("%s: switching pool to failure mode",
1402 dm_device_name(pool->pool_md));
Joe Thornber5383ef32013-12-04 16:30:01 -05001403 dm_pool_metadata_read_only(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01001404 pool->process_bio = process_bio_fail;
1405 pool->process_discard = process_bio_fail;
1406 pool->process_prepared_mapping = process_prepared_mapping_fail;
1407 pool->process_prepared_discard = process_prepared_discard_fail;
1408 break;
1409
1410 case PM_READ_ONLY:
Mike Snitzer4fa59712013-08-21 17:30:40 -04001411 DMERR("%s: switching pool to read-only mode",
1412 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001413 r = dm_pool_abort_metadata(pool->pmd);
1414 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04001415 DMERR("%s: aborting transaction failed",
1416 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001417 set_pool_mode(pool, PM_FAIL);
1418 } else {
1419 dm_pool_metadata_read_only(pool->pmd);
1420 pool->process_bio = process_bio_read_only;
1421 pool->process_discard = process_discard;
1422 pool->process_prepared_mapping = process_prepared_mapping_fail;
1423 pool->process_prepared_discard = process_prepared_discard_passdown;
1424 }
1425 break;
1426
1427 case PM_WRITE:
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001428 dm_pool_metadata_read_write(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01001429 pool->process_bio = process_bio;
1430 pool->process_discard = process_discard;
1431 pool->process_prepared_mapping = process_prepared_mapping;
1432 pool->process_prepared_discard = process_prepared_discard;
1433 break;
1434 }
1435}
1436
1437/*----------------------------------------------------------------*/
1438
Joe Thornber991d9fa2011-10-31 20:21:18 +00001439/*
1440 * Mapping functions.
1441 */
1442
1443/*
1444 * Called only while mapping a thin bio to hand it over to the workqueue.
1445 */
1446static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1447{
1448 unsigned long flags;
1449 struct pool *pool = tc->pool;
1450
1451 spin_lock_irqsave(&pool->lock, flags);
1452 bio_list_add(&pool->deferred_bios, bio);
1453 spin_unlock_irqrestore(&pool->lock, flags);
1454
1455 wake_worker(pool);
1456}
1457
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001458static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
Joe Thornbereb2aa482012-03-28 18:41:28 +01001459{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001460 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001461
1462 h->tc = tc;
1463 h->shared_read_entry = NULL;
Joe Thornbere8088072012-12-21 20:23:31 +00001464 h->all_io_entry = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001465 h->overwrite_mapping = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001466}
1467
Joe Thornber991d9fa2011-10-31 20:21:18 +00001468/*
1469 * Non-blocking function called from the thin target's map function.
1470 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001471static int thin_bio_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001472{
1473 int r;
1474 struct thin_c *tc = ti->private;
1475 dm_block_t block = get_bio_block(tc, bio);
1476 struct dm_thin_device *td = tc->td;
1477 struct dm_thin_lookup_result result;
Joe Thornber025b9682013-03-01 22:45:50 +00001478 struct dm_bio_prison_cell cell1, cell2;
1479 struct dm_bio_prison_cell *cell_result;
Joe Thornbere8088072012-12-21 20:23:31 +00001480 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001481
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001482 thin_hook_bio(tc, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001483
1484 if (get_pool_mode(tc->pool) == PM_FAIL) {
1485 bio_io_error(bio);
1486 return DM_MAPIO_SUBMITTED;
1487 }
1488
Joe Thornber104655f2012-03-28 18:41:28 +01001489 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001490 thin_defer_bio(tc, bio);
1491 return DM_MAPIO_SUBMITTED;
1492 }
1493
1494 r = dm_thin_find_block(td, block, 0, &result);
1495
1496 /*
1497 * Note that we defer readahead too.
1498 */
1499 switch (r) {
1500 case 0:
1501 if (unlikely(result.shared)) {
1502 /*
1503 * We have a race condition here between the
1504 * result.shared value returned by the lookup and
1505 * snapshot creation, which may cause new
1506 * sharing.
1507 *
1508 * To avoid this always quiesce the origin before
1509 * taking the snap. You want to do this anyway to
1510 * ensure a consistent application view
1511 * (i.e. lockfs).
1512 *
1513 * More distant ancestors are irrelevant. The
1514 * shared flag will be set in their case.
1515 */
1516 thin_defer_bio(tc, bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001517 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001518 }
Joe Thornbere8088072012-12-21 20:23:31 +00001519
1520 build_virtual_key(tc->td, block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001521 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
Joe Thornbere8088072012-12-21 20:23:31 +00001522 return DM_MAPIO_SUBMITTED;
1523
1524 build_data_key(tc->td, result.block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001525 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1526 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001527 return DM_MAPIO_SUBMITTED;
1528 }
1529
1530 inc_all_io_entry(tc->pool, bio);
Joe Thornber025b9682013-03-01 22:45:50 +00001531 cell_defer_no_holder_no_free(tc, &cell2);
1532 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001533
1534 remap(tc, bio, result.block);
1535 return DM_MAPIO_REMAPPED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001536
1537 case -ENODATA:
Joe Thornbere49e5822012-07-27 15:08:16 +01001538 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1539 /*
1540 * This block isn't provisioned, and we have no way
1541 * of doing so. Just error it.
1542 */
1543 bio_io_error(bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001544 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001545 }
1546 /* fall through */
1547
1548 case -EWOULDBLOCK:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001549 /*
1550 * In future, the failed dm_thin_find_block above could
1551 * provide the hint to load the metadata into cache.
1552 */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001553 thin_defer_bio(tc, bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001554 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001555
1556 default:
1557 /*
1558 * Must always call bio_io_error on failure.
1559 * dm_thin_find_block can fail with -EINVAL if the
1560 * pool is switched to fail-io mode.
1561 */
1562 bio_io_error(bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001563 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001564 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001565}
1566
1567static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1568{
1569 int r;
1570 unsigned long flags;
1571 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1572
1573 spin_lock_irqsave(&pt->pool->lock, flags);
1574 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1575 spin_unlock_irqrestore(&pt->pool->lock, flags);
1576
1577 if (!r) {
1578 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1579 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1580 }
1581
1582 return r;
1583}
1584
1585static void __requeue_bios(struct pool *pool)
1586{
1587 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1588 bio_list_init(&pool->retry_on_resume_list);
1589}
1590
1591/*----------------------------------------------------------------
1592 * Binding of control targets to a pool object
1593 *--------------------------------------------------------------*/
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001594static bool data_dev_supports_discard(struct pool_c *pt)
1595{
1596 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1597
1598 return q && blk_queue_discard(q);
1599}
1600
Joe Thornber58051b92013-03-20 17:21:25 +00001601static bool is_factor(sector_t block_size, uint32_t n)
1602{
1603 return !sector_div(block_size, n);
1604}
1605
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001606/*
1607 * If discard_passdown was enabled verify that the data device
Mike Snitzer0424caa2012-09-26 23:45:47 +01001608 * supports discards. Disable discard_passdown if not.
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001609 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01001610static void disable_passdown_if_not_supported(struct pool_c *pt)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001611{
Mike Snitzer0424caa2012-09-26 23:45:47 +01001612 struct pool *pool = pt->pool;
1613 struct block_device *data_bdev = pt->data_dev->bdev;
1614 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1615 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1616 const char *reason = NULL;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001617 char buf[BDEVNAME_SIZE];
1618
Mike Snitzer0424caa2012-09-26 23:45:47 +01001619 if (!pt->adjusted_pf.discard_passdown)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001620 return;
1621
Mike Snitzer0424caa2012-09-26 23:45:47 +01001622 if (!data_dev_supports_discard(pt))
1623 reason = "discard unsupported";
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001624
Mike Snitzer0424caa2012-09-26 23:45:47 +01001625 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1626 reason = "max discard sectors smaller than a block";
1627
1628 else if (data_limits->discard_granularity > block_size)
1629 reason = "discard granularity larger than a block";
1630
Joe Thornber58051b92013-03-20 17:21:25 +00001631 else if (!is_factor(block_size, data_limits->discard_granularity))
Mike Snitzer0424caa2012-09-26 23:45:47 +01001632 reason = "discard granularity not a factor of block size";
1633
1634 if (reason) {
1635 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1636 pt->adjusted_pf.discard_passdown = false;
1637 }
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001638}
1639
Joe Thornber991d9fa2011-10-31 20:21:18 +00001640static int bind_control_target(struct pool *pool, struct dm_target *ti)
1641{
1642 struct pool_c *pt = ti->private;
1643
Joe Thornbere49e5822012-07-27 15:08:16 +01001644 /*
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001645 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
Joe Thornbere49e5822012-07-27 15:08:16 +01001646 */
1647 enum pool_mode old_mode = pool->pf.mode;
Mike Snitzer0424caa2012-09-26 23:45:47 +01001648 enum pool_mode new_mode = pt->adjusted_pf.mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01001649
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001650 /*
1651 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1652 * not going to recover without a thin_repair. So we never let the
1653 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1654 * may have been due to a lack of metadata or data space, and may
1655 * now work (ie. if the underlying devices have been resized).
1656 */
1657 if (old_mode == PM_FAIL)
Joe Thornbere49e5822012-07-27 15:08:16 +01001658 new_mode = old_mode;
1659
Joe Thornber991d9fa2011-10-31 20:21:18 +00001660 pool->ti = ti;
1661 pool->low_water_blocks = pt->low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01001662 pool->pf = pt->adjusted_pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001663
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001664 set_pool_mode(pool, new_mode);
Mike Snitzerf4026932012-05-19 01:01:01 +01001665
Joe Thornber991d9fa2011-10-31 20:21:18 +00001666 return 0;
1667}
1668
1669static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1670{
1671 if (pool->ti == ti)
1672 pool->ti = NULL;
1673}
1674
1675/*----------------------------------------------------------------
1676 * Pool creation
1677 *--------------------------------------------------------------*/
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001678/* Initialize pool features. */
1679static void pool_features_init(struct pool_features *pf)
1680{
Joe Thornbere49e5822012-07-27 15:08:16 +01001681 pf->mode = PM_WRITE;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001682 pf->zero_new_blocks = true;
1683 pf->discard_enabled = true;
1684 pf->discard_passdown = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001685}
1686
Joe Thornber991d9fa2011-10-31 20:21:18 +00001687static void __pool_destroy(struct pool *pool)
1688{
1689 __pool_table_remove(pool);
1690
1691 if (dm_pool_metadata_close(pool->pmd) < 0)
1692 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1693
Mike Snitzer44feb382012-10-12 21:02:10 +01001694 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001695 dm_kcopyd_client_destroy(pool->copier);
1696
1697 if (pool->wq)
1698 destroy_workqueue(pool->wq);
1699
1700 if (pool->next_mapping)
1701 mempool_free(pool->next_mapping, pool->mapping_pool);
1702 mempool_destroy(pool->mapping_pool);
Mike Snitzer44feb382012-10-12 21:02:10 +01001703 dm_deferred_set_destroy(pool->shared_read_ds);
1704 dm_deferred_set_destroy(pool->all_io_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001705 kfree(pool);
1706}
1707
Mike Snitzera24c2562012-06-03 00:30:00 +01001708static struct kmem_cache *_new_mapping_cache;
Mike Snitzera24c2562012-06-03 00:30:00 +01001709
Joe Thornber991d9fa2011-10-31 20:21:18 +00001710static struct pool *pool_create(struct mapped_device *pool_md,
1711 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001712 unsigned long block_size,
1713 int read_only, char **error)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001714{
1715 int r;
1716 void *err_p;
1717 struct pool *pool;
1718 struct dm_pool_metadata *pmd;
Joe Thornbere49e5822012-07-27 15:08:16 +01001719 bool format_device = read_only ? false : true;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001720
Joe Thornbere49e5822012-07-27 15:08:16 +01001721 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001722 if (IS_ERR(pmd)) {
1723 *error = "Error creating metadata object";
1724 return (struct pool *)pmd;
1725 }
1726
1727 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1728 if (!pool) {
1729 *error = "Error allocating memory for pool";
1730 err_p = ERR_PTR(-ENOMEM);
1731 goto bad_pool;
1732 }
1733
1734 pool->pmd = pmd;
1735 pool->sectors_per_block = block_size;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +01001736 if (block_size & (block_size - 1))
1737 pool->sectors_per_block_shift = -1;
1738 else
1739 pool->sectors_per_block_shift = __ffs(block_size);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001740 pool->low_water_blocks = 0;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001741 pool_features_init(&pool->pf);
Mike Snitzer44feb382012-10-12 21:02:10 +01001742 pool->prison = dm_bio_prison_create(PRISON_CELLS);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001743 if (!pool->prison) {
1744 *error = "Error creating pool's bio prison";
1745 err_p = ERR_PTR(-ENOMEM);
1746 goto bad_prison;
1747 }
1748
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001749 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001750 if (IS_ERR(pool->copier)) {
1751 r = PTR_ERR(pool->copier);
1752 *error = "Error creating pool's kcopyd client";
1753 err_p = ERR_PTR(r);
1754 goto bad_kcopyd_client;
1755 }
1756
1757 /*
1758 * Create singlethreaded workqueue that will service all devices
1759 * that use this metadata.
1760 */
1761 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1762 if (!pool->wq) {
1763 *error = "Error creating pool's workqueue";
1764 err_p = ERR_PTR(-ENOMEM);
1765 goto bad_wq;
1766 }
1767
1768 INIT_WORK(&pool->worker, do_worker);
Joe Thornber905e51b2012-03-28 18:41:27 +01001769 INIT_DELAYED_WORK(&pool->waker, do_waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001770 spin_lock_init(&pool->lock);
1771 bio_list_init(&pool->deferred_bios);
1772 bio_list_init(&pool->deferred_flush_bios);
1773 INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber104655f2012-03-28 18:41:28 +01001774 INIT_LIST_HEAD(&pool->prepared_discards);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001775 pool->low_water_triggered = 0;
1776 pool->no_free_space = 0;
1777 bio_list_init(&pool->retry_on_resume_list);
Mike Snitzer44feb382012-10-12 21:02:10 +01001778
1779 pool->shared_read_ds = dm_deferred_set_create();
1780 if (!pool->shared_read_ds) {
1781 *error = "Error creating pool's shared read deferred set";
1782 err_p = ERR_PTR(-ENOMEM);
1783 goto bad_shared_read_ds;
1784 }
1785
1786 pool->all_io_ds = dm_deferred_set_create();
1787 if (!pool->all_io_ds) {
1788 *error = "Error creating pool's all io deferred set";
1789 err_p = ERR_PTR(-ENOMEM);
1790 goto bad_all_io_ds;
1791 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001792
1793 pool->next_mapping = NULL;
Mike Snitzera24c2562012-06-03 00:30:00 +01001794 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1795 _new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001796 if (!pool->mapping_pool) {
1797 *error = "Error creating pool's mapping mempool";
1798 err_p = ERR_PTR(-ENOMEM);
1799 goto bad_mapping_pool;
1800 }
1801
Joe Thornber991d9fa2011-10-31 20:21:18 +00001802 pool->ref_count = 1;
Joe Thornber905e51b2012-03-28 18:41:27 +01001803 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001804 pool->pool_md = pool_md;
1805 pool->md_dev = metadata_dev;
1806 __pool_table_insert(pool);
1807
1808 return pool;
1809
Joe Thornber991d9fa2011-10-31 20:21:18 +00001810bad_mapping_pool:
Mike Snitzer44feb382012-10-12 21:02:10 +01001811 dm_deferred_set_destroy(pool->all_io_ds);
1812bad_all_io_ds:
1813 dm_deferred_set_destroy(pool->shared_read_ds);
1814bad_shared_read_ds:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001815 destroy_workqueue(pool->wq);
1816bad_wq:
1817 dm_kcopyd_client_destroy(pool->copier);
1818bad_kcopyd_client:
Mike Snitzer44feb382012-10-12 21:02:10 +01001819 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001820bad_prison:
1821 kfree(pool);
1822bad_pool:
1823 if (dm_pool_metadata_close(pmd))
1824 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1825
1826 return err_p;
1827}
1828
1829static void __pool_inc(struct pool *pool)
1830{
1831 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1832 pool->ref_count++;
1833}
1834
1835static void __pool_dec(struct pool *pool)
1836{
1837 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1838 BUG_ON(!pool->ref_count);
1839 if (!--pool->ref_count)
1840 __pool_destroy(pool);
1841}
1842
1843static struct pool *__pool_find(struct mapped_device *pool_md,
1844 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001845 unsigned long block_size, int read_only,
1846 char **error, int *created)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001847{
1848 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1849
1850 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001851 if (pool->pool_md != pool_md) {
1852 *error = "metadata device already in use by a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001853 return ERR_PTR(-EBUSY);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001854 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001855 __pool_inc(pool);
1856
1857 } else {
1858 pool = __pool_table_lookup(pool_md);
1859 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001860 if (pool->md_dev != metadata_dev) {
1861 *error = "different pool cannot replace a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001862 return ERR_PTR(-EINVAL);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001863 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001864 __pool_inc(pool);
1865
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001866 } else {
Joe Thornbere49e5822012-07-27 15:08:16 +01001867 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001868 *created = 1;
1869 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001870 }
1871
1872 return pool;
1873}
1874
1875/*----------------------------------------------------------------
1876 * Pool target methods
1877 *--------------------------------------------------------------*/
1878static void pool_dtr(struct dm_target *ti)
1879{
1880 struct pool_c *pt = ti->private;
1881
1882 mutex_lock(&dm_thin_pool_table.mutex);
1883
1884 unbind_control_target(pt->pool, ti);
1885 __pool_dec(pt->pool);
1886 dm_put_device(ti, pt->metadata_dev);
1887 dm_put_device(ti, pt->data_dev);
1888 kfree(pt);
1889
1890 mutex_unlock(&dm_thin_pool_table.mutex);
1891}
1892
Joe Thornber991d9fa2011-10-31 20:21:18 +00001893static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1894 struct dm_target *ti)
1895{
1896 int r;
1897 unsigned argc;
1898 const char *arg_name;
1899
1900 static struct dm_arg _args[] = {
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001901 {0, 3, "Invalid number of pool feature arguments"},
Joe Thornber991d9fa2011-10-31 20:21:18 +00001902 };
1903
1904 /*
1905 * No feature arguments supplied.
1906 */
1907 if (!as->argc)
1908 return 0;
1909
1910 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1911 if (r)
1912 return -EINVAL;
1913
1914 while (argc && !r) {
1915 arg_name = dm_shift_arg(as);
1916 argc--;
1917
Joe Thornbere49e5822012-07-27 15:08:16 +01001918 if (!strcasecmp(arg_name, "skip_block_zeroing"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001919 pf->zero_new_blocks = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001920
Joe Thornbere49e5822012-07-27 15:08:16 +01001921 else if (!strcasecmp(arg_name, "ignore_discard"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001922 pf->discard_enabled = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001923
1924 else if (!strcasecmp(arg_name, "no_discard_passdown"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001925 pf->discard_passdown = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001926
1927 else if (!strcasecmp(arg_name, "read_only"))
1928 pf->mode = PM_READ_ONLY;
1929
1930 else {
1931 ti->error = "Unrecognised pool feature requested";
1932 r = -EINVAL;
1933 break;
1934 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001935 }
1936
1937 return r;
1938}
1939
Joe Thornberac8c3f32013-05-10 14:37:21 +01001940static void metadata_low_callback(void *context)
1941{
1942 struct pool *pool = context;
1943
1944 DMWARN("%s: reached low water mark for metadata device: sending event.",
1945 dm_device_name(pool->pool_md));
1946
1947 dm_table_event(pool->ti->table);
1948}
1949
Joe Thornberb17446d2013-05-10 14:37:18 +01001950static sector_t get_metadata_dev_size(struct block_device *bdev)
1951{
1952 sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1953 char buffer[BDEVNAME_SIZE];
1954
1955 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
1956 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1957 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
1958 metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
1959 }
1960
1961 return metadata_dev_size;
1962}
1963
Joe Thornber24347e92013-05-10 14:37:19 +01001964static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
1965{
1966 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
1967
1968 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
1969
1970 return metadata_dev_size;
1971}
1972
Joe Thornber991d9fa2011-10-31 20:21:18 +00001973/*
Joe Thornberac8c3f32013-05-10 14:37:21 +01001974 * When a metadata threshold is crossed a dm event is triggered, and
1975 * userland should respond by growing the metadata device. We could let
1976 * userland set the threshold, like we do with the data threshold, but I'm
1977 * not sure they know enough to do this well.
1978 */
1979static dm_block_t calc_metadata_threshold(struct pool_c *pt)
1980{
1981 /*
1982 * 4M is ample for all ops with the possible exception of thin
1983 * device deletion which is harmless if it fails (just retry the
1984 * delete after you've grown the device).
1985 */
1986 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
1987 return min((dm_block_t)1024ULL /* 4M */, quarter);
1988}
1989
1990/*
Joe Thornber991d9fa2011-10-31 20:21:18 +00001991 * thin-pool <metadata dev> <data dev>
1992 * <data block size (sectors)>
1993 * <low water mark (blocks)>
1994 * [<#feature args> [<arg>]*]
1995 *
1996 * Optional feature arguments are:
1997 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001998 * ignore_discard: disable discard
1999 * no_discard_passdown: don't pass discards down to the data device
Joe Thornber991d9fa2011-10-31 20:21:18 +00002000 */
2001static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2002{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002003 int r, pool_created = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002004 struct pool_c *pt;
2005 struct pool *pool;
2006 struct pool_features pf;
2007 struct dm_arg_set as;
2008 struct dm_dev *data_dev;
2009 unsigned long block_size;
2010 dm_block_t low_water_blocks;
2011 struct dm_dev *metadata_dev;
Joe Thornber5d0db962013-05-10 14:37:19 +01002012 fmode_t metadata_mode;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002013
2014 /*
2015 * FIXME Remove validation from scope of lock.
2016 */
2017 mutex_lock(&dm_thin_pool_table.mutex);
2018
2019 if (argc < 4) {
2020 ti->error = "Invalid argument count";
2021 r = -EINVAL;
2022 goto out_unlock;
2023 }
Joe Thornber5d0db962013-05-10 14:37:19 +01002024
Joe Thornber991d9fa2011-10-31 20:21:18 +00002025 as.argc = argc;
2026 as.argv = argv;
2027
Joe Thornber5d0db962013-05-10 14:37:19 +01002028 /*
2029 * Set default pool features.
2030 */
2031 pool_features_init(&pf);
2032
2033 dm_consume_args(&as, 4);
2034 r = parse_pool_features(&as, &pf, ti);
2035 if (r)
2036 goto out_unlock;
2037
2038 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2039 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002040 if (r) {
2041 ti->error = "Error opening metadata block device";
2042 goto out_unlock;
2043 }
2044
Joe Thornberb17446d2013-05-10 14:37:18 +01002045 /*
2046 * Run for the side-effect of possibly issuing a warning if the
2047 * device is too big.
2048 */
2049 (void) get_metadata_dev_size(metadata_dev->bdev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002050
2051 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2052 if (r) {
2053 ti->error = "Error getting data device";
2054 goto out_metadata;
2055 }
2056
2057 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2058 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2059 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002060 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002061 ti->error = "Invalid block size";
2062 r = -EINVAL;
2063 goto out;
2064 }
2065
2066 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2067 ti->error = "Invalid low water mark";
2068 r = -EINVAL;
2069 goto out;
2070 }
2071
Joe Thornber991d9fa2011-10-31 20:21:18 +00002072 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2073 if (!pt) {
2074 r = -ENOMEM;
2075 goto out;
2076 }
2077
2078 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
Joe Thornbere49e5822012-07-27 15:08:16 +01002079 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002080 if (IS_ERR(pool)) {
2081 r = PTR_ERR(pool);
2082 goto out_free_pt;
2083 }
2084
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002085 /*
2086 * 'pool_created' reflects whether this is the first table load.
2087 * Top level discard support is not allowed to be changed after
2088 * initial load. This would require a pool reload to trigger thin
2089 * device changes.
2090 */
2091 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2092 ti->error = "Discard support cannot be disabled once enabled";
2093 r = -EINVAL;
2094 goto out_flags_changed;
2095 }
2096
Joe Thornber991d9fa2011-10-31 20:21:18 +00002097 pt->pool = pool;
2098 pt->ti = ti;
2099 pt->metadata_dev = metadata_dev;
2100 pt->data_dev = data_dev;
2101 pt->low_water_blocks = low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002102 pt->adjusted_pf = pt->requested_pf = pf;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002103 ti->num_flush_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002104
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002105 /*
2106 * Only need to enable discards if the pool should pass
2107 * them down to the data device. The thin device's discard
2108 * processing will cause mappings to be removed from the btree.
2109 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002110 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002111 if (pf.discard_enabled && pf.discard_passdown) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002112 ti->num_discard_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002113
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002114 /*
2115 * Setting 'discards_supported' circumvents the normal
2116 * stacking of discard limits (this keeps the pool and
2117 * thin devices' discard limits consistent).
2118 */
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002119 ti->discards_supported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002120 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002121 ti->private = pt;
2122
Joe Thornberac8c3f32013-05-10 14:37:21 +01002123 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2124 calc_metadata_threshold(pt),
2125 metadata_low_callback,
2126 pool);
2127 if (r)
2128 goto out_free_pt;
2129
Joe Thornber991d9fa2011-10-31 20:21:18 +00002130 pt->callbacks.congested_fn = pool_is_congested;
2131 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2132
2133 mutex_unlock(&dm_thin_pool_table.mutex);
2134
2135 return 0;
2136
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002137out_flags_changed:
2138 __pool_dec(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002139out_free_pt:
2140 kfree(pt);
2141out:
2142 dm_put_device(ti, data_dev);
2143out_metadata:
2144 dm_put_device(ti, metadata_dev);
2145out_unlock:
2146 mutex_unlock(&dm_thin_pool_table.mutex);
2147
2148 return r;
2149}
2150
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002151static int pool_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002152{
2153 int r;
2154 struct pool_c *pt = ti->private;
2155 struct pool *pool = pt->pool;
2156 unsigned long flags;
2157
2158 /*
2159 * As this is a singleton target, ti->begin is always zero.
2160 */
2161 spin_lock_irqsave(&pool->lock, flags);
2162 bio->bi_bdev = pt->data_dev->bdev;
2163 r = DM_MAPIO_REMAPPED;
2164 spin_unlock_irqrestore(&pool->lock, flags);
2165
2166 return r;
2167}
2168
Joe Thornberb17446d2013-05-10 14:37:18 +01002169static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2170{
2171 int r;
2172 struct pool_c *pt = ti->private;
2173 struct pool *pool = pt->pool;
2174 sector_t data_size = ti->len;
2175 dm_block_t sb_data_size;
2176
2177 *need_commit = false;
2178
2179 (void) sector_div(data_size, pool->sectors_per_block);
2180
2181 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2182 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002183 DMERR("%s: failed to retrieve data device size",
2184 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01002185 return r;
2186 }
2187
2188 if (data_size < sb_data_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002189 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2190 dm_device_name(pool->pool_md),
Joe Thornberb17446d2013-05-10 14:37:18 +01002191 (unsigned long long)data_size, sb_data_size);
2192 return -EINVAL;
2193
2194 } else if (data_size > sb_data_size) {
2195 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2196 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002197 DMERR("%s: failed to resize data device",
2198 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01002199 set_pool_mode(pool, PM_READ_ONLY);
2200 return r;
2201 }
2202
2203 *need_commit = true;
2204 }
2205
2206 return 0;
2207}
2208
Joe Thornber24347e92013-05-10 14:37:19 +01002209static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2210{
2211 int r;
2212 struct pool_c *pt = ti->private;
2213 struct pool *pool = pt->pool;
2214 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2215
2216 *need_commit = false;
2217
Alasdair G Kergon610bba82013-05-19 18:57:50 +01002218 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
Joe Thornber24347e92013-05-10 14:37:19 +01002219
2220 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2221 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002222 DMERR("%s: failed to retrieve metadata device size",
2223 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01002224 return r;
2225 }
2226
2227 if (metadata_dev_size < sb_metadata_dev_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002228 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2229 dm_device_name(pool->pool_md),
Joe Thornber24347e92013-05-10 14:37:19 +01002230 metadata_dev_size, sb_metadata_dev_size);
2231 return -EINVAL;
2232
2233 } else if (metadata_dev_size > sb_metadata_dev_size) {
2234 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2235 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002236 DMERR("%s: failed to resize metadata device",
2237 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01002238 return r;
2239 }
2240
2241 *need_commit = true;
2242 }
2243
2244 return 0;
2245}
2246
Joe Thornber991d9fa2011-10-31 20:21:18 +00002247/*
2248 * Retrieves the number of blocks of the data device from
2249 * the superblock and compares it to the actual device size,
2250 * thus resizing the data device in case it has grown.
2251 *
2252 * This both copes with opening preallocated data devices in the ctr
2253 * being followed by a resume
2254 * -and-
2255 * calling the resume method individually after userspace has
2256 * grown the data device in reaction to a table event.
2257 */
2258static int pool_preresume(struct dm_target *ti)
2259{
2260 int r;
Joe Thornber24347e92013-05-10 14:37:19 +01002261 bool need_commit1, need_commit2;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002262 struct pool_c *pt = ti->private;
2263 struct pool *pool = pt->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002264
2265 /*
2266 * Take control of the pool object.
2267 */
2268 r = bind_control_target(pool, ti);
2269 if (r)
2270 return r;
2271
Joe Thornberb17446d2013-05-10 14:37:18 +01002272 r = maybe_resize_data_dev(ti, &need_commit1);
2273 if (r)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002274 return r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002275
Joe Thornber24347e92013-05-10 14:37:19 +01002276 r = maybe_resize_metadata_dev(ti, &need_commit2);
2277 if (r)
2278 return r;
2279
2280 if (need_commit1 || need_commit2)
Joe Thornber020cc3b2013-12-04 15:05:36 -05002281 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002282
2283 return 0;
2284}
2285
2286static void pool_resume(struct dm_target *ti)
2287{
2288 struct pool_c *pt = ti->private;
2289 struct pool *pool = pt->pool;
2290 unsigned long flags;
2291
2292 spin_lock_irqsave(&pool->lock, flags);
2293 pool->low_water_triggered = 0;
2294 pool->no_free_space = 0;
2295 __requeue_bios(pool);
2296 spin_unlock_irqrestore(&pool->lock, flags);
2297
Joe Thornber905e51b2012-03-28 18:41:27 +01002298 do_waker(&pool->waker.work);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002299}
2300
2301static void pool_postsuspend(struct dm_target *ti)
2302{
Joe Thornber991d9fa2011-10-31 20:21:18 +00002303 struct pool_c *pt = ti->private;
2304 struct pool *pool = pt->pool;
2305
Joe Thornber905e51b2012-03-28 18:41:27 +01002306 cancel_delayed_work(&pool->waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002307 flush_workqueue(pool->wq);
Joe Thornber020cc3b2013-12-04 15:05:36 -05002308 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002309}
2310
2311static int check_arg_count(unsigned argc, unsigned args_required)
2312{
2313 if (argc != args_required) {
2314 DMWARN("Message received with %u arguments instead of %u.",
2315 argc, args_required);
2316 return -EINVAL;
2317 }
2318
2319 return 0;
2320}
2321
2322static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2323{
2324 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2325 *dev_id <= MAX_DEV_ID)
2326 return 0;
2327
2328 if (warning)
2329 DMWARN("Message received with invalid device id: %s", arg);
2330
2331 return -EINVAL;
2332}
2333
2334static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2335{
2336 dm_thin_id dev_id;
2337 int r;
2338
2339 r = check_arg_count(argc, 2);
2340 if (r)
2341 return r;
2342
2343 r = read_dev_id(argv[1], &dev_id, 1);
2344 if (r)
2345 return r;
2346
2347 r = dm_pool_create_thin(pool->pmd, dev_id);
2348 if (r) {
2349 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2350 argv[1]);
2351 return r;
2352 }
2353
2354 return 0;
2355}
2356
2357static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2358{
2359 dm_thin_id dev_id;
2360 dm_thin_id origin_dev_id;
2361 int r;
2362
2363 r = check_arg_count(argc, 3);
2364 if (r)
2365 return r;
2366
2367 r = read_dev_id(argv[1], &dev_id, 1);
2368 if (r)
2369 return r;
2370
2371 r = read_dev_id(argv[2], &origin_dev_id, 1);
2372 if (r)
2373 return r;
2374
2375 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2376 if (r) {
2377 DMWARN("Creation of new snapshot %s of device %s failed.",
2378 argv[1], argv[2]);
2379 return r;
2380 }
2381
2382 return 0;
2383}
2384
2385static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2386{
2387 dm_thin_id dev_id;
2388 int r;
2389
2390 r = check_arg_count(argc, 2);
2391 if (r)
2392 return r;
2393
2394 r = read_dev_id(argv[1], &dev_id, 1);
2395 if (r)
2396 return r;
2397
2398 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2399 if (r)
2400 DMWARN("Deletion of thin device %s failed.", argv[1]);
2401
2402 return r;
2403}
2404
2405static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2406{
2407 dm_thin_id old_id, new_id;
2408 int r;
2409
2410 r = check_arg_count(argc, 3);
2411 if (r)
2412 return r;
2413
2414 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2415 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2416 return -EINVAL;
2417 }
2418
2419 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2420 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2421 return -EINVAL;
2422 }
2423
2424 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2425 if (r) {
2426 DMWARN("Failed to change transaction id from %s to %s.",
2427 argv[1], argv[2]);
2428 return r;
2429 }
2430
2431 return 0;
2432}
2433
Joe Thornbercc8394d2012-06-03 00:30:01 +01002434static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2435{
2436 int r;
2437
2438 r = check_arg_count(argc, 1);
2439 if (r)
2440 return r;
2441
Joe Thornber020cc3b2013-12-04 15:05:36 -05002442 (void) commit(pool);
Joe Thornber0d200ae2012-07-03 12:55:31 +01002443
Joe Thornbercc8394d2012-06-03 00:30:01 +01002444 r = dm_pool_reserve_metadata_snap(pool->pmd);
2445 if (r)
2446 DMWARN("reserve_metadata_snap message failed.");
2447
2448 return r;
2449}
2450
2451static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2452{
2453 int r;
2454
2455 r = check_arg_count(argc, 1);
2456 if (r)
2457 return r;
2458
2459 r = dm_pool_release_metadata_snap(pool->pmd);
2460 if (r)
2461 DMWARN("release_metadata_snap message failed.");
2462
2463 return r;
2464}
2465
Joe Thornber991d9fa2011-10-31 20:21:18 +00002466/*
2467 * Messages supported:
2468 * create_thin <dev_id>
2469 * create_snap <dev_id> <origin_id>
2470 * delete <dev_id>
2471 * trim <dev_id> <new_size_in_sectors>
2472 * set_transaction_id <current_trans_id> <new_trans_id>
Joe Thornbercc8394d2012-06-03 00:30:01 +01002473 * reserve_metadata_snap
2474 * release_metadata_snap
Joe Thornber991d9fa2011-10-31 20:21:18 +00002475 */
2476static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2477{
2478 int r = -EINVAL;
2479 struct pool_c *pt = ti->private;
2480 struct pool *pool = pt->pool;
2481
2482 if (!strcasecmp(argv[0], "create_thin"))
2483 r = process_create_thin_mesg(argc, argv, pool);
2484
2485 else if (!strcasecmp(argv[0], "create_snap"))
2486 r = process_create_snap_mesg(argc, argv, pool);
2487
2488 else if (!strcasecmp(argv[0], "delete"))
2489 r = process_delete_mesg(argc, argv, pool);
2490
2491 else if (!strcasecmp(argv[0], "set_transaction_id"))
2492 r = process_set_transaction_id_mesg(argc, argv, pool);
2493
Joe Thornbercc8394d2012-06-03 00:30:01 +01002494 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2495 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2496
2497 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2498 r = process_release_metadata_snap_mesg(argc, argv, pool);
2499
Joe Thornber991d9fa2011-10-31 20:21:18 +00002500 else
2501 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2502
Joe Thornbere49e5822012-07-27 15:08:16 +01002503 if (!r)
Joe Thornber020cc3b2013-12-04 15:05:36 -05002504 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002505
2506 return r;
2507}
2508
Joe Thornbere49e5822012-07-27 15:08:16 +01002509static void emit_flags(struct pool_features *pf, char *result,
2510 unsigned sz, unsigned maxlen)
2511{
2512 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2513 !pf->discard_passdown + (pf->mode == PM_READ_ONLY);
2514 DMEMIT("%u ", count);
2515
2516 if (!pf->zero_new_blocks)
2517 DMEMIT("skip_block_zeroing ");
2518
2519 if (!pf->discard_enabled)
2520 DMEMIT("ignore_discard ");
2521
2522 if (!pf->discard_passdown)
2523 DMEMIT("no_discard_passdown ");
2524
2525 if (pf->mode == PM_READ_ONLY)
2526 DMEMIT("read_only ");
2527}
2528
Joe Thornber991d9fa2011-10-31 20:21:18 +00002529/*
2530 * Status line is:
2531 * <transaction id> <used metadata sectors>/<total metadata sectors>
2532 * <used data sectors>/<total data sectors> <held metadata root>
2533 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002534static void pool_status(struct dm_target *ti, status_type_t type,
2535 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002536{
Joe Thornbere49e5822012-07-27 15:08:16 +01002537 int r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002538 unsigned sz = 0;
2539 uint64_t transaction_id;
2540 dm_block_t nr_free_blocks_data;
2541 dm_block_t nr_free_blocks_metadata;
2542 dm_block_t nr_blocks_data;
2543 dm_block_t nr_blocks_metadata;
2544 dm_block_t held_root;
2545 char buf[BDEVNAME_SIZE];
2546 char buf2[BDEVNAME_SIZE];
2547 struct pool_c *pt = ti->private;
2548 struct pool *pool = pt->pool;
2549
2550 switch (type) {
2551 case STATUSTYPE_INFO:
Joe Thornbere49e5822012-07-27 15:08:16 +01002552 if (get_pool_mode(pool) == PM_FAIL) {
2553 DMEMIT("Fail");
2554 break;
2555 }
2556
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002557 /* Commit to ensure statistics aren't out-of-date */
2558 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
Joe Thornber020cc3b2013-12-04 15:05:36 -05002559 (void) commit(pool);
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002560
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002561 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2562 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002563 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2564 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002565 goto err;
2566 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002567
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002568 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2569 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002570 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2571 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002572 goto err;
2573 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002574
2575 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002576 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002577 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2578 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002579 goto err;
2580 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002581
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002582 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2583 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002584 DMERR("%s: dm_pool_get_free_block_count returned %d",
2585 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002586 goto err;
2587 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002588
2589 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002590 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002591 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2592 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002593 goto err;
2594 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002595
Joe Thornbercc8394d2012-06-03 00:30:01 +01002596 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002597 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002598 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2599 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002600 goto err;
2601 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002602
2603 DMEMIT("%llu %llu/%llu %llu/%llu ",
2604 (unsigned long long)transaction_id,
2605 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2606 (unsigned long long)nr_blocks_metadata,
2607 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2608 (unsigned long long)nr_blocks_data);
2609
2610 if (held_root)
Joe Thornbere49e5822012-07-27 15:08:16 +01002611 DMEMIT("%llu ", held_root);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002612 else
Joe Thornbere49e5822012-07-27 15:08:16 +01002613 DMEMIT("- ");
2614
2615 if (pool->pf.mode == PM_READ_ONLY)
2616 DMEMIT("ro ");
2617 else
2618 DMEMIT("rw ");
2619
Mike Snitzer018debe2012-12-21 20:23:32 +00002620 if (!pool->pf.discard_enabled)
2621 DMEMIT("ignore_discard");
2622 else if (pool->pf.discard_passdown)
Joe Thornbere49e5822012-07-27 15:08:16 +01002623 DMEMIT("discard_passdown");
2624 else
2625 DMEMIT("no_discard_passdown");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002626
2627 break;
2628
2629 case STATUSTYPE_TABLE:
2630 DMEMIT("%s %s %lu %llu ",
2631 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2632 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2633 (unsigned long)pool->sectors_per_block,
2634 (unsigned long long)pt->low_water_blocks);
Mike Snitzer0424caa2012-09-26 23:45:47 +01002635 emit_flags(&pt->requested_pf, result, sz, maxlen);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002636 break;
2637 }
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002638 return;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002639
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002640err:
2641 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002642}
2643
2644static int pool_iterate_devices(struct dm_target *ti,
2645 iterate_devices_callout_fn fn, void *data)
2646{
2647 struct pool_c *pt = ti->private;
2648
2649 return fn(ti, pt->data_dev, 0, ti->len, data);
2650}
2651
2652static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2653 struct bio_vec *biovec, int max_size)
2654{
2655 struct pool_c *pt = ti->private;
2656 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2657
2658 if (!q->merge_bvec_fn)
2659 return max_size;
2660
2661 bvm->bi_bdev = pt->data_dev->bdev;
2662
2663 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2664}
2665
Mike Snitzer0424caa2012-09-26 23:45:47 +01002666static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
Joe Thornber104655f2012-03-28 18:41:28 +01002667{
Mike Snitzer0424caa2012-09-26 23:45:47 +01002668 struct pool *pool = pt->pool;
2669 struct queue_limits *data_limits;
2670
Joe Thornber104655f2012-03-28 18:41:28 +01002671 limits->max_discard_sectors = pool->sectors_per_block;
2672
2673 /*
Mike Snitzer0424caa2012-09-26 23:45:47 +01002674 * discard_granularity is just a hint, and not enforced.
Joe Thornber104655f2012-03-28 18:41:28 +01002675 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01002676 if (pt->adjusted_pf.discard_passdown) {
2677 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2678 limits->discard_granularity = data_limits->discard_granularity;
Mike Snitzerf13945d2013-03-01 22:45:44 +00002679 } else
Mike Snitzer0424caa2012-09-26 23:45:47 +01002680 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
Joe Thornber104655f2012-03-28 18:41:28 +01002681}
2682
Joe Thornber991d9fa2011-10-31 20:21:18 +00002683static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2684{
2685 struct pool_c *pt = ti->private;
2686 struct pool *pool = pt->pool;
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002687 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002688
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002689 /*
2690 * If the system-determined stacked limits are compatible with the
2691 * pool's blocksize (io_opt is a factor) do not override them.
2692 */
2693 if (io_opt_sectors < pool->sectors_per_block ||
2694 do_div(io_opt_sectors, pool->sectors_per_block)) {
2695 blk_limits_io_min(limits, 0);
2696 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2697 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002698
2699 /*
2700 * pt->adjusted_pf is a staging area for the actual features to use.
2701 * They get transferred to the live pool in bind_control_target()
2702 * called from pool_preresume().
2703 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002704 if (!pt->adjusted_pf.discard_enabled) {
2705 /*
2706 * Must explicitly disallow stacking discard limits otherwise the
2707 * block layer will stack them if pool's data device has support.
2708 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2709 * user to see that, so make sure to set all discard limits to 0.
2710 */
2711 limits->discard_granularity = 0;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002712 return;
Mike Snitzerb60ab992013-09-19 18:49:11 -04002713 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002714
2715 disable_passdown_if_not_supported(pt);
2716
2717 set_discard_limits(pt, limits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002718}
2719
2720static struct target_type pool_target = {
2721 .name = "thin-pool",
2722 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2723 DM_TARGET_IMMUTABLE,
Mike Snitzer94563ba2013-08-22 09:56:18 -04002724 .version = {1, 9, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002725 .module = THIS_MODULE,
2726 .ctr = pool_ctr,
2727 .dtr = pool_dtr,
2728 .map = pool_map,
2729 .postsuspend = pool_postsuspend,
2730 .preresume = pool_preresume,
2731 .resume = pool_resume,
2732 .message = pool_message,
2733 .status = pool_status,
2734 .merge = pool_merge,
2735 .iterate_devices = pool_iterate_devices,
2736 .io_hints = pool_io_hints,
2737};
2738
2739/*----------------------------------------------------------------
2740 * Thin target methods
2741 *--------------------------------------------------------------*/
2742static void thin_dtr(struct dm_target *ti)
2743{
2744 struct thin_c *tc = ti->private;
2745
2746 mutex_lock(&dm_thin_pool_table.mutex);
2747
2748 __pool_dec(tc->pool);
2749 dm_pool_close_thin_device(tc->td);
2750 dm_put_device(ti, tc->pool_dev);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002751 if (tc->origin_dev)
2752 dm_put_device(ti, tc->origin_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002753 kfree(tc);
2754
2755 mutex_unlock(&dm_thin_pool_table.mutex);
2756}
2757
2758/*
2759 * Thin target parameters:
2760 *
Joe Thornber2dd9c252012-03-28 18:41:28 +01002761 * <pool_dev> <dev_id> [origin_dev]
Joe Thornber991d9fa2011-10-31 20:21:18 +00002762 *
2763 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2764 * dev_id: the internal device identifier
Joe Thornber2dd9c252012-03-28 18:41:28 +01002765 * origin_dev: a device external to the pool that should act as the origin
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002766 *
2767 * If the pool device has discards disabled, they get disabled for the thin
2768 * device as well.
Joe Thornber991d9fa2011-10-31 20:21:18 +00002769 */
2770static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2771{
2772 int r;
2773 struct thin_c *tc;
Joe Thornber2dd9c252012-03-28 18:41:28 +01002774 struct dm_dev *pool_dev, *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002775 struct mapped_device *pool_md;
2776
2777 mutex_lock(&dm_thin_pool_table.mutex);
2778
Joe Thornber2dd9c252012-03-28 18:41:28 +01002779 if (argc != 2 && argc != 3) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002780 ti->error = "Invalid argument count";
2781 r = -EINVAL;
2782 goto out_unlock;
2783 }
2784
2785 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2786 if (!tc) {
2787 ti->error = "Out of memory";
2788 r = -ENOMEM;
2789 goto out_unlock;
2790 }
2791
Joe Thornber2dd9c252012-03-28 18:41:28 +01002792 if (argc == 3) {
2793 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2794 if (r) {
2795 ti->error = "Error opening origin device";
2796 goto bad_origin_dev;
2797 }
2798 tc->origin_dev = origin_dev;
2799 }
2800
Joe Thornber991d9fa2011-10-31 20:21:18 +00002801 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2802 if (r) {
2803 ti->error = "Error opening pool device";
2804 goto bad_pool_dev;
2805 }
2806 tc->pool_dev = pool_dev;
2807
2808 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2809 ti->error = "Invalid device id";
2810 r = -EINVAL;
2811 goto bad_common;
2812 }
2813
2814 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2815 if (!pool_md) {
2816 ti->error = "Couldn't get pool mapped device";
2817 r = -EINVAL;
2818 goto bad_common;
2819 }
2820
2821 tc->pool = __pool_table_lookup(pool_md);
2822 if (!tc->pool) {
2823 ti->error = "Couldn't find pool object";
2824 r = -EINVAL;
2825 goto bad_pool_lookup;
2826 }
2827 __pool_inc(tc->pool);
2828
Joe Thornbere49e5822012-07-27 15:08:16 +01002829 if (get_pool_mode(tc->pool) == PM_FAIL) {
2830 ti->error = "Couldn't open thin device, Pool is in fail mode";
2831 goto bad_thin_open;
2832 }
2833
Joe Thornber991d9fa2011-10-31 20:21:18 +00002834 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2835 if (r) {
2836 ti->error = "Couldn't open thin internal device";
2837 goto bad_thin_open;
2838 }
2839
Mike Snitzer542f9032012-07-27 15:08:00 +01002840 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2841 if (r)
2842 goto bad_thin_open;
2843
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002844 ti->num_flush_bios = 1;
Joe Thornber16ad3d12012-07-27 15:08:07 +01002845 ti->flush_supported = true;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002846 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002847
2848 /* In case the pool supports discards, pass them on. */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002849 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002850 if (tc->pool->pf.discard_enabled) {
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002851 ti->discards_supported = true;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002852 ti->num_discard_bios = 1;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002853 /* Discard bios must be split on a block boundary */
2854 ti->split_discard_bios = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002855 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002856
2857 dm_put(pool_md);
2858
2859 mutex_unlock(&dm_thin_pool_table.mutex);
2860
2861 return 0;
2862
2863bad_thin_open:
2864 __pool_dec(tc->pool);
2865bad_pool_lookup:
2866 dm_put(pool_md);
2867bad_common:
2868 dm_put_device(ti, tc->pool_dev);
2869bad_pool_dev:
Joe Thornber2dd9c252012-03-28 18:41:28 +01002870 if (tc->origin_dev)
2871 dm_put_device(ti, tc->origin_dev);
2872bad_origin_dev:
Joe Thornber991d9fa2011-10-31 20:21:18 +00002873 kfree(tc);
2874out_unlock:
2875 mutex_unlock(&dm_thin_pool_table.mutex);
2876
2877 return r;
2878}
2879
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002880static int thin_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002881{
Alasdair G Kergon6efd6e82012-03-28 18:41:28 +01002882 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002883
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002884 return thin_bio_map(ti, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002885}
2886
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002887static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
Joe Thornbereb2aa482012-03-28 18:41:28 +01002888{
2889 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002890 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01002891 struct list_head work;
Mike Snitzera24c2562012-06-03 00:30:00 +01002892 struct dm_thin_new_mapping *m, *tmp;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002893 struct pool *pool = h->tc->pool;
2894
2895 if (h->shared_read_entry) {
2896 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002897 dm_deferred_entry_dec(h->shared_read_entry, &work);
Joe Thornbereb2aa482012-03-28 18:41:28 +01002898
2899 spin_lock_irqsave(&pool->lock, flags);
2900 list_for_each_entry_safe(m, tmp, &work, list) {
2901 list_del(&m->list);
2902 m->quiesced = 1;
2903 __maybe_add_mapping(m);
2904 }
2905 spin_unlock_irqrestore(&pool->lock, flags);
2906 }
2907
Joe Thornber104655f2012-03-28 18:41:28 +01002908 if (h->all_io_entry) {
2909 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002910 dm_deferred_entry_dec(h->all_io_entry, &work);
Joe Thornber563af182012-12-21 20:23:31 +00002911 if (!list_empty(&work)) {
2912 spin_lock_irqsave(&pool->lock, flags);
2913 list_for_each_entry_safe(m, tmp, &work, list)
2914 list_add(&m->list, &pool->prepared_discards);
2915 spin_unlock_irqrestore(&pool->lock, flags);
2916 wake_worker(pool);
2917 }
Joe Thornber104655f2012-03-28 18:41:28 +01002918 }
2919
Joe Thornbereb2aa482012-03-28 18:41:28 +01002920 return 0;
2921}
2922
Joe Thornber991d9fa2011-10-31 20:21:18 +00002923static void thin_postsuspend(struct dm_target *ti)
2924{
2925 if (dm_noflush_suspending(ti))
2926 requeue_io((struct thin_c *)ti->private);
2927}
2928
2929/*
2930 * <nr mapped sectors> <highest mapped sector>
2931 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002932static void thin_status(struct dm_target *ti, status_type_t type,
2933 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002934{
2935 int r;
2936 ssize_t sz = 0;
2937 dm_block_t mapped, highest;
2938 char buf[BDEVNAME_SIZE];
2939 struct thin_c *tc = ti->private;
2940
Joe Thornbere49e5822012-07-27 15:08:16 +01002941 if (get_pool_mode(tc->pool) == PM_FAIL) {
2942 DMEMIT("Fail");
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002943 return;
Joe Thornbere49e5822012-07-27 15:08:16 +01002944 }
2945
Joe Thornber991d9fa2011-10-31 20:21:18 +00002946 if (!tc->td)
2947 DMEMIT("-");
2948 else {
2949 switch (type) {
2950 case STATUSTYPE_INFO:
2951 r = dm_thin_get_mapped_count(tc->td, &mapped);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002952 if (r) {
2953 DMERR("dm_thin_get_mapped_count returned %d", r);
2954 goto err;
2955 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002956
2957 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002958 if (r < 0) {
2959 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2960 goto err;
2961 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002962
2963 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2964 if (r)
2965 DMEMIT("%llu", ((highest + 1) *
2966 tc->pool->sectors_per_block) - 1);
2967 else
2968 DMEMIT("-");
2969 break;
2970
2971 case STATUSTYPE_TABLE:
2972 DMEMIT("%s %lu",
2973 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2974 (unsigned long) tc->dev_id);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002975 if (tc->origin_dev)
2976 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
Joe Thornber991d9fa2011-10-31 20:21:18 +00002977 break;
2978 }
2979 }
2980
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002981 return;
2982
2983err:
2984 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002985}
2986
2987static int thin_iterate_devices(struct dm_target *ti,
2988 iterate_devices_callout_fn fn, void *data)
2989{
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002990 sector_t blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002991 struct thin_c *tc = ti->private;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002992 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002993
2994 /*
2995 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2996 * we follow a more convoluted path through to the pool's target.
2997 */
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002998 if (!pool->ti)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002999 return 0; /* nothing is bound */
3000
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003001 blocks = pool->ti->len;
3002 (void) sector_div(blocks, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003003 if (blocks)
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003004 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003005
3006 return 0;
3007}
3008
Joe Thornber991d9fa2011-10-31 20:21:18 +00003009static struct target_type thin_target = {
3010 .name = "thin",
Mike Snitzer94563ba2013-08-22 09:56:18 -04003011 .version = {1, 9, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00003012 .module = THIS_MODULE,
3013 .ctr = thin_ctr,
3014 .dtr = thin_dtr,
3015 .map = thin_map,
Joe Thornbereb2aa482012-03-28 18:41:28 +01003016 .end_io = thin_endio,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003017 .postsuspend = thin_postsuspend,
3018 .status = thin_status,
3019 .iterate_devices = thin_iterate_devices,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003020};
3021
3022/*----------------------------------------------------------------*/
3023
3024static int __init dm_thin_init(void)
3025{
3026 int r;
3027
3028 pool_table_init();
3029
3030 r = dm_register_target(&thin_target);
3031 if (r)
3032 return r;
3033
3034 r = dm_register_target(&pool_target);
3035 if (r)
Mike Snitzera24c2562012-06-03 00:30:00 +01003036 goto bad_pool_target;
3037
3038 r = -ENOMEM;
3039
Mike Snitzera24c2562012-06-03 00:30:00 +01003040 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3041 if (!_new_mapping_cache)
3042 goto bad_new_mapping_cache;
3043
Mike Snitzera24c2562012-06-03 00:30:00 +01003044 return 0;
3045
Mike Snitzera24c2562012-06-03 00:30:00 +01003046bad_new_mapping_cache:
Mike Snitzera24c2562012-06-03 00:30:00 +01003047 dm_unregister_target(&pool_target);
3048bad_pool_target:
3049 dm_unregister_target(&thin_target);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003050
3051 return r;
3052}
3053
3054static void dm_thin_exit(void)
3055{
3056 dm_unregister_target(&thin_target);
3057 dm_unregister_target(&pool_target);
Mike Snitzera24c2562012-06-03 00:30:00 +01003058
Mike Snitzera24c2562012-06-03 00:30:00 +01003059 kmem_cache_destroy(_new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003060}
3061
3062module_init(dm_thin_init);
3063module_exit(dm_thin_exit);
3064
Alasdair G Kergon7cab8bf2012-05-12 01:43:19 +01003065MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003066MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3067MODULE_LICENSE("GPL");