blob: 2c0cf511ec2385fa5a558b5d2e1e1ed0c874c9f6 [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) {
Mike Snitzerc3977412012-12-21 20:23:34 +0000643 DMERR_LIMIT("dm_thin_insert_block() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000644 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100645 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000646 }
647
648 /*
649 * Release any bios held while the block was being provisioned.
650 * If we are processing a write bio that completely covers the block,
651 * we already processed it so can ignore it now when processing
652 * the bios in the cell.
653 */
654 if (bio) {
Joe Thornberf286ba02012-12-21 20:23:33 +0000655 cell_defer_no_holder(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000656 bio_endio(bio, 0);
657 } else
Joe Thornber2aab3852012-12-21 20:23:33 +0000658 cell_defer(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000659
Joe Thornber905386f2012-07-27 15:08:05 +0100660out:
Joe Thornber991d9fa2011-10-31 20:21:18 +0000661 list_del(&m->list);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000662 mempool_free(m, pool->mapping_pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000663}
664
Joe Thornbere49e5822012-07-27 15:08:16 +0100665static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber104655f2012-03-28 18:41:28 +0100666{
Joe Thornber104655f2012-03-28 18:41:28 +0100667 struct thin_c *tc = m->tc;
668
Joe Thornbere49e5822012-07-27 15:08:16 +0100669 bio_io_error(m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000670 cell_defer_no_holder(tc, m->cell);
671 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere49e5822012-07-27 15:08:16 +0100672 mempool_free(m, tc->pool->mapping_pool);
673}
Joe Thornber104655f2012-03-28 18:41:28 +0100674
Joe Thornbere49e5822012-07-27 15:08:16 +0100675static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
676{
677 struct thin_c *tc = m->tc;
678
Joe Thornbere8088072012-12-21 20:23:31 +0000679 inc_all_io_entry(tc->pool, m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000680 cell_defer_no_holder(tc, m->cell);
681 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere8088072012-12-21 20:23:31 +0000682
Joe Thornber104655f2012-03-28 18:41:28 +0100683 if (m->pass_discard)
684 remap_and_issue(tc, m->bio, m->data_block);
685 else
686 bio_endio(m->bio, 0);
687
Joe Thornber104655f2012-03-28 18:41:28 +0100688 mempool_free(m, tc->pool->mapping_pool);
689}
690
Joe Thornbere49e5822012-07-27 15:08:16 +0100691static void process_prepared_discard(struct dm_thin_new_mapping *m)
692{
693 int r;
694 struct thin_c *tc = m->tc;
695
696 r = dm_thin_remove_block(tc->td, m->virt_block);
697 if (r)
Mike Snitzerc3977412012-12-21 20:23:34 +0000698 DMERR_LIMIT("dm_thin_remove_block() failed");
Joe Thornbere49e5822012-07-27 15:08:16 +0100699
700 process_prepared_discard_passdown(m);
701}
702
Joe Thornber104655f2012-03-28 18:41:28 +0100703static void process_prepared(struct pool *pool, struct list_head *head,
Joe Thornbere49e5822012-07-27 15:08:16 +0100704 process_mapping_fn *fn)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000705{
706 unsigned long flags;
707 struct list_head maps;
Mike Snitzera24c2562012-06-03 00:30:00 +0100708 struct dm_thin_new_mapping *m, *tmp;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000709
710 INIT_LIST_HEAD(&maps);
711 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +0100712 list_splice_init(head, &maps);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000713 spin_unlock_irqrestore(&pool->lock, flags);
714
715 list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornbere49e5822012-07-27 15:08:16 +0100716 (*fn)(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000717}
718
719/*
720 * Deferred bio jobs.
721 */
Joe Thornber104655f2012-03-28 18:41:28 +0100722static int io_overlaps_block(struct pool *pool, struct bio *bio)
723{
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100724 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber104655f2012-03-28 18:41:28 +0100725}
726
Joe Thornber991d9fa2011-10-31 20:21:18 +0000727static int io_overwrites_block(struct pool *pool, struct bio *bio)
728{
Joe Thornber104655f2012-03-28 18:41:28 +0100729 return (bio_data_dir(bio) == WRITE) &&
730 io_overlaps_block(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000731}
732
733static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
734 bio_end_io_t *fn)
735{
736 *save = bio->bi_end_io;
737 bio->bi_end_io = fn;
738}
739
740static int ensure_next_mapping(struct pool *pool)
741{
742 if (pool->next_mapping)
743 return 0;
744
745 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
746
747 return pool->next_mapping ? 0 : -ENOMEM;
748}
749
Mike Snitzera24c2562012-06-03 00:30:00 +0100750static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000751{
Mike Snitzera24c2562012-06-03 00:30:00 +0100752 struct dm_thin_new_mapping *r = pool->next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000753
754 BUG_ON(!pool->next_mapping);
755
756 pool->next_mapping = NULL;
757
758 return r;
759}
760
761static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
Joe Thornber2dd9c252012-03-28 18:41:28 +0100762 struct dm_dev *origin, dm_block_t data_origin,
763 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100764 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000765{
766 int r;
767 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100768 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000769
770 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +0100771 m->quiesced = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000772 m->prepared = 0;
773 m->tc = tc;
774 m->virt_block = virt_block;
775 m->data_block = data_dest;
776 m->cell = cell;
777 m->err = 0;
778 m->bio = NULL;
779
Mike Snitzer44feb382012-10-12 21:02:10 +0100780 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
Joe Thornbereb2aa482012-03-28 18:41:28 +0100781 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000782
783 /*
784 * IO to pool_dev remaps to the pool target's data_dev.
785 *
786 * If the whole block of data is being overwritten, we can issue the
787 * bio immediately. Otherwise we use kcopyd to clone the data first.
788 */
789 if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000790 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100791
Joe Thornbereb2aa482012-03-28 18:41:28 +0100792 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000793 m->bio = bio;
794 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000795 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000796 remap_and_issue(tc, bio, data_dest);
797 } else {
798 struct dm_io_region from, to;
799
Joe Thornber2dd9c252012-03-28 18:41:28 +0100800 from.bdev = origin->bdev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000801 from.sector = data_origin * pool->sectors_per_block;
802 from.count = pool->sectors_per_block;
803
804 to.bdev = tc->pool_dev->bdev;
805 to.sector = data_dest * pool->sectors_per_block;
806 to.count = pool->sectors_per_block;
807
808 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
809 0, copy_complete, m);
810 if (r < 0) {
811 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000812 DMERR_LIMIT("dm_kcopyd_copy() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000813 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000814 }
815 }
816}
817
Joe Thornber2dd9c252012-03-28 18:41:28 +0100818static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
819 dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100820 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100821{
822 schedule_copy(tc, virt_block, tc->pool_dev,
823 data_origin, data_dest, cell, bio);
824}
825
826static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
827 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100828 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100829{
830 schedule_copy(tc, virt_block, tc->origin_dev,
831 virt_block, data_dest, cell, bio);
832}
833
Joe Thornber991d9fa2011-10-31 20:21:18 +0000834static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzera24c2562012-06-03 00:30:00 +0100835 dm_block_t data_block, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +0000836 struct bio *bio)
837{
838 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100839 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000840
841 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +0100842 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000843 m->prepared = 0;
844 m->tc = tc;
845 m->virt_block = virt_block;
846 m->data_block = data_block;
847 m->cell = cell;
848 m->err = 0;
849 m->bio = NULL;
850
851 /*
852 * If the whole block of data is being overwritten or we are not
853 * zeroing pre-existing data, we can issue the bio immediately.
854 * Otherwise we use kcopyd to zero the data first.
855 */
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100856 if (!pool->pf.zero_new_blocks)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000857 process_prepared_mapping(m);
858
859 else if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000860 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100861
Joe Thornbereb2aa482012-03-28 18:41:28 +0100862 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000863 m->bio = bio;
864 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000865 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000866 remap_and_issue(tc, bio, data_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000867 } else {
868 int r;
869 struct dm_io_region to;
870
871 to.bdev = tc->pool_dev->bdev;
872 to.sector = data_block * pool->sectors_per_block;
873 to.count = pool->sectors_per_block;
874
875 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
876 if (r < 0) {
877 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000878 DMERR_LIMIT("dm_kcopyd_zero() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000879 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000880 }
881 }
882}
883
Joe Thornbere49e5822012-07-27 15:08:16 +0100884static int commit(struct pool *pool)
885{
886 int r;
887
888 r = dm_pool_commit_metadata(pool->pmd);
889 if (r)
Mike Snitzer4fa59712013-08-21 17:30:40 -0400890 DMERR_LIMIT("%s: commit failed: error = %d",
891 dm_device_name(pool->pool_md), r);
Joe Thornbere49e5822012-07-27 15:08:16 +0100892
893 return r;
894}
895
896/*
897 * A non-zero return indicates read_only or fail_io mode.
898 * Many callers don't care about the return value.
899 */
900static int commit_or_fallback(struct pool *pool)
901{
902 int r;
903
904 if (get_pool_mode(pool) != PM_WRITE)
905 return -EINVAL;
906
907 r = commit(pool);
908 if (r)
909 set_pool_mode(pool, PM_READ_ONLY);
910
911 return r;
912}
913
Joe Thornber991d9fa2011-10-31 20:21:18 +0000914static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
915{
916 int r;
917 dm_block_t free_blocks;
918 unsigned long flags;
919 struct pool *pool = tc->pool;
920
Mike Snitzer94563ba2013-08-22 09:56:18 -0400921 /*
922 * Once no_free_space is set we must not allow allocation to succeed.
923 * Otherwise it is difficult to explain, debug, test and support.
924 */
925 if (pool->no_free_space)
926 return -ENOSPC;
927
Joe Thornber991d9fa2011-10-31 20:21:18 +0000928 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
929 if (r)
930 return r;
931
932 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
Joe Thornberb17446d2013-05-10 14:37:18 +0100933 DMWARN("%s: reached low water mark for data device: sending event.",
Joe Thornber991d9fa2011-10-31 20:21:18 +0000934 dm_device_name(pool->pool_md));
935 spin_lock_irqsave(&pool->lock, flags);
936 pool->low_water_triggered = 1;
937 spin_unlock_irqrestore(&pool->lock, flags);
938 dm_table_event(pool->ti->table);
939 }
940
941 if (!free_blocks) {
Mike Snitzer94563ba2013-08-22 09:56:18 -0400942 /*
943 * Try to commit to see if that will free up some
944 * more space.
945 */
946 (void) commit_or_fallback(pool);
947
948 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
949 if (r)
950 return r;
951
952 /*
953 * If we still have no space we set a flag to avoid
954 * doing all this checking and return -ENOSPC. This
955 * flag serves as a latch that disallows allocations from
956 * this pool until the admin takes action (e.g. resize or
957 * table reload).
958 */
959 if (!free_blocks) {
960 DMWARN("%s: no free space available.",
961 dm_device_name(pool->pool_md));
962 spin_lock_irqsave(&pool->lock, flags);
963 pool->no_free_space = 1;
964 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000965 return -ENOSPC;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000966 }
967 }
968
969 r = dm_pool_alloc_data_block(pool->pmd, result);
970 if (r)
971 return r;
972
973 return 0;
974}
975
976/*
977 * If we have run out of space, queue bios until the device is
978 * resumed, presumably after having been reloaded with more space.
979 */
980static void retry_on_resume(struct bio *bio)
981{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000982 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +0100983 struct thin_c *tc = h->tc;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000984 struct pool *pool = tc->pool;
985 unsigned long flags;
986
987 spin_lock_irqsave(&pool->lock, flags);
988 bio_list_add(&pool->retry_on_resume_list, bio);
989 spin_unlock_irqrestore(&pool->lock, flags);
990}
991
Joe Thornber6beca5e2013-03-01 22:45:50 +0000992static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000993{
994 struct bio *bio;
995 struct bio_list bios;
996
997 bio_list_init(&bios);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000998 cell_release(pool, cell, &bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000999
1000 while ((bio = bio_list_pop(&bios)))
1001 retry_on_resume(bio);
1002}
1003
Joe Thornber104655f2012-03-28 18:41:28 +01001004static void process_discard(struct thin_c *tc, struct bio *bio)
1005{
1006 int r;
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001007 unsigned long flags;
Joe Thornber104655f2012-03-28 18:41:28 +01001008 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001009 struct dm_bio_prison_cell *cell, *cell2;
Mike Snitzer44feb382012-10-12 21:02:10 +01001010 struct dm_cell_key key, key2;
Joe Thornber104655f2012-03-28 18:41:28 +01001011 dm_block_t block = get_bio_block(tc, bio);
1012 struct dm_thin_lookup_result lookup_result;
Mike Snitzera24c2562012-06-03 00:30:00 +01001013 struct dm_thin_new_mapping *m;
Joe Thornber104655f2012-03-28 18:41:28 +01001014
1015 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001016 if (bio_detain(tc->pool, &key, bio, &cell))
Joe Thornber104655f2012-03-28 18:41:28 +01001017 return;
1018
1019 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1020 switch (r) {
1021 case 0:
1022 /*
1023 * Check nobody is fiddling with this pool block. This can
1024 * happen if someone's in the process of breaking sharing
1025 * on this block.
1026 */
1027 build_data_key(tc->td, lookup_result.block, &key2);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001028 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
Joe Thornberf286ba02012-12-21 20:23:33 +00001029 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001030 break;
1031 }
1032
1033 if (io_overlaps_block(pool, bio)) {
1034 /*
1035 * IO may still be going to the destination block. We must
1036 * quiesce before we can do the removal.
1037 */
1038 m = get_next_mapping(pool);
1039 m->tc = tc;
Mike Snitzer17b7d632012-07-27 15:07:57 +01001040 m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
Joe Thornber104655f2012-03-28 18:41:28 +01001041 m->virt_block = block;
1042 m->data_block = lookup_result.block;
1043 m->cell = cell;
1044 m->cell2 = cell2;
1045 m->err = 0;
1046 m->bio = bio;
1047
Mike Snitzer44feb382012-10-12 21:02:10 +01001048 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001049 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001050 list_add(&m->list, &pool->prepared_discards);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001051 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001052 wake_worker(pool);
1053 }
1054 } else {
Joe Thornbere8088072012-12-21 20:23:31 +00001055 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001056 cell_defer_no_holder(tc, cell);
1057 cell_defer_no_holder(tc, cell2);
Joe Thornbere8088072012-12-21 20:23:31 +00001058
Joe Thornber104655f2012-03-28 18:41:28 +01001059 /*
Mikulas Patocka49296302012-07-27 15:08:03 +01001060 * The DM core makes sure that the discard doesn't span
1061 * a block boundary. So we submit the discard of a
1062 * partial block appropriately.
Joe Thornber104655f2012-03-28 18:41:28 +01001063 */
Mikulas Patocka650d2a02012-07-20 14:25:05 +01001064 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1065 remap_and_issue(tc, bio, lookup_result.block);
1066 else
1067 bio_endio(bio, 0);
Joe Thornber104655f2012-03-28 18:41:28 +01001068 }
1069 break;
1070
1071 case -ENODATA:
1072 /*
1073 * It isn't provisioned, just forget it.
1074 */
Joe Thornberf286ba02012-12-21 20:23:33 +00001075 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001076 bio_endio(bio, 0);
1077 break;
1078
1079 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001080 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1081 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001082 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001083 bio_io_error(bio);
1084 break;
1085 }
1086}
1087
Joe Thornber991d9fa2011-10-31 20:21:18 +00001088static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzer44feb382012-10-12 21:02:10 +01001089 struct dm_cell_key *key,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001090 struct dm_thin_lookup_result *lookup_result,
Mike Snitzera24c2562012-06-03 00:30:00 +01001091 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001092{
1093 int r;
1094 dm_block_t data_block;
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001095 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001096
1097 r = alloc_data_block(tc, &data_block);
1098 switch (r) {
1099 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001100 schedule_internal_copy(tc, block, lookup_result->block,
1101 data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001102 break;
1103
1104 case -ENOSPC:
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001105 no_space(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001106 break;
1107
1108 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001109 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1110 __func__, r);
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001111 set_pool_mode(pool, PM_READ_ONLY);
1112 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001113 break;
1114 }
1115}
1116
1117static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1118 dm_block_t block,
1119 struct dm_thin_lookup_result *lookup_result)
1120{
Mike Snitzera24c2562012-06-03 00:30:00 +01001121 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001122 struct pool *pool = tc->pool;
Mike Snitzer44feb382012-10-12 21:02:10 +01001123 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001124
1125 /*
1126 * If cell is already occupied, then sharing is already in the process
1127 * of being broken so we have nothing further to do here.
1128 */
1129 build_data_key(tc->td, lookup_result->block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001130 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001131 return;
1132
Joe Thornber60049702012-07-27 15:08:06 +01001133 if (bio_data_dir(bio) == WRITE && bio->bi_size)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001134 break_sharing(tc, bio, block, &key, lookup_result, cell);
1135 else {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001136 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornber991d9fa2011-10-31 20:21:18 +00001137
Mike Snitzer44feb382012-10-12 21:02:10 +01001138 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
Joe Thornbere8088072012-12-21 20:23:31 +00001139 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001140 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001141
Joe Thornber991d9fa2011-10-31 20:21:18 +00001142 remap_and_issue(tc, bio, lookup_result->block);
1143 }
1144}
1145
1146static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001147 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001148{
1149 int r;
1150 dm_block_t data_block;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001151 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001152
1153 /*
1154 * Remap empty bios (flushes) immediately, without provisioning.
1155 */
1156 if (!bio->bi_size) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001157 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001158 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001159
Joe Thornber991d9fa2011-10-31 20:21:18 +00001160 remap_and_issue(tc, bio, 0);
1161 return;
1162 }
1163
1164 /*
1165 * Fill read bios with zeroes and complete them immediately.
1166 */
1167 if (bio_data_dir(bio) == READ) {
1168 zero_fill_bio(bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001169 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001170 bio_endio(bio, 0);
1171 return;
1172 }
1173
1174 r = alloc_data_block(tc, &data_block);
1175 switch (r) {
1176 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001177 if (tc->origin_dev)
1178 schedule_external_copy(tc, block, data_block, cell, bio);
1179 else
1180 schedule_zero(tc, block, data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001181 break;
1182
1183 case -ENOSPC:
Joe Thornber6beca5e2013-03-01 22:45:50 +00001184 no_space(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001185 break;
1186
1187 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001188 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1189 __func__, r);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001190 set_pool_mode(pool, PM_READ_ONLY);
1191 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001192 break;
1193 }
1194}
1195
1196static void process_bio(struct thin_c *tc, struct bio *bio)
1197{
1198 int r;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001199 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001200 dm_block_t block = get_bio_block(tc, bio);
Mike Snitzera24c2562012-06-03 00:30:00 +01001201 struct dm_bio_prison_cell *cell;
Mike Snitzer44feb382012-10-12 21:02:10 +01001202 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001203 struct dm_thin_lookup_result lookup_result;
1204
1205 /*
1206 * If cell is already occupied, then the block is already
1207 * being provisioned so we have nothing further to do here.
1208 */
1209 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001210 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001211 return;
1212
1213 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1214 switch (r) {
1215 case 0:
Joe Thornbere8088072012-12-21 20:23:31 +00001216 if (lookup_result.shared) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001217 process_shared_bio(tc, bio, block, &lookup_result);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001218 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
Joe Thornbere8088072012-12-21 20:23:31 +00001219 } else {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001220 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001221 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001222
Joe Thornber991d9fa2011-10-31 20:21:18 +00001223 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001224 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001225 break;
1226
1227 case -ENODATA:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001228 if (bio_data_dir(bio) == READ && tc->origin_dev) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001229 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001230 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001231
Joe Thornber2dd9c252012-03-28 18:41:28 +01001232 remap_to_origin_and_issue(tc, bio);
1233 } else
1234 provision_block(tc, bio, block, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001235 break;
1236
1237 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001238 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1239 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001240 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001241 bio_io_error(bio);
1242 break;
1243 }
1244}
1245
Joe Thornbere49e5822012-07-27 15:08:16 +01001246static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1247{
1248 int r;
1249 int rw = bio_data_dir(bio);
1250 dm_block_t block = get_bio_block(tc, bio);
1251 struct dm_thin_lookup_result lookup_result;
1252
1253 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1254 switch (r) {
1255 case 0:
1256 if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1257 bio_io_error(bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001258 else {
1259 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001260 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001261 }
Joe Thornbere49e5822012-07-27 15:08:16 +01001262 break;
1263
1264 case -ENODATA:
1265 if (rw != READ) {
1266 bio_io_error(bio);
1267 break;
1268 }
1269
1270 if (tc->origin_dev) {
Joe Thornbere8088072012-12-21 20:23:31 +00001271 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001272 remap_to_origin_and_issue(tc, bio);
1273 break;
1274 }
1275
1276 zero_fill_bio(bio);
1277 bio_endio(bio, 0);
1278 break;
1279
1280 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001281 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1282 __func__, r);
Joe Thornbere49e5822012-07-27 15:08:16 +01001283 bio_io_error(bio);
1284 break;
1285 }
1286}
1287
1288static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1289{
1290 bio_io_error(bio);
1291}
1292
Joe Thornberac8c3f32013-05-10 14:37:21 +01001293/*
1294 * FIXME: should we also commit due to size of transaction, measured in
1295 * metadata blocks?
1296 */
Joe Thornber905e51b2012-03-28 18:41:27 +01001297static int need_commit_due_to_time(struct pool *pool)
1298{
1299 return jiffies < pool->last_commit_jiffies ||
1300 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1301}
1302
Joe Thornber991d9fa2011-10-31 20:21:18 +00001303static void process_deferred_bios(struct pool *pool)
1304{
1305 unsigned long flags;
1306 struct bio *bio;
1307 struct bio_list bios;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001308
1309 bio_list_init(&bios);
1310
1311 spin_lock_irqsave(&pool->lock, flags);
1312 bio_list_merge(&bios, &pool->deferred_bios);
1313 bio_list_init(&pool->deferred_bios);
1314 spin_unlock_irqrestore(&pool->lock, flags);
1315
1316 while ((bio = bio_list_pop(&bios))) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001317 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001318 struct thin_c *tc = h->tc;
1319
Joe Thornber991d9fa2011-10-31 20:21:18 +00001320 /*
1321 * If we've got no free new_mapping structs, and processing
1322 * this bio might require one, we pause until there are some
1323 * prepared mappings to process.
1324 */
1325 if (ensure_next_mapping(pool)) {
1326 spin_lock_irqsave(&pool->lock, flags);
1327 bio_list_merge(&pool->deferred_bios, &bios);
1328 spin_unlock_irqrestore(&pool->lock, flags);
1329
1330 break;
1331 }
Joe Thornber104655f2012-03-28 18:41:28 +01001332
1333 if (bio->bi_rw & REQ_DISCARD)
Joe Thornbere49e5822012-07-27 15:08:16 +01001334 pool->process_discard(tc, bio);
Joe Thornber104655f2012-03-28 18:41:28 +01001335 else
Joe Thornbere49e5822012-07-27 15:08:16 +01001336 pool->process_bio(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001337 }
1338
1339 /*
1340 * If there are any deferred flush bios, we must commit
1341 * the metadata before issuing them.
1342 */
1343 bio_list_init(&bios);
1344 spin_lock_irqsave(&pool->lock, flags);
1345 bio_list_merge(&bios, &pool->deferred_flush_bios);
1346 bio_list_init(&pool->deferred_flush_bios);
1347 spin_unlock_irqrestore(&pool->lock, flags);
1348
Joe Thornber905e51b2012-03-28 18:41:27 +01001349 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001350 return;
1351
Joe Thornbere49e5822012-07-27 15:08:16 +01001352 if (commit_or_fallback(pool)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001353 while ((bio = bio_list_pop(&bios)))
1354 bio_io_error(bio);
1355 return;
1356 }
Joe Thornber905e51b2012-03-28 18:41:27 +01001357 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001358
1359 while ((bio = bio_list_pop(&bios)))
1360 generic_make_request(bio);
1361}
1362
1363static void do_worker(struct work_struct *ws)
1364{
1365 struct pool *pool = container_of(ws, struct pool, worker);
1366
Joe Thornbere49e5822012-07-27 15:08:16 +01001367 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1368 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001369 process_deferred_bios(pool);
1370}
1371
Joe Thornber905e51b2012-03-28 18:41:27 +01001372/*
1373 * We want to commit periodically so that not too much
1374 * unwritten data builds up.
1375 */
1376static void do_waker(struct work_struct *ws)
1377{
1378 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1379 wake_worker(pool);
1380 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1381}
1382
Joe Thornber991d9fa2011-10-31 20:21:18 +00001383/*----------------------------------------------------------------*/
1384
Joe Thornbere49e5822012-07-27 15:08:16 +01001385static enum pool_mode get_pool_mode(struct pool *pool)
1386{
1387 return pool->pf.mode;
1388}
1389
1390static void set_pool_mode(struct pool *pool, enum pool_mode mode)
1391{
1392 int r;
1393
1394 pool->pf.mode = mode;
1395
1396 switch (mode) {
1397 case PM_FAIL:
Mike Snitzer4fa59712013-08-21 17:30:40 -04001398 DMERR("%s: switching pool to failure mode",
1399 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001400 pool->process_bio = process_bio_fail;
1401 pool->process_discard = process_bio_fail;
1402 pool->process_prepared_mapping = process_prepared_mapping_fail;
1403 pool->process_prepared_discard = process_prepared_discard_fail;
1404 break;
1405
1406 case PM_READ_ONLY:
Mike Snitzer4fa59712013-08-21 17:30:40 -04001407 DMERR("%s: switching pool to read-only mode",
1408 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001409 r = dm_pool_abort_metadata(pool->pmd);
1410 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04001411 DMERR("%s: aborting transaction failed",
1412 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001413 set_pool_mode(pool, PM_FAIL);
1414 } else {
1415 dm_pool_metadata_read_only(pool->pmd);
1416 pool->process_bio = process_bio_read_only;
1417 pool->process_discard = process_discard;
1418 pool->process_prepared_mapping = process_prepared_mapping_fail;
1419 pool->process_prepared_discard = process_prepared_discard_passdown;
1420 }
1421 break;
1422
1423 case PM_WRITE:
1424 pool->process_bio = process_bio;
1425 pool->process_discard = process_discard;
1426 pool->process_prepared_mapping = process_prepared_mapping;
1427 pool->process_prepared_discard = process_prepared_discard;
1428 break;
1429 }
1430}
1431
1432/*----------------------------------------------------------------*/
1433
Joe Thornber991d9fa2011-10-31 20:21:18 +00001434/*
1435 * Mapping functions.
1436 */
1437
1438/*
1439 * Called only while mapping a thin bio to hand it over to the workqueue.
1440 */
1441static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1442{
1443 unsigned long flags;
1444 struct pool *pool = tc->pool;
1445
1446 spin_lock_irqsave(&pool->lock, flags);
1447 bio_list_add(&pool->deferred_bios, bio);
1448 spin_unlock_irqrestore(&pool->lock, flags);
1449
1450 wake_worker(pool);
1451}
1452
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001453static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
Joe Thornbereb2aa482012-03-28 18:41:28 +01001454{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001455 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001456
1457 h->tc = tc;
1458 h->shared_read_entry = NULL;
Joe Thornbere8088072012-12-21 20:23:31 +00001459 h->all_io_entry = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001460 h->overwrite_mapping = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001461}
1462
Joe Thornber991d9fa2011-10-31 20:21:18 +00001463/*
1464 * Non-blocking function called from the thin target's map function.
1465 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001466static int thin_bio_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001467{
1468 int r;
1469 struct thin_c *tc = ti->private;
1470 dm_block_t block = get_bio_block(tc, bio);
1471 struct dm_thin_device *td = tc->td;
1472 struct dm_thin_lookup_result result;
Joe Thornber025b9682013-03-01 22:45:50 +00001473 struct dm_bio_prison_cell cell1, cell2;
1474 struct dm_bio_prison_cell *cell_result;
Joe Thornbere8088072012-12-21 20:23:31 +00001475 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001476
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001477 thin_hook_bio(tc, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001478
1479 if (get_pool_mode(tc->pool) == PM_FAIL) {
1480 bio_io_error(bio);
1481 return DM_MAPIO_SUBMITTED;
1482 }
1483
Joe Thornber104655f2012-03-28 18:41:28 +01001484 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001485 thin_defer_bio(tc, bio);
1486 return DM_MAPIO_SUBMITTED;
1487 }
1488
1489 r = dm_thin_find_block(td, block, 0, &result);
1490
1491 /*
1492 * Note that we defer readahead too.
1493 */
1494 switch (r) {
1495 case 0:
1496 if (unlikely(result.shared)) {
1497 /*
1498 * We have a race condition here between the
1499 * result.shared value returned by the lookup and
1500 * snapshot creation, which may cause new
1501 * sharing.
1502 *
1503 * To avoid this always quiesce the origin before
1504 * taking the snap. You want to do this anyway to
1505 * ensure a consistent application view
1506 * (i.e. lockfs).
1507 *
1508 * More distant ancestors are irrelevant. The
1509 * shared flag will be set in their case.
1510 */
1511 thin_defer_bio(tc, bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001512 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001513 }
Joe Thornbere8088072012-12-21 20:23:31 +00001514
1515 build_virtual_key(tc->td, block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001516 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
Joe Thornbere8088072012-12-21 20:23:31 +00001517 return DM_MAPIO_SUBMITTED;
1518
1519 build_data_key(tc->td, result.block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001520 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1521 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001522 return DM_MAPIO_SUBMITTED;
1523 }
1524
1525 inc_all_io_entry(tc->pool, bio);
Joe Thornber025b9682013-03-01 22:45:50 +00001526 cell_defer_no_holder_no_free(tc, &cell2);
1527 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001528
1529 remap(tc, bio, result.block);
1530 return DM_MAPIO_REMAPPED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001531
1532 case -ENODATA:
Joe Thornbere49e5822012-07-27 15:08:16 +01001533 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1534 /*
1535 * This block isn't provisioned, and we have no way
1536 * of doing so. Just error it.
1537 */
1538 bio_io_error(bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001539 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001540 }
1541 /* fall through */
1542
1543 case -EWOULDBLOCK:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001544 /*
1545 * In future, the failed dm_thin_find_block above could
1546 * provide the hint to load the metadata into cache.
1547 */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001548 thin_defer_bio(tc, bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001549 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001550
1551 default:
1552 /*
1553 * Must always call bio_io_error on failure.
1554 * dm_thin_find_block can fail with -EINVAL if the
1555 * pool is switched to fail-io mode.
1556 */
1557 bio_io_error(bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001558 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001559 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001560}
1561
1562static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1563{
1564 int r;
1565 unsigned long flags;
1566 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1567
1568 spin_lock_irqsave(&pt->pool->lock, flags);
1569 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1570 spin_unlock_irqrestore(&pt->pool->lock, flags);
1571
1572 if (!r) {
1573 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1574 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1575 }
1576
1577 return r;
1578}
1579
1580static void __requeue_bios(struct pool *pool)
1581{
1582 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1583 bio_list_init(&pool->retry_on_resume_list);
1584}
1585
1586/*----------------------------------------------------------------
1587 * Binding of control targets to a pool object
1588 *--------------------------------------------------------------*/
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001589static bool data_dev_supports_discard(struct pool_c *pt)
1590{
1591 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1592
1593 return q && blk_queue_discard(q);
1594}
1595
Joe Thornber58051b92013-03-20 17:21:25 +00001596static bool is_factor(sector_t block_size, uint32_t n)
1597{
1598 return !sector_div(block_size, n);
1599}
1600
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001601/*
1602 * If discard_passdown was enabled verify that the data device
Mike Snitzer0424caa2012-09-26 23:45:47 +01001603 * supports discards. Disable discard_passdown if not.
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001604 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01001605static void disable_passdown_if_not_supported(struct pool_c *pt)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001606{
Mike Snitzer0424caa2012-09-26 23:45:47 +01001607 struct pool *pool = pt->pool;
1608 struct block_device *data_bdev = pt->data_dev->bdev;
1609 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1610 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1611 const char *reason = NULL;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001612 char buf[BDEVNAME_SIZE];
1613
Mike Snitzer0424caa2012-09-26 23:45:47 +01001614 if (!pt->adjusted_pf.discard_passdown)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001615 return;
1616
Mike Snitzer0424caa2012-09-26 23:45:47 +01001617 if (!data_dev_supports_discard(pt))
1618 reason = "discard unsupported";
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001619
Mike Snitzer0424caa2012-09-26 23:45:47 +01001620 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1621 reason = "max discard sectors smaller than a block";
1622
1623 else if (data_limits->discard_granularity > block_size)
1624 reason = "discard granularity larger than a block";
1625
Joe Thornber58051b92013-03-20 17:21:25 +00001626 else if (!is_factor(block_size, data_limits->discard_granularity))
Mike Snitzer0424caa2012-09-26 23:45:47 +01001627 reason = "discard granularity not a factor of block size";
1628
1629 if (reason) {
1630 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1631 pt->adjusted_pf.discard_passdown = false;
1632 }
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001633}
1634
Joe Thornber991d9fa2011-10-31 20:21:18 +00001635static int bind_control_target(struct pool *pool, struct dm_target *ti)
1636{
1637 struct pool_c *pt = ti->private;
1638
Joe Thornbere49e5822012-07-27 15:08:16 +01001639 /*
1640 * We want to make sure that degraded pools are never upgraded.
1641 */
1642 enum pool_mode old_mode = pool->pf.mode;
Mike Snitzer0424caa2012-09-26 23:45:47 +01001643 enum pool_mode new_mode = pt->adjusted_pf.mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01001644
1645 if (old_mode > new_mode)
1646 new_mode = old_mode;
1647
Joe Thornber991d9fa2011-10-31 20:21:18 +00001648 pool->ti = ti;
1649 pool->low_water_blocks = pt->low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01001650 pool->pf = pt->adjusted_pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001651
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001652 set_pool_mode(pool, new_mode);
Mike Snitzerf4026932012-05-19 01:01:01 +01001653
Joe Thornber991d9fa2011-10-31 20:21:18 +00001654 return 0;
1655}
1656
1657static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1658{
1659 if (pool->ti == ti)
1660 pool->ti = NULL;
1661}
1662
1663/*----------------------------------------------------------------
1664 * Pool creation
1665 *--------------------------------------------------------------*/
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001666/* Initialize pool features. */
1667static void pool_features_init(struct pool_features *pf)
1668{
Joe Thornbere49e5822012-07-27 15:08:16 +01001669 pf->mode = PM_WRITE;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001670 pf->zero_new_blocks = true;
1671 pf->discard_enabled = true;
1672 pf->discard_passdown = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001673}
1674
Joe Thornber991d9fa2011-10-31 20:21:18 +00001675static void __pool_destroy(struct pool *pool)
1676{
1677 __pool_table_remove(pool);
1678
1679 if (dm_pool_metadata_close(pool->pmd) < 0)
1680 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1681
Mike Snitzer44feb382012-10-12 21:02:10 +01001682 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001683 dm_kcopyd_client_destroy(pool->copier);
1684
1685 if (pool->wq)
1686 destroy_workqueue(pool->wq);
1687
1688 if (pool->next_mapping)
1689 mempool_free(pool->next_mapping, pool->mapping_pool);
1690 mempool_destroy(pool->mapping_pool);
Mike Snitzer44feb382012-10-12 21:02:10 +01001691 dm_deferred_set_destroy(pool->shared_read_ds);
1692 dm_deferred_set_destroy(pool->all_io_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001693 kfree(pool);
1694}
1695
Mike Snitzera24c2562012-06-03 00:30:00 +01001696static struct kmem_cache *_new_mapping_cache;
Mike Snitzera24c2562012-06-03 00:30:00 +01001697
Joe Thornber991d9fa2011-10-31 20:21:18 +00001698static struct pool *pool_create(struct mapped_device *pool_md,
1699 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001700 unsigned long block_size,
1701 int read_only, char **error)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001702{
1703 int r;
1704 void *err_p;
1705 struct pool *pool;
1706 struct dm_pool_metadata *pmd;
Joe Thornbere49e5822012-07-27 15:08:16 +01001707 bool format_device = read_only ? false : true;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001708
Joe Thornbere49e5822012-07-27 15:08:16 +01001709 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001710 if (IS_ERR(pmd)) {
1711 *error = "Error creating metadata object";
1712 return (struct pool *)pmd;
1713 }
1714
1715 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1716 if (!pool) {
1717 *error = "Error allocating memory for pool";
1718 err_p = ERR_PTR(-ENOMEM);
1719 goto bad_pool;
1720 }
1721
1722 pool->pmd = pmd;
1723 pool->sectors_per_block = block_size;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +01001724 if (block_size & (block_size - 1))
1725 pool->sectors_per_block_shift = -1;
1726 else
1727 pool->sectors_per_block_shift = __ffs(block_size);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001728 pool->low_water_blocks = 0;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001729 pool_features_init(&pool->pf);
Mike Snitzer44feb382012-10-12 21:02:10 +01001730 pool->prison = dm_bio_prison_create(PRISON_CELLS);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001731 if (!pool->prison) {
1732 *error = "Error creating pool's bio prison";
1733 err_p = ERR_PTR(-ENOMEM);
1734 goto bad_prison;
1735 }
1736
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001737 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001738 if (IS_ERR(pool->copier)) {
1739 r = PTR_ERR(pool->copier);
1740 *error = "Error creating pool's kcopyd client";
1741 err_p = ERR_PTR(r);
1742 goto bad_kcopyd_client;
1743 }
1744
1745 /*
1746 * Create singlethreaded workqueue that will service all devices
1747 * that use this metadata.
1748 */
1749 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1750 if (!pool->wq) {
1751 *error = "Error creating pool's workqueue";
1752 err_p = ERR_PTR(-ENOMEM);
1753 goto bad_wq;
1754 }
1755
1756 INIT_WORK(&pool->worker, do_worker);
Joe Thornber905e51b2012-03-28 18:41:27 +01001757 INIT_DELAYED_WORK(&pool->waker, do_waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001758 spin_lock_init(&pool->lock);
1759 bio_list_init(&pool->deferred_bios);
1760 bio_list_init(&pool->deferred_flush_bios);
1761 INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber104655f2012-03-28 18:41:28 +01001762 INIT_LIST_HEAD(&pool->prepared_discards);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001763 pool->low_water_triggered = 0;
1764 pool->no_free_space = 0;
1765 bio_list_init(&pool->retry_on_resume_list);
Mike Snitzer44feb382012-10-12 21:02:10 +01001766
1767 pool->shared_read_ds = dm_deferred_set_create();
1768 if (!pool->shared_read_ds) {
1769 *error = "Error creating pool's shared read deferred set";
1770 err_p = ERR_PTR(-ENOMEM);
1771 goto bad_shared_read_ds;
1772 }
1773
1774 pool->all_io_ds = dm_deferred_set_create();
1775 if (!pool->all_io_ds) {
1776 *error = "Error creating pool's all io deferred set";
1777 err_p = ERR_PTR(-ENOMEM);
1778 goto bad_all_io_ds;
1779 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001780
1781 pool->next_mapping = NULL;
Mike Snitzera24c2562012-06-03 00:30:00 +01001782 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1783 _new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001784 if (!pool->mapping_pool) {
1785 *error = "Error creating pool's mapping mempool";
1786 err_p = ERR_PTR(-ENOMEM);
1787 goto bad_mapping_pool;
1788 }
1789
Joe Thornber991d9fa2011-10-31 20:21:18 +00001790 pool->ref_count = 1;
Joe Thornber905e51b2012-03-28 18:41:27 +01001791 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001792 pool->pool_md = pool_md;
1793 pool->md_dev = metadata_dev;
1794 __pool_table_insert(pool);
1795
1796 return pool;
1797
Joe Thornber991d9fa2011-10-31 20:21:18 +00001798bad_mapping_pool:
Mike Snitzer44feb382012-10-12 21:02:10 +01001799 dm_deferred_set_destroy(pool->all_io_ds);
1800bad_all_io_ds:
1801 dm_deferred_set_destroy(pool->shared_read_ds);
1802bad_shared_read_ds:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001803 destroy_workqueue(pool->wq);
1804bad_wq:
1805 dm_kcopyd_client_destroy(pool->copier);
1806bad_kcopyd_client:
Mike Snitzer44feb382012-10-12 21:02:10 +01001807 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001808bad_prison:
1809 kfree(pool);
1810bad_pool:
1811 if (dm_pool_metadata_close(pmd))
1812 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1813
1814 return err_p;
1815}
1816
1817static void __pool_inc(struct pool *pool)
1818{
1819 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1820 pool->ref_count++;
1821}
1822
1823static void __pool_dec(struct pool *pool)
1824{
1825 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1826 BUG_ON(!pool->ref_count);
1827 if (!--pool->ref_count)
1828 __pool_destroy(pool);
1829}
1830
1831static struct pool *__pool_find(struct mapped_device *pool_md,
1832 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001833 unsigned long block_size, int read_only,
1834 char **error, int *created)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001835{
1836 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1837
1838 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001839 if (pool->pool_md != pool_md) {
1840 *error = "metadata device already in use by a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001841 return ERR_PTR(-EBUSY);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001842 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001843 __pool_inc(pool);
1844
1845 } else {
1846 pool = __pool_table_lookup(pool_md);
1847 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001848 if (pool->md_dev != metadata_dev) {
1849 *error = "different pool cannot replace a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001850 return ERR_PTR(-EINVAL);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001851 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001852 __pool_inc(pool);
1853
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001854 } else {
Joe Thornbere49e5822012-07-27 15:08:16 +01001855 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001856 *created = 1;
1857 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001858 }
1859
1860 return pool;
1861}
1862
1863/*----------------------------------------------------------------
1864 * Pool target methods
1865 *--------------------------------------------------------------*/
1866static void pool_dtr(struct dm_target *ti)
1867{
1868 struct pool_c *pt = ti->private;
1869
1870 mutex_lock(&dm_thin_pool_table.mutex);
1871
1872 unbind_control_target(pt->pool, ti);
1873 __pool_dec(pt->pool);
1874 dm_put_device(ti, pt->metadata_dev);
1875 dm_put_device(ti, pt->data_dev);
1876 kfree(pt);
1877
1878 mutex_unlock(&dm_thin_pool_table.mutex);
1879}
1880
Joe Thornber991d9fa2011-10-31 20:21:18 +00001881static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1882 struct dm_target *ti)
1883{
1884 int r;
1885 unsigned argc;
1886 const char *arg_name;
1887
1888 static struct dm_arg _args[] = {
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001889 {0, 3, "Invalid number of pool feature arguments"},
Joe Thornber991d9fa2011-10-31 20:21:18 +00001890 };
1891
1892 /*
1893 * No feature arguments supplied.
1894 */
1895 if (!as->argc)
1896 return 0;
1897
1898 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1899 if (r)
1900 return -EINVAL;
1901
1902 while (argc && !r) {
1903 arg_name = dm_shift_arg(as);
1904 argc--;
1905
Joe Thornbere49e5822012-07-27 15:08:16 +01001906 if (!strcasecmp(arg_name, "skip_block_zeroing"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001907 pf->zero_new_blocks = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001908
Joe Thornbere49e5822012-07-27 15:08:16 +01001909 else if (!strcasecmp(arg_name, "ignore_discard"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001910 pf->discard_enabled = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001911
1912 else if (!strcasecmp(arg_name, "no_discard_passdown"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001913 pf->discard_passdown = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001914
1915 else if (!strcasecmp(arg_name, "read_only"))
1916 pf->mode = PM_READ_ONLY;
1917
1918 else {
1919 ti->error = "Unrecognised pool feature requested";
1920 r = -EINVAL;
1921 break;
1922 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001923 }
1924
1925 return r;
1926}
1927
Joe Thornberac8c3f32013-05-10 14:37:21 +01001928static void metadata_low_callback(void *context)
1929{
1930 struct pool *pool = context;
1931
1932 DMWARN("%s: reached low water mark for metadata device: sending event.",
1933 dm_device_name(pool->pool_md));
1934
1935 dm_table_event(pool->ti->table);
1936}
1937
Joe Thornberb17446d2013-05-10 14:37:18 +01001938static sector_t get_metadata_dev_size(struct block_device *bdev)
1939{
1940 sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1941 char buffer[BDEVNAME_SIZE];
1942
1943 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
1944 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1945 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
1946 metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
1947 }
1948
1949 return metadata_dev_size;
1950}
1951
Joe Thornber24347e92013-05-10 14:37:19 +01001952static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
1953{
1954 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
1955
1956 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
1957
1958 return metadata_dev_size;
1959}
1960
Joe Thornber991d9fa2011-10-31 20:21:18 +00001961/*
Joe Thornberac8c3f32013-05-10 14:37:21 +01001962 * When a metadata threshold is crossed a dm event is triggered, and
1963 * userland should respond by growing the metadata device. We could let
1964 * userland set the threshold, like we do with the data threshold, but I'm
1965 * not sure they know enough to do this well.
1966 */
1967static dm_block_t calc_metadata_threshold(struct pool_c *pt)
1968{
1969 /*
1970 * 4M is ample for all ops with the possible exception of thin
1971 * device deletion which is harmless if it fails (just retry the
1972 * delete after you've grown the device).
1973 */
1974 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
1975 return min((dm_block_t)1024ULL /* 4M */, quarter);
1976}
1977
1978/*
Joe Thornber991d9fa2011-10-31 20:21:18 +00001979 * thin-pool <metadata dev> <data dev>
1980 * <data block size (sectors)>
1981 * <low water mark (blocks)>
1982 * [<#feature args> [<arg>]*]
1983 *
1984 * Optional feature arguments are:
1985 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001986 * ignore_discard: disable discard
1987 * no_discard_passdown: don't pass discards down to the data device
Joe Thornber991d9fa2011-10-31 20:21:18 +00001988 */
1989static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1990{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001991 int r, pool_created = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001992 struct pool_c *pt;
1993 struct pool *pool;
1994 struct pool_features pf;
1995 struct dm_arg_set as;
1996 struct dm_dev *data_dev;
1997 unsigned long block_size;
1998 dm_block_t low_water_blocks;
1999 struct dm_dev *metadata_dev;
Joe Thornber5d0db962013-05-10 14:37:19 +01002000 fmode_t metadata_mode;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002001
2002 /*
2003 * FIXME Remove validation from scope of lock.
2004 */
2005 mutex_lock(&dm_thin_pool_table.mutex);
2006
2007 if (argc < 4) {
2008 ti->error = "Invalid argument count";
2009 r = -EINVAL;
2010 goto out_unlock;
2011 }
Joe Thornber5d0db962013-05-10 14:37:19 +01002012
Joe Thornber991d9fa2011-10-31 20:21:18 +00002013 as.argc = argc;
2014 as.argv = argv;
2015
Joe Thornber5d0db962013-05-10 14:37:19 +01002016 /*
2017 * Set default pool features.
2018 */
2019 pool_features_init(&pf);
2020
2021 dm_consume_args(&as, 4);
2022 r = parse_pool_features(&as, &pf, ti);
2023 if (r)
2024 goto out_unlock;
2025
2026 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2027 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002028 if (r) {
2029 ti->error = "Error opening metadata block device";
2030 goto out_unlock;
2031 }
2032
Joe Thornberb17446d2013-05-10 14:37:18 +01002033 /*
2034 * Run for the side-effect of possibly issuing a warning if the
2035 * device is too big.
2036 */
2037 (void) get_metadata_dev_size(metadata_dev->bdev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002038
2039 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2040 if (r) {
2041 ti->error = "Error getting data device";
2042 goto out_metadata;
2043 }
2044
2045 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2046 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2047 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002048 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002049 ti->error = "Invalid block size";
2050 r = -EINVAL;
2051 goto out;
2052 }
2053
2054 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2055 ti->error = "Invalid low water mark";
2056 r = -EINVAL;
2057 goto out;
2058 }
2059
Joe Thornber991d9fa2011-10-31 20:21:18 +00002060 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2061 if (!pt) {
2062 r = -ENOMEM;
2063 goto out;
2064 }
2065
2066 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
Joe Thornbere49e5822012-07-27 15:08:16 +01002067 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002068 if (IS_ERR(pool)) {
2069 r = PTR_ERR(pool);
2070 goto out_free_pt;
2071 }
2072
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002073 /*
2074 * 'pool_created' reflects whether this is the first table load.
2075 * Top level discard support is not allowed to be changed after
2076 * initial load. This would require a pool reload to trigger thin
2077 * device changes.
2078 */
2079 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2080 ti->error = "Discard support cannot be disabled once enabled";
2081 r = -EINVAL;
2082 goto out_flags_changed;
2083 }
2084
Joe Thornber991d9fa2011-10-31 20:21:18 +00002085 pt->pool = pool;
2086 pt->ti = ti;
2087 pt->metadata_dev = metadata_dev;
2088 pt->data_dev = data_dev;
2089 pt->low_water_blocks = low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002090 pt->adjusted_pf = pt->requested_pf = pf;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002091 ti->num_flush_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002092
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002093 /*
2094 * Only need to enable discards if the pool should pass
2095 * them down to the data device. The thin device's discard
2096 * processing will cause mappings to be removed from the btree.
2097 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002098 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002099 if (pf.discard_enabled && pf.discard_passdown) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002100 ti->num_discard_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002101
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002102 /*
2103 * Setting 'discards_supported' circumvents the normal
2104 * stacking of discard limits (this keeps the pool and
2105 * thin devices' discard limits consistent).
2106 */
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002107 ti->discards_supported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002108 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002109 ti->private = pt;
2110
Joe Thornberac8c3f32013-05-10 14:37:21 +01002111 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2112 calc_metadata_threshold(pt),
2113 metadata_low_callback,
2114 pool);
2115 if (r)
2116 goto out_free_pt;
2117
Joe Thornber991d9fa2011-10-31 20:21:18 +00002118 pt->callbacks.congested_fn = pool_is_congested;
2119 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2120
2121 mutex_unlock(&dm_thin_pool_table.mutex);
2122
2123 return 0;
2124
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002125out_flags_changed:
2126 __pool_dec(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002127out_free_pt:
2128 kfree(pt);
2129out:
2130 dm_put_device(ti, data_dev);
2131out_metadata:
2132 dm_put_device(ti, metadata_dev);
2133out_unlock:
2134 mutex_unlock(&dm_thin_pool_table.mutex);
2135
2136 return r;
2137}
2138
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002139static int pool_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002140{
2141 int r;
2142 struct pool_c *pt = ti->private;
2143 struct pool *pool = pt->pool;
2144 unsigned long flags;
2145
2146 /*
2147 * As this is a singleton target, ti->begin is always zero.
2148 */
2149 spin_lock_irqsave(&pool->lock, flags);
2150 bio->bi_bdev = pt->data_dev->bdev;
2151 r = DM_MAPIO_REMAPPED;
2152 spin_unlock_irqrestore(&pool->lock, flags);
2153
2154 return r;
2155}
2156
Joe Thornberb17446d2013-05-10 14:37:18 +01002157static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2158{
2159 int r;
2160 struct pool_c *pt = ti->private;
2161 struct pool *pool = pt->pool;
2162 sector_t data_size = ti->len;
2163 dm_block_t sb_data_size;
2164
2165 *need_commit = false;
2166
2167 (void) sector_div(data_size, pool->sectors_per_block);
2168
2169 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2170 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002171 DMERR("%s: failed to retrieve data device size",
2172 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01002173 return r;
2174 }
2175
2176 if (data_size < sb_data_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002177 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2178 dm_device_name(pool->pool_md),
Joe Thornberb17446d2013-05-10 14:37:18 +01002179 (unsigned long long)data_size, sb_data_size);
2180 return -EINVAL;
2181
2182 } else if (data_size > sb_data_size) {
2183 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2184 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002185 DMERR("%s: failed to resize data device",
2186 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01002187 set_pool_mode(pool, PM_READ_ONLY);
2188 return r;
2189 }
2190
2191 *need_commit = true;
2192 }
2193
2194 return 0;
2195}
2196
Joe Thornber24347e92013-05-10 14:37:19 +01002197static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2198{
2199 int r;
2200 struct pool_c *pt = ti->private;
2201 struct pool *pool = pt->pool;
2202 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2203
2204 *need_commit = false;
2205
Alasdair G Kergon610bba82013-05-19 18:57:50 +01002206 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
Joe Thornber24347e92013-05-10 14:37:19 +01002207
2208 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2209 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002210 DMERR("%s: failed to retrieve metadata device size",
2211 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01002212 return r;
2213 }
2214
2215 if (metadata_dev_size < sb_metadata_dev_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002216 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2217 dm_device_name(pool->pool_md),
Joe Thornber24347e92013-05-10 14:37:19 +01002218 metadata_dev_size, sb_metadata_dev_size);
2219 return -EINVAL;
2220
2221 } else if (metadata_dev_size > sb_metadata_dev_size) {
2222 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2223 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002224 DMERR("%s: failed to resize metadata device",
2225 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01002226 return r;
2227 }
2228
2229 *need_commit = true;
2230 }
2231
2232 return 0;
2233}
2234
Joe Thornber991d9fa2011-10-31 20:21:18 +00002235/*
2236 * Retrieves the number of blocks of the data device from
2237 * the superblock and compares it to the actual device size,
2238 * thus resizing the data device in case it has grown.
2239 *
2240 * This both copes with opening preallocated data devices in the ctr
2241 * being followed by a resume
2242 * -and-
2243 * calling the resume method individually after userspace has
2244 * grown the data device in reaction to a table event.
2245 */
2246static int pool_preresume(struct dm_target *ti)
2247{
2248 int r;
Joe Thornber24347e92013-05-10 14:37:19 +01002249 bool need_commit1, need_commit2;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002250 struct pool_c *pt = ti->private;
2251 struct pool *pool = pt->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002252
2253 /*
2254 * Take control of the pool object.
2255 */
2256 r = bind_control_target(pool, ti);
2257 if (r)
2258 return r;
2259
Joe Thornberb17446d2013-05-10 14:37:18 +01002260 r = maybe_resize_data_dev(ti, &need_commit1);
2261 if (r)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002262 return r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002263
Joe Thornber24347e92013-05-10 14:37:19 +01002264 r = maybe_resize_metadata_dev(ti, &need_commit2);
2265 if (r)
2266 return r;
2267
2268 if (need_commit1 || need_commit2)
Joe Thornbere49e5822012-07-27 15:08:16 +01002269 (void) commit_or_fallback(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002270
2271 return 0;
2272}
2273
2274static void pool_resume(struct dm_target *ti)
2275{
2276 struct pool_c *pt = ti->private;
2277 struct pool *pool = pt->pool;
2278 unsigned long flags;
2279
2280 spin_lock_irqsave(&pool->lock, flags);
2281 pool->low_water_triggered = 0;
2282 pool->no_free_space = 0;
2283 __requeue_bios(pool);
2284 spin_unlock_irqrestore(&pool->lock, flags);
2285
Joe Thornber905e51b2012-03-28 18:41:27 +01002286 do_waker(&pool->waker.work);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002287}
2288
2289static void pool_postsuspend(struct dm_target *ti)
2290{
Joe Thornber991d9fa2011-10-31 20:21:18 +00002291 struct pool_c *pt = ti->private;
2292 struct pool *pool = pt->pool;
2293
Joe Thornber905e51b2012-03-28 18:41:27 +01002294 cancel_delayed_work(&pool->waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002295 flush_workqueue(pool->wq);
Joe Thornbere49e5822012-07-27 15:08:16 +01002296 (void) commit_or_fallback(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002297}
2298
2299static int check_arg_count(unsigned argc, unsigned args_required)
2300{
2301 if (argc != args_required) {
2302 DMWARN("Message received with %u arguments instead of %u.",
2303 argc, args_required);
2304 return -EINVAL;
2305 }
2306
2307 return 0;
2308}
2309
2310static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2311{
2312 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2313 *dev_id <= MAX_DEV_ID)
2314 return 0;
2315
2316 if (warning)
2317 DMWARN("Message received with invalid device id: %s", arg);
2318
2319 return -EINVAL;
2320}
2321
2322static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2323{
2324 dm_thin_id dev_id;
2325 int r;
2326
2327 r = check_arg_count(argc, 2);
2328 if (r)
2329 return r;
2330
2331 r = read_dev_id(argv[1], &dev_id, 1);
2332 if (r)
2333 return r;
2334
2335 r = dm_pool_create_thin(pool->pmd, dev_id);
2336 if (r) {
2337 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2338 argv[1]);
2339 return r;
2340 }
2341
2342 return 0;
2343}
2344
2345static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2346{
2347 dm_thin_id dev_id;
2348 dm_thin_id origin_dev_id;
2349 int r;
2350
2351 r = check_arg_count(argc, 3);
2352 if (r)
2353 return r;
2354
2355 r = read_dev_id(argv[1], &dev_id, 1);
2356 if (r)
2357 return r;
2358
2359 r = read_dev_id(argv[2], &origin_dev_id, 1);
2360 if (r)
2361 return r;
2362
2363 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2364 if (r) {
2365 DMWARN("Creation of new snapshot %s of device %s failed.",
2366 argv[1], argv[2]);
2367 return r;
2368 }
2369
2370 return 0;
2371}
2372
2373static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2374{
2375 dm_thin_id dev_id;
2376 int r;
2377
2378 r = check_arg_count(argc, 2);
2379 if (r)
2380 return r;
2381
2382 r = read_dev_id(argv[1], &dev_id, 1);
2383 if (r)
2384 return r;
2385
2386 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2387 if (r)
2388 DMWARN("Deletion of thin device %s failed.", argv[1]);
2389
2390 return r;
2391}
2392
2393static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2394{
2395 dm_thin_id old_id, new_id;
2396 int r;
2397
2398 r = check_arg_count(argc, 3);
2399 if (r)
2400 return r;
2401
2402 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2403 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2404 return -EINVAL;
2405 }
2406
2407 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2408 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2409 return -EINVAL;
2410 }
2411
2412 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2413 if (r) {
2414 DMWARN("Failed to change transaction id from %s to %s.",
2415 argv[1], argv[2]);
2416 return r;
2417 }
2418
2419 return 0;
2420}
2421
Joe Thornbercc8394d2012-06-03 00:30:01 +01002422static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2423{
2424 int r;
2425
2426 r = check_arg_count(argc, 1);
2427 if (r)
2428 return r;
2429
Joe Thornbere49e5822012-07-27 15:08:16 +01002430 (void) commit_or_fallback(pool);
Joe Thornber0d200ae2012-07-03 12:55:31 +01002431
Joe Thornbercc8394d2012-06-03 00:30:01 +01002432 r = dm_pool_reserve_metadata_snap(pool->pmd);
2433 if (r)
2434 DMWARN("reserve_metadata_snap message failed.");
2435
2436 return r;
2437}
2438
2439static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2440{
2441 int r;
2442
2443 r = check_arg_count(argc, 1);
2444 if (r)
2445 return r;
2446
2447 r = dm_pool_release_metadata_snap(pool->pmd);
2448 if (r)
2449 DMWARN("release_metadata_snap message failed.");
2450
2451 return r;
2452}
2453
Joe Thornber991d9fa2011-10-31 20:21:18 +00002454/*
2455 * Messages supported:
2456 * create_thin <dev_id>
2457 * create_snap <dev_id> <origin_id>
2458 * delete <dev_id>
2459 * trim <dev_id> <new_size_in_sectors>
2460 * set_transaction_id <current_trans_id> <new_trans_id>
Joe Thornbercc8394d2012-06-03 00:30:01 +01002461 * reserve_metadata_snap
2462 * release_metadata_snap
Joe Thornber991d9fa2011-10-31 20:21:18 +00002463 */
2464static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2465{
2466 int r = -EINVAL;
2467 struct pool_c *pt = ti->private;
2468 struct pool *pool = pt->pool;
2469
2470 if (!strcasecmp(argv[0], "create_thin"))
2471 r = process_create_thin_mesg(argc, argv, pool);
2472
2473 else if (!strcasecmp(argv[0], "create_snap"))
2474 r = process_create_snap_mesg(argc, argv, pool);
2475
2476 else if (!strcasecmp(argv[0], "delete"))
2477 r = process_delete_mesg(argc, argv, pool);
2478
2479 else if (!strcasecmp(argv[0], "set_transaction_id"))
2480 r = process_set_transaction_id_mesg(argc, argv, pool);
2481
Joe Thornbercc8394d2012-06-03 00:30:01 +01002482 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2483 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2484
2485 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2486 r = process_release_metadata_snap_mesg(argc, argv, pool);
2487
Joe Thornber991d9fa2011-10-31 20:21:18 +00002488 else
2489 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2490
Joe Thornbere49e5822012-07-27 15:08:16 +01002491 if (!r)
2492 (void) commit_or_fallback(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002493
2494 return r;
2495}
2496
Joe Thornbere49e5822012-07-27 15:08:16 +01002497static void emit_flags(struct pool_features *pf, char *result,
2498 unsigned sz, unsigned maxlen)
2499{
2500 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2501 !pf->discard_passdown + (pf->mode == PM_READ_ONLY);
2502 DMEMIT("%u ", count);
2503
2504 if (!pf->zero_new_blocks)
2505 DMEMIT("skip_block_zeroing ");
2506
2507 if (!pf->discard_enabled)
2508 DMEMIT("ignore_discard ");
2509
2510 if (!pf->discard_passdown)
2511 DMEMIT("no_discard_passdown ");
2512
2513 if (pf->mode == PM_READ_ONLY)
2514 DMEMIT("read_only ");
2515}
2516
Joe Thornber991d9fa2011-10-31 20:21:18 +00002517/*
2518 * Status line is:
2519 * <transaction id> <used metadata sectors>/<total metadata sectors>
2520 * <used data sectors>/<total data sectors> <held metadata root>
2521 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002522static void pool_status(struct dm_target *ti, status_type_t type,
2523 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002524{
Joe Thornbere49e5822012-07-27 15:08:16 +01002525 int r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002526 unsigned sz = 0;
2527 uint64_t transaction_id;
2528 dm_block_t nr_free_blocks_data;
2529 dm_block_t nr_free_blocks_metadata;
2530 dm_block_t nr_blocks_data;
2531 dm_block_t nr_blocks_metadata;
2532 dm_block_t held_root;
2533 char buf[BDEVNAME_SIZE];
2534 char buf2[BDEVNAME_SIZE];
2535 struct pool_c *pt = ti->private;
2536 struct pool *pool = pt->pool;
2537
2538 switch (type) {
2539 case STATUSTYPE_INFO:
Joe Thornbere49e5822012-07-27 15:08:16 +01002540 if (get_pool_mode(pool) == PM_FAIL) {
2541 DMEMIT("Fail");
2542 break;
2543 }
2544
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002545 /* Commit to ensure statistics aren't out-of-date */
2546 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
2547 (void) commit_or_fallback(pool);
2548
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002549 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2550 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002551 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2552 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002553 goto err;
2554 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002555
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002556 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2557 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002558 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2559 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002560 goto err;
2561 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002562
2563 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002564 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002565 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2566 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002567 goto err;
2568 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002569
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002570 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2571 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002572 DMERR("%s: dm_pool_get_free_block_count returned %d",
2573 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002574 goto err;
2575 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002576
2577 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002578 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002579 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2580 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002581 goto err;
2582 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002583
Joe Thornbercc8394d2012-06-03 00:30:01 +01002584 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002585 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002586 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2587 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002588 goto err;
2589 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002590
2591 DMEMIT("%llu %llu/%llu %llu/%llu ",
2592 (unsigned long long)transaction_id,
2593 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2594 (unsigned long long)nr_blocks_metadata,
2595 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2596 (unsigned long long)nr_blocks_data);
2597
2598 if (held_root)
Joe Thornbere49e5822012-07-27 15:08:16 +01002599 DMEMIT("%llu ", held_root);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002600 else
Joe Thornbere49e5822012-07-27 15:08:16 +01002601 DMEMIT("- ");
2602
2603 if (pool->pf.mode == PM_READ_ONLY)
2604 DMEMIT("ro ");
2605 else
2606 DMEMIT("rw ");
2607
Mike Snitzer018debe2012-12-21 20:23:32 +00002608 if (!pool->pf.discard_enabled)
2609 DMEMIT("ignore_discard");
2610 else if (pool->pf.discard_passdown)
Joe Thornbere49e5822012-07-27 15:08:16 +01002611 DMEMIT("discard_passdown");
2612 else
2613 DMEMIT("no_discard_passdown");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002614
2615 break;
2616
2617 case STATUSTYPE_TABLE:
2618 DMEMIT("%s %s %lu %llu ",
2619 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2620 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2621 (unsigned long)pool->sectors_per_block,
2622 (unsigned long long)pt->low_water_blocks);
Mike Snitzer0424caa2012-09-26 23:45:47 +01002623 emit_flags(&pt->requested_pf, result, sz, maxlen);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002624 break;
2625 }
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002626 return;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002627
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002628err:
2629 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002630}
2631
2632static int pool_iterate_devices(struct dm_target *ti,
2633 iterate_devices_callout_fn fn, void *data)
2634{
2635 struct pool_c *pt = ti->private;
2636
2637 return fn(ti, pt->data_dev, 0, ti->len, data);
2638}
2639
2640static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2641 struct bio_vec *biovec, int max_size)
2642{
2643 struct pool_c *pt = ti->private;
2644 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2645
2646 if (!q->merge_bvec_fn)
2647 return max_size;
2648
2649 bvm->bi_bdev = pt->data_dev->bdev;
2650
2651 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2652}
2653
Mike Snitzer0424caa2012-09-26 23:45:47 +01002654static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
Joe Thornber104655f2012-03-28 18:41:28 +01002655{
Mike Snitzer0424caa2012-09-26 23:45:47 +01002656 struct pool *pool = pt->pool;
2657 struct queue_limits *data_limits;
2658
Joe Thornber104655f2012-03-28 18:41:28 +01002659 limits->max_discard_sectors = pool->sectors_per_block;
2660
2661 /*
Mike Snitzer0424caa2012-09-26 23:45:47 +01002662 * discard_granularity is just a hint, and not enforced.
Joe Thornber104655f2012-03-28 18:41:28 +01002663 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01002664 if (pt->adjusted_pf.discard_passdown) {
2665 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2666 limits->discard_granularity = data_limits->discard_granularity;
Mike Snitzerf13945d2013-03-01 22:45:44 +00002667 } else
Mike Snitzer0424caa2012-09-26 23:45:47 +01002668 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
Joe Thornber104655f2012-03-28 18:41:28 +01002669}
2670
Joe Thornber991d9fa2011-10-31 20:21:18 +00002671static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2672{
2673 struct pool_c *pt = ti->private;
2674 struct pool *pool = pt->pool;
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002675 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002676
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002677 /*
2678 * If the system-determined stacked limits are compatible with the
2679 * pool's blocksize (io_opt is a factor) do not override them.
2680 */
2681 if (io_opt_sectors < pool->sectors_per_block ||
2682 do_div(io_opt_sectors, pool->sectors_per_block)) {
2683 blk_limits_io_min(limits, 0);
2684 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2685 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002686
2687 /*
2688 * pt->adjusted_pf is a staging area for the actual features to use.
2689 * They get transferred to the live pool in bind_control_target()
2690 * called from pool_preresume().
2691 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002692 if (!pt->adjusted_pf.discard_enabled) {
2693 /*
2694 * Must explicitly disallow stacking discard limits otherwise the
2695 * block layer will stack them if pool's data device has support.
2696 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2697 * user to see that, so make sure to set all discard limits to 0.
2698 */
2699 limits->discard_granularity = 0;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002700 return;
Mike Snitzerb60ab992013-09-19 18:49:11 -04002701 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002702
2703 disable_passdown_if_not_supported(pt);
2704
2705 set_discard_limits(pt, limits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002706}
2707
2708static struct target_type pool_target = {
2709 .name = "thin-pool",
2710 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2711 DM_TARGET_IMMUTABLE,
Mike Snitzer94563ba2013-08-22 09:56:18 -04002712 .version = {1, 9, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002713 .module = THIS_MODULE,
2714 .ctr = pool_ctr,
2715 .dtr = pool_dtr,
2716 .map = pool_map,
2717 .postsuspend = pool_postsuspend,
2718 .preresume = pool_preresume,
2719 .resume = pool_resume,
2720 .message = pool_message,
2721 .status = pool_status,
2722 .merge = pool_merge,
2723 .iterate_devices = pool_iterate_devices,
2724 .io_hints = pool_io_hints,
2725};
2726
2727/*----------------------------------------------------------------
2728 * Thin target methods
2729 *--------------------------------------------------------------*/
2730static void thin_dtr(struct dm_target *ti)
2731{
2732 struct thin_c *tc = ti->private;
2733
2734 mutex_lock(&dm_thin_pool_table.mutex);
2735
2736 __pool_dec(tc->pool);
2737 dm_pool_close_thin_device(tc->td);
2738 dm_put_device(ti, tc->pool_dev);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002739 if (tc->origin_dev)
2740 dm_put_device(ti, tc->origin_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002741 kfree(tc);
2742
2743 mutex_unlock(&dm_thin_pool_table.mutex);
2744}
2745
2746/*
2747 * Thin target parameters:
2748 *
Joe Thornber2dd9c252012-03-28 18:41:28 +01002749 * <pool_dev> <dev_id> [origin_dev]
Joe Thornber991d9fa2011-10-31 20:21:18 +00002750 *
2751 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2752 * dev_id: the internal device identifier
Joe Thornber2dd9c252012-03-28 18:41:28 +01002753 * origin_dev: a device external to the pool that should act as the origin
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002754 *
2755 * If the pool device has discards disabled, they get disabled for the thin
2756 * device as well.
Joe Thornber991d9fa2011-10-31 20:21:18 +00002757 */
2758static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2759{
2760 int r;
2761 struct thin_c *tc;
Joe Thornber2dd9c252012-03-28 18:41:28 +01002762 struct dm_dev *pool_dev, *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002763 struct mapped_device *pool_md;
2764
2765 mutex_lock(&dm_thin_pool_table.mutex);
2766
Joe Thornber2dd9c252012-03-28 18:41:28 +01002767 if (argc != 2 && argc != 3) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002768 ti->error = "Invalid argument count";
2769 r = -EINVAL;
2770 goto out_unlock;
2771 }
2772
2773 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2774 if (!tc) {
2775 ti->error = "Out of memory";
2776 r = -ENOMEM;
2777 goto out_unlock;
2778 }
2779
Joe Thornber2dd9c252012-03-28 18:41:28 +01002780 if (argc == 3) {
2781 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2782 if (r) {
2783 ti->error = "Error opening origin device";
2784 goto bad_origin_dev;
2785 }
2786 tc->origin_dev = origin_dev;
2787 }
2788
Joe Thornber991d9fa2011-10-31 20:21:18 +00002789 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2790 if (r) {
2791 ti->error = "Error opening pool device";
2792 goto bad_pool_dev;
2793 }
2794 tc->pool_dev = pool_dev;
2795
2796 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2797 ti->error = "Invalid device id";
2798 r = -EINVAL;
2799 goto bad_common;
2800 }
2801
2802 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2803 if (!pool_md) {
2804 ti->error = "Couldn't get pool mapped device";
2805 r = -EINVAL;
2806 goto bad_common;
2807 }
2808
2809 tc->pool = __pool_table_lookup(pool_md);
2810 if (!tc->pool) {
2811 ti->error = "Couldn't find pool object";
2812 r = -EINVAL;
2813 goto bad_pool_lookup;
2814 }
2815 __pool_inc(tc->pool);
2816
Joe Thornbere49e5822012-07-27 15:08:16 +01002817 if (get_pool_mode(tc->pool) == PM_FAIL) {
2818 ti->error = "Couldn't open thin device, Pool is in fail mode";
2819 goto bad_thin_open;
2820 }
2821
Joe Thornber991d9fa2011-10-31 20:21:18 +00002822 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2823 if (r) {
2824 ti->error = "Couldn't open thin internal device";
2825 goto bad_thin_open;
2826 }
2827
Mike Snitzer542f9032012-07-27 15:08:00 +01002828 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2829 if (r)
2830 goto bad_thin_open;
2831
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002832 ti->num_flush_bios = 1;
Joe Thornber16ad3d12012-07-27 15:08:07 +01002833 ti->flush_supported = true;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002834 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002835
2836 /* In case the pool supports discards, pass them on. */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002837 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002838 if (tc->pool->pf.discard_enabled) {
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002839 ti->discards_supported = true;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002840 ti->num_discard_bios = 1;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002841 /* Discard bios must be split on a block boundary */
2842 ti->split_discard_bios = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002843 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002844
2845 dm_put(pool_md);
2846
2847 mutex_unlock(&dm_thin_pool_table.mutex);
2848
2849 return 0;
2850
2851bad_thin_open:
2852 __pool_dec(tc->pool);
2853bad_pool_lookup:
2854 dm_put(pool_md);
2855bad_common:
2856 dm_put_device(ti, tc->pool_dev);
2857bad_pool_dev:
Joe Thornber2dd9c252012-03-28 18:41:28 +01002858 if (tc->origin_dev)
2859 dm_put_device(ti, tc->origin_dev);
2860bad_origin_dev:
Joe Thornber991d9fa2011-10-31 20:21:18 +00002861 kfree(tc);
2862out_unlock:
2863 mutex_unlock(&dm_thin_pool_table.mutex);
2864
2865 return r;
2866}
2867
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002868static int thin_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002869{
Alasdair G Kergon6efd6e82012-03-28 18:41:28 +01002870 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002871
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002872 return thin_bio_map(ti, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002873}
2874
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002875static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
Joe Thornbereb2aa482012-03-28 18:41:28 +01002876{
2877 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002878 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01002879 struct list_head work;
Mike Snitzera24c2562012-06-03 00:30:00 +01002880 struct dm_thin_new_mapping *m, *tmp;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002881 struct pool *pool = h->tc->pool;
2882
2883 if (h->shared_read_entry) {
2884 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002885 dm_deferred_entry_dec(h->shared_read_entry, &work);
Joe Thornbereb2aa482012-03-28 18:41:28 +01002886
2887 spin_lock_irqsave(&pool->lock, flags);
2888 list_for_each_entry_safe(m, tmp, &work, list) {
2889 list_del(&m->list);
2890 m->quiesced = 1;
2891 __maybe_add_mapping(m);
2892 }
2893 spin_unlock_irqrestore(&pool->lock, flags);
2894 }
2895
Joe Thornber104655f2012-03-28 18:41:28 +01002896 if (h->all_io_entry) {
2897 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002898 dm_deferred_entry_dec(h->all_io_entry, &work);
Joe Thornber563af182012-12-21 20:23:31 +00002899 if (!list_empty(&work)) {
2900 spin_lock_irqsave(&pool->lock, flags);
2901 list_for_each_entry_safe(m, tmp, &work, list)
2902 list_add(&m->list, &pool->prepared_discards);
2903 spin_unlock_irqrestore(&pool->lock, flags);
2904 wake_worker(pool);
2905 }
Joe Thornber104655f2012-03-28 18:41:28 +01002906 }
2907
Joe Thornbereb2aa482012-03-28 18:41:28 +01002908 return 0;
2909}
2910
Joe Thornber991d9fa2011-10-31 20:21:18 +00002911static void thin_postsuspend(struct dm_target *ti)
2912{
2913 if (dm_noflush_suspending(ti))
2914 requeue_io((struct thin_c *)ti->private);
2915}
2916
2917/*
2918 * <nr mapped sectors> <highest mapped sector>
2919 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002920static void thin_status(struct dm_target *ti, status_type_t type,
2921 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002922{
2923 int r;
2924 ssize_t sz = 0;
2925 dm_block_t mapped, highest;
2926 char buf[BDEVNAME_SIZE];
2927 struct thin_c *tc = ti->private;
2928
Joe Thornbere49e5822012-07-27 15:08:16 +01002929 if (get_pool_mode(tc->pool) == PM_FAIL) {
2930 DMEMIT("Fail");
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002931 return;
Joe Thornbere49e5822012-07-27 15:08:16 +01002932 }
2933
Joe Thornber991d9fa2011-10-31 20:21:18 +00002934 if (!tc->td)
2935 DMEMIT("-");
2936 else {
2937 switch (type) {
2938 case STATUSTYPE_INFO:
2939 r = dm_thin_get_mapped_count(tc->td, &mapped);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002940 if (r) {
2941 DMERR("dm_thin_get_mapped_count returned %d", r);
2942 goto err;
2943 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002944
2945 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002946 if (r < 0) {
2947 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2948 goto err;
2949 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002950
2951 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2952 if (r)
2953 DMEMIT("%llu", ((highest + 1) *
2954 tc->pool->sectors_per_block) - 1);
2955 else
2956 DMEMIT("-");
2957 break;
2958
2959 case STATUSTYPE_TABLE:
2960 DMEMIT("%s %lu",
2961 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2962 (unsigned long) tc->dev_id);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002963 if (tc->origin_dev)
2964 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
Joe Thornber991d9fa2011-10-31 20:21:18 +00002965 break;
2966 }
2967 }
2968
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002969 return;
2970
2971err:
2972 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002973}
2974
2975static int thin_iterate_devices(struct dm_target *ti,
2976 iterate_devices_callout_fn fn, void *data)
2977{
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002978 sector_t blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002979 struct thin_c *tc = ti->private;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002980 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002981
2982 /*
2983 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2984 * we follow a more convoluted path through to the pool's target.
2985 */
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002986 if (!pool->ti)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002987 return 0; /* nothing is bound */
2988
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002989 blocks = pool->ti->len;
2990 (void) sector_div(blocks, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002991 if (blocks)
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002992 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002993
2994 return 0;
2995}
2996
Joe Thornber991d9fa2011-10-31 20:21:18 +00002997static struct target_type thin_target = {
2998 .name = "thin",
Mike Snitzer94563ba2013-08-22 09:56:18 -04002999 .version = {1, 9, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00003000 .module = THIS_MODULE,
3001 .ctr = thin_ctr,
3002 .dtr = thin_dtr,
3003 .map = thin_map,
Joe Thornbereb2aa482012-03-28 18:41:28 +01003004 .end_io = thin_endio,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003005 .postsuspend = thin_postsuspend,
3006 .status = thin_status,
3007 .iterate_devices = thin_iterate_devices,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003008};
3009
3010/*----------------------------------------------------------------*/
3011
3012static int __init dm_thin_init(void)
3013{
3014 int r;
3015
3016 pool_table_init();
3017
3018 r = dm_register_target(&thin_target);
3019 if (r)
3020 return r;
3021
3022 r = dm_register_target(&pool_target);
3023 if (r)
Mike Snitzera24c2562012-06-03 00:30:00 +01003024 goto bad_pool_target;
3025
3026 r = -ENOMEM;
3027
Mike Snitzera24c2562012-06-03 00:30:00 +01003028 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3029 if (!_new_mapping_cache)
3030 goto bad_new_mapping_cache;
3031
Mike Snitzera24c2562012-06-03 00:30:00 +01003032 return 0;
3033
Mike Snitzera24c2562012-06-03 00:30:00 +01003034bad_new_mapping_cache:
Mike Snitzera24c2562012-06-03 00:30:00 +01003035 dm_unregister_target(&pool_target);
3036bad_pool_target:
3037 dm_unregister_target(&thin_target);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003038
3039 return r;
3040}
3041
3042static void dm_thin_exit(void)
3043{
3044 dm_unregister_target(&thin_target);
3045 dm_unregister_target(&pool_target);
Mike Snitzera24c2562012-06-03 00:30:00 +01003046
Mike Snitzera24c2562012-06-03 00:30:00 +01003047 kmem_cache_destroy(_new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003048}
3049
3050module_init(dm_thin_init);
3051module_exit(dm_thin_exit);
3052
Alasdair G Kergon7cab8bf2012-05-12 01:43:19 +01003053MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003054MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3055MODULE_LICENSE("GPL");